Nature Neuroscience March 2005

www.nature.com/natureneuroscience

EDITORIAL OFFICE [email protected] Park Avenue South, New York, NY 10010-1707Tel: (212) 726 9319, Fax: (212) 696 0978Editor: Sandra Aamodt Senior Editor: Kalyani Narasimhan Associate Editors: I-han Chou, Annette Markus Assistant Editor: Cara AllenCopy 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 [email protected] (US/Canada) [email protected] (Europe) [email protected] (Japan)US Head of Display Advertising: Stephen Schwartz, Tel: (212) 726 9256, Fax: (212) 696 9481Global Head of Display Advertising Sales: John Michael, Tel: 44 207 843 4960, Fax: 44 207 843 4996Head of Display Advertising—Europe: Gerard Preston, Tel: 44 207 843 4960, Fax: 44 207 843 4996Business Development Manager: Claire Hines, Tel: 44 207 843 4960, Fax: 44 207 843 4996Asia-Pacific Sales Manager: Rinoko Asami, Tel: 81 3 3267 8751, Fax: 81 3 3267 8746Western Region Sales Manager: George Lui, Tel: (415) 781 3804, Fax: (415) 781 3805Sales Executives:New England: Sheila Reardon, Tel: (617) 399 4098, Fax: (617) 426 3717New York, Mid-Atlantic, Southeast: Jim Breault, Tel: (212) 726 9334, Fax: (212) 696 9481Midwest: Mike Rossi, Tel: (212) 726 9255, Fax: (212) 696 9481Northwest: Mathieu DesJardins, Tel: (415) 781 6422, Fax: (415) 781 3805Eastern England/Scotland/Italy/Spain/Israel: Matthew Wilkinson, Tel: 44 207 843 4960, Fax: 44 207 843 4749South/West United Kingdom/Scandinavia/Holland: Marianne Boulakas, Tel: 44 207 843 4969, Fax: 44 207 843 4749Northern Germany: Gerard Preston, Tel: 44 207 843 4960, Fax: 44 207 843 4749Southern Germany/Austria/Switzerland/France/Belgium: Sabine Hugi-Fürst, Tel: 41 52761 3386, Fax: 41 52761 3419

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

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

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

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

i

VOLUME 8 NUMBER 3 MARCH 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 © 2004 Nature Publishing Group. Printed in USA.

E D I TO R I A L253 Separating science from stereotype

B O O K R E V I E W

255 Brain and Visual Perception: The Story of a 25-Year Collaborationby David H Hubel and Torsten N WieselReviewed by Andrew J Parker

NE W S A N D V I E W S

257 Inhibitory synapses turn excitingJulie A Kauer see also p 332

259 Stem cells in the injured spinal cord: reducing the pain and increasing the gainSandra Klein and Clive N Svendsen see also p 346

261 Controlling stress: how the brain protects itself from depressionTrevor W Robbins see also p 365

262 Myelin repair: developmental myelination redux?Roumen Balabanov & Brian Popko

264 Zooming in on cortical mapsDavid Fitzpatrick

266 Disentangling simple from complex cellsI-han Chou see also p 372

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

267 Constant light desynchronizes mammalian clock neuronsH Ohta, S Yamazaki & D G McMahon

270 Restoration of spatial working memory by genetic rescue of GluR-A–deficient miceW B Schmitt, R Sprengel, V Mack, R W Draft, P H Seeburg, R M J Deacon, J N P Rawlins & D M Bannerman

273 Preserved spatial memory after hippocampal lesions: effects of extensive experience in a complex environmentG Winocur, M Moscovitch, S Fogel, R S Rosenbaum & M Sekeres

276 Attending to local form while ignoring global aspects depends on handedness: evidence from TMSC Mevorach, G W Humphreys & L Shalev

Elimination and strengthening of inhibitory synapses from the medial

nucleus of the trapezoid body (MNTB) are essential for the formation of a

precise tonotopic map in the lateral superior olive, but the mechanisms

behind this plasticity are unclear. Kandler and colleagues now find

that these inhibitory MNTB terminals co-release the excitatory transmitter

glutamate during the period of synapse elimination, which activates postsynaptic NMDA receptors. Here,

an axon terminal from a dye-filled GABA/glycinergic MNTB neuron

(red) is immunolabeled against the vesicular glutamate transporter

VGLUT3 (blue) and the synaptic vesicle protein SV2 (green).

(pp 257 and 332)

Kir channel domains for gating and inward rectification

(p 279)

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

iii

VOLUME 8 NUMBER 3 MARCH 2005

NATURE NEUROSCIENCE

A R T I C L E S279 Cytoplasmic domain structures of Kir2.1 and Kir3.1 show sites for modulating

gating and rectificationS Pegan, C Arrabit, W Zhou, W Kwiatkowski, A Collins, P A Slesinger & S Choe

288 Directed differentiation of telencephalic precursors from embryonic stem cellsK Watanabe, D Kamiya, A Nishiyama, T Katayama, S Nozaki, H Kawasaki, Y Watanabe, K Mizuseki & Y Sasai

297 Sonic hedgehog guides commissural axons along the longitudinal axis of the spinal cordD Bourikas, V Pekarik, T Baeriswyl, Å Grunditz, R Sadhu, M Nardó & E T Stoeckli

305 Local calcium transients regulate the spontaneous motility of dendritic filopodiaC Lohmann, A Finski & T Bonhoeffer

313 Histone deacetylase 9 couples neuronal activity to muscle chromatin acetylation and gene expressionA Méjat, F Ramond, R Bassel-Duby, S Khochbin, E N Olson & L Schaeffer

322 Soluble CPG15 expressed during early development rescues cortical progenitors from apoptosisU Putz, C Harwel & E Nedivi

332 Inhibitory synapses in the developing auditory system are glutamatergicD C Gillespie, G Kim & K Kandler see also p 257

339 Alcohol-induced motor impairment caused by increased extrasynaptic GABAA receptor activityH J Hanchar, P D Dodson, R W Olsen, T S Otis & M Wallner

346 Allodynia limits the usefulness of intraspinal neural stem cell grafts; directed differentiation improves outcomeC P Hofstetter, N A V Holmström, J A Lilja, P Schweinhardt, J Hao, C Spenger, Z Wiesenfeld-Hallin, S N Kurpad, J Frisén & L Olson see also p 259

354 Intraglomerular inhibition: signaling mechanisms of an olfactory microcircuitG J Murphy, D P Darcy & J S Isaacson

365 Medial prefrontal cortex determines how stressor controllability affects behavior and dorsal raphe nucleusJ Amat, M V Baratta, E Paul, S T Bland, L R Watkins & S F Maier see also p 261

372 Receptive field structure varies with layer in the primary visual cortexL M Martinez, Q Wang, R C Reid, C Pillai, J-M Alonso, F T Sommer & J A Hirsch see also p 266

380 Multiple periods of functional ocular dominance plasticity in mouse visual cortexY Tagawa, P O Kanold, M Majdan & C J Shatz

389 Hierarchical and asymmetric temporal sensitivity in human auditory corticesA Boemio, S Fromm, A Braun & D Poeppel

396 CO R R I G E N DA

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.

Extrasynaptic GABA receptors and alcohol-induced motor impairment

(p 339)

Asymmetric sensitivity to temporal structure in auditory cortex

(p 389)

Dendrodendritic signaling in the olfactory bulb

(p 354)

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 3 | MARCH 2005 253

E D I TO R I A L

Separating science from stereotype

Harvard University president Lawrence Summers stirred up a hornet’s nest in January when he suggested that innate biologi-cal differences may help explain why men have more career suc-

cess in science and mathematics than women. His speech at a National Bureau of Economics conference, which was off the record, set off a storm of protest, and Summers has spent the last several weeks clarifying and apologizing for his remarks. Journalists have had a field day debating the incident, and although some have criticized Summers as being almost neanderthal in his thinking, others have portrayed him as a victim of political correctness.

There is no doubt that Summers’ comments were impolitic. There is no official record of what he said, but his reported implication that men are biologically predisposed to outperform women at the upper end of the math and science spectrum has captured the media’s fancy. Unfortunately, however, most of his supporters have not been preoc-cupied with evaluating whether this argument has any scientific merit. Are there neurobiological differences between men and women that may explain the gender gap in science and mathematics? The evidence to sup-port this hypothesis of ‘innate difference’ turns out to be quite slim.

In his talk, Summers apparently cited a gender difference in Scholastic Aptitude Test (SAT) mathematics scores: boys are more likely than girls to score on the tail ends of the bell curve. (That is, the worst performers and the highest scorers tend to be male.) Similarly, in a 2003 study of teenagers by the Programme for International Student Assessment, boys outperformed girls in math by a statistically small margin (http://www.pisa.oecd.org/dataoecd/1/63/34002454.pdf). However, in 7 of 43 coun-tries, boys and girls had similar scores, and in Iceland girls outscored boys, suggesting that cultural factors can influence this gender difference.

The meaning of this overrepresentation at the high end is anyone’s guess. Summers apparently mused that it may explain why there are many more gifted male mathematicians than female. However, apti-tude tests are not very good predictors of future educational success. In particular, for reasons that are unclear, the SAT tends to underpredict female and overpredict male academic performance. On average, males score 33 points higher on the math section of the SAT than females who earn the same grades in the same college math courses (www.FairTest.org). Similarly, Chinese nationals tend to do very well on the Graduate Record Examination (GRE) subject tests, although they do not perform much differently in graduate school from their American counterparts, indicating that test score differences do not necessarily translate to meaningful professional distinctions.

In terms of neuroscience, there is evidence that male and female brains differ anatomically in subtle ways, but no one knows how (or even if) these anatomical differences relate to cognitive performance. Women have greater gyrification of the brain surface (and by inference, increased cortical surface area) in frontal and parietal regions1. Boys were reported to have an increased gray matter volume relative to girls (even correcting for total cerebral volume), but other authors2 have contradicted these findings, reporting that this difference in gray matter volume can be

accounted for by differences in brain size and is unrelated to gender. A related problem is that studies reporting gender differences in anatomy, no matter how small, are more likely to be published and reported in the press than those that fail to find such differences. And despite these differences, males and females score equally well on IQ tests3.

One of the clearest cognitive gender differences is in spatial reasoning and navigation, which some media reports have linked to mathematical ability. Spatial cognition is organized differently in male and female rats. In the Morris water maze, female rats rely more on frontal cortex for spatial navigation, whereas male rats rely more on entorhinal cortex. Human males and females use different neural strategies to maneuver through unfamiliar environments as well, with men showing a greater activation of left hippocampus4. Men and women also use different behavioral strate-gies—women are thought to focus on landmarks, whereas men tend to assess the euclidean properties of the environment. Similarly, men outper-form women on mental rotation tasks. Some of these differences are linked to hormones—a single testosterone injection improves women’s perfor-mance on a visuospatial task5. Although this difference may well have a biological basis, it seems much too narrow to account for the dramatic overrepresentation of men in science departments at top universities.

Social scientists find that changing a female name to a male name on otherwise identical work increases its perceived value. In addition, female and minority students who are aware of gender and racial ste-reotypes score lower on tests such as the SAT. In the early years of the SAT, females scored higher on the verbal section, until male test scores were raised by selective inclusion of questions on which males per-formed better, such as those on politics, business and sports6. No similar attempt has been made to ‘balance’ the math section of the SAT. In light of such evidence that gender bias influences test scores and academic success, it is difficult to take seriously the enshrinement of the test score gap as reflecting biological differences.

In a world of perfectly equal opportunity, what proportion of Harvard’s mathematics professors would be female? No one knows, and no studies can be done to find out because humans cannot be examined in a culture-free state. What does seem clear is that we do not live in such a perfect world. In this one, Summers’ comments as Harvard’s president—and the resulting media hype—are likely to make the road tougher for aspiring female mathematicians and scientists, who now must confront the additional handicap of being told that they are at a biological disadvantage. Putting less faith in aptitude differences and more belief in hard work and individual evaluation of performance seems like a more productive way forward.

1. Luders, E. et al. Nat. Neurosci. 7, 799–800 (2004).2. Luders, E., Steinmetz, H. & Jancke, L. Neuroreport 13, 2371–2374 (2002).3. Haier, R.J., Jung, R.E., Yeo, R.A., Head, K. & Alkire, M.T. Neuroimage 23, 425–433

(2004).4. Gron, G. et al. Nat. Neurosci. 3, 404–408 (2000).5. Aleman, A., Bronk, E., Kessels, R.P., Koppeschaar, H.P. & van Honk, J.

Psychoneuroendocrinology 29, 612-617 (2004).6. Tavris, C. The Mismeasure of Woman (Simon and Schuster, New York, 1992).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


B O O K R E V I E W

Functional interactions of a scientific collaborationBrain and Visual Perception: The Story of a 25-Year Collaboration

by David H Hubel and Torsten N Wiesel

Oxford University Press, 2004744 pp., hardcover, $49.50ISBN 0195176189

Reviewed by Andrew J Parker

Brain and Visual Perception is an elegantly presented and well-organized collection of critical and important papers from the 25-year collabora-tion of Hubel and Wiesel. The writing is interspersed with informal observations, almost exclusively written by Hubel, about the motiva-tions, thoughts and afterthoughts surrounding these experiments. The aim has been to convey the atmosphere and personal interactions associated with their research and to illustrate how their early scien-tific development shaped the collaboration and possibly even led to its ultimate fission. Fascinating though these commentaries are, they must be understood as a present-day perspective on events rather than a contemporary record of what happened. As such, for any future his-torian of science, they will be the starting point for inquiry rather than an answer to the question of what really happened.

The ‘slightly less than young’ reader should be warned against look-ing at the official bibliographic classification of this book, where two of the three main categorizations are ‘Biomedical Research—History’ and ‘History of Medicine, Twentieth Century’. This is certainly a sting for anyone who has grown up with these papers.

Indeed, the book arrived for review as I tidied up from last year’s teach-ing; the early papers from Hubel and Wiesel were there, with the long 1962 paper lying next to the 1959 paper, which introduced orientation selectivity as a property of visual cortical neurons. Among this bundle also lives the 1966 paper from Enroth-Cugell and Robson, describing the first use of sinusoidal visual grating patterns to investigate the cat retina. This paper must be the serpent in the garden of Eden in Hubel’s view, simply because of its use of sinusoidal gratings and linear systems analysis. This is the kind of controversy that draws readers to the com-mentaries to gain further insight. Here, the book is enormously helpful in helping people to relive these debates at first hand.

It is remarkable that a burning dispute could be set alight by this differ-ence in methodology. Both papers arose from honest and valid attempts to develop concepts from the pioneering work of Hartline, Barlow and

Kuffler. For example, both correctly claim to better define the visual recep-tive field of neurons; both use an explanatory framework in which the visual receptive field may be divided into regions, in some of which the presence of light causes the neuron to fire more action potentials, whereas in others light suppresses firing; and both consider that these regions are mutually antagonistic, so that excitation generated within one region can be cancelled by the suppressive effect of another. The 1959 Hubel and Wiesel paper is a classic teaching tool for introducing these concepts; it is always an interesting moment when students, who have read only textbooks, confront Figure 1. Here, they realize it is not obligatory to use an oriented bar or edge stimulus to reveal the orientation selectivity of a V1 simple cell, even though the commentaries tell us that this is how the property of orienta-tion selectivity was first discovered. Finally, the 1962 Hubel and Wiesel paper and the 1966 Enroth-Cugell and Robson paper both break this model by identifying types of neuron (complex cortical cells and Y-type retinal ganglion cells respectively) that fail to conform to the earlier conceptions.

Other themes run through the commentaries: for instance, Hubel’s conscious repudiation of all forms of administration and Wiesel’s will-ingness to undertake it, at least later in his career, particularly to give opportunities to younger scientists. However, the most striking comment for anyone, already working in the field or aspiring to do so, is the dif-ference in the nature of scientific writing between now and 40 years ago. Hubel thinks that reading modern papers is like ‘eating sawdust’. This comment provokes a lot of thought (and not a little defensiveness) as to the reasons for such a difference.

One issue must be the increased demand for quantification of results, which is surely a good thing. However, another issue is that, when the early papers were written, they were still the main method of commu-nicating with other scientists. Meeting face to face was difficult: one colleague said that to transport him and his family by air to the USA at that time cost no less than three months’ salary. So papers themselves had to fulfil the present-day roles of conference presentations, talks, seminars and Internet chit-chat, as well as an archival written record.

Just how the audience for papers differs is illustrated by the anecdote from the editor of the Journal of Physiology who dealt with the 1962 paper. The paper arrived close to a university vacation, and the editor realized that it would not be easy to find referees for it. Accordingly, as he was about to disappear on holiday himself, he decided he would simply read through the paper overnight; he had no background in vision, although he was a neurophysiologist. By the next morning, he had decided that it was clearly a very important paper, so without being deeply concerned about whether he had grasped all the detail, he accepted the paper without further editing. Such, I assume, is the version that we read nowadays.

This is a valuable volume. Librarians will want to acquire it, as it will save finger marks on bound journals and the ‘razoring out’ of journal pages by modern-day bounty hunters. Scientists will want it because, whether or not you like the commentaries, you have to admit that they are fascinating reading in themselves. Will it replace my reprint collec-tion? Probably not. Although it is far too early to consign these papers to the category of history, there is a great deal of the historic about them, which is best grasped by handling a venerable photocopy of rather poor quality or diving back into the original leather-bound volume, even if it does annoy the librarian.

Andrew J. Parker is in the University Laboratory of Physiology, University of

Oxford, Parks Road, Oxford OX1 3PT, UK.

e-mail: [email protected]

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


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

Inhibitory synapses turn excitingJulie A Kauer

How activity-dependent synaptic plasticity shapes the development of inhibitory synapses has remained unclear. In this issue, Gillespie et al. show that in the developing rat auditory system, inhibitory synapses transiently co-release glutamate. The consequent activation of postsynaptic NMDA receptors may be critical for the plasticity mechanisms that determine tonotopic sharpening.

How do the connections between neurons get correctly wired in the developing nervous system? Over the past 20 years, experimental support has accumulated for the idea that activity-dependent refinement of synaptic connections involves an increase or decrease in synaptic strength, mediated by forms of synaptic plasticity such as long-term poten-tiation (LTP) and long-term depression (LTD). However, such mechanisms of plas-ticity depend on the regulation of glutama-tergic NMDA receptors (NMDARs), and so the question remains as to how inhibitory synapses can be strengthened and maintained in an activity-dependent manner. How can a neurotransmitter whose job it is to inhibit and dampen excitability be used during development to signal synaptic stabilization or elimination?

In this issue, Gillespie et al.1 provide intrigu-ing data suggesting a possible mechanism: early in the development of an inhibitory synapse in the auditory system, not only are glycine and GABA released, but glutamate is released along with them, activating postsynaptic NMDARs and AMPA receptors. This glutamatergic component of synaptic transmission is pres-ent only during an early period critical for the activity-dependent development of the tono-topic map. These data suggest a previously unknown mechanism by which GABAergic or glycinergic synapses may activate postsynaptic NMDARs, which may be crucial for the refine-ment of appropriate synaptic connections.

LTP and LTD were first described at hip-pocampal and cerebellar synapses but are expressed by glutamatergic synapses through-

out the brain, where they are thought to constitute a basic property of excitatory syn-apses2. Perhaps the most prevalent and well-studied forms of LTP and LTD are NMDAR dependent. Although most of the glutamater-gic synaptic current is carried by postsynaptic AMPA receptors, NMDARs (often present at the same synapses) can be activated by con-comitant glutamate binding and significant intracellular depolarization. When NMDARs are activated, extracellular Ca2+ enters the neuron, enabling multiple downstream effec-tors and ultimately strengthening or weak-ening synaptic connections. The beauty of this system is its ability to provide synapse specificity: if a particular synapse is not driven sufficiently, neurotransmission will occur without local Ca2+ entry, and synaptic strength will remain unchanged. In contrast, with sufficient depolarization accompanying synaptic activation, Ca2+ will enter the local postsynaptic region, gating either synaptic depression or potentiation.

In the mature brain, synaptic strengthening and weakening have been proposed to underlie various forms of neuroadaptation and learn-ing, as well as contributing to pathological phenomena such as epileptiform firing, drug addiction and hyperalgesia. Furthermore, the same mechanisms are used during nervous sys-tem development to refine specific axonal path-ways in an activity-dependent manner. This idea was first proposed by Hebb, who suggested that during development, coincident activity in a synaptically coupled neuron pair should lead to synaptic strengthening; conversely, out-of-phase activity causes synaptic weakening. Most of the fast synaptic transmission in the cen-tral nervous system is glutamatergic, and the best-studied examples of activity-dependent pathway refinement have also focused on glu-tamatergic synapses3–7. In most cases, NMDAR activation is critical to each of these examples

of activity-dependent pathway refinement.GABA and glycine are the neurotransmit-

ters in the vast majority of non-glutamatergic synapses. In the mature nervous system, these transmitters directly gate chloride channels and inhibit postsynaptic neurons by direct hyperpolarization and current shunt. In some parts of the nervous system, synapses using these inhibitory transmitters are refined in response to activity. One place where this has been studied is in the central auditory nuclei, including the lateral superior olive8 (LSO). To assess interaural sound intensity differences, the LSO integrates excitatory information from the ipsilateral ear with inhibitory information from the contralateral ear. Contralateral infor-mation is relayed to the LSO by neurons in the medial nucleus of the trapezoid body (MNTB) that are glycinergic and, during development, also GABAergic (Fig. 1a). A tonotopic organi-zation and alignment of excitatory and inhibi-tory inputs allows LSO neurons to compare the activity levels that come from both ears (which reflect interaural intensity differences) in a fre-quency-specific manner. The afferent inputs from the MNTB to the LSO are heavily refined during development in response to spontane-ous neural and auditory inputs, resulting in a sharpening of the tonotopic organization of this pathway. Yet the afferents from the MNTB are inhibitory, raising a perplexing question: how can hyperpolarization and membrane stabilization, which depress neuronal activ-ity and Ca2+ entry through voltage-gated ion channels, be used to signal whether a synapse should be maintained or winnowed out in an activity-dependent manner?

A potential clue to the process was observed years ago: in many parts of the CNS, intracel-lular chloride levels in immature neurons differ from those in mature neurons, so that at resting membrane potentials, GABAA or glycine recep-tor activation causes depolarization rather than

Julie A. Kauer is at the Department of Molecular

Pharmacology, Physiology and Biotechnology,

Brown University, MPPB Box G B4, Providence,

Rhode Island 02192, USA.


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

258 VOLUME 8 | NUMBER 3 | MARCH 2005 NATURE NEUROSCIENCE


hyperpolarization9,10. This is also observed in the developing LSO in the first postnatal week11. This observation explains how the release of GABA or glycine onto their cognate receptors can provide depolarization. However, the post-synaptic response will still depend on chloride channels, which are only capable of allowing chloride to leave the neuron and cannot initiate further intracellular signaling. As in glutama-tergic synapses, one potential signal that could influence synaptic strength is calcium entry through NMDARs. However, although the depolarization evoked by Ca2+ flux in the imma-ture neuron could in theory suffice to remove the Mg2+ block from NMDA channels, without glutamate binding occurring at the same time as the depolarization provided by the GABA/gly-cine receptors, the NMDARs do not open.

This scenario with NMDARs seemed unnecessarily complex: would the strength-ening or weakening of GABA/glycine syn-apses depend upon simultaneous activity in glutamatergic pathways? This cannot be the case, as in some studies functional glu-tamatergic afferents are not even present when GABA/glycine synapses are stabilizing. Nonetheless, there is evidence in several sys-tems, notably the developing hippocampus, supporting the involvement of NMDARs in LTP of nascent GABAergic synapses12. At MNTB-LSO synapses, Ca2+-dependent LTD has been observed in response to low-frequency afferent stimulation in vitro13.

The paper by Gillespie et al . in this issue now sheds light on this paradox by reporting that when the glycine/GABA postsynaptic currents in the LSO are blocked with appropriate antagonists, a postsynaptic current mediated by glutamate receptors remains. Although both AMPA and NMDA responses can be observed in most synapses, the NMDA component dominates. This suggests that these immature inhibi-

tory neurons in the LSO also release g lutamate (Fig. 1b).

How do the authors show that glutamate as well as GABA and glycine are released from the same synapse? Because they record postsynaptic currents from a brain slice (evoking responses by electrical stimula-tion), one obvious explanation is that the electrical stimulation activates a disyn-aptic pathway, one arm of which releases glutamate. However, the authors’ experi-ments argue against this hypothesis. First, the latency of both the glutamate and the GABA/glycine synaptic currents are similar, as expected if they arise from a single set of synapses. Second, even with the minimal stimulation likely to activate a single nerve terminal, there is evidence for both types of neurotransmitter release. Third, the authors uncage glutamate in the MNTB some 500 µm from the LSO, thus avoiding acti-vating stray glutamatergic fibers, a potential problem with nonspecific electrical stimu-lation. Uncaging glutamate in the MNTB evokes a synaptic current in LSO neurons with a significant glutamatergic component. Finally, the authors examine the distribution of vesicular glutamate transporters, mol-ecules required for filling synaptic vesicles with glutamate, and find that the vesicular glutamate transporter 3 colocalizes with the vesicular GABA transporter in the terminals on LSO cells, again supporting the idea that single GABA/glycinergic terminals also co-release glutamate.

Intriguingly, in the LSO, the glutamatergic component of synaptic transmission seems to decline sharply in the second postnatal week, coincident with the switch of the chloride gra-dient from depolarizing to hyperpolarizing. It is tantalizing to speculate that the presence of NMDARs during this early postnatal period is essential for synaptic plasticity that may presage the enormous loss and refinement of

GABA/glycinergic MNTB synapses during the same period14.

Many other questions are also raised by this work. Do the NMDARs at MNTB-LSO syn-apses contribute to synapse elimination or to strengthening of connections? Is functional synaptic strengthening or weakening observed in vitro at this synapse13 followed by synapse elimination? How do GABAA receptors , gly-cine receptors, AMPA receptors, NMDARs and associated scaffolding proteins all coexist at the same synaptic site? Do multiple GABA/glycin-ergic synapses throughout the CNS co-release glutamate during a critical period, as suggested by the presence of the glutamate transporter VGLUT3 at many inhibitory nerve terminals15? Finally, does the shift in chloride equilibrium trigger a loss of glutamate receptors and trans-mitter release? The well-defined structure and relative simplicity of the LSO makes it an excel-lent system in which to examine such questions and will provide further insights into the devel-opment of inhibitory synapses.

1. Gillespie et al. Nat. Neurosci. 8, 332–338 (2005).2. Malenka, R.C. & Bear, M.F. Neuron 44, 5–21 (2004).3. Hubel, D.H. & Wiesel, T.N. J. Physiol. (Lond.) 160,

106–154 (1962).4. Constantine-Paton, M. & Cline, H.T. Curr. Opin.

Neurobiol. 8, 139–148 (1998).5. Ruthazer, E.S. & Cline, H.T. J. Neurobiol. 59, 134–

146 (2004).6. Takahashi, T., Svoboda, K. & Malinow, R. Science 299,

1585–1588 (2003).7. Bear, M.F. & Rittenhouse, C.D. J. Neurobiol. 41, 83–

91 (1999).8. Kandler, K. Curr. Opin. Neurobiol. 14, 96–104

(2004).9. Obata, K., Oide, M. & Tanaka, H. Brain Res. 144,

179–184 (1978).10. Ben-Ari, Y. Nat. Rev. Neurosci. 3, 728–739 (2002).11. Kandler, K. & Friauf, E. J. Neurosci. 15, 6890–6904

(1995).12. Gaiarsa, J.L., Caillard, O. & Ben-Ari, Y. Trends Neurosci.

25, 564–570 (2002).13. Kotak, V.C & Sanes, D.H. J. Neurosci. 20, 5820–5826

(2000).14. Kim, G. & Kandler, K. Nat. Neurosci. 6, 282–290

(2003).15. Fremeau, R.T. Jr, Voglmaier, S., Seal, R.P. & Edwards,

R.H. Trends Neurosci. 27, 98–103 (2004).

Figure 1 Anatomy of the mammalian auditory brainstem. (a) Information from the right cochlear nucleus is sent to the MNTB, which projects via glycinergic/GABAergic afferents to the left LSO. The left LSO compares this information with that arriving via excitatory inputs from the left cochlear nucleus. A tonotopic alignment here is crucial for processing frequency-specific interaural intensity differences used for binaural sound localization. (b) During a critical period in development, afferents from the MNTB release glutamate as well as glycine and GABA. Postsynaptic AMPA and NMDARs on the LSO neurons are activated in response to glutamate release1.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



Stem cells in the injured spinal cord: reducing the pain and increasing the gainSandra Klein and Clive N Svendsen

Stem cells transplanted after spinal cord injury mostly generate astrocytes, which can promote aberrant sprouting of sensory neurons, leading to allodynia. Making cells produce more oligodendrocytes reduces allodynia and improves functional recovery.

Over 10,000 Americans sustain spinal cord injuries (SCI) every year. The major symp-tom is the loss of movement and sensation in areas of the body innervated from areas of the spinal cord below the site of injury. However, some patients also develop a mod-erate or severe hypersensitivity to stimuli that were not previously painful—a condi-tion termed allodynia. This type of pain is often unresponsive to traditional treatments, and its origin is not completely understood. In this issue, Hofstetter et al.1 describe how neural stem cells (NSCs) injected into the spinal cord after SCI in the rodent can have significant positive effects on both move-ment and sensation below the site of injury. However, if naive stem cells are transplanted into injured animals, they also result in a significant increase in allodynia associated with sprouting of sensory fibers within the spinal cord. By forcing the stem cells down a distinct lineage in the culture dish before transplantation, Hofstetter et al.1 were able to avoid allodynia while further improving functional recovery. These thoughtful and rigorous observations highlight both the amazing potential benefits and the possible dangers of using neural stem cells to treat the damaged nervous system.

In SCI, the connection between the brain and spinal cord is essentially lost through damage (Fig. 1a,b). Above the lesion, descending pathways that were completely severed retract and regroup within the proximal stump. Damaged but intact path-ways that survive are subject to inflammation and secondary chemical damage associated with the trauma, and they often lose myelin, leading to poor signal transduction2. Current treatments for SCI are limited to the use of steroids, such as methylprednisolone, which must be administered soon after the injury, although their use and effectiveness remain controversial3. More recently there has been a

large interest in the possibility of using stem cells to treat spinal cord injury, fueled by the pioneering spirit of Christopher Reeve and his foundation. But can stem cells be used to treat SCI?

One approach is to transplant new neu-rons above the site of the lesion, which may sprout connections across to the other side. If these connect to any of the original out-put neurons below the lesion and are inner-vated by remaining fibers in the proximal stump, there may be some motor function improvement4. The power of stem cells lies in their ability to generate many different cell types; neural stem cells derived from the developing or adult brain can produce neu-rons, astrocytes and oligodendrocytes5. Yet when transplanted into the brain or spinal cord, they generate mainly non-neuronal cells, including astrocytes6. In an attempt to suppress differentiation into astrocytes, Hofstetter et al.1 virally induced adult spi-nal cord–derived NSCs to express the helix-loop-helix neurogenic transcription factor neurogenin-2 (Ngn-2; ref. 7) and the marker green fluorescent protein (GFP). These Ngn2-NSCs were then transplanted into a rat model of SCI and compared with naive NSCs that only expressed GFP. However, the authors had a number of surprises in store. Although Ngn-2 promotes neuronal differ-entiation in vitro, this was not true follow-ing transplantation. Instead, the stem cells differentiated into mainly oligodendrocytes, whereas naive NSCs became GFP-positive astrocytes, as expected from previous trans-plant studies.

What was the functional outcome? Transplants of naive NSCs did lead to some motor recovery (Fig. 1c), but Ngn2-NSC grafts had significantly greater effects (Fig. 1d). In addition, Ngn2-NSC–transplanted animals also showed a 50% improve-ment in hindlimb sensation based on hot-plate withdrawal response latency. The authors also used fMRI to look for evidence of functional recovery in somatosensory cortex. Activation in response to hindlimb stimulation was partially restored in animals transplanted with Ngn2-NSCs but not in those receiving vehicle or naive NSC transplants.

As the Ngn2-NSCs generated more oligodendro-cytes, the improved recovery may have been due to increased myelination of axons within the injured white matter tracts. Remyelination using a variety of cell types, including olfactory ensheathing cells, has been beneficial in models of SCI8,9. Indeed, Ngn2-NSCs–transplanted animals in the current study showed significantly greater white matter area than naive NSC– or vehicle-treated animals (Fig. 1d). However, without electron microscopy, the gold standard for establishing axon myelina-tion, the exact amount of remyelination and how it correlated with recovery is difficult to determine from this study.

There is yet a further twist in this tale. The authors observed that transplants of naive NSCs resulted in significantly more allodynia than Ngn2-NSCs or the lesion alone. This induced allodynia was found rostral to the injury site, in the unaffected forelimbs. What might be the mechanism? The authors hypoth-esized that the increased number of astrocytes produced by the naive NSCs may have been responsible for the allodynia. NSCs and astro-cytes in lesion models are known to release a range of trophic factors, including NGF10, which has direct effects on axonal sprouting and cell survival11. Although these functions are generally regarded as favorable for recovery, they could lead to negative consequences.

Hofstetter et al.1 found increased nociceptive fiber sprouting into inappropriate regions of the dorsal horn at spinal cord levels above the lesion. The amount of sprouting correlated well with the number of NSC-derived astrocytes in the transplant and with allodynia-related behavior (Fig. 1c). The authors therefore concluded that transplantation of naive NSCs and their sub-sequent differentiation into astrocytes led to decreased pain thresholds in the transplanted animals, which could present a severe limita-tion on moving stem cell therapy for spinal cord injury to the clinic. Fortunately, animals trans-planted with Ngn2-NSCs did not show increased allodynia or fiber sprouting. Therefore, direct-ing differentiation of NSCs away from astrocytes and into oligodendrocytes produced a ‘double whammy’ effect of increasing functional recov-ery while also reducing allodynia.

Sandra Klein and Clive N. Svendsen are at the

Waisman Center at the University of Wisconsin,

Madison, Wisconsin 53705, USA.


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



SCI leads to large changes in spinal cord anatomy, and it creates a dynamic environ-ment in which active regenerative processes might be diverted in a pathological direc-tion. This highly reactive state, in addition to enabling cortical sensorimotor remodel-ing in response to the lesion, may produce a unique situation in which astrocytes facilitate sensory sprouting, leading to allo-dynia. Hofstetter et al.1 also found exten-sive sprouting of lesioned corticospinal tract fibers in the proximal stump of animals transplanted with naive NSCs as opposed to Ngn2-NSCs, but this did not correlate with any functional recovery. Presumably, these extra fibers could not connect with any sys-tems below the level of injury.

This study highlights a number of impor-tant issues. In some cases, it will be vital to optimize the differentiation of stem cells before transplantation to allow good func-tional outcomes while reducing possible side effects such as allodynia. Furthermore, modulation of NSC migration after trans-plantation may also be crucial. Although it

was not clear from the present study1 exactly how many astrocytes migrated into the dor-sal horn to induce sprouting, it would be wise to target specific regions of the CNS for therapy rather than allowing widespread migration to occur.

Does this report mean that NSC differ-entiation into astrocytes may lead to allo-dynia in the intact spinal cord or other types of spinal cord damage or pathology? The authors did not include a control group of animals without lesions that received astrocytes to see if they also showed allo-dynia. Nevertheless, the relevance of this study to other disorders of the spinal cord needs to be examined. A difference in the baseline activity of dorsal horn neu-rons separates SCI from other diseases. Hyperactivity of sensory neurons rostral to the lesion has been observed and may contribute to the hypersensitivity to pain in SCI12. Neurodegenerative diseases in which populations of neurons die slowly may not produce changes in neuronal activ-ity, circuitry or extracellular environment

similar to those seen in SCI. One example is amyotrophic lateral sclerosis (ALS), in which motor neurons are selectively and progressively lost, leading to paralysis and death. Recent data suggest underlying astrocyte pathology may result in the sec-ondary death of motor neurons13. Thus, in this condition, replacement of astrocytes seems warranted. Alterations in sensory neuron activity have not been reported in ALS, and therefore stem cell–derived astro-cytes may not influence normal sensory neurons as they influence neurons with a predisposition to hyperactivity. Given the different disease course in ALS as compared with SCI, it will be of great interest to see whether NSC-derived astrocytes also cause allodynia here.

NSCs are promising treatments for central nervous system disease, including neurode-generation, ischemia and spinal cord injury. Treating SCI will require both repairing the initial injury of severed fiber tracts and also fighting widespread secondary damage. Stem cells may provide new oligodendrocytes for remyelination of injured fiber tracts, and through trophic support and extracellular matrix changes, they may counteract fac-tors in the lesioned environment that inhibit axonal regeneration. However, great power often comes at a high price. Although spi-nal cord injury warrants radical approaches, every effort must be made to optimize stem cell transplantation strategies. If only modest functional benefits are accompanied by unin-tended and serious negative side effects, such as the increased allodynia Hofstetter et al.1 describe, the first clinical transplants may be heavily criticized. Careful studies such as this one are now necessary to establish potential side effects and, in some cases, fine-tune stem cell differentiation before transplantation.

1. Hofstetter, C.P. et al. Nat. Neurosci. 8, 346–353 (2005).

2. Waxman, S.G. J. Neurol. Sci. 91, 1–14 (1989).3. Short, D.J., El Masry, W.S. & Jones, P.W. Spinal Cord

38, 273–286 (2000).4. Bareyre, F.M. et al. Nat. Neurosci. 7, 269–277

(2004).5. Gage, F.H. Science 287, 1433–1439 (2000).6. Winkler, C. et al. Mol. Cell. Neurosci. 11, 99–116

(1998).7. Sommer, L., Ma, Q. & Anderson, D.J. Mol. Cell.

Neurosci. 8, 221–241 (1996).8. Keyvan-Fouladi, N., Raisman, G. & Li, Y. J. Neurosci.

23, 9428–9434 (2003).9. Ramon-Cueto, A., Cordero, M.I., Santos-Benito, F.F.

& Avila, J. Neuron 25, 425–435 (2000).10. Brown, A., Ricci, M.J. & Weaver, L.C. Exp. Neurol.

188, 115–127 (2004).11. Llado, J. et al. Mol. Cell. Neurosci. 27, 322–331

(2004).12. Hoheisel, U. et al. Brain Res. 974, 134–145 (2003).13. Clement, A.M. et al. Science 302, 113–117 (2003).

Figure 1 Recovery from SCI after stem cell transplantation. (a) Intact spinal cord with pain (blue) and motor (green) tracts highlighted. (b) SCI causes cyst development (dark red) and loss of descending motor output to the muscle and ascending sensory input to the brain (dotted lines). Minor forelimb allodynia is found rostral to lesion (thicker line). (c) NSC transplantation to sites just rostral and caudal to injury. NSC group showed hindlimb motor improvement (solid line) without recovery of white matter (box). Ascending hindlimb sensory deficits persisted (dotted line). Significant forelimb allodynia correlated with sprouting of nociceptive fibers into lamina III (LIII) rostral to the lesion (thick line and box). (d) NSCs expressing Ngn2 promote additional hindlimb motor and sensory improvement (solid lines). Hindlimb recovery may be due to increased myelination of spared fiber tracts (box). Significantly less forelimb allodynia is reported than with naive NSC transplants, and this is correlated with reduced sprouting (solid line and box).

Intact SCI

SCI + NSC SCI + Ngn2-NSC

MotorSensory

L IIIL III

Deb

bie

Mai

zels

a b

c d

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



Controlling stress: how the brain protects itself from depressionTrevor W Robbins

Having control over a stressful situation can reduce its negative physiological and cognitive consequences. In this issue, a new study in rats suggests that descending inputs from the prefrontal cortex to the serotonergic midbrain signal the controllability of stress.

The impact of traumatic life events is often gauged by how well people appear to cope with the experience. This suggests that there are psy-chological control processes that can buffer the adverse consequences of stress. Exactly how these elusive cognitive factors interact with the often unconscious, vegetative consequences of stress has been a central question in neurosci-ence, representing yet another take on the famil-iar ‘mind-body’ problem. In this issue, Amat et al.1 suggest that such an interaction might occur between the prefrontal cortex (PFC), one of the more recently evolved ‘executive centers’ of the brain, and the serotonergic system, which proj-ects diffusely from the phylogenetically ancient dorsal raphé nucleus (DRN).

Early work by Weiss and others2 helped to define the conditions under which psycho-logical control over stress could be inferred in experimental animals. Rats exposed to inescap-able shocks showed significantly more gastric ulceration and higher levels of stress hormone (corticosterone) than rats receiving the same quantity, intensity and scheduling of shock but with the opportunity to voluntarily turn a wheel and stop the shocks. The ability of the animal to exert control over its environment seems to act as a ‘coping behavior’ that somehow combats the deleterious effects of stress.

Such control can also reverse the deleterious effects (depletion or overactivity) of uncontrol-lable stress on those central neurotransmitter systems implicated in anxiety and depression, such as serotonin (or 5-hydroxytryptamine, 5-HT) and noradrenaline3. The dramatic func-tional consequences of uncontrollable shock have served as a model for the symptoms of depression. Uncontrollable shock leads not only to increased ulceration and increased fear, but also to a range of depressive symptoms, including weight loss and sleep abnormalities. Moreover, animals exposed to uncontrollable shock show an apparent inertia in learning new

environmental contingencies, including those relating to rewards such as food. Seligman memorably described this effect as ‘learned helplessness’, based on the idea that perceived lack of control over one’s environment leads to future inaction4.

Although the theory of learned helplessness was advanced to explain how exposure to stress can prevent some people from learning how to cope with their trauma, thus leading to depres-sion, it is now clear that learned helplessness may be even more important in other affective disor-ders, including post-traumatic stress disorder4. Nevertheless, it is significant that many of the adverse consequences of uncontrollable stress in animals can be ameliorated by treatment with antidepressant drugs, including those affect-ing the central monoamine (5-HT and nor-adrenaline) systems3.

The various somatic, affective and cognitive effects of inescapable stress all may potentially result from dysfunction of central monoamine

systems, because of their widely ramifying innervation of diverse brain regions. Yet, given their location in brainstem and midbrain and their presumed impoverished access to infor-mation required for cognitive control, the question remains of how controllability could possibly regulate these ‘dumb’ chemical systems and lead to protective effects. The new findings of Amat et al.1 suggest a role for descending inputs from the PFC5–7, which could sense the environmental contingencies of behavioral control and use this information to regulate the activity of serotonergic neurons in the DRN.

Amat et al.1 carefully based their study on several well-known effects of inescapable stress on the 5-HT system3, as well as on a burgeoning understanding of the cognitive functions of the rat PFC8, which may represent the initial building blocks of the more complex executive functions of the human frontal lobes. Certain areas of the rat PFC, such as the infralimbic cortex, are prob-ably homologous to the human ventromedial

Trevor W Robbins is in the Department of

Experimental Psychology and the MRC Centre

for Behavioural and Clinical Neuroscience at the

University of Cambridge, Cambridge, UK.


Figure 1 Sensing control over stress. The dorsal raphé nucleus (DRN) is a major source of ascending serotonergic (5-HT) input to forebrain structures such as the neocortex, the dorsal (DS) and ventral (VS) striatum, and the amygdala (Amyg)11. The ventromedial prefrontal cortex (PFC) (including the infralimbic and prelimbic regions in the rat brain) is a major source of descending input to the DRN5,6. Stressors can elevate the activity of 5-HT neurons in the DRN through a number of other inputs to this structure (not shown)3. One potential consequence of uncontrollable, chronic stress is a dysregulation of activity in the ascending 5-HT system3, which likely impairs information processing in its diverse terminal domains, possibly leading to depression and other affective disorders. By sensing the capacity to exert control over stress through instrumental behavior, the mPFCv may modulate the activity of 5-HT neurons in the DRN through descending excitatory afferents (purple), either directly or through inhibitory GABAergic (G) interneurons5. Amat et al.1 show that inactivation of the mPFCv in rats eliminates the protective effect of cognitive control on 5-HT activity and behavior.

Neocortex

DRN

PFC

DS

G

VS

5-HT

'Control'

Stress

–

Amyg

Debbie Maizels

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



PFC. This area is importantly implicated in the integration of cognitive and emotional influ-ences8 and may be one of the key sites where antidepressant drugs affecting the 5-HT system act9. Amat et al. capitalized on neuroanatomical and neurophysiological findings5–7 that the rat infralimbic cortex sends descending projections to the serotonergic DRN, where they synapse onto 5-HT neurons and exert inhibitory control indirectly via GABAergic interneurons5 (Fig. 1). The rat medial PFC is not generally involved in associative learning per se but rather in learning about certain types of instrumental contingencies by which a voluntary behavior leads to a goal such as food delivery or shock escape10. Thus, it is likely that some mechanism in this region enables the animal to sense and realize its capacity for con-trolling its environment and to use this informa-tion for a variety of functions.

One such function seems to be ‘instructing’ DRN cells about the controllability of stress. Inescapable shock activates 5-HT DRN neurons more than escapable shock, and exposure to inescapable shock, but not to escapable shock, potentiates subsequent fear conditioning and impairs subsequent escape learning. When the descending influence of the ventromedial pre-frontal cortex (mPFCv) was eliminated tem-porarily by local infusion of the GABAergic agonist muscimol, escapable (or controllable) shock heightened activity of the DRN cells to a similar degree as inescapable shock, as indi-cated by increased c-fos expression and elevated extracellular 5-HT in the vicinity of the DRN neurons1. Even more impressively, without the influence of mPFCv, exposure to controllable shock potentiated the behavioral expression of fear and impaired effective escape learning. Thus, it appears that the normal susceptibil-ity of the 5-HT neurons to stress, which may lead to adverse functional consequences, can be moderated by the influence of cognitive influ-

ences relayed from the PFC. When this influ-ence is blocked, the normal protective effects of learned controllability are also blocked.

Like other important neuromodulators such as noradrenaline, dopamine, histamine and ace-tylcholine, serotonin is important in optimizing the output of forebrain cortical networks that are necessary for adaptive behavior. Although these neuromodulatory systems have been implicated in more general physiological processes such as arousal and stress, it seems likely that, in subtly different ways, they enable the animal to finely tune the activity of diverse forebrain regions with complementary functions, including such structures as the hippocampus and the PFC itself11. Thus, the control function exerted by the mPFCv on ascending 5-HT activity may indirectly affect processing in other forebrain regions to which the DRN projects (Fig. 1). These regions might be involved in perceptual, motor, mnemonic and cognitive functions that are relevant to depression.

Control principles similar to those demon-strated by Amat et al. for the 5-HT system may also apply to the ascending noradrenergic7,12, dopaminergic13 and cholinergic14 systems, as the PFC also sends descending projections to each of these modulatory systems (perhaps from different PFC regions). For example, the midbrain dopaminergic system may be involved in error prediction for reward learning, which also implies some ‘education’ of those cells about the nature of behavioral contingencies13. Sensitization of the response to repeated doses of stimulant drugs such as amphetamine depends on the influence of descending input from the PFC on midbrain dopamine neurons15. There is also clear neurophysiological evidence for descending PFC influence on the functioning of the noradrenergic locus coeruleus in the rat12. However, the findings of Amat et al. for the DRN 5-HT neurons provide the most compelling evi-

dence to date of the capacity of the PFC to medi-ate the influence of cognitive control.

Future studies will clarify how this array of adaptations, orchestrated by the PFC, optimizes the response to stress. The present discovery of one mechanism by which cog-nitive control regulates serotonergic activity reveals yet another aspect of the executive functions the PFC, better known for its coor-dination of cortical processes and its control over the behavioral and cognitive output of the striatum and motor cortex. This alterna-tive form of executive control through the regulation of the chemically defined systems of the brain’s reticular core enables the PFC to more generally modulate a range of fore-brain areas by adjusting the ‘tone’ of their subcortical inputs, an interaction that may correspond to the influence of mood states over cognition.

1. Amat, J. et al. Nat. Neurosci. 8, 365–371 (2005).2. Weiss, J.M. & Simson, P.G. Ciba Found. Symp. 123,

191–215 (1986).3. Anisman, H. et al. in Animal Models in Psychiatry II (ed.

Martin-Iverson, M.T.) 1–55 (Humana, Clifton, New Jersey, USA, 1992).

4. LoLordo, V.M. in Animal Research and Human Health (eds. Carroll, M.E. & Overmier, J.B.) 63–77 (American Psychological Association, Washington DC, 2001).

5. Celada, P. et al. J. Neurosci. 21, 9917–9929 (2001).6. Hajos, M., Gartside, S.E., Varga, V. & Sharp, T. Neuroscience

87, 95–108 (1998).7. Arnsten, A.F.T. & Goldman-Rakic, P.S. Brain Res. 306,

9–18 (1984).8. Vertes, R.P. Synapse 51, 32–58 (2004).9. Mayberg, H.S. Br. Med. Bull. 65, 193–207 (2003).10. Corbit, L.H. & Balleine, B.W. Behav. Brain Res. 146,

145–157 (2003).11. Azmitia, E.Z. & Whitaker-Azmitia, P.M. in

Psychopharmacology, The Fourth Generation of Progress (eds. Bloom, F.E. & Kupfer, D.J.) 443–480 (Raven, New York, 1995).

12. Jodo, E., Chiang, C. & Aston-Jones, G.J. Neuroscience 83, 63–79 (1998).

13. Hollerman, J.R., Tremblay, L.& Schultz, W. Prog. Brain Res. 126,193–215 (2000).

14. Gaykema, R.P.A., VanWeeghel, R., Hersh, L.B. & Luiten, P.G.M. J. Comp. Neurol. 303, 563–583 (1991).

15. Li, Y. et al. Synapse 34, 169–180 (1999).

Myelin repair: developmental myelination redux?Roumen Balabanov & Brian Popko

A recent study of an Olig1 knockout mouse concludes that remyelination after injury may occur by a different mechanism from myelination during normal development, but another report suggests that this mouse model should be interpreted cautiously.

Roumen Balabanov and Brian Popko are at the

Jack Miller Center for Peripheral Neuropathy and

the Department of Neurology, The University of

Chicago, 5841 South Maryland Avenue, MC2030,

Chicago, Illinois 60637, USA.

e–mail: [email protected]

Just as the restoration of a deteriorated home presents challenges not encountered during original construction, the repair of a dam-aged adult nervous system faces obstacles distinct from the critical issues of develop-ment. Normal developmental attractive and

repulsive guidance cues are generally no lon-ger present; damage has typically resulted in a multitude of changes in the nervous system’s cytoarchitecture, and the damaging agents frequently are still present, creating an unfa-vorable environment. Nevertheless, we often

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



use developmental models in an effort to bet-ter understand repair. But is remyelination the same as developmental myelination? A study by Arnett et al. in a recent issue of Science sug-gests that, at least when considering the repair of damaged myelin sheaths, we need to be careful when extrapolating from developmen-tal systems1. Their work, which used a mouse model with a targeted mutation in a myelinat-ing cell transcription factor, indicated that the process of remyelination in response to demy-elinating insults is distinct from developmental myelination. However, a separate recent report from Xin et al. in the Journal of Neuroscience suggests that there may be a complication with the knockout mouse model2.

Myelin, the multilayered membrane sur-rounding nerve axons, is produced by oligoden-drocytes in the CNS and Schwann cells in the PNS3. Myelin sheath formation causes the parti-tioning of axonal voltage-gated sodium channels to the node of Ranvier, a process that is essential for the rapid saltatory propagation of the action potential4. Abnormal myelination during devel-opment can cause a developmental delay, func-tional incapacitation, and/or premature death (as seen in vanishing white matter disease5). Moreover, the demyelination of properly formed

myelin occurs in a number of adult-onset disor-ders, including Guillain-Barré syndrome in the PNS and multiple sclerosis in the CNS, causing a severe disruption of nervous system function6. Prolonged myelin loss can also potentially result in axonal degeneration7.

A comprehensive understanding of the remyelination process is thus an area of con-siderable scientific interest and utmost clinical importance. Remyelination is a natural regen-erative process causing symptomatic improve-ment, but it is unfortunately restricted and inefficient in chronic demyelinating disor-ders8. Therefore, although enhancement of the myelin repair process may not provide a cure, it does offer a therapeutic opportunity to improve a patient’s functional recovery and quality of life. Thus far, our understanding of remyelination has been heavily influenced by insights gained from studies of myelination during development.

However, Arnett and colleagues now present intriguing genetic evidence that remyelination may differ in critical ways from developmen-tal myelination1. The central finding in their study is that the effects of the inactivation of the gene (Olig1) encoding oligodendrocyte transcription factor 1 (Olig1) are markedly

dissimilar during myelination and remyelin-ation. Normal controls and Olig1 knockout mice were subjected to chemical demyelin-ation (either by cuprizone or lysolecithin) and examined during the remyelination phase. As expected, wild-type mice demonstrated robust and efficient remyelination. Olig1 mutant mice, however, showed normal myelination patterns during development prior to the demyelinating insult but strikingly ineffi-cient remyelination. The Olig1 mutant mice generated a strong oligodendrocyte prolifera-tive and recruitment response, but these cells subsequently did not differentiate into mature oligodendrocytes or efficiently remyelinate axons. This study suggests that Olig1 is essen-tial for the remyelination process but not for developmental myelination.

Olig1, together with Olig2, belongs to a fam-ily of proteins recently described as sonic hedge-hog–induced basic helix-loop-helix (bHLH) transcription factors9,10. Initial studies implicated Olig1 and Olig2 in motor neuron generation and oligodendrogenesis during development. Double Olig1/Olig2 knockout mice and Olig2 knockout mice die during the perinatal period. In contrast, the Olig1 mutant mice do not show significant phenotypic abnormalities, except for a subtle delay in oligodendrocyte matura-tion11,12. Despite this maturation problem, nor-mal oligodendrocytes and myelin are generated by postnatal day 30 in the Olig1 mutants.

Arnett et al. also describe the cellular expres-sion and the dynamic subcellular localization of Olig1 during remyelination in experimental animal models and in specimens from human patients with multiple sclerosis1. In normal con-trols, Olig1 was localized to the cytoplasm of the mature oligodendrocytes but to the nuclei of the oligodendrocyte progenitors involved in remy-elination. Similar patterns of Olig1 expression were also detected in multiple sclerosis lesions. Cytoplasmic oligodendrocyte localization of Olig1 was detected in the normal-appearing white matter of multiple sclerosis specimens, whereas nuclear Olig1 localization was seen in the oligodendrocyte progenitor cells specifically at the edges of the active lesions (the site of pre-sumptive remyelination). This dynamic local-ization of Olig1 contrasted with the continuous nuclear localization of Olig2, which remained unchanged regardless of the maturational state of the oligodendrocyte.

A study by Xin et al. published recently in the Journal of Neuroscience adds an interesting twist to the story2. The original Olig1 knockout mouse used by Arnett and colleagues was created using the fairly typical technique of using a PGKneo cassette as the insertional mutagen, which allowed the authors to also use neomycin resistance as the selectable marker for the targeted embryonic stem

Figure 1 Olig1 and oligodendrocyte differentiation. Normal oligodendrocyte differentiation (top) requires the expression of the Olig1 (red) and Olig2 (blue) transcription factors. In Olig1 knockout mice, oligodendrocyte differentiation seems to be arrested, and no myelin is formed. Note the dynamic localization of Olig1 during differentiation, in the cell nucleus and, after myelination, in the cytoplasm, as opposed to the continuous localization of Olig2 in the cell nucleus.

Oligodendrocyteprogenitor

Unmyelinated axon

Oligodendrocyte maturation arrest

Myelinated axon

Mature oligodendrocyteImmature oligodendrocyte

Immature oligodendrocyte

Olig2

Olig2Olig1

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



cells. The transcriptional control region of the PGK (phosphoglycerate kinase) gene, however, also contains a strong transcriptional enhancer that has the capacity to influence the expression of neighboring genes, as previously described in the Hox gene complex, among others13,14. Furthermore, the Olig2 gene, encoding a pro-tein with 80% homology to Olig1, is located just 40 kb away from Olig1 on mouse chromosome 16. Xin et al. speculated that the presence of the PGK transcriptional control sequences in the Olig1 locus might enhance the transcriptional activity of the linked Olig2 gene, enabling the Olig2 protein to compensate for the Olig1 muta-tion2. When they eliminated the PGKneo cassette from the targeted Olig1 locus, leaving behind a different insertional mutagen, the Olig1 mutant mice now developed severe myelination abnor-malities and died during the third postnatal week. Myelination in these mice was essentially arrested at the level of oligodendrocyte progenitor cell differentiation. This step of oligodendrocyte differentiation arrest seems to be similar to the one described by Arnett and colleagues in their remyelination experiments (Fig. 1).

The results of Xin et al. therefore call into questions the conclusion of Arnett et al., that remyelinating oligodendrocytes are more

sensitive to the loss of Olig1 than are devel-oping oligodendrocytes. Nevertheless, the two studies demonstrate that PGK sequences inserted into the Olig1 locus have a differing ability to compensate for the loss of Olig1 during development versus repair. Adult oli-godendroglial progenitor cells may be less sensitive than their developmental counter-parts to the potentially compensatory activ-ity of Olig2 or to other unidentified genes in the region of the PGK sequences inserted into Olig1. Alternatively, enhancement of transcription by PGK sequences may be reduced in adult progenitor cells relative to developing cells. Another possibility is that the unique extracellular environment encountered by remyelinating oligodendro-cytes might influence the ability of Olig2 or other genes in the region to compensate for the absence of Olig1.

Regardless of the molecular mechanism, the observations described in these two reports are consistent with the overall conclu-sion of Arnett et al. that remyelinating adult oligodendrocyte progenitors display charac-teristics distinct from those of developing oli-godendroglial progenitor cells. These studies highlight the need for a better understand-

ing of the molecular and cellular processes involved in myelin repair. They also call into consideration the degree to which a particu-lar knockout mouse phenotype is influenced by the activities of untargeted neighboring genes. Ultimately, the question of the neces-sity of Olig1, as well as other genes, in devel-opmental myelination versus myelin repair will require the use of conditionally targeted alleles in combination with temporally controllable and oligodendrocyte-specific systems of gene inactivation15.

1. Arnett, H. et al. Science 306, 2111–2115 (2004).2. Xin, M. et al. J. Neurosci. 25, 1354–1365 (2005).3. Morell, P. et al. in Basic Neurochemistry 6th edn. (eds.

Siegel, G., Agranoff, B., Albers, W., Fisher, S. & Uhler, M.) 69–94 (Lippincott-Raven, New York, 1999).

4. Salzer, J.L. Neuron 40, 297–318 (2003).5. Di Rocco, M. et al. Am. J. Med. Genet. 129B, 85–93

(2004).6. Liblau, R. et al. Trends Neurosci. 24, 134–135

(2001).7. Waxman, S. N. Engl. J. Med. 338, 323–325

(1998).8. Franklin, R. Nat. Rev. Neurosci. 3, 705–714 (2002).9. Lu, Q.R. et al. Neuron 25, 317–329 (2000).10. Zhou, Q. et al. Neuron 31, 791–807 (2001).11. Zhou, Q. & Anderson, D. Cell 109, 61–73 (2002).12. Lu, Q.R. et al. Cell 109, 75–86 (2002).13. Olson, E. et al. Cell 85, 1–4 (1996).14. Pham, C. et al. Proc. Natl. Acad. Sci. USA 93,

13090–13095 (1996).15. Doerflinger, N. et al. Genesis 35, 63–72 (2003).

Zooming in on cortical mapsDavid Fitzpatrick

In a technical tour de force, Okhi et al. image the activity of thousands of visual cortical neurons in vivo at a single-cell resolution, and examine their orientation and direction selectivity. Their results show that cortical maps can be built with single-cell precision.

The author is at the Department of Neurobiology,

Duke University Medical Center, Durham,

North Carolina 27710, USA.


Neurons in the visual cortex are grouped together into radial columns that share similar response properties. In turn, when viewed in the plane of the cortical surface, cortical col-umns are arranged in an orderly fashion form-ing functional maps that represent stimulus features such as edge orientation, direction of motion, and position in space. Much has been learned about the basic topological features of cortical maps, but the finest details of map organization—the arrangement of the indi-vidual neurons whose activity forms the basis for the maps—have remained frustratingly elu-sive. A new report from Okhi et al.1 provides a stunning view of the cellular architecture of cortical maps using two-photon imaging and

demonstrates a surprising degree of precision in the way that cortical maps are constructed.

Current views of cortical maps are derived largely from optical imaging techniques that monitor small changes in intrinsic signals or in the fluorescence of voltage-sensitive dyes as measures of neuronal activity. Orientation or direction preference maps are constructed by determining the optimal stimulus for each point on the cortical surface, and the maps are dominated by regions where the preferred stimulus parameter changes progressively and at a relatively slow rate. This pattern is punc-tuated by small regions where the preferred stimulus changes abruptly, producing dis-continuities in an otherwise smooth map2,3 (Fig. 1). Discontinuities in orientation prefer-ence maps generally take the form of points called singularities or pinwheel centers; discon-tinuities in direction preference maps generally take the form of lines called fractures4,5.

To date, optical imaging techniques have been limited to measures that reflect net changes in the activity of neuronal cell bodies and processes that occupy a given volume of cortical tissue. As a result, it has been unclear whether the smooth and continuous appear-ance of functional maps accurately portrays the layout of individual neurons. Much like a pointillist painting seen at a distance, opti-cal maps may seem smooth and continuous as a result of spatial averaging; in contrast, maps that are generated at a resolution suf-ficient to depict the properties of individual cortical neurons could be much less orderly. The problem of spatial averaging is especially acute for the discontinuities that are activated equivalently by a broad range of stimuli. This pattern is consistent with a variety of single cell arrangements: individual neurons in the discontinuities could (i) respond to all stim-uli (that is, the neurons are broadly tuned);

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



Figure 1 Visualizing the cellular architecture of direction maps. (a) Typical map of direction preference in visual cortex based on optical imaging of intrinsic signals. At this scale, direction preference (depicted with pseudocolors and arrows) is mapped in a continuous fashion, interrupted by occasional fractures (dark regions) where direction preference changes abruptly, often by 180° (dashed white line). However, the properties of individual neurons that contribute to the map cannot be resolved. (b) Direction preference maps viewed with cellular-level resolution show remarkable precision. Ohki et al.1 combined calcium imaging with two-photon microscopy to resolve the direction preferences of individual neurons in a cortical region that included a direction fracture (boxed region in a). Individual cells are color-coded according to their direction preference (see a); the degree of saturation is proportional to the magnitude of direction selectivity. (c,d) Quantification of direction-selective responses of individual cortical neurons as a function of distance from the center of a direction fracture. Panels modified from ref. 1; reproduced with permission from the author. The magnitude of direction selectivity declines in the region of the fracture (c); nevertheless, continuity prevails such that the border between neurons with opposite direction preference (d) is razor sharp.

(ii) be narrowly tuned but arranged in a ran-dom fashion; or (iii) be narrowly tuned and precisely arranged, but with a precision that exceeds imaging resolution. In principle, finely spaced microelectrode recordings or tetrode recordings can help in addressing this problem by sampling the activity of individual corti-cal neurons6,7, but the nature of the approach makes it virtually impossible to reconstruct the properties of even a small percentage of the neurons whose activity contributes to a func-tional map. Indeed, the pioneers of functional architecture in visual cortex, frustrated by the difficulty in using a one-dimensional weapon to attack this three-dimensional problem, lik-ened it to trying to “cut the back lawn with a pair of nail scissors”8.

Thanks to two-photon imaging and cal-cium-sensitive dyes, it is now possible to sample from each blade of grass, and doing so reveals a level of precision in the architecture of the cortex that few would have imagined. Okhi et al. approached this problem by using calcium-sensitive fluorescent dyes to measure the activity of individual neurons in visual cortex while the animal viewed various visual stimuli. To detect stimulus-induced changes in fluorescence, they used two-photon imaging, a technique that makes it possible to obtain a sharp image of the cell bodies of neurons in a narrow focal plane at distances up to 400 µm below the cortical surface without altering cortical function. By presenting a range of stimulus conditions, they were able to derive tuning functions for large populations of single neurons and probe the organization of cortical architecture at a scale never before achieved.

Most of their analysis focused on area 18 of the cat, where functional maps for orientation and direction have been thoroughly explored with optical imaging techniques3,5. Overall, a high percentage of neurons were found to be responsive to visual stimuli (in some cases 100%) and virtually all of these (97%) were tuned to the orientation and/or direction of motion of the stimulus. Moreover, the tuning preferences of adjacent neurons were similar, creating a smooth and continuous progression of preference values across the 300 µm diameter of the imaging field. Even more remarkable was the appearance of single-cell maps in regions where tuning preference changes abruptly: for example, in fractures that lie between popula-tions of neurons that prefer opposite directions of motion (Fig. 1). The magnitude of the direc-tion-selective responses of individual neurons grows progressively weaker within the 30–50 µm width of the fracture. Nonetheless, continuity prevails, such that weakly biased cells show the same direction preference as their neighbors. As a result, the boundary between cells that have

opposite direction preferences is razor sharp. The authors provide several examples where a straight line is sufficient to cleave the active neurons into two populations with preferences for opposite directions of motion. Along this line, adjacent neurons show biases for opposite directions of motion. Thus, the basic features of cortical maps that have been discerned at a coarse scale with optical imaging techniques are evident at the finest scale in the spatial arrange-ment of individual neurons.

Future studies will no doubt take advantage of the power of this approach to resolve other controversies of map organization. For exam-ple, the orientation tuning properties of indi-

vidual neurons in regions of pinwheel centers remain to be explored at this fine scale9, as does the mapping of visual space and its relation to the mapping of orientation preference10,11. Likewise, combining this approach with cellu-lar markers for different classes of cortical neu-rons defined by morphology, neurotransmitter or channel expression phenotype should yield a wealth of new information about the micro-architecture of cortical circuits that has been impossible to obtain with other techniques.

As much as it enlightens, however, the study by Okhi et al. also exposes some of the more profound gaps in our understanding of corti-cal structure-function relationships. How is

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



such single-cell map precision instantiated in the spatial arrangement of neural connec-tions? For orientation tuning, at least, several lines of evidence suggest that an orientation-specific bias in the spatial arrangement of feed-forward axon arbors plays a significant role12,13. But the scale of precision in axonal arrays—that is, the degree of bias in the axon arbors that converge at a given point in the cortex—remains unclear. Moreover, it is not yet known how the broad and overlapping dendritic processes of cortical neurons sample from this array, as well as those supplied by other axonal populations, and how they do so in a fashion that yields such precision.

At a more fundamental level, the function served by cortical maps remains a puzzle—one that is deepened further by Okhi and col-leagues’ observations on the microarchitec-ture of orientation tuning in rat visual cortex. Despite the presence of neurons that are well tuned for orientation and/or direction of motion in rat visual cortex, there is no sign of the orderly arrangements that are so apparent in cat visual cortex: the orientation preferences of nearby cells are uncorrelated. The lack of

maps in the rat is consistent with results from other studies in rodents and indicates that the orderly mapping of stimulus features is not a prerequisite for constructing tuned cortical responses14. As the authors suggest, orderly mapping might contribute to differences in tuning sharpness, but this seems a modest gain for such a major difference in cortical architec-ture. Perhaps the significance lies less in gener-ating the mapped properties and more in how additional dimensions of the visual scene are encoded in the responses of individual neu-rons. Although they respond similarly to grat-ing stimuli, adjacent neurons in cat area 18 are likely to differ in their response to a variety of other properties of the stimulus (for example, phase, contrast or temporal properties). Such diversity of functionally distinct cell types may be significantly reduced in rodent visual cor-tex, obviating advantages for connectivity or pooling that clustering of neurons according to similar properties may endow7,15.

Whatever the reason for such species dif-ferences, new approaches for simultaneously visualizing the properties of large numbers of individual neurons in functioning circuits

promise to add appreciably to our understand-ing of the complexities of cortical architecture. Exploring how the world is mapped in the brain just got a lot more exciting.

1. Ohki, K., Chung, S., Ch’ng, Y.H., Kara, P. & Reid, R.C. Nature 433, 597–603 (2005).

2. Blasdel, G.G. & Salama, G. Nature 321, 579–585 (1986).

3. Bonhoeffer, T. & Grinvald, A. Nature 353, 429–431 (1991).

4. Weliky, M., Bosking, W.H. & Fitzpatrick, D. Nature 379, 725–728 (1996).

5. Shmuel, A. & Grinvald, A. J. Neurosci. 16, 6945–6964 (1996).

6. Swindale, N.V., Matsubara, J.A. & Cynader, M.S. J. Neurosci. 7, 1414–1427 (1987).

7. DeAngelis, G.C., Ghose, G.M., Ohzawa, I. & Freeman, R.D. J. Neurosci. 19, 4046–4064 (1999).

8. Hubel, D.H. & Wiesel, T.N. Proc. R. Soc. Lond. B 198, 1–59 (1977).

9. Maldonado, P.E., Godecke, I., Gray, C.M. & Bonhoeffer, T. Science 276, 1551–1555 (1997).

10. Das, A. & Gilbert, C.D. Nature 387, 594–598 (1997).11. Bosking, W.H., Crowley, J.C. & Fitzpatrick, D. Nat.

Neurosci. 5, 874–882 (2002).12. Chapman, B., Zahs, K.R. & Stryker, M.P. J. Neurosci.

11, 1347–1358 (1991).13. Mooser, F., Bosking, W.H. & Fitzpatrick, D. Nat.

Neurosci. 7, 872–879 (2004).14. Van Hooser, S.D., Heimel, J.A., Chung, S., Nelson,

S.B. & Toth, L.J. J. Neurosci. 25, 19–28 (2005).15. Koulakov, A.A. & Chklovskii, D.B. Neuron 29, 519–

527 (2001).

Disentangling simple from complex cellsA long-standing view of receptive field mechanisms in primary visual cortex (V1) neurons has been challenged recently; a paper in this issue now provides strong evidence to suggest that the textbooks have it right. Hubel and Wiesel were the first to describe neurons in cat V1 with elongated receptive fields, which respond to stimuli such as bars in a particular orientation. In 1962, they also described classes of neurons with distinctive receptive field structures. Simple-cell receptive fields are made up of nonoverlapping subregions; either bright or dark bars in the appropriate location excite a particular cell. For complex cells, these subregions are overlapping. Hubel and Wiesel hypothesized that simple-cell receptive fields are constructed from systematically arranged inputs from the lateral geniculate nucleus of the thalamus, whereas complex-cell receptive fields are constructed by combining inputs from multiple simple cells.

More recent work, however, has led to the suggestion that two distinct classes of neural circuitry are not necessary to account for the different properties of simple and complex cells. Such views have been supported by evidence that whereas simple and complex cells seem to form two classes when evaluated by their spiking behavior, the synaptic input responses of both cell types form a unimodal distribution. It has been proposed that spike threshold differences could lead to this apparent separation of the cell types by their spiking behavior (N. Priebe et al., Nat. Neurosci. 7, 1113–1122, 2004). Now, Martinez and colleagues (p. 372) combine intracellular recording (to characterize the spatial distribution and interaction of responses to bright and dark stimuli; red and blue regions in the figure) with anatomy (to locate neurons within different cortical layers). They found that simple cells were exclusively located in layer 4 and upper layer 6, the region where thalamic inputs enter the cortex. In contrast, complex cells were found in all layers, with cells in different layers having different receptive field structures. These results support the original proposal of Hubel and Wiesel that simple and complex cells belong to distinct neural circuits and that simple cells are an exclusive feature of an earlier stage of visual processing.

I-han Chou

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


Constant light desynchronizes mammalian clock neuronsHidenobu Ohta, Shin Yamazaki & Douglas G McMahon

Circadian organization can be disrupted by constant light, resulting in behavioral arrhythmicity or ‘splitting’ of rhythms of activity and rest. By imaging molecular rhythms of individual clock neurons in explanted mouse clock nuclei, we now find that constant light desynchronizes clock neurons but does not compromise their ability to generate circadian rhythms. Cellular synchrony within clock nuclei is disrupted during arrhythmicity, whereas neurons in the left and right clock nuclei cycle in antiphase during ‘splitting.’

Neural circadian clocks are composed of populations of circadian pace-maker neurons organized to drive coherent daily rhythms in behavior and physiology. An unresolved question is how external stimuli that disrupt behavioral and physiological rhythmicity (such as shift work and jet lag in humans) affect the function and organization of central neural pacemakers. Exposure to constant light (LL) disrupts overt rhythms and induces circadian arrhythmicity in mammals1 and other species. Two general hypotheses have been advanced to explain this phenomenon: light either stops the cell-autonomous molecular oscillations that gener-ate circadian rhythms, or it desynchronizes the ongoing rhythms of the individual oscillator neurons that make up neural circadian clocks. We have now successfully tested these hypotheses in the mammalian brain circadian clock and have examined the cellular basis for the ‘splitting’ of behavioral rhythms into two bouts of activity and rest per 24-h interval as occurs in some individuals exposed to LL1.

For this study we used Per1:GFP transgenic mice in which rhyth-mic activation of the Period1 (Per1) clock gene promoter can be monitored at the single cell level in real time using time-lapse confocal microscopy2. Mice were exposed to LL while their locomotor activity rhythms were assayed by wheel running (see Supplementary Methods online for details). Individual mice in LL either (i) became behavior-ally arrhythmic (ii) remained rhythmic with lengthened free run-ning period or (iii) showed split locomotor activity rhythms with two bouts of activity per 24-h interval1. To study the organization of the biological clock in the behaviorally arrhythmic state, we selected mice that had been rendered behaviorally arrhythmic but that continued to show robust overall levels of activity (Fig. 1a, n = 5). We then analyzed the Per1 gene transcription dynamics of their hypothalamic suprachiasmatic nuclei (SCN), which make up the circadian clock for locomotor activity3.

SCN of arrhythmic mice were dissected and explanted into organ-otypic culture and their Per1 promoter–driven GFP fluorescence rhythms recorded during the initial 4 d in vitro. SCN Per1:GFP signals from behaviorally arrhythmic mice showed significant levels of Per1 promoter activity, as indicated by fluorescence intensities well above background, but SCN circadian rhythmicity was severely disrupted (Fig. 1b). The weak oscillations we observed in SCN from arrhythmic mice showed trough-to-peak amplitudes of only 1.09-fold on average (that is, peaks only 9% greater than troughs, ±0.02% s.e.m.), whereas the prominent SCN Per1:GFP rhythms from mice maintained on nor-mal light-dark (LD) cycles are typically 2- to 3-fold in amplitude2.

In contrast, imaging of individual cellular rhythms within SCN from behaviorally arrhythmic mice revealed the persistence of robust neuronal circadian rhythms in Per1 promoter activation that were 1.5- to 3-fold in amplitude, similar to the amplitudes of cellular Per1:GFP rhythms recorded in SCN from LD mice2. The SCN

Department of Biological Sciences, Vanderbilt University, VU Station B, Box 35-1634, Nashville, Tennessee 37235-1634, USA. Correspondence should be addressed to D.G.M. ([email protected]).

Published online 30 January 2005; doi:10.1038/nn1395

Figure 1 Behavioral and SCN rhythms from an arrhythmic constant light–treated mouse. (a) Actogram of wheel running activity. Black marks indicate wheel revolutions. Note loss of temporal organization in latter portion of record. (b) Time-lapse SCN Per1:GFP fluorescence signals for 3.5 d in vitro. (c) Individual SCN neuronal Per1:GFP rhythms from SCN in b. Four representative cells are plotted for clarity (colored lines). (d) Peak time histograms of individual neuronal rhythms. Peak times of neurons in the right SCN are plotted with black bars, whereas those in the left SCN are plotted with open bars. Histograms for this and the following figures are for hours 12–36 in vitro. n = 193 cells. Animal care and use was reviewed and approved by the Vanderbilt University IACUC.

B R I E F COM M U N I C AT I O N S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e



neuronal rhythms of arrhythmic LL mice were, however, notably desynchronized in phase (Fig. 1c). Peak times of individual neuronal rhythms from arrhythmic mice were widely dispersed (Fig. 1d) with a mean standard deviation of 6.1 ± 0.5 h (mean ± s.e.m., throughout) (n = 10 nuclei). Similar robust but desynchronized neuronal rhythms were observed in the SCN of all arrhythmic mice assayed (n = 843 cells, 10 nuclei, 5 mice).

To test whether SCN cellular desynchronization in LL was specific to behaviorally arrhythmic mice, we also assayed SCN molecular rhythms of mice in which significant behavioral rhythmicity persisted in our LL conditions (Fig. 2a, n = 5). SCN from rhythmic LL mice showed clear circadian rhythms in gene expression averaging 2.57-fold (± 0.69) in amplitude (Fig. 2b), similar in amplitude to SCN rhythms from LD entrained mice2. In addition, individual neurons within the SCN of rhythmic LL mice also showed robust Per1:GFP rhythms, but with a much greater degree of synchrony than those of arrhythmic LL mice (Fig. 2c). Peak time histograms of individual cell rhythms from rhyth-mic LL mice SCN showed the majority of cells synchronized in a peak centered near the projected onset time of the locomotor rhythm, with a few cell rhythms in antiphase to the main cluster (Fig. 2d). The stan-dard deviation of cell peak times was significantly less in SCN from LL rhythmic mice than in SCN from arrhythmic mice (3.6 ± 1.0 h, n = 10 nuclei, P < 0.01), indicating greater phase coherence. Similar results were observed in SCN of all rhythmic LL mice (n = 881 cells, 10 nuclei, 5 mice).

Whereas constant light desynchronized the neuronal populations within the SCN of arrhythmic animals, it did not significantly affect the generation and expression of circadian rhythmicity by individual neurons. A high proportion of imaged neurons in SCN from arrhyth-mic mice were rhythmic (735/843, 87%), similar to LL rhythmic mice (783/881, 89%) and LD mice (89%, ref. 2). In addition, the calculated free running periods of individual neurons in SCN from arrhyth-mic LL mice and from rhythmic LL mice were similar (23.0 h versus 23.2 h, P > 0.05), as were their amplitudes and their standard devia-tions in period (±0.5 h versus ±0.5 h, n = 735 versus 783 neurons). Thus, constant light primarily affects phase organization among SCN neurons, not their properties as individual circadian oscillators.

Some of the mice that remained rhythmic in LL showed ‘split’ activity rhythms (Fig. 3a). In SCN from behaviorally split mice (n = 5), indi-vidual SCN neuronal rhythms were in synchrony within each SCN nucleus, but the cells in the left and right SCN oscillated in approximate antiphase, peaking about 12 h apart, similar to the activity bouts (Fig. 3b–d). Notably, each nucleus within the paired SCN structure showed coherent 24-h rhythmicity, and individual cell rhythms were never ‘split’, showing only a single peak of Per1-driven GFP fluorescence during each 24-h interval. The standard deviation of cell peak times in

individual SCN nuclei from LL ‘split’ animals was 3.2 ± 1.5 h (n = 10 nuclei), similar to that of LL rhythmic animals and significantly less than that of LL arrhythmic animals (P < 0.01), again indicating a greater phase coherence within the individual nuclei of ‘split’ rhythmic mice compared to nuclei of arrhythmic mice. The calculated free running period of cellular rhythms in SCN from ‘split’ animals was also similar to that of the arrhythmic and rhythmic groups (22.7 ± 0.6 h, P > 0.05), as was the overall proportion of rhythmic neurons (936/1,033 neurons, 91%). These data demonstrating ongoing ‘split’ in vitro rhythms from SCN of animals that showed ‘split’ locomotor activity rhythms indicate that our imaging protocol faithfully preserves behaviorally relevant aspects of SCN function during the transition from in vivo to in vitro conditions. In addition, they reveal that the basis for the ‘split’ circadian organization is intercellular antiphasic synchrony between neurons of the paired SCN nuclei, as has been inferred from static gene expression assays of ‘split’ hamster SCN4, rather than intra-SCN or intracellular splitting. The differences in neuronal synchrony in SCN from arrhythmic, rhythmic and split mice are illustrated in time-lapse videos (see Supplementary Videos 1–3 online and high-resolution videos at http://vvrc.vanderbilt.edu/NatNeurosci2005OhtaYamazakiMcMahon).

Our findings resolve a basic question regarding the mammalian brain biological clock: at what level of organization is rhythmicity disrupted by external stimuli? Clearly, constant light disrupts cir-cadian behavioral rhythms by disrupting the cellular organization of the SCN clock. The asynchronous but robust individual cellular rhythms in the SCN from arrhythmic LL mice indicate that dis-ruption of behavioral and SCN tissue-level rhythmicity5–7 is not the result of stopping the core molecular clock mechanism of indi-vidual neuronal oscillators. Circadian rhythm generation by mam-malian biological clock neurons apparently persists at the cellular and molecular levels even as behavioral rhythmicity is blunted or reorganized by constant light; however, normal temporal organiza-tion of cellular rhythms within the clock nuclei is lost under these conditions. Coherent organization of neuronal population rhythms within the SCN is critical for driving robust circadian locomotor rhythms, and each nucleus of the SCN pair is evidently capable of independently driving a component of locomotor behavior. Loss of cellular synchrony is also a mechanism for damping circadian molecular oscillations in peripheral tissue circadian oscillators. At least some peripheral circadian oscillators can show self-sustained individual cell rhythms but lack coupling mechanisms that main-tain tissue-level temporal organization (for example, fibroblast cell rhythms8,9). Thus, the SCN is distinguished from peripheral tis-sue oscillators by its ability to sustain phase coherence among its constituent neuronal oscillators through strong coupling interactions under normal circumstances. This ability is critical for the role of the

Figure 2 Behavioral and SCN rhythms from a rhythmic constant light–treated mouse. (a) Actogram of wheel running activity from a mouse that remained rhythmic in LL. (b) Time-lapse SCN Per1:GFP fluorescence signals. (c) Individual SCN neuronal Per1:GFP rhythms from SCN in b. (d) Peak time histograms of individual neuronal rhythms. n = 154 cells.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



SCN as a master pacemaker for orchestrating normal behavioral and physiological rhythmicity. The human circadian system can be sub-ject to internal desynchronization in constant-light environments10 that might be experienced in hospital intensive care units, during shift work or during prolonged space travel, as on planned Mars missions. Our results show the potential for mammalian biological clocks to remain functional during exposure to such environments and for chronotherapy to focus on issues of synchronization rather than rhythm generation.

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

ACKNOWLEDGMENTSThe authors thank T. Page and C. Johnson for thoughtful comments and discussion of the manuscript. Supported by National Institutes of Health grant R01 MH63341 to D.G.M.

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

Received 17 December 2004; accepted 11 January 2005Published online at http://www.nature.com/natureneuroscience/

1. Pittendrigh, C.S. & Daan, S. J. Comp. Physiol. A 106, 333–355 (1976).2. Quintero, J.E., Kuhlman, S.J. & McMahon, D.G. J. Neurosci. 23, 8070–8076 (2003).3. Reppert, S.M. & Weaver, D.R. Annu. Rev. Physiol. 63, 647–676 (2001).4. Mason, R. Neurosci. Lett. 123, 160–163 (1991).5. Zlomanczuk, P., Margraf, R.R. & Lynch, G.R. Brain Res. 559, 94–99 (1991).6. Granados-Fuentes, D., Prolo, L.M., Abraham, U. & Herzog, E.D. J. Neurosci. 24, 615–

619 (2004).7. de la Iglesia, H.O., Meyer, J., Carpino, A. Jr. & Schwartz, W.J. Science 290, 799–801

(2000).8. Nagoshi, E. et al. Cell 119, 693–705 (2004).9. Welsh, D. et al. Curr. Biol. 14, 2289–2295 (2004).10. Aschoff, J. & Wever, R.A. Fed. Proc. 35, 2326–2332 (1976).

Figure 3 Behavioral and SCN rhythms from a ‘split’ rhythmic constant light–treated mouse. (a) Actogram of wheel running activity from a mouse that showed split rhythms in LL. (b,c) Individual SCN neuronal Per1:GFP rhythms in the left (b) and right (c) SCN. (d) Peak time histograms of individual neuronal rhythms. n = 204 cells.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



Restoration of spatial working memory by genetic rescue of GluR-A–deficient miceW B Schmitt1, R Sprengel2, V Mack2, R W Draft2, P H Seeburg2, R M J Deacon1, J N P Rawlins1 & D M Bannerman1

Gene-targeted mice lacking the AMPA receptor subunit GluR-A (also called GluR1 encoded by the gene Gria1,) have deficits in hippocampal CA3–CA1 long-term potentiation (LTP) and have profoundly impaired hippocampus-dependent spatial working memory (SWM) tasks, although their spatial reference memory remains normal. Here we show that forebrain-localized expression of GFP-tagged GluR-A subunits in GluR-A–deficient mice rescues SWM, paralleling its rescue of CA3–CA1 LTP. This provides powerful new evidence linking hippocampal GluR-A–dependent synaptic plasticity to rapid, flexible memory processing.

Induction of hippocampal LTP requires NMDA receptor activation. Its expression depends in part on recruiting to activated synapses AMPA receptors that contain GluR-A subunits1,2. Gene-targeted mice lack-ing GluR-A AMPA receptor subunits do not show tetanus-induced, rapid-onset hippocampal CA3–CA1 LTP3 but can show a form of LTP

after theta-burst stimulation4. In parallel, GluR-A–deficient mice are impaired in hippocampus-dependent SWM tasks but show normal hippocampus-dependent spatial reference memory (SRM) perfor-mance5,6. Together, these results suggest that GluR-A–independent neuronal mechanisms within the hippocampus can support gradual acquisition of a fixed-location, hidden-platform water maze task5 or discrimination of initially baited arms from never-baited arms on a radial maze6. Nonetheless, GluR-A–dependent hippocampal synaptic plasticity seems necessary for rapid, flexible, trial-specific memory7–9 such as keeping track of which initially baited arms have been chosen within a given trial.

We tested this by transferring a GFP-tagged GluR-A expression sys-tem into GluR-A–deficient mice. This produces a mosaic pattern of subunit expression across the CA1 cell population such that LTP can be induced in CA1 neurons expressing GFP–GluR-A, but not in GluR-A–deficient cells10. This results in partial recovery of tetanus-induced field LTP. We assessed whether this could restore SWM performance.

Constitutively rescued GluR-A–deficient mice expressing GFP–GluR-A (n = 10) were compared with GluR-A–deficient knockout mice (n = 10) and wild-type controls (n = 10) on a radial maze task. The same three out of six arms were always baited, but milk rewards were not replaced within trials6, allowing assessment of SWM and SRM. The animals collected rewards from the ends of the maze arms guided by distal extramaze cues. As rewards were not replaced, the animal had to adopt a ‘win-shift’ strategy and remember which arms it had already visited on that trial. This provided a test of SWM. By only baiting three

1Department of Experimental Psychology, University of Oxford, South Parks Road, Oxford, OX1 3UD, UK. 2Max-Planck-Institute of Medical Research, Department of Molecular Neurobiology, D-69120 Heidelberg, Jahnstrasse 29, Germany. Correspondence should be addressed to D.M.B. ([email protected])

Published online 20 February 2005; doi:10.1038/nn1412

Figure 1 GFP-tagged GluR-A expression rescues spatial working memory performance in GluR-A–deficient mice. (a) Acquisition phase: reference memory (RM) errors (all s.e.m. < 0.18) for wild-type (WT; open circles), GluR-A–deficient knockout (KO; filled circles) and GluR-A–deficient mice expressing GFP–GluR-A (Rescue mice; filled triangles) during 6 blocks of training on a 3 out of 6 RM radial maze task. (b) Test phase: working memory (WM) errors (± s.e.m., otherwise all s.e.m. < 0.49) during 8 blocks of testing on a 3 out of 6 RM and WM radial maze task. (c) Test phase: reference memory errors (all s.e.m. < 0.12) during 8 blocks of testing on a 3 out of 6 RM and WM radial maze task. (d) Rewarded alternation: mean percentage of correct responses (± s.e.m.) for wild-type (WT), GluR-A–deficient knockout (KO) mice and GluR-A–deficient mice expressing GFP–GluR-A (Rescue) for 60 trials of spatial non-matching to place testing on the T-maze. All experiments were conducted under the auspices of the UK Home Office Project and personal licenses held by the authors (UK) and the Regierungspräsidium Karlsruhe.

30

40

50

60

70

80

90

100

Perc

en

t a

lte

rna

tio

n

WT KO Rescue

d

0

2

4

6

8

10

Err

ors

1 2 3 4 5 6 7 8

Block

Rescue

KOWT

b

0

1

2

3

Err

ors

1 2 3 4 5 6 7 8

Block

RescueKOWT

c

3

0

1

2

Err

ors

1 2 3 4 5 6

Block

RescueKOWT

a©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



of the six arms, but always baiting the same three arms, we were able to assess SRM simultaneously.

Mice were first trained solely on the reference memory (RM) com-ponent of the task. Doors prevented the mice from reentering an arm that they had already visited on that trial, precluding working memory (WM) errors. RM errors were scored when a mouse entered an arm that had never been baited. All mice learned the RM task at the same rate (main effect of block, F5,135 = 125.0, P < 0.0001; main effect of group and groups by blocks interaction, both F < 1; Fig. 1a), confirming that GluR-A–deficient mice show normal SRM acquisition3,5,6.

The WM component was then introduced. All doors were open for every choice, so it was now possible to re-enter previously visited arms. WM errors were scored if a mouse revisited an initially baited arm within a trial. Wild-type mice made virtually no WM errors. In con-trast, GluR-A–deficient mice made numerous WM errors (main effect F2,27 = 1,588.2; P < 0.0001; Duncan’s P < 0.01; Fig. 1b). The GluR-A–deficient mice rescued by GFP–GluR-A performed at a level intermedi-ate between the wild-type and knockout animals (Duncan’s P < 0.01 for both comparisons). RM errors remained very low in all three groups (Fig. 1c), although wild-type mice showed a slightly increased RM error rate during blocks 3 and 4 (groups × blocks interaction, F14,189 = 2.0; P < 0.05; simple main effects P < 0.01). The intact RM performance indicates that GluR-A deletion induces a specific WM deficit rather than a non-mnemonic impairment of general maze behavior.

Additional SWM testing on a discrete trial, rewarded alternation task (spatial non-matching-to-position on an elevated T-maze11,12) showed excellent alternation in wild-type but chance performance in GluR-A–deficient mice (main effect of group, F2,27 = 83.8; P < 0.0001; Duncan’s, P < 0.01; Fig. 1d). The GluR-A–deficient mice rescued by GFP–GluR-A again showed intermediate performance (Duncan’s P < 0.01 versus both wild type and knockout).

Expression of the bidirectionally transgenic-encoded GFP–GluR-A subunit and nucleus-localized β-galactosidase (β-gal) was controlled by a regionally specific CaMKII promoter10, limiting the GluR-A res-cue to forebrain areas such as the hippocampus (Fig. 2). Expression

was especially high in dorsal CA1 (shown for β-gal), which has been specifically implicated in spatial memory13,14. In contrast, constitutive GluR-A–deletion affects the entire brain3, not only impairing spatial working memory but also inducing a range of behavioral pheno-types including hyperactivity, a subtle motor coordination deficit, changes in aspects of emotionality, and disruption of some species-typical behaviors15. It is likely, therefore, that some of these behav-ioral phenotypes result from deletion of GluR-A subunits from brain areas outside the hippocampus. Those should be unaffected by more region-specific rescue. We therefore also tested emotionality (hypo-neophagia), motor behavior (motor activity and coordination) and species-typical behaviors (nesting)15. On each test, GluR-A–defi-cient mice that had and had not been rescued by GFP–GluR-A were equally impaired relative to wild-type mice (Supplementary Table 1 online). There were no significant differences between the two geneti-cally modified groups. The rescue was specific to the hippocampal-dependent SWM tasks.

These results demonstrate genetic rescue of a high-level cognitive ability and provide a new link between hippocampal GluR-A–dependent LTP and SWM, a rapid, flexible form of memory7–9. The partial behavioral rescue suggests that the mosaic of rescued and dysfunctional neurons may act like a neural network that shows graceful degradation after partial dele-tion of key connections. The rescue was specific to the SWM impairment. This is consistent with the hypothesis that the learning impairment in the knockout mice results from GluR-A deletion in the dorsal hippocampus and that emotional or non-cognitive effects of the knockout are unlikely to contribute to the memory deficit.


ACKNOWLEDGMENTSThis work was supported by the Wellcome Trust (65298 and 074385), by a European Union Framework V grant (QLG3-CT-1999-01022) awarded to P.H.S., R.S. and J.N.P.R. and by Deutsche Forschungsgemeinschaft grants to R.S. (SP602/1 and SFB633/4).


Received 29 September 2004; accepted 24 January 2005Published online at http://www.nature.com/natureneuroscience/

Figure 2 Immunohistochemical comparison of β-gal expression in dorsal and ventral hippocampal CA1 and CA3 of GluR-A–deficient mice expressing GFP–GluR-A. The expression of GFP–GluR-A and β-gal is coregulated. The nucleus-localized β-gal was used as reporter for the presence of GFP–GluR-A in hippocampal pyramidal neurons10. (a) Schematic sagittal views show the position of horizontal cuts for dorsal (left) and ventral (right) hippocampus (dorsal: bregma –2.8 through –3.3 mm; ventral: –4.0 through –5.0 mm). (b) Confocal images of dorsal (left) and ventral (right) horizontal slices from a GluR-A–deficient mouse expressing GFP–GluR-A stained with primary anti–β-gal (rabbit, 1:8,000; ICN) and secondary FITC-labeled antibody (goat, 1:200; Jackson ImmunoResearch; green) and counterstained with propidium iodide (1 µM, Sigma) to label nuclei from all cells (red). CB, cerebellum; OB, olfactory bulb; DG, dentate gyrus; Cx, cortex. Boxed regions represent a single panel in the hippocampal pyramidal cell layer where pyramidal cells were counted. Six or seven such panels were evaluated per slice. (c) Dorsal CA1 and CA3, and ventral CA1 and CA3 regions, in higher magnifications. Images with the sharpest and most compact pyramidal cell layering were selected. For quantitative analysis, brains of five GluR-A–deficient mice expressing GFP–GluR-A (age P44–P47) were perfused with paraformaldehyde, sliced horizontally and histochemically stained (see above). CA1 and CA3 images were recorded within a single optical plane using confocal microscopy for five slices per animal at various depths. In total, about 9,000 cells expressing the αCaMKII-driven nucleus-localized β-gal were counted. The number of β-gal–positive cells was slightly variable between slices of different animals (Supplementary Fig. 1 online). The average percentage of CA1 and CA3 pyramidal cells rescued with GFP–GluR-A and expressing nucleus-localized β-gal is given on the top of panels in c.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



1. Bliss, T.V. & Collingridge, G.L. Nature 361, 31–39 (1993).2. Malinow, R. & Malenka, R.C. Annu. Rev. Neurosci. 25, 103–126 (2002).3. Zamanillo, D. et al. Science 284, 1805–1811 (1999).4. Hoffman, D., Sprengel, R. & Sakmann, B. Proc. Natl. Acad. Sci. USA 99, 7740–7745

(2002).5. Reisel, D. et al. Nat. Neurosci. 5, 868–873 (2002).6. Schmitt, W.B., Deacon, R.M.J., Seeburg, P.H., Rawlins, J.N.P. & Bannerman, D.M.

J. Neurosci. 23, 3953–3959 (2003).7. Eichenbaum, H. & Fortin, N. Curr. Dir. Psychol. Sci. 12, 53–57 (2003).8. Morris, R.G.M. et al. Phil. Trans. R. Soc. Lond. B. 358,773–786 (2003).

9. Nakazawa, K., McHugh, T.J., Wilson, M.A. & Tonegawa, S. Nat. Rev. Neurosci. 5, 361–372 (2004).

10. Mack, V. et al. Science 292, 2501–2504 (2001).11. Deacon, R.M.J., Bannerman, D.M., Kirby, B.P., Croucher, A. & Rawlins, J.N.P. Behav.

Brain Res. 133, 57–68 (2002).12. Rawlins, J.N. & Olton, D.S. Behav. Brain Res. 5, 331–358 (1982).13. Bannerman, D.M. et al. Behav. Neurosci. 113, 1170–1188 (1999).14. Moser, M.-B., Moser, E.I., Forrest, E., Andersen, P. & Morris, R.G.M. Proc. Natl. Acad.

Sci. USA 92, 9697–9701 (1995).15. Bannerman, D.M. et al. Behav. Neurosci. 118, 643–647 (2004).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



Preserved spatial memory after hippocampal lesions: effects of extensive experience in a complex environmentGordon Winocur1–3, Morris Moscovitch2–4, Stuart Fogel3, R Shayna Rosenbaum1 & Melanie Sekeres3

Damage to the hippocampus typically impairs spatial learning and memory in animals, but humans with hippocampal lesions retain spatial memories of premorbidly familiar environments. We showed that, like humans, normal rats reared in a complex environment and then given hippocampal lesions retained allocentric spatial memory for that environment. These results, which ruled out dependency on single cues, landmarks or specific routes, suggest that extensive premorbid experience leads to spatial representations that are independent of the hippocampus.

Converging evidence from lesion, electrophysiological and neuroimaging studies implicates the hippocampus in spatial memory and navigation1–4. Less is known about the effects of hippocampal damage on premorbidly acquired spatial memories. In animals, hippocampal lesions typically lead to memory loss, extending as far back as 9 months preoperatively4,5. In contrast, patients with bilateral hippocampal lesions retain accurate allocentric spatial representations of familiar environments6,7.

A crucial difference between the studies is that the humans had extensive experience with the environment before hippocampal dam-age, whereas the animals did not. To model the human condition, we reared rats in a complex environment or ‘village’ and studied the effects of hippocampal lesions on allocentric spatial memory (Fig. 1a).

The village (1.2 × 1.2 × 1.2 m) was located in the center of a room with standard laboratory furniture (for example, desks and book shelves) and pictures on the walls. The room was dimly and uniformly illuminated by overhead lighting. The village contained two levels, with interconnected walkways within and between the levels. Two walkways leading to the lower levels were situated across from the entrance to the reward compartments in the northeast and northwest corners. The walls and ceiling were made of wire mesh, and the walkways of aluminum sheet metal. The upper level, also constructed of sheet metal, consisted of a gathering area in the middle of the upper level with four walls each containing a central opening. This area served as a start box for training and test trials. A compartment containing food (southeast corner), water (northwest corner), an assortment of toys (northeast corner) or a female rat (southwest corner) was attached to each of four

corners on the lower level (Fig. 1a). The compartment containing the female rat was separated from the village by a wire mesh screen, whereas the other compartments could be accessed freely.

Preoperative training consisted of five daily trials in which rats were placed individually in the start area and allowed to find the reward compartment. On each trial, the rat was forced to enter the village through a different doorway, thereby ensuring the use of different routes to the reward. The floors and walkways of the village were divided into imaginary zones demarcated by intersections, from which the rat could move closer to the reward compartment or farther from it. A rat was considered to have made an error when it arrived at a choice point and turned away from the reward compartment (Supplementary Methods , online). After each training session, the rats were returned to individual cages where they had access to food or water for 1 h.

Three-month-old male Long-Evans rats were assigned to Rearing + Training (RT) and Training-Only (TO) conditions. For 3 months, rats in the RT condition spent 8 h per day in groups of five or six in the village during their high-activity cycle. The rats were allowed to explore the entire village, with access to various reward sites, including food and water, which were always available at the same locations. After each session, the rats were returned to individual cages, where they were deprived of food and water. TO rats were on the same deprivation schedule, but they remained in individual cages except for brief daily periods during which they were handled and given food and water.

After this three-month period, RT and TO rats were placed on a 23-h food or water deprivation schedule. Preoperative training to find the appropriate reward compartment began 1 week later. All rewards were available in the compartments where they had been found during rear-ing. Rats required 9.86 d (RT) and 9.21 d (TO) to reach a criterion of 80% errorless trials over 2 d, t < 1. There were no differences between rats motivated to find food or water on these or subsequent tests, so this variable was collapsed into a single reward condition.

Within 48 h of reaching criterion, rats received bilateral NMDA hippocampal lesions or a surgical control procedure. At this time, sur-gery was also done on a third group (Postoperative testing–Only (PO)) with no previous exposure to the village. Fifty-one rats survived and met inclusion criteria. Twenty-seven rats had an acceptable hippocampal lesion, at least 50% damage to dorsal and ventral regions (Fig. 2), and completed the study. There was no relationship between extent of lesion and performance (Supplementary Results, online). Seven days postop-eratively, rats were deprived of food or water, and their ability to find the reward was assessed. Testing procedures were the same as in preoperative training in the RT and TO conditions, except that they lasted 10 d.

Rats with hippocampal lesions in the RT condition found the appro-priate reward compartment efficiently from the beginning, and made no more errors than controls (Fig. 1b). This was confirmed by a significant Lesion × Rearing interaction (F2,53 = 3.17, P = 0.050). Hippocampal and

1Rotman Research Institute, Baycrest Centre for Geriatric Care, Toronto, Ontario M6A 2E1, Canada. 2Department of Psychology, University of Toronto, Toronto, Ontario M5S 1A1, Canada. 3Department of Psychology, Trent University, Peterborough, Ontario, K9J 7B8, Canada. Correspondence should be addressed to G.W. ([email protected]).


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



control RT animals did not differ significantly (F < 1), but there were sig-nificant group differences in TO (F1,53 = 5.00, P = 0.030) and PO (F1,53 = 13.22, P < 0.001) animals. Rats with more experience in the environment, including those with lesions, performed well initially; those with less experience started poorly and then improved (Rearing × Days interaction, F2,477 = 7.92, P < 0.001).

To ensure that rats were relying on allocentric spatial cues, we did experiments on a subset of hippocampal and control rats from the RT condition (Supplementary Results). We first moved the village to a new room with entirely different cues and tested the animals as before. The reward compartments remained in the same relationship to the village and to each other. The room change led to severe impairment in the hippocampal group (F1,6 = 19.10, P = 0.005), indicating that the rats had used distal cues rather than local village cues to navigate in the familiar environment (Fig. 3a). Hippocampal and control groups were similarly affected by the change, as shown by Day 1, Trial 1 responses. Both groups showed substantial increases in average errors (hippocam-pal, 7; control, 5), over the last test in the original environment. Only one animal in each group selected the correct reward compartment on the first trial, indicating that, initially, both groups responded randomly in the new environment. The overall Group × Days interaction was not significant (F9,54 = 1.51, P = 0.17), but by the seventh day of testing, the lesioned rats had improved substantially and performed as well as controls. Both groups were impaired after room change, relative to the original test, but only the control group showed some savings compared with their performance in the preoperative training condition.

If successful postoperative navigation depended on configural, allocentric representations, then some of the environment could be altered without affecting performance. Thus, we returned the rats to the initial environment, but removed most of the distal cues (for example, pictures and chairs) and introduced new ones, while retain-ing elements of the original array in the same relationship to each

other (Supplementary Methods). All rats with hippocampal lesions continued to perform well (Fig. 3b; F < 1), indicating that perfor-mance in the original condition was attributable to a minimal con-figuration of cues, including room geometry, and their relationships to the reward compartments.

To determine whether the rats could learn new spatial relationships between the village and the familiar, external environment, we rotated the village and the attached compartments by 180° relative to the dis-tal cues and tested the rats as before. Whereas controls learned quickly, a significant Group × Days interaction (F9,54 = 2.07, P = 0.048) indi-cated that the hippocampal group was severely impaired in locating the correct compartment (Fig. 3c). This deficit was related to a strong tendency in lesioned, but not control, rats to continue visiting the com-partment’s original location, especially at the beginning of each test trial (Supplementary Results). Thus, both groups retained considerable spa-tial memory of the familiar environment, but rats with hippocampal lesions could not assimilate the changed location of the compartment within their retained knowledge. This deficit was also apparent when we compared rats’ preoperative performance with their scores on the village rotation (as well as the room-change test). Although both groups’ perfor-mance declined postoperatively, only the control group showed savings relative to preoperative training (Supplementary Results).

Two additional experiments ruled out the possibility that rats relied on sensory cues within the village or olfactory cues outside the village. Rotating the floor of the village by 180°, while holding constant the location of the compartments, had no effect on performance (F < 1), confirming that neither lesioned nor control rats depended on sensory cues within the village (Supplementary Results, online). Rendering rats anosmic with zinc sulfate nasal treatment did not affect either group’s ability to locate the reward compartment (Fig. 3c; all t values < 1.0), indicating that navigation was not guided by olfactory cues (Supplementary Methods and Supplementary Results).

Figure 1 The village and postoperative test results for the first series of experiments. (a) The village apparatus. (b) Performance of hippocampal (HPC) and control groups in the three rearing conditions, tested in the familiar environment. Error bars indicate s.e.m. RT: Rearing + Training (HPC, n = 8; control, n = 6). TO, Training-Only (HPC, n = 10; control, n = 9). PO, Postoperative testing–Only (HPC, n = 9; Control, n = 9). This study was approved by the Trent University Animal Care Committee, and the rats were cared for in accordance with the ethical standards of that committee.

Figure 2 Maximal and minimal extents of hippocampal lesions.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



Efficient performance of the hippocampal rats in the RT condition could be related to preservation of spatial learning abilities, rather than spatial memories. We ruled this out by finding that hippocampal, RT rats were severely impaired when tested in a new room on the radial arm maze, a test of spatial learning and memory8 (Supplementary Methods; F1,6 = 25.75, P = 0.002, Supplementary Results).

The results provide strong evidence that a map-like or allocentric spa-tial representation of a complex environment, gained through extensive preoperative experience, can survive hippocampal damage and support navigation. Because some hippocampal tissue was spared, we cannot rule out the possibility that the residual tissue mediated the spared perfor-mance. We consider this highly unlikely because there was no correlation between performance and lesion size, and no difference in performance between rats with more than 80% of the hippocampus removed and those with smaller lesions in any of the conditions (Supplementary Results). These results show a notable parallel between humans and rats with hippocampal lesions in the retention of spatial memories of a familiar environment, and point to similar underlying mechanisms.

There was a noteworthy finding in the room-change and village-rotation conditions, where the hippocampal group, despite their ini-tial impairment, eventually improved to the level of controls. These findings are analogous to patterns observed in the amnesic patients H.M. and K.C., both of whom have extensive hippocampal damage but eventually learned to navigate in new environments as they became increasingly familiar7,9. Future work will determine whether the repre-sentations acquired with difficulty after hippocampal lesions resemble those acquired before the hippocampus was damaged.

The distinction between premorbid spatial memories acquired through extensive experience and newly acquired spatial memories has received scant attention in theories of hippocampal function. Insights into this issue come from recent considerations of anterograde and retrograde amnesia after hippocampal lesions in humans and rats4. In humans, old semantic memories (such as general knowledge con-cerning facts, events and public figures) are retained normally after hippocampal lesions, whereas postmorbid semantic memories are acquired in adulthood with difficulty9–11. By analogy, the preserved spatial memories may also be considered ‘semantic’ in the sense that they represent the gist or core of spatial knowledge without the extra-neous details that accompany its acquisition. The core of a well-estab-lished spatial memory, therefore, may consist of the bare topographical elements of the environment such as the geometry of a room or a small array of specific landmarks: a schematic cognitive map that can support navigation and that is represented in extrahippocampal struc-tures (for example, the superior-medial parietal, posterior cingulate, retrosplenial, frontal and parahippocampal cortices, the latter some-times bordering on the hippocampus12,13). This indeed is what our

experiments showed: changing details had little effect on performance, whereas altering the topography entirely led to severe impairment.

It has yet to be determined how experience alters both the nature of the spatial representation and their neural substrates. One possibility is that spatial memories are laid down independently in the hippo-campus and in extrahippocampal structures, with the hippocampus reinforcing the weaker connections until this structure is no longer needed6,10. Another alternative is that the allocentric representations in the hippocampus may be different in nature from those in extra-hippocampal structures4. Whereas hippocampal representations may code information about details of the environment in an allocentric framework and details of the events that occurred there3, extrahippo-campal representations may simply abstract coarse information that is sufficient to support navigation but not sufficient to support a rich re-experiencing of the environment (in humans) or detailed recogni-tion (in rats)7. The mechanisms needed to transform spatial memories from detailed to schematic representations have yet to be elucidated, though conscious recollection or rehearsal combined with offline reac-tivation of hippocampal-cortical networks may be involved14,15.


ACKNOWLEDGMENTSThis work was supported by National Science and Engineering Research Council of Canada grants to G.W. and M.M. R.S.R. is supported by a Heart and Stroke Foundation/Canadian Institutes of Health Research postdoctoral fellowship. We thank H. Eichenbaum, P. Frankland and L. Nadel for their comments.


Received 6 October 2004; accepted 24 January 2005Published online at http://www.nature.com/natureneuroscience/

1. O’Keefe, J. & Nadel, L. The Hippocampus as a Cognitive Map (Oxford Univ. Press, Clarendon, 1978).

2. Muller, R. Neuron 17, 813–822 (1996).3. Burgess, N., Maguire, E.A. & O’Keefe, J. Neuron 35, 625–641 (2002).4. Rosenbaum, R.S., Winocur, G. & Moscovitch, M. Behav. Brain Res. 127, 183–197

(2001).5. Clark, R.E., Broadbent, N.J. & Squire, L.R. Hippocampus (in the press).6. Teng, E. & Squire, L.R. Nature 400, 675–677 (1999).7. Rosenbaum, R.S. et al. Nat. Neurosci. 3, 1044–1048 (2000).8. Olton, D.S., Becker, J.T. & Handelman, G.H. Behav. Brain Sci. 2, 313 (1979).9. Corkin, S. Nat. Rev. Neurosci. 3, 153–160 (2002).10. Manns, J.R., Hopkins, R.O. & Squire, L.R. Neuron 38, 127–133 (2003).11. Moscovitch, M. et al. in Dynamic Cognitive Processes (eds. Ohta, N., MacLeod, C.M.,

& Uttl, B.) 333–380 (Springer-Verlag, Tokyo, 2005).12. Rosenbaum, R.S., Ziegler, M., Winocur, G., Grady, C.L. & Moscovitch, M. Hippocampus

14, 826–835 (2004).13. Maviel, T., Durkin, T.P., Menzaghi, F. & Bontempi, B. Science 305, 96–99 (2004).14. Wilson, M.A. & McNaughton, B.L. Science 265, 676–679 (1994).15. Nadel, L. & Moscovitch, M. Curr. Opin. Neurobiol. 7, 217–227 (1997).

Figure 3 Test results for second series of experiments. (a–c) Performance of RT hippocampal (HPC; n = 4) and control (n = 4) groups, tested in different environmental conditions.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

534 VOLUME 8 | NUMBER 4 | APRIL 2005 NATURE NEUROSCIENCE

E R R AT U M

Erratum: Preserved spatial memory after hippocampal lesions: effects of extensive experience in a complex environment Winocur, G., Moscovitch, M., Fogel, S., Rosenbaum, R.S. & Sekeres, M. Nat. Neurosci. 8, 273–275 (2005).

In the published version of this article, the y-axis of the right panel in Figure 1b on page 274 was obscured by a fragment of another graphic. The corrected figure is reproduced below.

Figure 1 The village and postoperative test results for the first series of experiments. (a) The village apparatus. (b) Performance of hippocampal (HPC) and control groups in the three rearing conditions, tested in the familiar environment. Error bars indicate s.e.m. RT: Rearing + Training (HPC, n = 8; control, n = 6). TO, Training-Only (HPC, n = 10; control, n = 9). PO, Postoperative testing–Only (HPC, n = 9; Control, n = 9). This study was approved by the Trent University Animal Care Committee, and the rats were cared for in accordance with the ethical standards of that committee.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



Attending to local form while ignoring global aspects depends on handedness: evidence from TMSCarmel Mevorach1, Glyn W Humphreys1 & Lilach Shalev2

Our perceptions of the whole and of the parts of a visual stimulus are mediated by different brain regions. We used low-frequency transcranial magnetic stimulation (TMS) to show for the first time that opposite, homologous regions in the two hemispheres are involved in attending to local parts for left- and right-handed individuals. The brain regions that focus on the ‘trees’ while ignoring the ‘forest’ are switched as a function of handedness.

The ability to attend to and identify global (the ‘forest’) or local (the ‘trees’) aspects of a hierarchical object1 is lateralized between the two hemispheres in the human brain; the left hemisphere is biased toward the local level, whereas the right hemisphere is biased toward the global level2. Hence, patients with unilateral damage may have difficulty in identifying local or global information according to the site of lesion3. However, this assignment is not absolute, as damage to the right hemisphere can disrupt global4 as well as local5 identification.

No previous studies have assessed whether the brain regions that focus on parts and wholes differ in left- and right-handers, though handedness is known to influence language lateralization6 and some aspects of spatial perception7. Here we show differential lateralization of function for attention to local aspects of form in homologous areas in the right and left cerebral hemispheres.

Transcranial magnetic stimulation (TMS) is a technique in which transient disruption of normal brain activity is induced by the application of focal magnetic pulses to specific regions on the scalp8. We used TMS to assess the role of the left and right posterior parietal lobes (PPL; electrode sites P3 and P4) for right- (n = 11, mean age = 29.18) and left-handed participants (n = 11, mean age = 25.36; handedness was assessed using the Edinburgh Handedness Inventory) when they were required to identify global and local forms in two stimulus categories (letters and shapes) under conditions of focal attention (see Fig. 1).

Statistical analysis showed contrasting effects of TMS for right- and left-handed participants specifically when identifying local congruent displays as compared with incongruent displays (the global interference effect; see Fig. 2). This effect did not interact with stimulus category (letters or shapes). For right-handers, TMS over the left PPL increased global interference as compared with TMS over the right PPL (Fig. 2a,c). For left-handers, TMS over the right PPL resulted

in larger global-to-local interference than did TMS over the left PPL (Fig. 2b,c). There were no effects on the global identification task. Notably, TMS affected performance primarily when attentional con-trol was required in order to ignore an irrelevant level (on incongruent relative to congruent trials).

These findings fit with previous studies pointing to an attentional involvement in global and local identification and in the apparent lat-eralization of these processes in right-handed individuals4,9,10. More than this, though, our data show that opposite, homologous regions in the left and right hemispheres control attention to local form in left- and right-handed individuals. In right-handers, the left PPL controls

1Behavioural Brain Sciences Centre, School of Psychology, University of Birmingham, Edgbaston, Birmingham B15 2TT, UK. 2Department of Education and Psychology, The Open University Israel, 108 Ravutski St., Ra'anana, Israel. Correspondence should be addressed to C.M. ([email protected]).


Figure 1 Typical display sequence showing congruent and incongruent compound stimuli. A red compound figure was presented either to the left, center or right of fixation and was composed of either ‘H’ and ‘D’ (letter condition) or ‘crosses’ and ‘boxes’ (shape condition). On different blocks of trials, participants were asked to identify the global or the local elements while ignoring information on the other level. In half of the trials, the compound figures consisted of the same global and local elements (congruent trials), and in the other half, of different global and local elements (incongruent trials). For TMS, a 70-mm figure-eight coil connected to a MagStim Rapid stimulator (MagStim) was positioned over the left or right posterior parietal lobe (P3 or P4 on the 10–20 EEG coordinate system). TMS trains consisted of 600 1-Hz pulses with intensity set to 90% of the participants’ motor threshold (this can produce prolonged inhibition of the stimulated cortical site11 including P3 and P4; ref. 12). Each block consisted of 96 trials and was given six times to each participant (three blocks per category: letters or shapes). In a first session, a pre-TMS run was followed by one of the two TMS trains, which was immediately followed by another run of the task; in a second session there was only the TMS train followed by the task. The order of both the stimulus category and the TMS site was counterbalanced across participants. As participants initially performed the pre-TMS condition, differences between pre- and post-TMS could be masked by practice. Consequently our focus was on the two post-TMS conditions. The experiment was approved by the Ethics Committee of the School of Psychology, Birmingham University. Written informed consent was obtained from all participants.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



the ability to focus on local form and to ignore global information. In left-handers, the right PPL mediates this function. The brain regions controlling attention to local form are cross-lateralized in left- and right-handers. As performance was not affected by stimulus category, we con-clude that this difference could not be attributed to different language lateralization in the two groups. Rather, in left- and right-handers, the PPL in opposite hemispheres modulates global-to-local interference irrespective of the category to which the stimulus belongs.

ACKNOWLEDGMENTSThis work was supported by a grant from the UK Medical Research Council.



1. Navon, D. Cog. Psych. 9, 353–383 (1977).2. Fink, G. et al. Nature 382, 626–628 (1996).3. Delis, D.C., Robertson, L.C. & Efron, R. Neuropsychologia 24, 205–214 (1986).4. Robertson, L.C., Lamb, M. & Knight, R. J. Neurosci. 8, 3757–3769 (1988).5. Marshall, J.C. & Halligan, P.W. Nature 373, 521–523 (1995).6. Knecht, S. et al. Brain 123, 2512–2518 (2000).7. Laeng, B. & Peters, M. Neuropsychologia 33, 421–439 (1995).8. Pascual-Leone, A., Walsh, V. & Rothwell, J. Curr. Opin. Neurobiol. 10, 232–237

(2000).9. Fink, G. et al. Brain 120, 1779–1791 (1997).10. Shalev, L., Humphreys, G.W. & Mevorach, C. Cogn. Neuro. (in the press).11. Chen, R. et al. Neurology 48, 1398–1403 (1997).12. Hilgetag, C.C., Theorat, H. & Pascual-Leone, A. Nat. Neurosci. 4, 953–957 (2001).

Figure 2 Performance (mean reaction time (RT) ± s.e.m.) for global and local identification pre- and post-TMS trains across stimulus category. A mixed-design ANOVA with group (right- and left-handers) as a between-subject factor and stimulus category (shapes and letters), TMS (pre, P3 and P4), level (global and local), location (left, center and right) and congruency (congruent and incongruent) as within-subject factors showed a significant interaction of group, TMS, level and congruency (F1,41 = 3.74, P < 0.05). The magnitude of the congruency effect (RTs (incongruent) – RTs (congruent)) varied between the groups as a function of TMS site and level of task. No significant interactions involving stimulus category (and either group or TMS) were found. In addition, a main effect of TMS (F1,41 = 5.26, P < 0.05) indicated that performance in the pre-TMS condition was slower than performance in both post-TMS conditions (which may reflect a practice effect). (a) Mean RTs for right-handed participants. (b) Mean RTs for left-handed participants. (c) Increase or decrease in interference (± s.e.m.) after TMS. The change in the congruency effect following TMS trains was analyzed separately for each level of task. There was a significant interaction of group and TMS for local (F1,20 = 6.743, P < 0.05) but not for global identification (F < 1.0). Analysis of simple main effects of TMS on local identification showed that for right-handers, global-to-local interference increased after left (P3; 21 ms) compared to a decrease after right (P4; –7 ms) stimulation (t10 = 1.985, P < 0.05); the reverse pattern (a decrease in interference after P3 stimulation compared to an increase after P4 stimulation) was evident for left-handers (–6 ms and 7 ms for P3 and P4, respectively; t10 = 1.80, P = 0.052). Furthermore, simple main effects of group, analyzed for each TMS site separately, showed that left stimulation (P3) induced an increase in global-to-local interference for right-handers (21 ms), contrasting to a decrease for left-handers (–6 ms; t20 = 1.98, P < 0.05). A tendency for the reverse pattern (a decrease for right-handers compared to an increase for left-handers) followed TMS trains over the right PPL (P4; –6 ms and 7 ms for left- and right-handers, respectively; t20 = 1.3, P = 0.10).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


Cytoplasmic domain structures of Kir2.1 and Kir3.1 show sites for modulating gating and rectificationScott Pegan1, Christine Arrabit2, Wei Zhou1, Witek Kwiatkowski1, Anthony Collins3, Paul A Slesinger2 & Senyon Choe1

N- and C-terminal cytoplasmic domains of inwardly rectifying K (Kir) channels control the ion-permeation pathway through diverse interactions with small molecules and protein ligands in the cytoplasm. Two new crystal structures of the cytoplasmic domains of Kir2.1 (Kir2.1L) and the G protein–sensitive Kir3.1 (Kir3.1S) channels in the absence of PIP2 show the cytoplasmic ion-permeation pathways occluded by four cytoplasmic loops that form a girdle around the central pore (G-loop). Significant flexibility of the pore-facing G-loop of Kir2.1L and Kir3.1S suggests a possible role as a diffusion barrier between cytoplasmic and transmembrane pores. Consistent with this, mutations of the G-loop disrupted gating or inward rectification. Structural comparison shows a di-aspartate cluster on the distal end of the cytoplasmic pore of Kir2.1L that is important for modulating inward rectification. Taken together, these results suggest the cytoplasmic domains of Kir channels undergo structural changes to modulate gating and inward rectification.

The family of inwardly rectifying potassium (Kir) channels of eukaryotic cells is unique in that the channels conduct K+ ions better in the inward direction than in the outward direction. In native tissues, the small outward K+ current through Kir channels influences the resting membrane potential and membrane excitability. The major structural mechanism underlying inward rectification involves a physical occlusion of the pore by polyamines and Mg2+ from the cytoplasmic side of the channel1,2. In addition to being inwardly rectifying, Kir channels respond to a variety of intracellular messengers, including G proteins (Kir3 channels) and ATP (Kir6 channels), and respond to changes in pH (Kir1 channels)3. Aberrant activity of Kir channels has been linked to a variety of endocrine, cardiac and neurological disorders: for instance, the loss of Kir3 channels leads to hyperexcitability and seizures in the brain4, cardiac abnormalities5, hyperactivity and reduced anxiety. Mutations in Kir1 and Kir 2.1 channels have been implicated for Bartter syndrome6 and Andersen syndrome7, respectively. Therefore, elucidating the molec-ular mechanisms of inward rectification and gating will be instrumental to discovering means to treat these diseases.

The Kir family is composed of seven different subfamilies, Kir1–Kir7 (ref. 3). All Kir channels are tetrameric and contain two transmembrane helix domains (M1 and M2), the ion-selective P-loop between M1 and M2, and cytoplasmic N- and C-terminal domains. The extent of rectification varies among the different subfamilies, ranging from weak (Kir1) to strong (Kir2). The stronger rectification of Kir2 can be partly explained by charge negativity in the M2 domain (Kir2.1 Asp172)8 and a pair of acidic amino acids (Kir2.1 Glu224 and Glu299) in the cytoplasmic domain9,10. However, the degree of rectification among different Kir channels does not directly relate to the presence of these negative charges.

For example, Kir7.1 shows weak rectification but has the negative charge in the M2 domain11. Kir3.2 shows strong rectification but lacks the same negative charge12. There is also little consensus on the mechanism by which Mg2+ and polyamines produce inward rectification13. Together, these studies raise the possibility that other previously unknown sites may be involved in regulating inward rectification.

In addition to rectification, another common feature of Kir channels is regulation by the membrane phospholipid PIP2. Opening of Kir channels requires PIP2 binding to basic and polar amino acids in the cytoplasmic domains, whereas depletion of PIP2 seems to close the channel14–16. The high degree of sequence similarity in the cytoplasmic domains suggests there should be a common structural means for gating among Kir chan-nels, yet the cytoplasmic domains respond differently to intracellular regulatory signals. For example, Kir2.1 channels remain constitutively open through endogenous PIP2 binding, whereas Kir3 channels are opened by G proteins and Kir6 channels are closed by intracellular ATP, respectively14.

Recently, structures of two Kir channels, rat Kir3.1 (ref. 17) and bacterial KirBac1.1 (ref. 18), have been reported; they show a central water-filled ‘cytoplasmic pore’ that extends ∼60 Å coaxially from the transmembrane pore domain17. The structure of KirBac1.1, which contains both trans-membrane and cytoplasmic pore domains, also shows a long pore18. In the KirBac1.1 structure, a phenylalanine (Phe146) located within the pore-facing M2 helices is postulated to form a barrier to ion permeation, similar to the ‘helix bundle crossing’ in KcsA or in Kir3.4, to control the pore aperture. However, gating of eukaryotic Kir channels (Kir1, Kir3 and Kir6) requires additional elements19–21 such as ATP and Gβγ G pro-tein subunits binding to the channels' cytoplasmic domain. To better

1Structural Biology and 2Peptide Biology Laboratories, The Salk Institute, La Jolla, California 92037, USA. 3Department of Pharmaceutical Sciences, College of Pharmacy, Oregon State University, Corvallis, Oregon 97331, USA. Correspondence should be addressed to S.C. ([email protected]) or P.A.S. ([email protected]).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e


A R T I C L E S

understand mechanisms underlying inward rectification and gating, a structural comparison of different mammalian Kir channels is needed.

Here, we have determined the crystal structures of the cytoplasmic domains of mouse Kir2.1 and rat Kir3.1. Structural comparison has showed an intrinsically flexible cytoplasmic pore–facing loop. These loops constrict the cytoplasmic pore to ∼3 Å, forming a girdle around the central pore axis. This girdle, which we refer to as the G-loop, forms the narrowest portion of the cytoplasmic pathway, reminiscent of the narrow pore-forming girdle of the acetylcholine receptor channel22. Additionally, we have discovered that mutating two aspartate amino acids near the end of the cytoplasmic pore, opposite the G-loop, changes inward rectifica-tion. We propose a model in which the di-aspartate cluster and the cyto-plasmic G-loop act as cytoplasmic regulatory elements for rectification and gating, respectively.

RESULTSN- and C-terminal domains fused to form a tetramer in solutionBoth N- and C-terminal domains of Kir channels are involved in bind-ing PIP2 and G proteins12,14–16,23,24. We therefore first investigated the coexpression of N- and C-terminal domains of Kir3.2 or Kir3.4 chan-nels, which form homotetramers unlike heterotetrameric Kir3.112,25,26. Expression of the C-terminal domain alone produces insoluble protein. Complexes of N-terminal (residues 1–99 for Kir3.2 and Kir3.4) and C-terminal (Kir3.2198–414 and Kir3.4193–419) domains were soluble, how-ever, when coexpressed in a dicistronic vector. These non- covalent N- and

C-terminal domain complexes of both Kir3.2 and Kir3.4 eluted as dimers upon sizing chromatography (Supplementary Fig. 1). However, the com-plexes were relatively unstable over time, l eading to an unstable C-terminal domain protein with varying amounts of associated N-terminal domain.

To overcome the instability of the complex, we subsequently adopted the fusion strategy used successfully for the Kir3.1 domain structure17, the GluR extracellular domain27 and the KChIP1/Kv4.2 complex28. In contrast to the non-covalent complexes, Kir3.1S contains a smaller N-terminal segment (amino acids 41–63) that is fused directly to a truncated C-terminal domain (amino acids 189–371) of rat Kir3.1 (see Fig. 1; ref. 17). Kir2.1L, on the other hand, contains the N-terminal domain (residues 41–64) of mouse Kir2.1 fused to the entire C-terminal domain (residues 189–428), including 57 additional amino acids in the C-terminal end. Both Kir3.1S and Kir2.1L were expressed efficiently in bacteria, and they migrated as a stable tetramer in solution (Supplementary Fig. 1) and were successfully crystallized for structural analyses. Notably, fusion complexes of Kir3.2 or Kir3.4 were not successfully produced, unlike the aforementioned non-covalent complexes. It remains to be understood whether the dimeric form of the non-covalent complex reflects an alter-nate oligomeric state of the cytoplasmic domains resulting from the pres-ence of the entire N-terminal segment (such as residues 1–99).

Structures of Kir2.1L and Kir3.1S are similarWe determined the structures of Kir2.1L and Kir3.1S by molecular replacement methods (Table 1 and Fig. 2), using the published rKir3.1

Figure 1 Sequence alignment and size-exclusion chromatograms of Kir3.1S, Kir2.1L, and KirBac1.1 with secondary structure elements noted. Pore-facing residues within the cytoplasmic domains based on the structure of Kir2.1L are boxed in red. The selectivity filter residues GYG are highlighted in pink. Mutations known to be related to Andersen syndrome (blue) and six positions implicated for inward rectification (green) are highlighted.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

structure as a search model17. We found that Kir2.1L forms a non- crystallographic homo-tetramer with a long cytoplasmic pore in the space group C2221 (Fig. 2a). Kir3.1S forms a homotetrameric assembly in the space group P4212 (Fig. 2b). Despite different crystalli-zation conditions, its crystal-packing mode is identical to that for the published rKir3.1 crystals in the same space group17, which we refer to as Kir3.1S-Met. The core structures of Kir2.1L and Kir3.1S are readily superimposed (Fig. 2c). Briefly, the core structural elements of three β sheets and two α helices contain the first β sheet (βΙ, βΗ, βD and βE) and the sec-ond β sheet (βJ, βB, βC and βG), forming the protein interior of the monomer subunit. The third β sheet (βΑ, βM and βL) on the protein exterior faces the inner leaflet of the cell mem-brane with α helices (αA and αB). Notably, the marked similarity between Kir3.1S and Kir2.1L and the full-length KirBac1.1 indi-cates that the fusion strategy is a reasonable approach for studying these channels.

A short β strand, βA(N), of the N-terminal domain interacts extensively with the C-terminal domain of a neighboring subunit (Fig. 3a), indicative of its structural role in the subunit assembly by intersubunit inter-action. The additional 57 C-terminal amino acids of Kir2.1L were disordered in the struc-ture. This suggests that the C-terminal end is highly mobile and may require the full N-terminal domain or other cytoplasmic proteins to gain structural rigidity. All three proteins (Kir2.1L, Kir3.1S and Kir3.1S-Met) crystallized in the absence of PIP2; thus, they are most likely to reflect the PIP2-depleted closed conformation.

Structural flexibility in the cytoplasmic pore-facing G-loopsTwo pore-facing loop regions, located between the βC and βD strands (CD loop) and the βH and βI strands (HI loop), show a high degree of sequence conservation between Kir3.1 and Kir2.1 (red-lined boxes in Fig. 1a). These loops form the narrowest part of the cytoplasmic struc-tures, forming a girdle around the central pore axis. We refer to this region, in particular the HI loop, as the ‘G-loop’. Closer inspection showed that the G-loop in Kir2.1L (residues 300–315) deviates substantially from that in Kir3.1S (blue in Fig. 2d) and also from Kir3.1S-Met (gold in Fig. 2e), with a minor shift in the CD loop. A significant conformational difference in the pore-facing HI loop between Kir2.1L, Kir3.1S-Met and Kir3.1S indi-cates that these loops are intrinsically flexible, to possibly accommodate even larger conformational changes during ion conduction or activation with PIP2. It is noteworthy that these flexible loops are anchored by con-served and backbone-flexible glycines at positions 216 and 301 of Kir3.1, or 215 and 300 of Kir2.1 (Fig. 2e). An overlay of the structure of Kir2.1L tetramer on the C-terminal domain of the KirBac1.1 (Fig. 3) suggests that the pore-facing HI-equivalent loop of KirBac1.1 abuts the proposed path-blocking element (Phe146) of the M2 of KirBac1.1 (ref. 18).

Functional role for the G-loop in Kir2.1L and Kir3.1SThe G-loop contains several small or hydrophobic residues, including Gly300, Met301, Ala306 and Met307 in Kir2.1 (Fig. 4a). In Kir3.1S, I302 forms the narrowest region of the pore with an opening of 9.0 Å (the diagonal distance between closest atom centers). The G-loop of Kir2.1L

is narrower, with a width of 5.7 Å formed by Ala306. Given approxi-mately 2.8 Å for van der Waals radii of two opposing atoms, we estimate that the narrowest actual opening is ∼2.9 Å (5.7–2.8 Å) between the four Cβ atoms of Ala306 (Fig. 3a). Met301 (or Ile302 of Kir3.1) adjacent to Ala306 forms another hydrophobic ring near the G-loop that is com-parable to the opening of Kir3.1 of 6.2 Å diameter (9.0–2.8 Å). Thus, the physical opening formed by four opposing hydrophobic G-loops is too narrow to accommodate the passage of a hydrated potassium ion. Examination of the space-filled models for Kir2.1L and Kir3.1S viewed from the cytoplasmic side (Fig. 3b) makes this readily apparent: the water-filled pore is nearly completely occluded in Kir2.1L. The variation between Kir3.1S and Kir2.1L in the size of the opening suggests that the pore-facing G-loop is intrinsically flexible.

To assess whether the G-loop forms an important gating structure in native channels, we studied the effect of mutations in the G-loop of full-length Kir2.1 channels. Xenopus oocytes were injected with the cRNA for mutant channels, and macroscopic currents were then recorded in response to a series of voltage steps from –100 mV to +100 mV. To moni-tor K+ selectivity and possible changes in rectification, currents were recorded with either 2 or 95 mM extracellular KCl (Fig. 4). For wild-type Kir2.1, the zero current potential shifted from ∼0 mV in 95 mM KCl to approximately –100 mV in 2 mM KCl and there was little outward cur-rent at potentials positive to EK (Fig. 4b). The narrowest opening of the cytoplasmic pore is formed by Cβ atoms of Ala306 and to a lesser extent Glu299, Gly300, Met301 and Met307 (Fig. 4a). Ala306 is located at the apex of the G-loop. We hypothesized that a side chain larger than alanine would not permit the physical closure of the G-loop without changes in its backbone conformation and would lead to functional disruption. We found that only a glycine substitution (A306G) was well tolerated in Kir2.1 (Fig. 4a, Fig. 5). By contrast, glutamate, cysteine and threo-nine mutations did not produce functional channels (Fig. 4h). Similar

Table 1 Crystallographic data

Data collection

Protein / data set Kir2.1L Kir3.1S

Space group C2221 P4212

Cell constants a = 112.46 a = 76.36

b = 138.29 b = 76.37

c = 138.75 c = 92.16

α = β = γ = 90 α = β = γ = 90

Wavelength (Å) 1.033 1.080

Source ALS SSRL

Resolution (Å) 2.40 2.09

Total observations / total reflections 125,410 / 37,871 109,243 / 16,100

Completeness (highest-resolution shell) 91.7 96.0

I / σ (highest-resolution shell) 16.9 (6.0) 23.0 (2.8)

Rsyma 0.062 0.070

Model refinement

Total reflections (reflections for test) 38,814 (1,977) 15,270 (813)

Rwork(%) / Rfree (%)b 22.5 / 28.1 19.5 / 23.5

Protein atoms / water atoms 6,332 / 213 1,596 / 132

R.m.s. deviation of bond lengths (Å) 0.007 0.002

R.m.s. deviation of bond angles (˚) 1.3 1.2

aRsym = ΣhΣI |ΙI(h) − <Ι(h)>| / ΣhΣi ΙI(h), where ΙI(h) is the ith measurement and <Ι(h)> is the weighted mean of all measurements of I(h). bRwork and Rfree = h(|F(h)obs| – |F(h)calc|) / h|F(h)obs| for reflections in the working and test sets, respectively. R.m.s., root mean square.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

to A306G, the reverse substitution at Gly300 (G300A) is also tolerated and leads to functional channels, whereas the bulkier G300V does not give rise to appreciable current. These results suggest that the closure of the G-loop (Fig. 3b and Fig. 4a) reflects a physiologically relevant, closed conformational state during the gating of the channels.

We next examined the effect of mutations at Met301, which points to the cytoplasmic pore. Substituting alanine at Met301 did not affect functional expression of Kir2.1, although it did alter inward rectification (Fig. 4d). A positive charge introduced at Met301 (M301R), however, yields no functional channels. Like the M301A substitution, substituting alanine at Met307 also did not affect functional expression. Previously, it was found that coexpressing G300V, which is not functional on its own, with wild-type channels produced heteromeric channels with reduced PIP2 affinity, suggesting this region may be involved in gating by PIP2

15. As both A306G and G300A were expressed as functional homomers, we next investigated whether these mutants had altered affinity for PIP2. Intracellular application of polylysine closes Kir2.1 channels by antagonizing the interaction of endogenous PIP2 with the channel15. The time course of current rundown (t50) produced by intracellular polylysine reflects the appar-ent binding affinity of PIP2.. Whereas G300A showed a t50 that was ∼80% faster than wild type, the t50 for A306G was indistinguishable from that of the wild-type control (Fig. 5).

The finding that the affinity for PIP2 is unchanged in A306G indicates that both the open and closed (PIP2-depleted) states of the channel tolerate the subtle change from ala-nine to glycine without disruption of the flex-ibility in the G-loop. In contrast, the alanine substitution at Gly300 decreases PIP2 affinity (shorter t50). This affinity change is likely to be the result of an allosteric change in the PIP2 binding site15. This difference in PIP2 affin-ity with mutations at Gly300 and Ala306 is not entirely unexpected, as the positions of Gly300 and Ala306 in the structure are quite distinct. Gly300 is buried in the closed struc-ture, whereas Ala306 forms the narrow apex

ring around the central pore (see Figs. 3c and 4a). Thus, we conclude that the G-loop is a gating element that is structurally distinct from but functionally coupled to the PIP2-binding site. PIP2 binding most likely modulates allosterically the conformational switch of the flexible G-loop for gating.

Di-aspartate cluster is implicated for inward rectificationUsing both Kir3.1S and Kir2.1L structures, we next performed a side-by-side comparison of the surface charges in the cytoplasmic pore. The Kir2.1L structure shows a very high degree of electronegative surface potential compared to that of Kir3.1S (Fig. 3b). Notably, the KirBac1.1 cytoplasmic pore shows even less electronegative surface than Kir3.1S (Fig. 3B). Previously, the strong rectification of Kir2.1 has been attributed to two principal electronegative regions: Asp172 in the M2 domain8 and Glu224/Glu299 in the cytoplasmic domains9,10. Using the structure of Kir2.1L as a guide, we identified four new

Figure 2 Structures of Kir3.1S and Kir2.1L. (a) Standard view of Kir2.1L ribbon diagram with secondary structure labels consistent with those of Kir3.1S-Met

17. N-terminal domain has been removed for clarity. (b) Top-down (membrane-to-cytoplasm) view of Kir2.1L tetramer with four subunits labeled A (red), B (gold), C (green) and D (blue). Note that the N-terminal segment of the Kir2.1L interfaces with the C-terminal domain of the neighboring subunit (clockwise). (c) Stereo view of Kir2.1L Cα backbone (red) overlaid onto Kir3.1S (blue). (d) Stereo view of Kir2.1L Cα (red) and Kir3.1S (blue) near their G-loops. HI loop is labeled. (e) Stereo view of Kir3.1S Cα (blue) and Kir3.1S-Met (gold). Gold balls denote two methionine residues at positions 223 and 308 of Kir3. 1S-Met. Black balls highlight four anchoring residues of Kir3.1 (216 and 227 for CD, 301 and 310 for HI loops), including two flexible residues, Gly216 and Gly301. Views in d and e are very close to that in a but are slightly rotated vertically for clarity.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

pore-facing charged amino acids (Arg228, Asp255, Asp259 and Arg260), and examined the effect of charge reversal or neutralization at these residues (Fig. 6).

To quantify the differences in rectification among channels with different amplitude currents, we normalized the current measured at +100 mV to that measured at –100 mV (the ‘rectification ratio’). We also normalized the current at +30 mV for mutants that might show a change in voltage-dependent rectification, such as D172N. Thus, a value of unity at +100 mV or 0.4 at +30 mV indicates a complete loss of rectification (that is, linear I-V), whereas a value close to zero indicates strong rectification. As shown previously, E224G and E299S mutations reduced inward rectification of Kir2.1 (refs. 9,10). Charge reversal at Arg255 (D255R) or at Arg259 (D259R) also decreased inward rectification (Fig. 6d–6g). Notably, charge neutralization at Arg255 (D255A mutation) maintains strong inward rectification whereas that at Arg259 (D259A mutation) produces a large decrease in rectification. Thus, mutation of each of the four acidic amino acids in the cytoplasmic domain significantly reduces inward rectification (Fig. 6f). Mutation of the two pore-facing basic residues (R228A and R260A), on the other hand, produces little change in inward rectifica-tion (Fig. 6g). Taken together, these results suggest that a di-aspartate cluster (Arg255/Arg259) is also important in determining the extent of inward rectification of Kir2.1 channels.

DISCUSSIONComparison of Kir3.1 and Kir2.1 structuresComparison of two different types of cardiac inward rectifiers, Kir2.1 (IK1) and Kir3.1 (IKACh), has shown two new important gating features of Kir channels. First, a narrow girdle formed by G-loops poised near

the junction between cytoplasmic and transmembrane pore domains is very flexible, suggesting its conformational change could be a means to modulate gating. Second, a di-aspartate cluster in Kir2.1, located oppo-site from the G-loop, is important in controlling inward rectification.

Consistent with the conformational change envisioned for G-loops, cysteine-substituted accessibility studies on Kir2.1 using E224C, H226C and R228C mutants, all modified with multiple, large methane thiosul-fonate (MTS) reagents (12 × 10 × 6 Å), suggest the inner vestibule must be >20 Å in the open pore21. By contrast, the diagonal distance between Glu224 of Kir2.1L is 9.2 Å and must therefore widen to accommodate large MTS reagents. Second, intracellular MTS treatment modifies sites buried in the Kir2.1L, as indicated by analysis of G215C, N216C and V223C mutants, as well as sites at the top surface of the cytoplasmic pore (as indicated by analysis of L217C, S220C and H221C mutants)29. Third, FRET spectroscopy studies with Kir3 channels have demonstrated that binding of G proteins triggers a conformational change of the cyto-plasmic domain30. Finally, the Kir2.1L and Kir3.1S structures have been solved in the absence of PIP2 and are likely to reflect a closed state of the full-length channel. Taken together, these findings suggest that the cytoplasmic domain undergoes a conformational change in the presence of PIP2 or G proteins, thereby changing the accessibility of these amino acids and widening the pore.

Whether the additional 60 amino acids in the C-terminal domain of Kir2.1 also are involved in gating remains unclear. The observation that this segment of Kir2.1L was disordered in our crystals was unexpected; one possible explanation is that the C-terminal end of the channel is flex-ible because it lacks molecular association with cytoplasmic proteins. For

Figure 3 Kir3.1S and Kir2.1L in comparison to KirBac1.1. (a) Stereo overlay of Kir2.1L Cα trace (red) overlaid onto KirBac1.1 Cα trace (black). Side-chains of three key residues forming the closest points between M2 (Phe146 of KirBac1.1) and Met301 and Ala306 of Kir2.1L of the transmembrane and cytoplasmic pores, respectively, are highlighted in blue. (b) Electrostatic potential rendered on molecular surface of Kir2.1L (left), Kir3.1S (middle), and KirBac1.1 (right) viewed from the cytoplasmic side into the pore. Note highly positive (blue) cytoplasmic vestibule of KirBac1.1 in contrast to highly negative (red) vestibule of Kir2.1L. (c) Top-down (membrane-to-cytoplasm) view of molecular surface of Kir2.1L. Andersen syndrome mutations and putative PIP2-binding sites cluster on surface of Kir2.1L. In ‘A’ subunit, mutations implicated in changing PIP2 affinity are highlighted. A PIP2-binding site Arg228 is buried. In the ‘C’ subunit, amino acids known for Andersen syndrome are shown for those that are exposed to the surface (Arg189, Thr192, Arg218, Gly300, Val302, Glu303, Tyr315 of ∆314–315). Gly215 and Asn217 are buried in protein interface and invisible from this viewpoint. Also, disordered side-chain atoms of Arg189 of the ‘C’ subunit are not shown except for its Cα. Amino acids Gly300–Gln310 encompass the βH-βI or 'G-loop' (Met301 is not visible in this view). Gly215–Val227 comprise the βC-βD loop (Gly215, Asn216, Val223–Ala225, Val227 are not visible in this view).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

example, the C-terminus of Kir2.1 possesses a sequence motif binding to postsynaptic density proteins (PSD-95; ref. 31) with which structural rigidity may be achieved.

Role for electronegative charge in Kir2 cytoplasmic poreInward rectification of Kir channels depends on the binding of intracel-lular polyamines and Mg2+ to the cytoplasmic pore and to the inner vestibule of the channel. Previously, three negatively charged amino acids in Kir2.1 (Asp172, Glu224 and Glu299)8–10 were found to be important for the binding of polyamines/Mg2+ to the cytoplasmic pore and inner vestibule of the channel for inward rectification. Here, we find that a di-aspartate cluster (Arg255 and Arg259) is also involved in controlling inward rectification. This brings a total charge of –5 for the family of Kir2 channels (Table 2), which may account for the strong electronegativity observed in the Kir2.1L structure. Comparing the different types of Kir channels shows that the total number of acidic residues correlates well with the strength of inward rectification but cannot account completely for strong rectification. For example, Kir4.1 shows strong rectification but only possesses three of the five acidic residues32. The aspartate in the M2 domain might contribute more than the cytoplasmic domains to controlling the strength of rectifica-tion, although Kir7.1 contains a glutamate and does not show strong rectification11. Nonpolar interactions are also likely to be important for regulating rectification.

In the Kir2.1L structure, the di-aspartate cluster is located farther away from the pore cavity than is the Glu224-Glu299 pair (Fig. 6). Given the distance from the pore, what mechanistic role could the

di-aspartate cluster have in controlling inward rectification? Several different mechanisms for inward rectification have been proposed13. In general, negative charges can attract cations such as polyamines, Mg2+ and K+. Various studies10,13 suggest that the negative charge of Glu224/Glu299 serves as a docking site for polyamines and Mg2+ such that binding to these sites influences both the entry and exit rates of polyamines in the deep pore formed by M2 domain and selectivity filter. The strong voltage dependence on polyamine inhibi-tion arises from the interaction of amine group both with the Asp172 in the pore cavity and with the selectivity filter33. The presence of an additional pair of acidic residues in Kir2.1 could also serve to attract polyamines to the cytoplasmic pore. Consistent with this idea, mutation of Arg259 showed a slow component of inhibition, which could reflect a slow rate of polyamine entry into the deep pore. Recently, it was proposed that polyamines may regulate high- or low-affinity conformational states34. Binding of polyamines to the di-aspartate cluster could be also involved in producing this type of conformational change. Additional studies examining the affinity for polyamines and Mg2+ in these mutant channels should provide more insights into the mechanism of inward rectification for Kir channels.

Notably, KirBac1.1 contains only two (Glu229 and Glu300) of the five acidic residues implicated in inward rectification of eukaryotic Kir channels (Table 2). KirBac1.1 differs in two ways in regards to its cytoplasmic pore. First, the loop between βE and βF residing near the entryway of the cytoplasmic pore bends away from the central pore axis, creating a wider vestibule. Second, the vestibule of the KirBac1.1's cytoplasmic pore is positively charged overall. The less pronounced electronegative surface of KirBac1.1 predicts that KirBac1.1 may not rectify as strongly as its eukaryotic counterparts. Recent expression studies of purified KirBac1.1 demonstrate much less K+ selectivity than do Kir channels.

Structural insights into Andersen syndrome mutantsFourteen of eighteen amino acids that are mutated in Andersen syn-drome are found in the cytoplasmic domains of Kir2.1 (mutants known to be associated with Anderson syndrome are R67W, D71V/N, T75R, P186L, R189I, T192A, G215D, N216H, R218W/Q, G300V/D, V302M, E303K, R312C and ∆314–315). These Andersen syndrome–related mutations generally result in a loss of function via dominant negative interactions and heteromeric assembly7,35–37. Some decreased functions are directly attributed to a change in PIP2 sensitivity15. Of the 18 posi-tions affected by mutations, 10 are visualized in the Kir2.1L structure (Fig. 3c), with 8 located on the top surface of the cytoplasmic structure (Arg189, Thr192, Arg218, Gly300, Val302, Glu303, Arg312, ∆314–315), which may be near the PIP2-binding site, and the other two buried in the protein interface (Gly215, Asn216). Four locations, residues Gly300, Val302 and Glu303 and the site of the deletion mutant ∆314–315, are clustered as part of the G-loop region (Fig. 3c).

Figure 4 Mutations in G-loop disrupt gating and inward rectification. (a) Side (left) and top-down (right; membrane-to-cytoplasm) views of the Kir2.1L structure. Cβ atoms (Ala306) of the G-loop form the narrowest ∼3-Å opening of the pore and are shown as open circles for clarity. Other residues near the G-loop are labeled. (b–g) Examples of inwardly rectifying current for wild-type and mutant Kir2.1 channels. Macroscopic current traces show the response to voltage steps from –100 mV to +100 mV in an extracellular containing 95 mM K+. Dashed line indicates zero current. Current-voltage relation is shown to the right for the currents measured in 2 mM K+ or 95 mM extracellular K+. Asp172 is located in the M2 transmembrane domain. (h) Bar graphs shows the average current at –100 mV (6.8 ms after the voltage step) for each mutant (n = 9–51).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Inspection of the Kir2.1L structure showed that two particular sites of Andersen-syndrome mutations (Arg218 and Glu303) form favorable charged and polar interactions with Thr309 and Arg312, respectively. This suggested that second-site mutations might rescue the Andersen muta-tions. Both single-point mutants Kir2.1L(R218Q) and Kir2.1L(E303K) aggregated in aqueous buffer, but Kir2.1L(R218Q) was rescued by an additional change (T309R). The double mutant Kir2.1L(R218Q/T309R) formed a stable tetramer, whereas a single-point mutant Kir2.1L(T309R) aggregates, suggesting that the basis of the Andersen mutation of R218Q is the misfolded cytoplasmic domain (data not shown) that is distin-guished by its altered PIP2 affinity at other sites. Notably, Arg218 and

Thr309 reside within the CD and HI loops, respectively, thus provid-ing additional evidence that interloop interaction may be important for global conformational coupling of the G-loop.

Functional implications of flexible G-loop in gating controlThe movement in the G-loop observed in the different structures of the cytoplasmic domains is notable. From structural perspectives, the different backbone conformations of the G-loops between Kir2.1L and Kir3.1S are further highlighted by the contrast between Kir3.1S and Kir3.1S-Met

17. The narrow opening formed by the G-loop indicates that the G-loops should undergo a conformational change during gating by PIP2 binding, Gβγ binding, or both. Consistent with this data, muta-tions at Cys311 of Kir2.1 alter the kinetics of single-channel activity38, Kir3.1/Kir3.2-E315A heteromeric channels show abnormal gating39, and mutation of Glu224 near the G-loop alters single-channel conductance and gating9,40. G-loops may contribute to conductance and gating by triggering allosteric changes in the selectivity filter20 or PIP2 affinity38 of Kir channels. Other studies have also substantiated a functional role of the G-loop region. For example, regulation of Kir3 channels by G proteins’ Gα subunits involves the region near the G-loop (Glu311-Gly335 of Kir3.2)41,42, pH-dependent gating of Kir1.1 channels involves

Figure 6 Mutation of di-aspartate cluster in Kir2.1 changes inward rectification. Oocytes were injected with the cRNA for mutant Kir2.1 channels and whole-cell macroscopic currents recorded with two-electrode voltage-clamp. (a) Structure of Kir2.1L (three of four subunits are shown), highlighting amino acids lining permeation pathway. Bottom panel is a view from the axis of the permeation pathway. (b–e) Examples of inwardly rectifying current for wild-type and mutant Kir2.1 channels. Macroscopic current traces show the response to voltage steps from –100 mV to +100 mV in an extracellular solution containing 95 mM K+. Current-voltage relation is shown to the right for the currents measured in 2 mM K+ or 95 mM K+, measured 6.8 ms after the voltage step. Dashed line indicates zero current. (f) Normalized currents show change in rectification for the indicated Kir2.1 mutants. The ratio at +30 mV (gray bar) and +100 mV (black bar) indicate the rectification ratio used in g. (g) Comparison of the rectification ratios at +30 (gray) and +100 mV (black). A value of unity at +100 mV or 0.4 (dashed line) at +30 mV indicates a complete loss of rectification. R228A, D255A, R260A and M307A were not statistically different from Kir2.1 wild-type at +30 or at +100 mV (n = 9–51).

Figure 5 Substitution of glycine at Ala306 of Kir2.1 does not alter PIP2 affinity. Oocytes were injected with the cRNA for mutant Kir2.1 channels and macroscopic patch currents recorded with giant patch technique50. (a–c) Inward currents recorded through wild-type (a), G300A (b) and A306G (c) are shown before and during exposure to intracellular polylysine (0.3 mg ml–1). Note the decrease in current occurs with an instantaneous (non-PIP2: ref. 15) and a slow phase, which represents the removal of PIP2 from the channel. Intracellular solution (‘FVPP’) contained 96 mM KCl, 5 mM Na2EDTA, 10 mM HEPES, 5 mM KF, 3 mM NaO3V, 10 mM K4P2O7 at pH 7.4 (HCl). (d) Average t50 for wild-type, G300A and A306G channels (n = 5–7). ** indicates statistical difference between G300A and wild-type as determined using one-way ANOVA (P < 0.05).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

movement of the N terminus and the G-loop in the C terminal domain24, and MTS modification of endogenous cysteines within the G-loop of Kir3.1/Kir3.2 channels can completely abolish basal cur-rent43.

An emerging theme is that functional gates are located in different parts of the channel including the selectivity filter (P-loop) and trans-membrane helices19,44,45, and these can in principle operate on different timescales. In a broad sense, multiple gates can operate on at least two different timescales: a ‘fast’ gate in the selectivity filter and a ‘slow’ gate as a diffusion barrier46. The fast gate may involve a stochastic switch between different conformational states. The slow gate may involve a larger conformational change similar to that of the octameric gate ring of MthK47, the dimer-tetramer conversion of KTN domain48, the outwardly sliding helix of KirBac1.1 (ref. 18), the twisted S6 helices of Kv channels, or (in the case of Kir channels) the proposed flexible G-loops of the cytoplasmic pore.

METHODSMolecular biology. N-terminal domains (amino acids 1–99) of mouse Kir3.2 (ref. 12) and Kir3.4 (ref. 25) and C-terminal domains (amino acids 198–414 and 193–419, respectively) were expressed dicistronically using pHis8, a modi-fied pET28a vector (Novagen). For fusion protein constructs, cytoplasmic N-terminal and C-terminal domains of mouse Kir2.1 (ref. 49) and rat Kir3.1 (ref. 12) were linked directly in frame by PCR to clone into the pHis8 vector. BL21 (DE3) cells were used to express the proteins. Kir2.1L mutants were created using the QuikChange Site-Directed Mutagenesis Kit (Stratagene).

For protein purification, cells were grown at 37 °C to 0.6 OD induced by 0.4 mM IPTG, and lysed in 0.5 M NaCl, 5 mM Tris-HCl, pH 8.5, 10% glycerol, 7 mM β-mercaptoethanol (lysis buffer) and 1 mg lysozyme per 100 ml lysate. The supernatant from the lysate was loaded on a nickel-affinity column (Qiagen) and eluted with 200 mM imidazole. Thrombin-cleaved protein samples were separated by S200 and S75 Sepharose chromatography, and concentrated to 15 mg ml–1 Kir2.1L and 12 mg ml–1 Kir3.1S in the presence of 10 mM DTT for analyses.

Crystallography. Kir2.1L was crystallized in the space group C2221 at 4 °C with vapor diffusion hanging drops of 30% isopropanol, 0.1 M Tris-HCl, pH 8.5, and 0.2 M MgCl2. Crystals were treated with 10% glycerol before flash freezing. Kir3.1S crystallized in the space group P4212, the same as for Kir3.1S-Met

17, but under different conditions (15% ethanol, 0.1 M HEPES, pH 7.5, and 0.2 M MgCl2). Kir3.1S crystals were treated with 20% glycerol before flash freezing. Data were collected for Kir2.1L and Kir3.1S at Advanced Light Source

(ALS), Berkeley, and Stanford Linear Accelerator Center (SSRL), respectively. Molecular replacement solutions were derived using models of Kir3.1S-Met and Kir3.1S for Kir3.1S and Kir2.1L, respectively, by AMORE and MOLREP and refinement by Crystallography and NMR System (CNS).

Electrophysiology. Kir2.1 and its mutants were constructed in pBSK49 using PCR. In vitro, methyl-capped cRNA was made from linear cDNA and T3 or T7 RNA polymerase (Stratagene). The quality of cRNA was estimated using an ethidium-stained formaldehyde gel. Xenopus oocytes were isolated as described previously12. Oocytes were injected with a 46 nl Kir2.1 cRNA solution (0.5–5 ng) and incubated in ND96 (96 mM NaCl, 2 mM KCl, 1 mM CaCl2, 1 mM MgCl2, 5 mM HEPES, pH adjusted to 7.6 with NaOH) for 1–4 days at 16 °C. Macroscopic currents were recorded with a two-electrode volt-age-clamp amplifier as described previously (ref. 12). Mutants were studied in two or three dif-ferent batches of oocytes and produced similar changes in rectification. In one case, M301A con-sistently showed less rectification at +30 mV, but the extent of rectification was more variable at

+100 mV in different batches of oocytes.To study PIP2 affinity, Kir2.1 currents were recorded in ‘giant’ membrane

patches from Xenopus oocytes50. Patch pipettes had inner tip diameters of 10–20 µm. Inside-out patches were perfused through a manifold connected to solution reservoirs by polyethylene tubes and switching perfusion solutions con-trolled by a valve system (ALA Scientific). The electrodes, recording chamber and perfusion lines were filled with 'FVPP' solution: (in mM) 96 KCl, 5 Na2EDTA, 10 HEPES, 5 KF, 3 NaO3V, 10 K4P2O7, pH 7.4 (HCl)15. Polylysine (average molecular weight 7 kDa, 50 mg ml–1; P6403, Sigma) was added to the FVPP per-fusion solution to 0.3 mg ml–1. The patch-clamp amplifier was an Axopatch 200B (Axon Instruments). Currents were filtered at 1 kHz, and data acquired at 5 kHz with a Digidata 1320A computer interface and pClamp 8 (Axon Instruments) for analyses. Membrane voltage was held at 0 mV and stepped every 1 or 5 s to 50 mV for 400 ms and then to –50 mV for 400 ms. Current was averaged over the last 10 ms of the –50 mV step.

Data analysis. Macroscopic currents were elicited with voltage steps from –100 mV to +100 mV. The current amplitude was measured at the beginning (6.8 ms) or end (146.8 ms) of a 150-ms step pulse. The 6.8-ms time point was selected to ensure that the oocyte membrane capacitative current did not contribute to the current measurement and was used for calculating the mean current and rectification ratio. For quantifying change in PIP2 affinity, t50 was calculated by measuring the time taken for the current to decrease by one-half, after the instantaneous change in current produced with intracel-lular polylysine. The instantaneous decrease in current is not due to loss of PIP2 (ref. 15). All values are given as mean ± s.e.m. Statistical differences were determined using one-way ANOVA, followed by Bonferroni post hoc using Kir2.1 as control.


ACKNOWLEDGMENTSWe thank D. Clapham, M. Lazdunski and S. Hebert for GIRK4, GIRK2 and ROMK1 cDNAs, respectively. We also thank D. Kaiser for analytical ultracentrifugation, C. Park for mass spectroscopy, and the staff at ALS and SSRL for X-ray data collection. This work was supported by grants from the National Institutes of Health (P.A.S & S.C.) and the McKnight Endowment for Neuroscience (P.A.S). S.C. acknowledges the support from the American Heart Association.



Table 2 Summary of cytoplasmic pore mutations involved in rectification

M2 C-terminal domaina

Kir channel Rectification Charge 172 224 228 255 259 260 299 301 307

Kir2.1 **** –5 D E R D D R E M M

Kir2.2 **** –5 D E R D D R E M M

Kir2.3 **** –5 D E R D D R E M M

Kir4.1 **** –3 E G T D D S S T A

Kir3.1 *** –3 D S R S D Q E I M

Kir3.2 *** –3 N E R Y D R E M M

Kir3.3 N.D. –4 N E R D D R E M M

Kir3.4 ** –4 N E R D D R E M M

Kir1.1 * –3 N G Y D E N D T A

Kir6.2 * –1 N S H G N G E V I

Kir7.1 * –3 E S S D E C S M E

KirBac1.1 N.D. –2 I E K N H P E S Q

References 1 2 3 3 3 3 4 3 3

aShaded cells indicate amino acids in Kir2.1 involved in rectification. Number at top indicates amino acid position for Kir2.1. Qualitative difference in rectification among Kir channels ranging from weak (*) to strong (****). N.D. denotes ‘not determined’. Note the general correlation of strength of inward rectification with number of acidic amino acids at sites implicated in rectification, though there are exceptions (such as Kir7.1). References for mutations demonstrating change in rectification: 1: refs. 8,32; 2: ref. 9; 3: this study; 4: ref. 10.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

1. Lopatin, A.N., Makhina, E.N. & Nichols, C.G. Potassium channel block by cytoplas-mic polyamines as the mechanism of intrinsic rectification. Nature 372, 366–369 (1994).

2. Matsuda, H., Saigusa, A. & Irisawa, H. Ohmic conductance through the inwardly rectifying K channel and blocking by internal Mg2+. Nature 325, 156–159 (1987).

3. Nichols, C.G. & Lopatin, A.N. Inward rectifier potassium channels. Annu. Rev. Physiol. 59, 171–191 (1997).

4. Signorini, S., Liao, Y.J., Duncan, S.A., Jan, L.Y. & Stoffel, M. Normal cerebellar development but susceptibility to seizures in mice lacking G protein-coupled, inwardly rectifying K+ channel GIRK2. Proc. Natl. Acad. Sci. USA 94, 923–927 (1997).

5. Wickman, K., Nemec, J., Gendler, S.J. & Clapham, D.E. Abnormal heart rate regulation in GIRK4 knockout mice. Neuron 20, 103–114 (1998).

6. Derst, C. et al. Mutations in the ROMK gene in antenatal Bartter syndrome are asso-ciated with impaired K+ channel function. Biochem. Biophys. Res. Commun. 230, 641–645 (1997).

7. Plaster, N.M. et al. Mutations in Kir2.1 cause the developmental and episodic electri-cal phenotypes of Andersen’s syndrome. Cell 105, 511–519 (2001).

8. Lu, Z. & MacKinnon, R. Electrostatic tuning of Mg2+ affinity in an inward-rectifier K+ channel. Nature 371, 243–246 (1994).

9. Yang, J., Jan, Y.N. & Jan, L.Y. Control of rectification and permeation by residues in two distinct domains in an inward rectifier K+ channel. Neuron 14, 1047–1054 (1995).

10. Kubo, Y. & Murata, Y. Control of rectification and permeation by two distinct sites after the second transmembrane region in Kir2.1 K+ channel. J. Physiol. (Lond.) 531, 645–660 (2001).

11. Döring, F. et al. The epithelial inward rectifier channel Kir7.1 displays unusual K+ permeation properties. J. Neurosci. 18, 8625–8636 (1998).

12. Slesinger, P.A. et al. Functional effects of the mouse weaver mutation on G protein-gated inwardly rectifying K+ channels. Neuron 16, 321–331 (1996).

13. John, S.A., Xie, L-H. & Weiss, J.N. Mechanism of inward rectification in Kir channels. J. Gen. Physiol. 123, 623–625 (2004).

14. Huang, C-L., Feng, S. & Hilgemann, D.W. Direct activation of inward rectifier potassium channels by PIP2 and its stabilization by Gβγ. Nature 391, 803–806 (1998).

15. Lopes, C.M. et al. Alterations in conserved Kir channel-PIP2 interactions underlie channelopathies. Neuron 34, 933–944 (2002).

16. Shyng, S-L., Cukras, C.A., Harwood, J. & Nichols, C.G. Structural determinants of PIP2 regulation of inward rectifier KATP channels. J. Gen. Physiol. 116, 599–607 (2000).

17. Nishida, M., and MacKinnon, R. Structural basis of inward rectification. Cytoplasmic pore of the G protein-gated inward rectifier GIRK1 at 1.8 A resolution. Cell 111, 957–965 (2002).

18. Kuo, A. et al. Crystal structure of the potassium channel KirBac1.1 in the closed state. Science 300, 1922–1926 (2003).

19. Proks, P., Antcliff, J.F. & Ashcroft, F.M. The ligand-sensitive gate of a potassium channel lies close to the selectivity filter. EMBO Rep. 4, 70–75 (2003).

20. Lu, T. et al. Probing ion permeation and gating in a K+ channel with backbone muta-tions in the selectivity filter. Nat. Neurosci. 4, 239–246 (2001).

21. Lu, T., Zhu, Y.-G. & Yang, J. Cytoplasmic amino and carboxyl domains form a wide intracellular vestibule in an inwardly rectifying potassium channel. Proc. Natl. Acad. Sci. USA 96, 9926–9931 (1999).

22. Miyazawa, A., Fujiyoshi, Y. & Unwin, N. Structure and gating mechanism of the acetylcholine receptor pore. Nature 423, 949–955 (2003).

23. Huang, C.L., Slesinger, P.A., Casey, P.J., Jan, Y.N. & Jan, L.Y. Evidence that direct binding of Gβγ to the GIRK1 G protein-gated inwardly rectifying K+ channel is impor-tant for channel activation. Neuron 15, 1133–1143 (1995).

24. Schulte, U., Hahn, H., Wiesinger, H., Ruppersberg, J.P. & Fakler, B. pH-dependent gating of ROMK (Kir1.1) channels involves conformational changes in both N and C termini. J. Biol. Chem. 273, 34575–34579 (1998).

25. Corey, S. & Clapham, D.E. Identification of native atrial G-protein-regulated inwardly rectifying K+ (GIRK4) channel homomultimers. J. Biol. Chem. 273, 27499–27504 (1998).

26. Inanobe, A. et al. Characterization of G- protein-gated K+ channels composed of Kir3.2 subunits in dopaminergic neurons of the substantia nigra. J. Neurosci. 19, 1006–1017 (1999).

27. Armstrong, N., Sun, Y., Chen, G.Q. & Gouaux, E. Structure of a glutamate-receptor ligand-binding core in complex with kainate. Nature 395, 913–917 (1998).

28. Zhou, W., Qian, Y., Kunjilwar, K., Pfaffinger, P.J. & Choe, S. Structural insights into the functional interaction of KChIP1 with Shal-type K channels. Neuron 41, 573–586 (2004).

29. Lu, T., Nguyen, B., Zhang, X. & Yang, J. Architecture of a K+ channel inner pore revealed by stoichiometric covalent modification. Neuron 22, 571–580 (1999).

30. Riven, I., Kalmanzon, E., Segev, L. & Reuveny, E. Conformational rearrangements associated with the gating of the G protein-coupled potassium channel revealed by FRET microscopy. Neuron 38, 225–235 (2003).

31. Cohen, N.A., Brenman, J.E., Snyder, S.H. & Bredt, D.S. Binding of the inward recti-fier K+ channel Kir 2.3 to PSD-95 is regulated by protein kinase A phosphorylation. Neuron 17, 759–767 (1996).

32. Fakler, B., Bond, C.T., Adelman, J.P. & Ruppersberg, J.P. Heterooligomeric assembly of inward-rectifier K+ channels from subunits of different subfamilies: Kir2.1 (IRK1) and Kir4.1 (BIR10). Pflugers Archiv. 433, 77–83 (1996).

33. Pearson, W.L. & Nichols, C.G. Block of the Kir2.1 channel pore by alkylamine ana-logues of endogenous polyamines. J. Gen. Physiol. 112, 351–363 (1998).

34. Ishihara, K. & Ehara, T. Two modes of polyamine block regulating the cardiac inward rectifier K+ current IK1 as revealed by a study of the Kir2.1 channel expressed i7n a human cell line. J. Physiol. (Lond.) 556, 61–78 (2004).

35. Bendahhou, S. et al. Defective potassium channel Kir2.1 trafficking underlies ander-sen-tawil syndrome. J. Biol. Chem. 278, 51779–51785 (2003).

36. Hosaka, Y. et al. Function, subcellular localization and assembly of a novel mutation of KCNJ2 in Andersen’s syndrome. J. Mol. Cell. Cardiol. 35, 409–415 (2003).

37. Preisig-Muller, R. et al. Heteromerization of Kir2.x potassium channels contributes to the phenotype of Andersen’s syndrome. Proc. Natl. Acad. Sci. USA 99, 7774–7779 (2002).

38. Garneau, L., Klein, H., Parent, L. & Sauve, R. Contribution of cytosolic cysteine resi-dues to the gating properties of the Kir2.1 inward rectifier. Biophys. J. 84, 3717–3729 (2003).

39. Chen, L. et al. A glutamate residue at the C terminus regulates activity of inward rectifier K+ channels: implication for Andersen’s syndrome. Proc. Natl. Acad. Sci. USA 99, 8430–8435 (2002).

40. Chang, H.-K., Yeh, S.-H., Shieh, R-C. (2005) A ring of negative charges in the intracel-lular vestibule of Kir2.1 channel modulates K+ permeation. Biophys. J. 88, 243–254 (2005).

41. Clancy, S.M. et al. Pertussis-toxin-sensitive G(alpha) subunits selectively bind to C-terminal domain of neuronal GIRK channels: evidence for a heterotrimeric G-protein-channel complex. Mol. Cell. Neurosci. 28, 375–389 (2005).

42. Ivanina, T. et al. Gi1 and Gi3 differentially interact with, and regulate, the G Protein-activated K+ channel. J. Biol. Chem. 279, 17260–17268 (2004).

43. Guo, Y., Waldron, G.J. & Murrell-Lagnado, R.D. A role for the middle C-terminus of GIRK channels in regulating gating. J. Biol. Chem. 277, 48289–48294 (2002).

44. Xiao, J., Zhen, X.G. & Yang, J. Localization of PIP2 activation gate in inward rectifier K+ channels. Nat. Neurosci. 6, 811–818 (2003).

45. Phillips, L.R., Enkvetchakul, D. & Nichols, C.G. Gating dependence of inner pore access in inward rectifier K+ channels. Neuron 37, 953–962 (2003).

46. Roosild, T.P., Le, T.-K. & Choe, S. Cytoplasmic gatekeepers of K channel flux: a structural perspective. Trends Biochem. Sci. 29, 39–45 (2004).

47. Jiang, Y. et al. The open pore conformation of potassium channels. Nature 417, 523–526 (2002).

48. Roosild, T.P., Miller, S., Booth, I.R. & Choe, S. A mechanism of regulating transmem-brane potassium flux through a ligand-mediated conformational switch. Cell 109, 781–791 (2002).

49. Kubo, Y., Baldwin, T.J., Jan, Y.N. & Jan, L.Y. Primary structure and functional expres-sion of a mouse inward rectifier potassium channel. Nature 362, 127–133 (1993).

50. Hilgemann, D. The giant membrane patch. in Single-Channel Recording 2nd edn. (eds. Sakmann, B. & Neher, E.) 307–327 (Plenum, New York, 1995).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Directed differentiation of telencephalic precursors from embryonic stem cellsKiichi Watanabe1,2, Daisuke Kamiya1,2, Ayaka Nishiyama1, Tomoko Katayama1, Satoshi Nozaki3, Hiroshi Kawasaki4, Yasuyoshi Watanabe3, Kenji Mizuseki1 & Yoshiki Sasai1,2

We demonstrate directed differentiation of telencephalic precursors from mouse embryonic stem (ES) cells using optimized serum-free suspension culture (SFEB culture). Treatment with Wnt and Nodal antagonists (Dkk1 and LeftyA) during the first 5 d of SFEB culture causes nearly selective neural differentiation in ES cells (∼90%). In the presence of Dkk1, with or without LeftyA, SFEB induces efficient generation (∼35%) of cells expressing telencephalic marker Bf1. Wnt3a treatment during the late culture period increases the pallial telencephalic population (Pax6+ cells yield up to 75% of Bf1+ cells), whereas Shh promotes basal telencephalic differentiation (into Nkx2.1+ and/or Islet1/2+ cells) at the cost of pallial telencephalic differentiation. Thus, in the absence of caudalizing signals, floating aggregates of ES cells generate naive telencephalic precursors that acquire subregional identities by responding to extracellular patterning signals.

Over the last few years, several reports have introduced methods for the in vitro generation of neural cells from mouse ES cells1–6. Floating aggregates of ES cells cultured in the presence of serum (embryoid bodies) contain various kinds of cells derived from the three germinal layers7,8, although the content of neural tissues is not generally high. When retinoic acid is added to the culture medium, relatively high proportions of neural cells are generated in embryoid bodies1. However, retinoic acid treatment evokes caudalization of induced neural tissues and inhibits the production of rostral CNS tissues2,3.

We have previously established an efficient protocol for inducing neural differentiation using a coculture system with PA6 stromal cells. The inducing activity on the surface of PA6 cells has been named SDIA (stromal cell–derived inducing activity4); its molecular nature remains to be understood. When cultured on PA6 cells under serum-free condi-tions, over 90% of mouse ES cells differentiate into neural cells within a week. Notably, neural tissues induced from ES cells by the SDIA method contain a high percentage of midbrain dopaminergic neurons4.

Although ES cell–derived neural precursors have been shown to have the competence to generate a wide range of dorsal-ventral neural tissues in response to patterning signals3, relatively little is understood about the differentiation control of rostral CNS tissues in ES cells. Our previous3 and current analyses with rostral-caudal neural markers suggest that SDIA-induced neural tissues express forebrain- midbrain-hindbrain CNS markers, but not caudal spinal cord markers. Notably, we show here that telencephalic cells are generated only at a low frequency in SDIA-induced neural tissues, although the SDIA method involves no exogenous factors with caudalizing activity such as retinoic

acid and FGF (fibroblast growth factor) in the culture medium4. This observation prompted us to test whether the inefficient telencephalic differentiation was caused by the presence of inhibitory signals or by the lack of inductive signals. We optimized the ES cell culture conditions for neural differentiation without feeder cells and serum. We report the first demonstration of telencephalic differentiation and subregional specification in ES cells using soluble signaling factors and an optimized culture method.

RESULTSSuspension culture of ES cell aggregates without serumIt has been shown that ES cells undergo neural conversion at a reasonable efficiency in serum-free adherent monoculture without feeder cells6. In an attempt to further optimize the feeder- and serum-free ES cell culture for neural conversion, we investigated suspension culture conditions without the use of retinoic acid and growth factors. We tested a number of serum-free media and found that at least three supported cell growth reasonably well (see Methods). In particular, a knockout serum replace-ment (KSR)-supplemented medium (the same used for the SDIA cul-ture) effectively supported the growth of floating ES cell aggregates (giving a 5.3-fold increase in cells in 3 d and a 10-fold increase in 7 d) and was used in the studies reported here (‘differentiation medium’; see Methods). In this report, the suspension culture conditions using the serum-free medium are referred to as ‘SFEB’ (serum-free, floating culture of embryoid body–like aggregates).

Notably, after 8 d of SFEB culture and 2 d of adherent culture, mouse ES cell aggregates (initial cell density of 5 × 104 cells ml–1)

1Organogenesis and Neurogenesis Group, Center for Developmental Biology, RIKEN, Kobe 650-0047, Japan. 2Department of Medical Embryology, Graduate School of Medicine, Kyoto University, Kyoto 606-8501, Japan. 3Department of Physiology, Graduate School of Medicine, Osaka City University, Osaka 545-8585, Japan. 4Department of Molecular and System Neurobiology, Graduate School of Medicine, University of Tokyo, Tokyo 113-0033, Japan. Correspondence should be addressed to Y.S. ([email protected]).


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

contained a large number of TuJ1+ neurons and Nestin+ neural precursors (∼10% and ∼70% of total cells, respectively; Fig. 1a–c). For quantitative analysis, we next used ES cells in which GFP cDNA was knocked in at the locus of the early neuroectodermal marker gene Sox1 (line 46C, gift from A. Smith; Sox1 transcripts are detectable at E7.5 in vivo)6,9,10. FACS analysis (Fig. 1d,e) showed that the proportion of Sox1-GFP+ cells increased to 70–80% of total cells during days 3–5 of culture. In contrast, E-cadherin+ and Oct3/4+ cells gradually decreased in number over the first 5 d (Fig. 1e and Supplementary Fig. 1). Immunocytochemical analyses showed that Sox1-GFP+ cells expressed the neural precursor markers Nestin and RC2, but not E-cadherin, Oct3/4 or Nanog (Supplementary Fig. 1 and data not shown). RT-PCR analysis indicated induction of the primitive ectoder-mal marker gene Fgf5 (refs. 11,12) during days 1–3 (Supplementary Fig. 1). We next optimized the initial cell density for maximal neural cell generation. The highest percentage of neural cell generation was observed when culture was started with a range of 3 × 104 to 1 × 105 cells ml–1; ∼80% of the cells became Sox1-GFP+ on day 5 under these conditions (Fig. 1f).

Wnt and TGFβ signals have negative effects on mammalian neural differentiation4,13–15. Consistent with these reports, addition of the Wnt3a protein (50 ng ml–1), the Nodal protein (5 µg ml–1) or the BMP4 protein (0.5 nM) during days 0–5, and also during days 2–5, suppressed the accumulation of Sox1-GFP+ cells (Fig. 1g and data not shown). Conversely, treatment with a combination of the anti-Wnt reagent Dickkopf-1 (Dkk1)16 and the anti-Nodal reagent LeftyA17 caused a clear and statistically significant increase in the proportion of Sox1-GFP+ cells (to 90.2 ± 1.6%, P < 0.05 versus the control; Fig. 1h). Although Dkk1 and LeftyA alone caused only a marginal increase, the addition of either Dkk1 or LeftyA (particularly the latter) tended to reduce the fluctuation of the percentages, making induction even more reproducible (error bars in Fig. 1h). Notably, treatment with Dkk1 and LeftyA facilitated neural conversion even under two types of unfavorable conditions that would normally result in low percentages of neural cell generation: excessively high cell density (2 × 105 ml–1) and medium containing serum (5%) (∼60% and <20% without treatment, respectively, and ∼80% and ∼70% with treatment, respectively; Supplementary Fig. 2). Taken together, these results suggest that blockades of endogenous Wnt and Nodal signals enhance and stabilize the generation of neural cells in ES cell aggregates.

Figure 1 Efficient neural conversion of mouse ES cells in the SFEB culture with Dkk1 and LeftyA. (a–c) Immunostaining of class III tubulin (TuJ1) and Nestin and nuclear staining with TOTO-3 in SFEB-induced cells (8 d of floating culture followed by 2 d of adherent culture on dishes coated with poly-D-lysine, laminin and fibronectin). (d) Flow cytometry profiles of Sox1-GFP and E-cadherin–APC fluorescence. (e) Percentage of Sox1-GFP+ and E-cadherin+ cells during the first 6 d in the SFEB culture. (f) Effects of the initial cell density on percentage of Sox1-GFP+ cells. (g) Suppression of neural conversion by Wnt3a, Nodal and BMP4 recombinant proteins. Strong suppression by BMP4 was reversed by adding BMPRIA-Fc or Noggin proteins. (h) Facilitated generation of neural precursors from SFEB-treated ES cells resulting from addition of Dkk1 and LeftyA recombinant proteins. (f–h) Cells were analyzed on day 5. *P < 0.05 versus control (Dunnett test).

TOTO-3NestinTuJ1

Undifferentiated ES cells

Sox1-GFP fluorescence

Cou

nts

Sox1-GFP+

Day 3

Cou

nts


Day 4

Cou

nts


Day 5

Cou

nts


Cou

nts

E-Cad-APC fluorescence

Cou

nts


Cou

nts



Cou

nts

E-Cad+

d

h

60 1 2 3 4 5

e

f

g

Sox1-GFP+

Sox1-GFP+

Sox1-GFP+

E-Cad+

E-Cad+

E-Cad+

(Pos

itive

cel

ls /

tota

l cel

ls)

Days after differentiation

100

75

50

25

0

Sox1-GFPE-Cad -APC

Per

cent

age

Sox

1-G

FP

+ c

ells

60

70

80

90

100

50100100.1

Cell density (x 104 cells/ml)1

Per

cent

age

of

100

0

80

60

40

20Sox

1-G

FP

+ c

ells

Wnt3a Nodal BMP4

Con

trol

5 ng

/ml

50 n

g/m

l

0.5

µg/m

l

5 µg

/ml

0.5

nM+

BM

PR

IA -

Fc

+N

oggi

n

Per

cent

age

of

100

60

40

20

0

Sox

1-G

FP

+ c

ells

Con

trol

Dkk

1 10

0 ng

/ml

Dkk

1 1

µg/m

l

Lefty

A 1

µg/

ml

Lefty

A 5

µg/

ml

Dkk

1 1

µg/m

l+

Lefty

A 5

µg/

ml

80

*

Per

cent

age

of

a b c

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Treatment with anti-BMP reagents (200 ng ml–1 BMPRIA-Fc recep-tobody or 500 ng ml–1 of recombinant Noggin protein) did not subs-tantially affect the Sox1-GFP+ percentage (Supplementary Fig. 2), although both could reverse the anti-neuralizing effects of 0.5 nM BMP4 completely (Fig. 1g). These findings are in contrast to the previous report showing that Noggin treatment enhances neural differentiation in human ES cells cultured in monolayer18.

Unlike embryoid bodies cultured in a serum-containing medium, which efficiently produce both primitive endodermal and mesodermal cells19, in situ hybridization analysis showed that only a small population of SFEB aggregates expressed the primitive endodermal markers Afp and Hnf4a (also called Hnf4; ref. 20) or the mesodermal marker T (also called Brachyury), and even then, only faintly (Supplementary Fig. 3; weak Brachyury expression was detectable in the FACS-enriched Sox1-GFP– population of SFEB on day 5 by RT-PCR). These findings sug-gest that the generation of neural precursors in SFEB occurs largely without concomitant induction of mesodermal and endodermal tissues. Notably, in embryoid bodies cultured with serum, treatment with Dkk1 and LeftyA, which enhances neural conversion even in the presence of serum (Supplementary Fig. 2), suppressed Afp and Hnf4a expression (to ∼50% and ∼20% of the control non-treated levels, respectively; data not shown), suggesting that enhanced neural conversion by Dkk1 and LeftyA was not accompanied by an increase of mesodermal or endodermal tissues in serum-treated embryoid bodies.

Neural conversion with little selective bias in SFEBNext, we analyzed the possible contribution of preferential apoptosis and proliferation to the selective accumulation of neural cells in the SFEB culture. First, we examined the presence of enhanced cell death in the non-neural population during days 3 and 4, when the propor-tion of Sox1-GFP+ cells was rapidly increasing (Fig. 1e). The percent-age of TUNEL-labeled apoptotic cells was generally low and showed no substantial difference between the Sox1-GFP– (5.5 ± 0.7% on day 3 and 4.8 ± 1.4% on day 4, n = 3) and Sox1-GFP+ (4.8 ± 0.9% on day 3 and 4.6 ± 1.5% on day 4, n = 3) populations (Fig. 2a). The proportion of cells labeled with annexin V (which also marks apoptotic cells) was 3.3 ± 0.6% on day 3 and 3.6 ± 0.7% on day 4 in the Sox1-GFP– population, compared to 4.5 ± 1.5% on day 3 and 5.8 ± 1.8% on day 4 in the Sox1-GFP+ popula-tion (n = 3; Fig. 2b). These observations show that Sox1-GFP– cells are not actively eliminated by apoptosis in SFEB.

We next compared the cell proliferation parameters in the Sox1-GFP– and Sox1-GFP+ populations. The percentage of cells labeled by bromodeoxyuridine (BrdU; 30-min exposure at the end of culture) did not differ greatly between the Sox1-GFP– (45.3 ± 0.4% on day 3 and 40.9 ± 1.2% on day 4, n = 3) and Sox1-GFP+ (49.6 ± 2.6% on day 3 and 42.4 ± 1.1% on day 4, n = 3) populations (Fig. 2c). The ratio of cells positive for phospho-histone H3 (which marks cells in the G2-M phases) was similar: 1.2 ± 0.5% on day 3 and 1.0 ± 0.7% on day 4 in the Sox1-GFP– population, compared to 1.5 ± 0.7% on day 3 and 1.3 ± 0.6% on day 4 in the Sox1-GFP+ population (n = 3; Fig. 2d). Taken together, these observations suggest that preferential apoptosis and cell growth are not major causes for the selective accumulation of neural precursors in SFEB.

In contrast, the suppression of neural precursor generation by BMP4 may at least in part involve a selective mechanism. Addition of 0.5 nM BMP4 on day 3 decreased the accumulation of Sox1-GFP+ cells in SFEB on day 5 (percentage of Sox1-GFP+ cells <10%). Under this condition, a moderate but statistically significant increase in the percentage of TUNEL+ cells was seen in the Sox1-GFP+ population on day 4 as compared to the Sox1-GFP– population (7.8 ± 1.1% and 3.7 ± 0.8%, respectively, P < 0.05), whereas no significant difference in

the amount of BrdU uptake was observed between the two populations (Supplementary Fig. 4).

Telencephalic differentiation in SFEB-induced neural cellsNext, we compared the frequencies of rostral-caudal marker expression between the feeder-dependent (SDIA) and feeder-free (SFEB) neural induction systems (using the same serum-free culture medium) by quantitative immunostaining. Bf1 (also known as Foxg1) is a transcription factor specifically expressed in the ventricular (VZ) and subventricular (SVZ)-intermediate zones of the telencephalon (Fig. 3a,b) that is essential for proper development of the region21,22. Although immunostaining detected Bf1+ telencephalic cells in neural tissues generated from ES cells on PA6 cells (10 or 15 d of SDIA culture), the frequency was quite low (1.6 ± 0.2% on day 10 and even lower on day 15). SDIA-treated cells more frequently expressed Otx1 (a marker of forebrain and midbrain), Pax2 (typically a marker of midbrain and caudal CNS) and tyrosine hydroxylase (TH; a dopaminergic and noradrenergic neuron marker that is typically a midbrain and hindbrain marker) (Fig. 3c). In contrast, under feeder-free condi-tions (SFEB for 5 d followed by adherent culture for 5 or 10 d on dishes coated with poly-D-lysine, laminin and fibronectin), telen-cephalic differentiation was markedly increased (Fig. 3d). A high percentage of cells expressed Bf1 (15.1 ± 2.1% on day 10) in SFEB-induced neural cells, whereas the midbrain–spinal cord markers (Pax2, TH and Hoxb4) were expressed in fewer cells; in addition, RT-PCR showed strong Bf1 (Foxg1) expression (Fig. 3h). Consistently, a large majority of SFEB-cultured ES cell aggregates strongly expressed Six3 (an early marker for the telencephalic and diencephalic regions rostral to the zona limitans23), but only a minority of SDIA-treated ES cell colonies did so (day 5; Supplementary Fig. 5).

The percentage of cells expressing Bf1 was high (11.4 ± 2.9%) when ES cells were cultured in SFEB for 5 d and then on PA6 cells for the next 5 d (Fig. 3e, second row). This is comparable to the percentage of cells expressing Bf1 after 10 d in feeder-free culture (Fig. 3e, top row). Conversely, few Bf1+ cells were observed (<1%) when cells were cultured on PA6 cells for the first 5 d and then on dishes coated with laminin, poly-D-lysine and fibronectin for the next 5 d (Fig. 3e, third

Figure 2 Selective biases do not have a major role in neural induction in SFEB. (a,b) Percentages of TUNEL+ and annexin V+ cells in the Sox1-GFP+ or Sox1-GFP– populations on differentiation days 3 and 4, respectively. (c,d) Percentages of BrdU-labeled and phospho-histone H3+ cells in the Sox1-GFP+ or Sox1-GFP– populations on differentiation days 3 and 4, respectively.

Per

cent

age

of

phos

pho-

hist

one

H3+

cel

ls

Day 3 Day 4

4

6

2

0

8

10

Per

cent

age

of a

nnex

in V

+ c

ells

Day 3 Day 4

50

40

30

20

10

0

50

100

75

25

0

Day 3 Day 4

Per

cent

age

of B

rdU

+ c

ells

50

40

30

20

10

0

Day 3 Day 4

Sox1-GFP+Sox1-GFP–

Per

cent

age

of T

UN

EL+

cel

ls

a b

c d

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

row). These results indicate that the first 5 d of the induction period are decisive in telencephalic induction. Consistently, a time-course analysis of immunostaining showed that cells positive for Bf1 and Nkx2.1 (another rostral CNS marker) appeared by day 6 and gradually increased in number during the next few days in the feeder-free culture (data not shown).

To further enhance telencephalic differentiation, we next investigated culture conditions during the first 5 d. Wnt and Nodal signals are also implicated in the rostral-caudal specification24,25. Notably, the percentage of Bf1+ cells substantially increased (to 35% of total cells) when both Dkk1 and LeftyA were added to the SFEB culture during

days 0–5 (Fig. 3f,g; this increase in Bf1 induction was ∼2.5-fold, whereas the increase in neural conversion was 1.2-fold, as shown in Fig. 1h). By itself, Dkk1 seemed to have a clear and statistically significant enhan-cing effect (Fig. 3f), whereas LeftyA (5 µg ml–1) alone caused a marginal increase. These findings suggest that endogenous Wnt signals (and possibly Nodal signals) have an inhibitory role in Bf1 induction under the SFEB condition during days 0–5. In contrast, treatment with any of Dkk1, LeftyA, Wnt3a or Nodal proteins during days 6–10 did not substantially affect Bf1 induction (data not shown).

Treatment with the caudalizing factor retinoic acid almost com-pletely eliminated induction of Bf1+ cells in SFEB (<1%; 0.2 µM retinoic acid during days 3–10; not shown). Accordingly, RT-PCR analysis showed that expression of the rostral markers Bf1 and Otx2 was completely suppressed by retinoic acid in SFEB-induced neural cells (days 3–5 or 3–10), but expression of the caudal markers Hlxb9 (also called HB9) and Hoxb9 was induced (Fig. 3h). In contrast, reti-noic acid treatment during days 6–10 suppressed Bf1 and Otx2 expres-sion only partially (if at all) and induced caudal marker expression (Supplementary Fig. 5).

Taken together, these results suggest that the SFEB culture efficiently induces telencephalic differentiation in the absence of exogenous caudalizing factors or endogenous Wnt signals during the first 5 d. The enhancement of Bf1 induction by Dkk1 (and by Dkk1 + LeftyA) in SFEB prompted us to test the roles of Wnt and Nodal signals in the SDIA culture. Addition of Dkk1, LeftyA, or Dkk1 + LeftyA during the first 5 d significantly increased the percentage of Bf1+ cells in SDIA-treated ES cells (Fig. 3i). These increases were counteracted by the addition of Wnt3a and Nodal. Dkk1 and LeftyA (separately or in combination) did not substantially affect the percentages of cells positive for TH, 5-HT (a marker of the hindbrain) and the neuronal marker TuJ1 (data not shown). These observations suggest that the SDIA system requires Wnt and Nodal specifically for inhibition of rostral CNS differentiation.

As for endogenous gene expression, RT-PCR analysis showed that differentiating ES cells expressed both Wnt genes (Wnt1, Wnt2b, Wnt3,

Dkk1+LeftyA

Bf1 /TuJ1 /TOTO-3

h

c30

25

20

15

10

5

0expr

essi

ng in

dica

ted

mar

ker

Bf1

Nkx

2.1

Otx

1

Pax

2

TH

5HT

Hox

b1

Hox

b4

d16

12

8

4

expr

essi

ng in

dica

ted

mar

ker

Bf1

Nkx

2.1

Otx

1

Pax

2

TH

5HT

Hox

b1

Hox

b4

0

i

Per

cent

age

of B

f1+ c

ells

SD

IA

Dkk

1 10

0 ng

/ml

Dkk

1 1

µg/m

l

Dkk

1 10

0 ng

/ml

+W

nt3a

50

ng/m

l

Lefty

A 1

µg/

ml

Dkk

1 1

µg/m

l+

Lefty

A 5

µg/

ml

Lefty

A 5

µg/

ml

Lefty

A 1

µg/

ml

+N

odal

5 µ

g/m

l0

5

10

15

20

e Percentage of Bf1+ cells

PA6 (days 0–10)

Feeder-free (days 0–10)

Feeder-free (days 0–5)/PA6 (days 6–10)

PA6 (days 0–5)/feeder-free (days 6–10)

0 4 8 12 16

MGE

LGE

ctx

AEP/POA

LV

Bf1 DAPIms

mt

my

sc

ole

LV

di

MGELGE

ctx

Bf1 DAPI

E12.5

**

**

*

**

R C R C

**

**

** P < 0.01vs. PA6 (days 0–10)

* P < 0.05** P < 0.01vs. SDIA

Bf1

HB9

En2

RT–

Otx2

NCAM

Hoxb9

G3PDH

Em

bryo

SF

EB

RA 0.2 µM RA 2 µM

Day

s 3–

5

Day

s 3–

10

Day

s 3–

5

Day

s 3–

10

1 2 3 4 5 6

SD

IA

7

f40

30

SF

EB

Dkk

1 10

0 ng

/ml

Dkk

1 1

µg/m

l

Lefty

A 1

µg/

ml

Dkk

1 1

µg/m

l+

Lefty

A 5

µg/

ml

Lefty

A 5

µg/

ml

Per

cent

age

of B

f1+ c

ells

0

10

20

****

Per

cent

age

of to

tal c

ells

Per

cent

age

of to

tal c

ellsSDIA SFEB

Parasagittal Coronal

a b

g

Figure 3 Differentiation of Bf1+ cells in SFEB-treated ES cells and further enhancement by Dkk1 and LeftyA. (a,b) Expression of the telencephalic marker protein Bf1 (red) in E12.5 mouse embryos. Cells were counterstained with DAPI (blue). (c,d) Percentage of cells expressing rostral and caudal markers in SDIA-induced cells or SFEB-induced cells. R, rostral; C, caudal. (e) Percentage of Bf1+ cells in swapping experiments between the SFEB and SDIA procedures. ‘Feeder-free (days 0–5)/PA6 (days 6–10)’ cells were cultured in SFEB for first 5 d and then in SDIA on PA6 cells for 5 d. ‘PA6 (days 0–5)/feeder-free (days 6–10)’ cells were cocultured on PA6 cells for the first 5 d and then cultured in adherent monoculture on dishes coated with poly-D-lysine, laminin and fibronectin for the next 5 d. (f) Effects of Dkk1 and LeftyA (days 0–5) on differentiation of Bf1+ cells in the SFEB culture. (g) Expression of Bf1 (red) and TuJ1 (green) in ES cells treated with SFEB, Dkk1 (1 µg ml–1) and LeftyA (5 µg ml–1). Cells were counterstained with TOTO-3. (h) RT-PCR analysis of rostral-caudal marker genes. Lane 1, E12.5 whole embryo; lanes 2 and 3, ES cells treated with SFEB and 0.2 µM all-trans retinoic acid (RA; days 3–5 and 3–10); lanes 4 and 5, ES cells treated with SFEB and 2 µM RA (days 3–5 and 3–10); lane 6, control SFEB-treated ES cells; lane 7, SDIA-treated ES cells. En2, marker of the midbrain-hindbrain boundary; NCAM (Ncam1), pan-neural marker; G3PDH (Gapd), control housekeeping gene; RT–, control without reverse transcriptase. (i) Increase in percentage of Bf1+ cells resulting from addition of Dkk1 and LeftyA (days 0–5) in the SDIA culture. In c–i, cells were analyzed on day 10. ctx, cortex; LV, lateral ventricle; LGE, lateral ganglionic eminence; MGE, medial ganglionic eminence; ole, olfactory epithelium; AEP, anterior entopeduncular area; POA, anterior preoptic area; di, diencephalon; ms, mesencephalon; mt, metencephalon; my, myelencephalon; sc, spinal cord. *P < 0.05 versus control; **P < 0.01 versus control (Dunnett test).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Wnt3a and Wnt10b) and Nodal (Supplementary Fig. 6). In contrast, PA6 cells expressed Wnt genes (Wnt1, Wnt2b and Wnt10b) but not Nodal. These findings suggest that the effect of LeftyA is likely to reflect the inhibition of endogenous low-level Nodal signals in differentiating ES cells and imply that the role of PA6 cells in Nodal signaling is indirect. One interpretation could be that endogenous Nodal signaling exerts a permissive effect that is required for other PA6-derived caudalizing factors to act efficiently. Another possibility is that PA6 cells enhance the responsiveness of ES cells to Nodal signals. Such a dynamic change of the Nodal response is seen in the mouse epiblast, in which the expression of the Nodal co-receptor gene Tdgf1 (also called Cripto) is transcriptionally regulated in a region-specific manner26. However, we observed no increase of Tdgf1 (Cripto) expression in the presence of PA6 cells (Supplementary Fig. 6).

Regional specification of SFEB-induced telencephalic cellsWe next investigated the subregional nature of SFEB-induced Bf1+ cells (Fig. 4a). Among Bf1+ cells induced by SFEB culture alone (‘control’ row), 36.5 ± 3.7% were also positive for Pax6, whose strong expression marks the dorsal or 'pallial' telencephalon27 (top row); 23.8 ± 0.9% were positive for Gsh2, which marks the ventral or 'basal' telencephalon and is strongest in the lateral ganglionic eminence (LGE)28; and 15 ± 6.3% were positive for Nkx2.1, which marks the ventral part of the basal telencephalon with the exception of the LGE29.

Recent studies have shown that Wnt genes (expressed in the dorsal pallium) positively control pallial telencephalic specification in vivo 30–32. This effect of Wnt occurs during relatively late developmental stages, as compared with the effect of Wnt signaling in the inhibition of neural differentiation and forebrain development, which occurs during earlier stages13,24,33. This prompted us to examine the effect of Wnt treat-ment on subregional specifications of ES cell– derived telencephalic cells during the late phase of SFEB culture. When SFEB-induced neu-ral cells were treated with Wnt3a during days 6–10, Pax6+ Bf1+ cells substantially increased in a dose- dependent manner, whereas Nkx2.1+ Bf1+ and Gsh2+ Bf1+ cells decreased (Fig. 4a,b). The proportion of

Bf1+ cells in SFEB-induced cells was not substantially changed by late Wnt treatment (data not shown). These findings indicate that late Wnt treatment promotes differentiation of pallial telencephalic cells at the cost of the basal cells. Consistently, when Wnt signals were blocked during the late culture period by adding Dkk1 (days 5–10), the Pax6+ population decreased significantly in Bf1+ cells (Fig. 4b), indicating that late (endogenous) Wnt signaling is essential for pallial telencephalon differentiation in SFEB-induced neural cells.

Treatment with Wnt3a also increased the number of cells positive for the cerebral cortical marker Emx127,34 (Fig. 4c–f), providing further evidence for pallial differentiation in SFEB-induced, Wnt-treated telencephalic tissues. Addition of FGF8b (5–50 ng ml–1) to Wnt3a did not further enhance Emx1 induction, in contrast to its effect on chick telencephalic tissues reported previously32 (data not shown).

We next investigated induction of basal telencephalic differentiation (Fig. 5). Shh has been implicated in the ventral specification of the forebrain35–37. When applied to SFEB-induced neural tissues, Shh (days 4–10) increased the percentage of Nkx2.1+ cells among Bf1+ cells but decreased the percentages of Pax6+ and Emx1+ cells (Fig. 4a, ‘Shh’ row; Fig. 4f and Fig. 5a). Shh did not cause substantial differences in the percentage of cells positive for Gsh2 and Mash1, a marker for the SVZ of the basal telencephalon (Fig. 5a and data not shown). The proportion of Bf1+ cells was not greatly affected by Shh treatment (data not shown). These observations show that Shh promotes the ventral specification of SFEB-induced telencephalic cells.

SFEB induces the generation of telencephalic precursors capable of responding to embryologically relevant dorsal and ventral patterning signals. Although the proportion of Pax6+ cells in SFEB-induced, Wnt-treated Bf1+ cells was quite high (∼75%, Fig. 4b), the percentage of Nkx2.1+ cells was moderate even at the highest Shh dose (∼40%; Fig. 5a). This observation led us to reexamine the distribution of Nkx2.1, Pax6 and Bf1 in the embryonic brain (E12.5). Bf1 expres-sion was detected both in the VZ and in a large part of the SVZ and intermediate zones along the dorsal-ventral axis (Figs. 3a,b, 4a and 5b). Pax6 expression covered the majority of the Bf1+ area in the pallial

bPax6 Bf1 Gsh2 Bf1 Nkx2.1 Bf1

E12

.5W

nt3a

Con

trol

Shh

ctx

LGE

MGEAEP/POA

a

Emx1 Bf1 TuJ1

E12.5

eWnt3a

f

expr

essi

ng E

mx1

10

7.5

5.0

2.5

0505.0

Con

trol330300

Shh (nM) Wnt3a(ng/ml)

Wnt3a (ng/ml)505.0

expr

essi

ng in

dica

ted

mar

ker

0

60

40

20

80

100

Dkk1100 ng/ml

Control

Pax6

Gsh2

Nkx2.1

Per

cent

age

of B

f1+ c

ells

Per

cent

age

of B

f1+ c

ells

c d

e

Figure 4 Differential control of dorsal and ventral telencephalic specification by soluble patterning signals. (a) Expression of Pax6 (red, left), Gsh2 (red, middle), Nkx2.1 (red, right) and Bf1 (green) in coronal sections of the E12.5 mouse telencephalon (top) and in SFEB-treated ES cells (differentiation day 10; second from bottom). Second row, SFEB with 50 ng ml–1 Wnt3a (days 6–10); bottom row, SFEB with 300 nM Shh (days 4–10). (b) Percentages of Bf1+ cells expressing dorsal (Pax6) and ventral (Gsh2, Nkx2.1) telencephalic markers in the SFEB culture. Wnt3a (5 ng ml–1 or 50 ng ml–1) and Dkk1 (100 ng ml–1) were added during days 6–10 and days 5–10, respectively. Cells were analyzed on day 10. (c) Expression of Emx1 (red), Bf1 (green) and TuJ1 (blue) in the telencephalic region of the E12.5 mouse embryo. Higher-magnification view of the boxed area is shown in d. (d) In the pallial region, nearly all Emx1+ cells (red) coexpressed Bf1 (green). (e) Expression of Emx1 (red), Bf1 (green) and TuJ1 (blue) in ES cells treated with SFEB in the presence of Wnt3a (50 ng ml–1, days 6–10). (f) Percentages of Emx1+ cells among ES cell–derived Bf1+ cells in the presence of Wnt3a (5 ng ml–1 or 50 ng ml–1, days 6–10) or Shh (3 nM, 30 nM or 300 nM, days 4–10). (e,f) Cells were analyzed on day 10.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

region (Fig. 4a). By contrast, in the MGE and the telencephalic stalk domain (the anterior entopeduncular and preoptic areas (AEP-POA)), Nkx2.1 expression was limited to the VZ and the adjacent part of the SVZ (Figs. 4a and 5b), leaving a substantial Nkx2.1– Bf1+ area in the ventral telencephalon. In the AEP-POA, Islet1 was expressed in a large proportion of the Nkx2.1– Bf1+ area (ref. 38; Fig. 5b,c). In SFEB-induced Bf1+ telencephalic cells, Shh induced expression of Islet1 in 25–30% of cells (Fig. 5d,e). Only one-fifth of Islet1+ cells were also positive for Nkx2.1, consistent with the small overlap of these markers in the embryonic brain (Fig. 5c and data not shown). Taken together with the data in Figure 5a, these results suggest that a large proportion (roughly 60–70%) of Bf1+ cells express ventral markers (Nkx2.1 and/or Islet1) after Shh treatment.

DISCUSSIONIn this work, we have successfully demonstrated effi-cient telencephalic differentiation from ES cells using opti-mized suspension culture conditions that support selective neural differentiation of ES cell aggregates in vitro. The temporal expression pattern of early ectodermal differentiation markers such as E-cadherin, Oct3/4, Fgf5 and Sox1 in SFEB generally mimicked that in the embryo (Fig. 1 and Supplementary Fig. 1), although more detailed comparative analysis of the speed of differentiation in vivo and in vitro is needed. Neural differentiation was further enhanced (to 90%) in SFEB when endogenous Wnt and Nodal signals were blocked by Dkk1 and LeftyA (Fig. 1).

Autonomous neural differentiation in the absence of exogenous inhibitory factors could be consistent with the so-called ‘neural default’ model39. In line with this idea, a previous study has reported that ES cells, when cultured at low density, generate sphere-forming neural stem cells,

although the forming efficiency seems low40. However, our preliminary study suggests that the use of the term ‘default state’ for ES cell culture may require more careful consideration. At least in the SDIA culture, inhibiting E-cadherin–dependent cell contact by adding the E-cadherin–blocking antibody ECCD1 (ref. 41) strongly attenuated neural differentiation of ES cells (Supplementary Fig. 6). Therefore, in addition to soluble or paracrine signals, the role of cell adhesion during early phases of ES cell differentiation must be studied in depth to understand the nature of the ground state.

Telencephalic differentiation occurred in a large population of SFEB-induced neural precursors (Fig. 3d). The temporal order of marker expression (Sox1 starting on day 3, Six3 starting on days 4–5, and Bf1 starting on day 6) in SFEB is again reminiscent of that in the embry-onic CNS10,22,23. Notably, the addition of Dkk1 with or without LeftyA during the first 5 d increased the percentage of Bf1+ cells to 35% of total SFEB cells (Fig. 3f). As some parts of the telencephalic region are Bf1– (that is, mature cells in the mantle layer and the dorsal-most pal-lium; Fig. 3b), the actual percentage of total telencephalic cells could be even higher in SFEB-treated ES cells. This is the first quantitative report of such efficient in vitro induction of telencephalic precursors from ES cells (qualitative detection of Bf1 induction has also been reported42,43).

These observations indicate that ES cell–derived neural precur-sors tend to form the rostral-most tissues in the absence of caudal-izing signals. In amphibian research on early neural patterning, a similar concept is widely accepted as the ‘two-signal model’, first pos-tulated by Nieuwkoop44. In this model, the ectoderm differentiates into archencephalic (rostral-most) neural tissues upon initial neural induction (the activation step); caudal neural tissues are subsequently induced by the next ‘caudalizing’ signals (the transformation step). The present study suggests that a similar principle is largely applicable to neural differentiation of mammalian ES cells.

The subregional specifications of telencephalic tissues can be reproduced in the in vitro system of SFEB using embryologically- relevant patterning factors at the right time (Supplementary Fig. 7). Three-quarters of Bf1+ cells showed the pallial phenotype (Pax6+, Emx1+) upon late Wnt3a treatment (Fig. 4), whereas two-thirds of Bf1+ cells treated with Shh expressed the subpallial markers Nkx2.1+ and Islet1+ (Fig. 5).

In addition to ventral telencephalic differentiation, our preliminary studies showed that SFEB culture with Shh promotes ventral diencephalic differentiation. Nkx2.1+ Bf1– cells are found in the embryonic ventral diencephalon21,29. In SFEB-induced cells, Shh treatment increased the percentages of both Nkx2.1+ Bf1+ and Nkx2.1+ Bf1– (corresponding to ventral diencephalic tissues) populations (Supplementary Fig. 8). The relative proportion of Bf1– cells in the Nkx2.1+ population increased moderately when neural precursors induced by SFEB culture with Shh (30 nM) were treated with Wnt3a or Nodal during days 3–5 (25.6 ± 5.8% for the control and 47.3 ± 2.9% with 50 ng ml–1 Wnt3a, P < 0.01 versus the control; 46.1 ± 3.6% with 5 µg ml–1 Nodal, P < 0.01

Nkx2.1 /Islet1

expr

essi

ng in

dica

ted

mar

ker

b

b'E12.5

Shh

Islet1/Bf1

0

5

10

15

20

25

30

35

Bf1

+ c

ells

exp

ress

ing

Isle

t1

3 30 30050 5

Wnt3a(ng/ml)

Shh (nM)

Con

trol

e

a

Control 3 30 300Shh concentration (nM)

b

0

50

30

40

10

20

Pax6Gsh2Nkx2.1

Per

cent

age

of B

f1+ c

ells

Nkx2.1 /Islet1 /Bf1

Per

cent

age

of

b c d

Figure 5 Expression of ventral telencephalic markers by Wnt and Shh signals. (a) Percentages of SFEB-induced Bf1+ cells expressing dorsal and ventral telencephalic markers in the presence of Shh (3 nM, 30 nM or 300 nM, days 4–10). (b,c) Expression of Nkx2.1 (red), Islet1 (green) and Bf1 (blue) in the coronal section across the ventral telencephalic region of the E12.5 mouse embryo. Higher-magnification view of the boxed area is shown in c (Nkx2.1 and Islet1). (d) Islet1 (red) and Bf1 (green) in ES cell–derived cells induced by SFEB and 300 nM Shh (days 4–10). (e) Percentages of Islet1+ cells among SFEB-induced Bf1+ cells in the presence of Wnt3a (5 or 50 ng ml–1, days 6–10) or Shh (3, 30 or 300 nM, days 4–10).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

versus the control). A small portion of SFEB-induced Nkx2.1+ cells coexpressed Nkx2.2 (<17%; Supplementary Fig. 8). This is consistent with the partial overlap of two markers in the ventral diencephalon.

It remains unclear how individual cells acquire their respective positional identity during ES cell differentiation. Further investigation is needed to determine how heterogeneous positional information is assigned within the apparently homogenous ES cell aggregate. For instance, in the absence of strong patterning signals, does a neural precursor cell acquire final positional identities stochastically within certain limitations? Also, it will be important to search for more complete sets of region-specific markers, as the number of bona fide markers suitable for in vitro analysis is limited at present (as exemplified in Fig. 3c,d) and insufficient to cover all CNS regions.

The present and previous studies have shown that two complementary systems, SFEB and SDIA, can be used to generate a wide range of rostral and caudal brain tissues (from telencephalic to brainstem cells; Fig. 3c,d). The SFEB method enables investigation of the molecular and cellular regulation of telencephalic development using in vitro ES cell culture. Other topics in this area are in vitro neurogenesis of the basal telencephalic nuclei, such as the striatum (derivatives of Gsh2+ Bf1+ cells) and the basal nucleus of Meynert (from Islet1+ Bf1+ cells). Our preliminary study has shown that SFEB- cultured ES cells treated with Dkk1, LeftyA and Shh express Gsh2 (Fig. 5) and the striatal mark-ers45,46 Foxp1 and Zfp503 (also called Nolz1) (Supplementary Fig. 8). After long-term culture, these cells generated mature neurons, such as functional GABAergic neurons (GAD65/67+ and GABA+ cells in 20–30% of TuJ1+ neurons on day 20; HPLC analysis showed the release of ∼2 nmol of GABA from 106 ES-derived cells in response to high K+; Supplementary Fig. 8). To understand the in vivo transplantability and functionality of the derivatives of SFEB-induced telencephalic precursors, a future challenge will be to use FACS to purify subregion-specific intermediate progenitors using ES cells knocked in with multicolored GFP genes at marker loci.

METHODSCell culture and treatment with soluble factors. Mouse ES cells (EB5), Sox1-GFP ES cells (46C) and PA6 cells were maintained and used for the SDIA culture as described4,6. Immunostaining showed that ∼95% of ES cells used in this study expressed E-cadherin, Oct3/4 and Nanog during maintenance culture.

'Differentiation medium' was prepared as follows: G-MEM supplemented with 5% KSR (up to 10% can be used; Gibco), 2 mM glutamine, 1 mM pyruvate, 0.1 mM nonessential amino acids, and 0.1 mM 2-mercaptoethanol (2-ME) (G-MEM can be replaced with αMEM). For the SFEB culture, ES cells were dissociated (0.25% trypsin–EDTA) to single cells, and 5 × 104 cells per 1 ml differentiation medium were seeded into bacterial-grade dishes (10 ml). ES cell aggregates were generated spontaneously in a suspension culture within 1 d. When 5 × 105 cells were initially seeded (4.9 ± 1.5) × 105 cells were found in the floating aggregates after a 24-h culture. Cell aggregates were cultured in differentiation medium unless stated otherwise. The day on which ES cells were seeded to differentiate was defined as differentiation day 0. Medium was changed to fresh differentiation medium (with factors added where applicable) on day 3. For experiments on telencephalic differentiation, ES cell aggregates were replated en bloc on dishes coated with poly-D-lysine, laminin and fibronectin, at a density of 1 × 102 to 2 × 102 aggregates per cm2, on differentiation day 5 of SFEB culture. At differentiation day 7, the medium was changed to GMEM-N2: G-MEM supplemented with N2 (Gibco), 2 mM glutamine, 1 mM pyruvate, 0.1 mM nonessential amino acids and 0.1 mM 2-ME. Recombinant human BMP-4 protein, human Dkk-1 protein, human Lefty-A protein, mouse Nodal protein, mouse Shh-N (25–198) peptide and mouse Wnt-3A protein (R&D Systems) were freshly added at each medium change. All-trans retinoic acid was purchased from Sigma. ECCD1 (Takara) was added to the culture at 0.2 mg ml–1 during days 0–5. Activities of BMPR-Fc and Noggin proteins (R&D) were confirmed by testing their ability to suppress the induction of phosphorylated Smad1, Smad5 and Smad8 by 0.5 nM BMP4 in EB5 cells.

Serum-free media. At the beginning of this study, we tested several serum-free culture media to establish a feeder-free suspension culture system. For instance, ES cell aggregates did not survive well in SFOIII47, SFOIII + Chemically Defined Lipid Concentrate (Gibco), Neurobasal + N2 + B27 (Gibco) + 2-ME, DMEM/F12 + N2 + 2-ME or differentiation medium (above) without KSR. In addition to KSR-containing differentiation medium, Chemically Defined Medium (containing 1 U ml–1 LIF)48 and DMEM/F12 + N2 + B27 + 2-ME6 supported both the growth of floating ES cell aggregates and neural differentiation. All three media required thiol-containing substances such as 2-ME or thioglycerol for support activities of cell growth and differentiation. For telencephalic induction, KSR-containing differentiation medium proved to be the most effective (at least four times more effective than the other two media) and was used throughout this study.

Immunocytochemistry, generation of antibodies and statistical quantification. For the observation of immunostained cells in colonies and aggregates, confocal microscopy was used unless stated otherwise.

Polyclonal antiserum against Bf1 was obtained by immunizing rabbits with the synthetic peptide CTHQNQGSSSNPLIH (corresponding to the C-terminal 14 residues of mouse Bf1 protein) and was used after affinity purification on a Bf1 peptide–Sepharose column. Polyclonal antiserum against Gsh2 was obtained by immunizing guinea pigs with the synthetic peptide CANEDKEISPL (corre-sponding the C-terminal 10 residues of mouse Gsh2 protein) and was used after affinity purification. Polyclonal antiserum against mouse Emx1 was obtained by immunizing guinea pigs with the synthetic peptide SFFSAQHRDPLHC and was used after protein A purification. Polyclonal antiserum against mouse Nkx2.1 was obtained by immunizing rabbits with a mixture of three peptides (SMSPKHTTPFSVSC, GNMSELPPYQDTMC and GPGWYGANPDPRFC) and was used after protein A purification. Primary antibody dilutions were as fol-lows: anti-Bf1, 1:1,000; anti-Gsh2, 1:1,000; anti-Emx1, 1:1,000; and anti-Nkx2.1 (rabbit-raised polyclonal), 1:1,000. Immunoreactivity of each antibody was con-firmed with appropriate embryonic CNS tissues as positive and negative controls under the same conditions. The specificity was also confirmed by observing the loss of staining signals after pre-immunoabsorption with corresponding anti-gen peptides (the example of Bf1 antibody is shown in Supplementary Fig. 5). Immunohistochemistry with the Bf1 antibody used in this study could detect the expression of Bf1 proteins in the telencephalon at E9.5, but only faintly at E9.0 (by in situ hybridization, Bf1 expression is detectable at E8.5).

The following commercial antibodies were used: anti-Hoxb4 rat monoclonal (I12) at 1:200, anti-Islet1/2 mouse monoclonal (39.4D5) at 1:200, anti-Nkx2.2 mouse monoclonal (74.5A5) at 1:40, anti-Otx1 mouse monoclonal (Otx–5F5) at 1:1,000, anti-Pax6 mouse monoclonal at 1:200 and anti-RC2 mouse monoclonal at 1:1,000 (Developmental Studies Hybridoma Bank); anti-TH sheep polyclonal at 1:100, anti-GAD65 mouse monoclonal at 1:1,000 and anti-GAD67 mouse monoclonal at 1:2,000 (Chemicon); anti-Nkx2.1 mouse monoclonal (8G7G3/1) at 1:100 and anti-Serotonin rabbit polyclonal at 1:50 (Zymed); anti-Mash1 mouse monoclonal (24B72D11.1) at 1:20 and anti-Nestin rat monoclonal (Rat401) at 1:300 (BD Pharmingen); anti-class III β-tubulin rabbit polyclonal at 1:600, anti-class III β-tubulin mouse monoclonal (TuJ1) at 1:300 and anti-Pax2 rabbit polyclonal at 1:200 (Babco); anti-GAD65/67 rabbit polyclonal at 1:5,000 and anti-GABA rabbit polyclonal at 1:10,000 (Sigma); anti–E-cadherin rat monoclonal (ECCD2) at 1:50 (Takara); and anti-GFP rabbit polyclonal at 1:150 (MBL).

For statistical analysis, 100–200 colonies were examined in each experiment, and each experiment was performed at least three times. P values for statistical significance in two-sample comparison to the control (Student’s t-test) and in multiple comparisons to the control (Dunnett test) are described in the corre-sponding figure legends. Values shown on graphs represent the mean ± s.d.

RT-PCR analysis and western blot. RT-PCR was performed as described3. The other primers used in this study were as follows: Oct3/4 (forward: 5′-AGGGATGGCATACTGTGGAC-3′, reverse: 5′-CCTGGGAAAGGTGTCCTGTA-3′); Zfp42 (also called Rex1) (forward: 5′-GGCCAGTCCAGAATACCAGA-3′, reverse: 5′-TTGAAATCCAGGGAGAAACG-3′); Fgf5 (forward: 5′-AAAGTCAATGGCTCCCACGAA-3′, reverse: 5′-CTTCAGTCTGTACTTCACTGG-3′; T/Brachyury (forward: 5′-CCGGTGCTG-AAGGTAAATGT-3′, reverse: 5′-TGACCGGTGGTTCCTTAGAG-3′); Sox1

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

(forward: 5′-GCACACAGCGTTTTCTCGG-3′, reverse: ACATCCGAC-TCCTCTTCCC); Nodal (forward: 5′-ACGTTCACCGTCATTCCTTC-3′, reverse: 5′-GCTCTGGATGTAGGCATGGT-3′); Tdgf1/Cripto (forward: 5′-TTTTACGAGCCGTCGAAGAT-3′, reverse: 5′-GCGCCAGCTAGCATAAAAGT-3′); Wnt1 (forward: 5′-TGCACCTGCGACTACCGGCG-3′, reverse: 5′-GTGCGCGGGGTCTTCGGGCT-3′); Wnt2b (forward: 5′-TGTGTC-AACGCTACCCACAC-3′, reverse: 5′-AGGGAACCTGAAGCCTTGTC-3′); Wnt3 (forward: 5′-ATGGAGCCCCACCTGCT-3′, reverse: 5′-TGCAGGTGTGCACATCGTAG-3′); Wnt3a (forward: 5′-GTAGCTTTCGCAGTGACACG-3′, reverse: 5′-CTGACGTAGCAGCACCAATG-3′); Wnt10b (forward: 5′-GGATGG-AAGGGTAGTGGTGA-3′, reverse: 5′-TAACAGCACCAGTGGAAACG-3′); Foxp1 (forward: 5′-CTGGAAAACAGCCGAAAGAG-3′, reverse: 5′-GGCAGCTTTGGGTTCTGTAG-3′); Zfp503 (Nolz1) (forward: 5′-CAAGAA-AGATCCGGAAGCTG-3′, reverse: 5′-GGTCGTTAGGGAGCATGAAA-3′) and Dlx5 (forward: 5′-TCTCAGGAATCGCCAACTTT-3′, reverse: 5′-CTGGTG-ACTGTGGCGAGTTA-3′).

Antibodies to Smad1, Smad5 and Smad8 and to phospho-Smad1, phospho-Smad5 and phospho-Smad8 for western blot analysis were obtained from Santa Cruz Biotechnology and Cell Signaling Technology, respectively.

Quantification by FACS and assays for apoptosis and proliferation. In all FACS analyses, cells were counted by FACSAria (BD Biosciences) and the data analyzed with FACSDiva software (BD Biosciences). Cells were dissociated to single cells by trypsin-EDTA treatment, incubated in medium for 37 °C for 1 h to recover surface E-cadherin proteins, and then labeled with apophycocyanin (APC)-labeled anti E-cadherin antibody (ECCD2; ref. 47). Dead cells were excluded by gating on forward and side scatter and by staining with propidium iodide or 7-aminoactinomycin D (7-AAD), a fluorescent dye that labels only dead cells. For Oct3/4 staining, cells were dissociated to single cells, fixed in 4% paraformaldehyde and then stained with anti-Oct3/4 antibody (BD Biosciences) and APC-labeled anti-mouse secondary antibody (BD Pharmingen).

For the TUNEL method, cells were dissociated to single cells, fixed in 4% paraformaldehyde, and analyzed for apoptosis by using the MEBSTAIN Apoptosis Kit II (MBL) according to the manufacturer’s instructions. For annexin V staining, cells were dissociated to single cells and stained with annexin V–PE apoptosis detection Kit I (BD Pharmingen) according to the manufacturer’s instructions. Cells were counted using FACS.

Cells in S-phase were analyzed by 30-min labeling with 5 µM BrdU (Nacalai Tesque). Labeled cells were dissociated to single cells, fixed in 4% paraformaldehyde and stained with anti-BrdU antibody (Sigma, 1:50). For phospho-histone H3 staining, cells were dissociated to single cells, fixed in 4% paraformaldehyde and then stained with anti–phospho-histone H3 primary antibody (Cell Signaling Technology, 1:50). In both assays, cells were incubated with APC-labeled anti-mouse secondary antibody (BD Pharmingen) and counted using FACS.


ACKNOWLEDGMENTSWe are grateful to members of the Sasai lab and to T. Era, S. Kuratani, R. Ladher, I. Matsuo, F. Matsuzaki, H. Niwa, T. Watabe, R. Shigemoto and S. Kaneko for invaluable comments and advice on this work; to S. Nishikawa, S.-i. Nishikawa and M. Takeichi for labeled E-cadherin antibody; to A. Smith for Sox1-GFP ES cells and to N. Sasai for advice on antibody production. K.W. is thankful to S. Iwamizu-Watanabe for discussion and constant encouragement. This work was supported by grants-in-aid from the Ministry of Education, Culture, Sports, Science and Technology (Y.S., Y.W.), the Kobe Cluster Project (Y.S.) and the Leading Project (Y.S.).


Received 17 November 2004; accepted 14 January 2005Published online at http://www.nature.com/natureneuroscience/

1. Bain, G., Kitchens, D., Yao, M., Huettner, J.E. & Gottlieb, D.I. Embryonic stem cells express neuronal properties in vitro. Dev. Biol. 168, 342–357 (1995).

2. Wichterle, H., Lieberam, I., Porter, J.A. & Jessell, T.M. Directed differentiation of embryonic stem cells into motor neurons. Cell 110, 385–397 (2002).

3. Mizuseki, K. et al. Generation of neural crest-derived peripheral neurons and floor

plate cells from mouse and primate embryonic stem cells. Proc. Natl. Acad. Sci. USA 100, 5828–5833 (2003).

4. Kawasaki, H. et al. Induction of midbrain dopaminergic neurons from ES cells by stromal cell-derived inducing activity. Neuron 28, 31–40 (2000).

5. Lee, S.H., Lumelsky, N., Studer, L., Auerbach, J.M. & McKay, R.D. Efficient generation of midbrain and hindbrain neurons from mouse embryonic stem cells. Nat. Biotechnol. 18, 675–679 (2000).

6. Ying, Q.L., Stavridis, M., Griffiths, D., Li, M. & Smith, A. Conversion of embryonic stem cells into neuroectodermal precursors in adherent monoculture. Nat. Biotechnol. 21, 183–186 (2003).

7. Evans, M.J. & Kaufman, M.H. Establishment in culture of pluripotential cells from mouse embryos. Nature 292, 154–156 (1981).

8. Martin, G.R. Isolation of a pluripotent cell line from early mouse embryos cultured in medium conditioned by teratocarcinoma stem cells. Proc. Natl. Acad. Sci. USA 78, 7634–7638 (1981).

9. Aubert, J. et al. Screening for mammalian neural genes via fluorescence-activated cell sorter purification of neural precursors from Sox1-gfp knock-in mice. Proc. Natl. Acad. Sci. USA 100, 11836–11841 (2003).

10. Wood, H.B. & Episkopou, V. Comparative expression of the mouse Sox1, Sox2 and Sox3 genes from pre-gastrulation to early somite stages. Mech. Dev. 86, 197–201 (1999).

11. Haub, O. & Goldfarb, M. Expression of the fibroblast growth factor-5 gene in the mouse embryo. Development 112, 397–406 (1991).

12. Hebert, J.M., Boyle, M. & Martin, G.R. mRNA localization studies suggest that murine FGF-5 plays a role in gastrulation. Development 112, 407–415 (1991).

13. Aubert, J., Dunstan, H., Chambers, I. & Smith, A. Functional gene screening in embry-onic stem cells implicates Wnt antagonism in neural differentiation. Nat. Biotechnol. 20, 1240–1245 (2002).

14. Parisi, S. et al. Nodal-dependent Cripto signaling promotes cardiomyogenesis and redi-rects the neural fate of embryonic stem cells. J. Cell Biol. 163, 303–314 (2003).

15. Finley, M.F., Devata, S. & Huettner, J.E. BMP-4 inhibits neural differentiation of murine embryonic stem cells. J. Neurobiol. 40, 271–287 (1999).

16. Glinka, A. et al. Dickkopf-1 is a member of a new family of secreted proteins and functions in head induction. Nature 391, 357–362 (1998).

17. Sakuma, R. et al. Inhibition of Nodal signalling by Lefty mediated through interaction with common receptors and efficient diffusion. Genes Cells 7, 401–412 (2002).

18. Pera, M.F. et al. Regulation of human embryonic stem cell differentiation by BMP-2 and its antagonist noggin. J. Cell Sci. 117, 1269–1280 (2004).

19. Doetschman, T.C., Eistetter, H., Katz, M., Schmidt, W. & Kemler, R. The in vitro development of blastocyst-derived embryonic stem cell lines: formation of visceral yolk sac, blood islands and myocardium. J. Embryol. Exp. Morphol. 87, 27–45 (1985).

20. Duncan, S.A., Nagy, A. & Chan, W. Murine gastrulation requires HNF-4 regulated gene expression in the visceral endoderm: tetraploid rescue of Hnf-4(−/−) embryos. Development 124, 279–287 (1997).

21. Tao, W. & Lai, E. Telencephalon-restricted expression of BF-1, a new member of the HNF-3/fork head gene family, in the developing rat brain. Neuron 8, 957–966 (1992).

22. Xuan, S. et al. Winged helix transcription factor BF-1 is essential for the development of the cerebral hemispheres. Neuron 14, 1141–1152 (1995).

23. Oliver, G. et al. Six3, a murine homologue of the sine oculis gene, demarcates the most anterior border of the developing neural plate and is expressed during eye develop-ment. Development 121, 4045–4055 (1995).

24. Nordstrom, U., Jessell, T.M. & Edlund, T. Progressive induction of caudal neural character by graded Wnt signaling. Nat. Neurosci. 5, 525–532 (2002).

25. Piccolo, S. et al. The head inducer Cerberus is a multifunctional antagonist of Nodal, BMP and Wnt signals. Nature 397, 707–710 (1999).

26. Ding, J. et al. Cripto is required for correct orientation of the anterior-posterior axis in the mouse embryo. Nature 395, 702–707 (1998).

27. Puelles, L. et al. Pallial and subpallial derivatives in the embryonic chick and mouse telencephalon, traced by the expression of the genes Dlx-2, Emx-1, Nkx-2.1, Pax-6, and Tbr-1. J. Comp. Neurol. 424, 409–438 (2000).

28. Hsieh-Li, H.M. et al. Gsh-2, a murine homeobox gene expressed in the developing brain. Mech. Dev. 50, 177–186 (1995).

29. Lazzaro, D., Price, M., de Felice, M. & Di Lauro, R. The transcription factor TTF-1 is expressed at the onset of thyroid and lung morphogenesis and in restricted regions of the foetal brain. Development 113, 1093–1104 (1991).

30. Lee, S.M., Tole, S., Grove, E. & McMahon, A.P. A local Wnt-3a signal is required for development of the mammalian hippocampus. Development 127, 457–467 (2000).

31. Galceran, J., Miyashita-Lin, E.M., Devaney, E., Rubenstein, J.L. & Grosschedl, R. Hippocampus development and generation of dentate gyrus granule cells is regulated by LEF1. Development 127, 469–482 (2000).

32. Gunhaga, L. et al. Specification of dorsal telencephalic character by sequential Wnt and FGF signaling. Nat. Neurosci. 6, 701–707 (2003).

33. Mukhopadhyay, M. et al. Dickkopf1 is required for embryonic head induction and limb morphogenesis in the mouse. Dev. Cell 1, 423–434 (2001).

34. Simeone, A., Acampora, D., Gulisano, M., Stornaiuolo, A. & Boncinelli, E. Nested expression domains of four homeobox genes in developing rostral brain. Nature 358, 687–690 (1992).

35. Ericson, J. et al. Sonic hedgehog induces the differentiation of ventral forebrain neu-rons: a common signal for ventral patterning within the neural tube. Cell 81, 747–756 (1995).

36. Chiang, C. et al. Cyclopia and defective axial patterning in mice lacking Sonic hedge-hog gene function. Nature 383, 407–413 (1996).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

37. Rallu, M. et al. Dorsoventral patterning is established in the telencephalon of mutants lacking both Gli3 and Hedgehog signaling. Development 129, 4963–4974 (2002).

38. Wang, H.F. & Liu, F.C. Developmental restriction of the LIM homeodomain transcrip-tion factor Islet-1 expression to cholinergic neurons in the rat striatum. Neuroscience 103, 999–1016 (2001).

39. Munoz-Sanjuan, I. & Brivanlou, A.H. Neural induction, the default model and embry-onic stem cells. Nat. Rev. Neurosci. 3, 271–280 (2002).

40. Tropepe, V. et al. Direct neural fate specification from embryonic stem cells: a primitive mammalian neural stem cell stage acquired through a default mechanism. Neuron 30, 65–78 (2001).

41. Shirayoshi, Y., Okada, T.S. & Takeichi, M. The calcium-dependent cell-cell adhesion system regulates inner cell mass formation and cell surface polarization in early mouse development. Cell 35, 631–638 (1983).

42. Kim, J.H. et al. Dopamine neurons derived from embryonic stem cells function in an animal model of Parkinson’s disease. Nature 418, 50–56 (2002).

43. Barberi, T. et al. Neural subtype specification of fertilization and nuclear transfer

embryonic stem cells and application in parkinsonian mice. Nat. Biotechnol. 21, 1200–1207 (2003).

44. Nieuwkoop, P.D. Activation and organization of the central nervous system in amphib-ians. J. Exp. Zool. 120, 1–81 (1952).

45. Takahashi, K., Liu, F.C., Hirokawa, K. & Takahashi, H. Expression of Foxp2, a gene involved in speech and language, in the developing and adult striatum. J. Neurosci. Res. 73, 61–72 (2003).

46. Chang, C.W. et al. Identification of a developmentally regulated striatum-enriched zinc-finger gene, Nolz-1, in the mammalian brain. Proc. Natl. Acad. Sci. USA 101, 2613–2618 (2004).

47. Hirashima, M. et al. A chemically defined culture of VEGFR2+ cells derived from embryonic stem cells shows the role of VEGFR1 in tuning the threshold for VEGF in developing endothelial cells. Blood 101, 2261–2267 (2003).

48. Johansson, B.M. & Wiles, M.V. Evidence for involvement of activin A and bone mor-phogenetic protein 4 in mammalian mesoderm and hematopoietic development. Mol. Cell. Biol. 15, 141–151 (1995).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Sonic hedgehog guides commissural axons along the longitudinal axis of the spinal cordDimitris Bourikas1,4, Vladimir Pekarik3,4, Thomas Baeriswyl1, Åsa Grunditz2, Rejina Sadhu1, Michele Nardó1 & Esther T Stoeckli1

Dorsal commissural axons in the developing spinal cord cross the floor plate, then turn rostrally and grow along the longitudinal axis, close to the floor plate. We used a subtractive hybridization approach to identify guidance cues responsible for the rostral turn in chicken embryos. One of the candidates was the morphogen Sonic hedgehog (Shh). Silencing of the gene SHH (which encodes Shh) by in ovo RNAi during commissural axon navigation demonstrated a repulsive role in post-commissural axon guidance. This effect of Shh was not mediated by Patched (Ptc) and Smoothened (Smo), the receptors that mediate effects of Shh in morphogenesis and commissural axon growth toward the floor plate. Rather, functional in vivo studies showed that the repulsive effect of Shh on postcommissural axons was mediated by Hedgehog interacting protein (Hip).

During the development of neural circuits, growing axons respond to both attractive and repulsive guidance cues to navigate to their target cells, where they establish synaptic contacts. Guidance cues act either over distance to outline the target, or locally at choice points to specify the pathway taken by extending axons1,2. One of the best-understood model systems for axonal pathfinding is the navigation of axons from commissural neurons located in the dorsal spinal cord3,4. These axons extend toward the floor plate in response to the chemoat-tractant netrin-1 (refs. 3–6). Recently, the morphogen Shh has been identified as an additional chemoattractant for commissural axons7. In vivo experiments demonstrated a requirement for the cell adhesion molecules axonin-1, expressed on commissural axons, and NrCAM, expressed on floor-plate cells, for commissural axons to enter the floor plate and cross the midline8,9. F-spondin has been shown to restrict the turning angle of commissural axons into the longitudinal axis at the contralateral floor-plate border without affecting the direction of the turn10. Guidance cues directing postcommissural axons rostrally instead of caudally remained elusive until recently, when the presence of a wnt4 gradient was demonstrated along the longitudinal axis of the spinal cord, attracting postcommissural axons toward higher Wnt4 concentrations at more rostral levels11.

In contrast to the candidate-based approach taken to describe the role of Wnt4 (ref. 11), we carried out a screen based on subtractive hybridization to identify guidance cues for postcommissural axons. In the screen, we focused on genes expressed in the floor plate at the time when commissural axons turn into the longitudinal axis. The floor plate is the source of the morphogen Shh that is involved in pattern-ing the spinal cord during early stages of development12–15. At later

stages, the floor plate is the source of trophic and tropic factors, such as the chemoattractant netrin-1 (refs. 5,6). Because axons turn into the longitudinal axis in close contact with the floor-plate surface, a role of floor plate–associated cues in directing c ommissural axons along the longitudinal axis of the spinal cord seemed very likely.

Using in vitro experiments and in ovo RNAi, a loss-of-function approach that allowed us to silence candidate genes in a temporally and spatially controlled manner in vivo16–18, we identified Shh as a guidance cue for postcommissural axons. Notably, the effect of Shh was not mediated by its well-characterized receptors Ptc and Smo but rather by Hip that is expressed transiently by commissural neurons.

RESULTSA candidate guidance cue for postcommissural axonsWe decided to screen for differentially expressed floor-plate genes as candidate guidance cues directing commissural axons along the longitudinal axis of the spinal cord based on two observations. First, commissural axons turn in close contact with the floor-plate border (Fig. 1). Second, commissural axons do not turn into the longitudinal axis in the absence of a floor plate in mouse19,20, chick21, zebrafish22 and frog23. At the lumbosacral level of the spinal cord, in which we carried out all our analyses, dorsal commissural axons have turned into the longitudinal axis at stage 25, and, therefore, concentrations of the putative guidance cue(s) were expected to be the highest. At stage 20, when commissural axons have just started to grow in the dorsal spinal cord and have not yet reached the floor plate, we expected the guidance cue(s) not to be expressed or to be expressed only at low levels. Because of the limited amount of starting material, we used a

1University of Zurich, Institute of Zoology, Winterthurerstrasse 190, CH-8057 Zurich, Switzerland. 2Friedrich Miescher Institute, Maulbeerstrasse 66, CH-4058 Basel, Switzerland. 3Current address: Cardiff School of Biosciences, Biomedical Building, Museum Avenue, P.O. Box 911, Cardiff CF10 3US, UK. 4These authors contributed equally to this work. Correspondence should be addressed to E.T.S. ([email protected]).


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

PCR-based approach for subtractive hybridization producing cDNA fragments with a length between several hundred and more than 1,000 base pairs that could be used directly to synthesize digoxigenin (DIG)-labeled in situ probes.

For further analysis, we selected those clones whose expression was predominantly restricted to the floor plate during the relevant time window, stages 23–26. One clone was identified as containing the gene encoding F-spondin10. For several clones we did not find any matches in the NCBI data bank using the BLAST program, presumably because we had sequence information only from the 3′-untranslated region (UTR). But full-length sequences of our candidate clones were not required for further analysis, as the isolated cDNA fragments could be used not only for the generation in situ probes but directly for functional analyses using in ovo RNAi17.

Functional analysis of candidates by in ovo RNAiThe injection of dsRNA derived from the cDNA fragments of candidate genes into the central canal of the spinal cord, in combina-tion with in ovo electroporation, resulted in specific downregulation of the targeted genes16–18. Using this assay, we found that one of our candidate genes indeed interfered with the decision of postcommis-

sural axons to turn rostrally (Fig. 2). We did not detect any effects on axon growth and pathfinding toward and into the floor plate area; however, most axons lingered at the exit site from the floor plate. The majority of the growth cones did not have the bias to grow

in the rostral direction or even pointed caudally. Relatively few axons extended along the longitudinal axis. Among those, many errone-ously turned caudally instead of rostrally. A total of 29 embryos were treated with dsRNA from the identified clone. On average, nine injec-tion sites were analyzed in each spinal cord. An abnormal phenotype (stalling with lack of rostral bias, caudal turn of axons, or both) was seen in 90% of the embryos and at least 78% of the injection sites per embryo. In age-matched control embryos (n = 27), commissural axons did not linger at the floor-plate exit site (average of nine injec-tion sites per spinal cord). Even when commissural axons were ana-lyzed in younger control embryos, axons never showed the lingering morphology seen in experimental embryos that is characterized by enlarged growth cones and/or orientation in any direction other than rostral. Therefore, we concluded that we had identified a guidance cue providing an instructive signal for the growth along the longitudinal axis. In its absence, commissural axons either stalled at the floor-plate exit site or randomly chose in which direction to turn.

Shh identified as guidance cue for postcommissural axonsTo identify the candidate gene that was shown to interfere with the rostral turn of postcommissural axons, we used the cDNA fragment from our screen as a probe to search a cDNA library derived from E14 chicken brain. The resulting cDNA was sequenced and found to encode Shh.

The role of Shh as guidance cue for postcommissural axons was confirmed by in ovo RNAi with a second, nonoverlapping fragment

Figure 1 Commissural axons turn into the longitudinal axis of the spinal cord in close contact with the contralateral floor-plate border. (a) Spinal cords were dissected and cut along the roof plate. (b) The trajectory of commissural axons in a control embryo was visualized by application of Fast DiI into the area of the cell body. The floor plate is indicated by dashed lines in b. Bar, 100 µm.

Figure 2 Shh directs postcommissural axons rostrally. (a–d) Images of four chick embryos in which SHH function was abolished by in ovo RNAi. The phenotype reflecting the lack of SHH function was found with two independent fragments of SHH that were used for dsRNA production. One fragment was derived from the 3′-UTR of SHH (a and b) and the other from the fragment encoding the N-terminal part of Shh (c and d). Axons stalled at the contralateral floor-plate border or even turned caudally (arrows in a and c). Axons showed no difference in their growth toward, into, and across the floor plate as compared to controls (compare to Fig. 1b), but they did not turn rostrally along the contralateral floor-plate border. The majority of the axons stalled at the floor-plate exit site (a,b,d). Growth cones clearly lacked the rostral bias that was seen in control embryos analyzed at earlier stages (data not shown) and explored movement in all possible directions (b). In some cases, axons initially turned caudally but corrected their pathway by forming loops (arrowheads in d). Corrections of initial pathway choices were never seen in control embryos, in which axons were already biased in the rostral direction upon floor-plate exit. The floor plate is indicated by dashed lines. Rostral direction is toward the top in all panels. Bar, 100 µm in a,c, 30 µm in b,d.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

of dsRNA that was derived from the N-terminal part of the SHH cDNA (Fig. 2c,d). The phenotypes obtained by in ovo RNAi using the two different dsRNA fragments were indistinguishable with respect to both quality and penetrance. As shown before with the dsRNA derived from the 3′-UTR of SHH, we injected embryos at stages 18–19 (n = 10) and analyzed them at stages 25–26. Eight of the ten embryos showed the abnormal phenotype at 92.5% of the injection sites (range, 78–100%; average, ten injection sites per spinal cord).

Blocking Shh confirms its role in axon guidanceAs a different means to induce SHH loss-of-function phenotypes, we used function-blocking antibodies. In contrast to previous in vivo assays

at the protein level8,9,24, we did not only inject purified antibodies but also grafted hybridoma cells (clone 5E1) directly into the central canal of the spinal cord. The interference with SHH function at the protein level resulted in the same phenotype that was observed after silencing SHH by in ovo RNAi (Supplementary Fig. 1 online). Commissural axons were no longer instructed to turn rostrally after midline crossing and therefore stalled at the floor-plate exit site or chose randomly to grow either rostrally or caudally. The phenotype was observed in eight of ten embryos injected with 5E1 antibodies (average of eight injection sites (range, 5–10), 90% showing the indicated phenotype). Sixteen embryos received a graft of 5E1 hybridoma cells. As a control, hybridoma cells producing an antibody against c-Myc (9E10 cells) or against P0, an epitope not expressed in the spinal cord (1E8 cells), were used (n = 5 and 4 embryos, respectively). None of the control embryos showed either stalling or caudal turns at any of the injection sites (9 or 10 injection sites per spinal cord).

Shh has a direct effect on postcommissural axonsShh is well known for its effect on spinal cord patterning13,25,26. It acts in a graded manner to establish different populations of neural progenitors, as defined by the expression of homeodomain transcription factors that are either repressed (class I) or induced (class II) by Shh.

To exclude the possibility that the patterning of the spinal cord was changed by our interference with SHH function starting at stage 18–19, we analyzed the expression pattern of sample class I and class II genes and compared them to control embryos. We did not detect any differ-ences in spinal cord patterning with respect to PAX7 (class I)25 or ISL1 and NKX2.2 (class II)27 (Supplementary Fig. 2 online). Therefore, we concluded that SHH did not affect commissural axon guidance indi-rectly by changing cell differentiation or the patterning of the spinal cord. However, downregulation of SHH in an earlier time window (before stage 14) using the same dsRNA fragment did result in a decrease (at lower tho-racic levels; data not shown) or even in the absence of ISL1 and NKX2.2 expression (at lumbosacral levels; Supplementary Fig. 3 online). When SHH was downregulated by in ovo RNAi at stage 8, even HNF3β, a marker of floor-plate cells known to be a target of Shh signaling, was reduced, in line with the notion that Shh has a role in floor plate induction as well as patterning of the spinal cord along the dorsoventral axis25.

Shh’s effect on axon guidance not mediated by Ptc and SmoTo obtain further evidence for a direct effect of Shh on commissural axon guidance, we tried to identify its receptor on commissural neu-rons. In both invertebrates and vertebrates Ptc and Smo have been identified as coreceptors for Shh, mediating its inducing activities28,29. Upon binding of Shh, Ptc releases and derepresses Smo30,31

, which, in turn, is responsible for signaling32.

To test whether Ptc and Smo were mediating Shh’s effect on axon guidance, we looked at their expression pattern in the developing chicken spinal cord (Fig. 3). Throughout the relevant time window

Figure 3 Neither Ptc nor Smo are expressed by commissural neurons when axons turn into the longitudinal axis. (a–c) Transverse sections of stage 23 spinal cords shown at low magnification. (d–f) Transverse sections at high magnification. Sections were used for in situ hybridization to demonstrate expression of Axonin-1 (a,d), Smo (b,d) and Ptc (c,f). Commissural axons at the lumbosacral level of the spinal cord cross the floor plate at stage 23. By stage 24 they have reached the contralateral floor-plate border and turn rostrally into the longitudinal axis. During that time commissural neurons in the dorsolateral spinal cord expressed axonin-1 (arrow in a,d) but they did not express Smo (open arrow in b,e) or Ptc (open arrow in c,f). The expression domains of both Smo and Ptc retracted to the ventricular zone. Bar, 200 µm in a–c, 100 µm in d–f.

Figure 4 Hip is expressed transiently by commissural neurons during the time when their axons turn into the longitudinal axis. (a–c) Hip expression was probed by in situ hybridization in transverse spinal cord sections at stage 23 (a), stage 24 (b) and stage 26 (c). At stage 23, when commissural axons are crossing the floor plate, Hip was not expressed by commissural neurons (open arrow in a). At this stage, its expression in the spinal cord was restricted to two areas of the ventricular zone. At stage 24, when commissural axons have reached the contralateral border of the floor plate and turned into the longitudinal axis, commissural neurons transiently expressed Hip (arrow in b). At stage 26, when commissural axons have extended along the longitudinal axis for some distance, Hip was downregulated to very low levels (open arrow in c). Bar, 200 µm in a, 300 µm in b,c.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

during which commissural axons cross the midline and turn ros-trally along the contralateral floor-plate border, neither PTC nor SMO are expressed by commissural neurons, suggesting that they are not required for SHH function in postcommissural axon guidance.

Because expression levels of Ptc are influenced by Shh28,32, we also compared PTC expression by in situ hybridization between control embryos and embryos treated with SHH dsRNA (Supplementary Fig. 4 online). No changes in PTC or SMO expression in commissural neurons were detected at stage 25.

Additional evidence for a receptor other than the complex of Ptc and Smo was obtained in functional experiments using cyclopamine, a small molecular inhibitor of Smo33. The injection of cyclopamine between stages 19 and 20 did not cause any changes in the turning behavior of commissural axons after midline crossing (n = 16; data not shown). Furthermore, downregulation of SMO by in ovo RNAi did not interfere with commissural axon guidance (n = 16 embryos; Supplementary Fig. 4 online). Therefore, we concluded that the effect

of Shh on postcommissural axon guidance was not mediated by Ptc and Smo.

Shh’s effect on postcommissural axons is mediated by HipVertebrates, unlike invertebrates, express an additional receptor for Shh, Hip. Hip is a transmembrane protein that has been shown to bind directly to all vertebrate hedgehog pro-teins: Indian, Desert and Sonic hedghogs34. Because its distribution as well as functional evidence excluded Ptc as the Shh receptor involved in axon guidance, we analyzed the

expression of HIP during the time window of commissural axon navi-gation (Fig. 4). HIP was regulated very dynamically and expressed by relatively few cells at any given time. Commissural neurons express HIP very briefly at stage 24, which is the time point at which axons have reached the contralateral floor-plate border and turn into the longitu-dinal axis. The expression of HIP in the spinal cord was not changed in response to silencing of SHH (Supplementary Fig. 5 online).

Direct evidence for an involvement of Hip as mediator of the Shh signal was found by in ovo RNAi. Perturbation of HIP function in com-missural axons resulted in the same turning phenotype as seen after downregulation of SHH (Fig. 5; n = 19 embryos, phenotype at 95% of the injection sites). Thus, we concluded that Hip was the receptor that mediated the effect of Shh on postcommissural axons.

Graded SHH expression suggests a repulsive mechanismA graded distribution of a guidance cue could explain the rostral turn of postcommissural axons at the lumbosacral level of the embryonic

Figure 5 Hip is the receptor that mediates the effect of Shh on postcommissural axons. (a–d) Images of open-book preparations from four different embryos in which HIP function was abolished by in ovo RNAi. Loss of HIP function resulted in the same phenotype as loss of SHH function (compare to Fig. 2). Commissural axons did not turn rostrally along the contralateral floor-plate border. Most of them stalled at the floor-plate exit site (a) but some of them grew caudally (arrows in b–d). As seen after perturbation of SHH function, some axons corrected their aberrant initial pathway to grow rostrally by forming loops (arrowheads in d). Loop formation and pathway corrections were never observed in control-injected embryos (compare Fig. 1b). Rostral is toward the top in all panels. The floor plate is indicated by dashed lines in a–c. Bar, 100 µm in a–c, 50 µm in d.

Figure 6 SHH is expressed in a gradient along the rostro-caudal axis of the lumbosacral spinal cord. (a,b) In situ hybridization reveals a graded expression of SHH in an open-book preparation of a stage 25 spinal cord with higher levels caudally. High (a) and low (b) magnification of the same spinal cord is shown. (c) In situ hybridization in open-book preparation with corresponding sense probe. Rostral is at the top of all panels. Only the lumbosacral level of the spinal cord is shown. The expression of Shh in open-book preparations is consistent with the expression of SHH in transverse sections (Supplementary Fig. 6 online). Bar, 500 µm.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

chicken spinal cord. Depending on the mechanism, higher expression of an attractive cue would be expected rostrally, whereas a repellent cue should be expressed at higher levels caudally. As shown in open-book preparations, the expression of SHH was higher caudally (Fig. 6 and Supplementary Fig. 6 online), suggesting that Shh was providing a repellent signal for commissural axons that was mediated by Hip.

SHH gain-of-function consistent with a repulsive roleTo provide evidence for a repulsive activity of Shh on postcommis-sural axons, we used in ovo electroporation to selectively express SHH in the thoracic and upper lumbosacral levels on one side of the spinal cord (n = 23 embryos; Fig. 7). In areas with high Shh expression, commissural axons mostly did not leave the floor plate and stalled at the exit site (Fig. 7e,f). Some axons encountering a reversed gradient upon floor-plate exit—that is, axons turning slightly caudally of the electroporated area—showed the expected pathfinding errors: they stalled or turned caudally to avoid high concentrations of Shh (n = 20; Fig. 7g,h). As a control, commis-sural axons ipsilateral to the ectopic SHH expression were traced. As expected, no effect on their turning behavior was detected, as

they grew through the area of high Shh before they expressed HIP, but encountered a normal Shh gradient upon floor-plate exit, that is, when they expressed HIP (Fig. 7j).

Shh’s repulsive activity confirmed by in vitro experimentsFurther evidence for a repulsive role of Shh on postcommissural axons was found in an explant assay similar to the one described previously for rat tissue35. We cultured spinal cord explants from stage 24–25 embryos in collagen gels in the presence of beads soaked in Shh or control beads soaked in bovine serum albumin (Fig. 8). To identify postcommis-sural axons, we labeled dorsal commissural neurons with the dye DiI in open-book preparations of dissected spinal cords before cutting and culturing explants (Fig. 8a). Unlabeled axons, therefore, represent pre-dominantly motor axons (which grow well in collagen gels), unlabeled postcommissural axons from more ventral areas of the spinal cord, or dorsal postcommissural axons that were not labeled with DiI. No DiI-labeled axons were found contacting Shh-treated beads (Fig. 8b–d). Control beads were neither attractive nor repulsive for postcommissural axons (Fig. 8e,f). Most often (89.5% of the explants; n = 19), axons did not grow out on the explant side facing a Shh bead, or they turned away from the bead as soon as they entered the collagen gel (Fig. 8b,c). In one case the Shh bead was placed 0.9 mm away from the edge of the explant (Fig. 8d). In this situation axons started to grow toward the bead but stalled or turned approximately 300 µm away from the bead.

Taken together, our results demonstrate that postcommissural axons avoid territories with high Shh both in vivo and in vitro, indicating that Shh acts as a repellent for postcommissural axons. The repulsive activity of Shh is mediated by Hip, in contrast to its attractive effect that is mediated by Ptc and Smo7.

DISCUSSIONReceptor switch at the midline for response to ShhUsing in ovo RNAi, a technique recently developed in our lab to specifi-cally silence candidate genes during commissural axon pathfinding17,18, we identified Shh as a guidance cue that directs these axons rostrally along the longitudinal axis of the spinal cord after they have crossed the midline. In contrast to its earlier effects, such as the induction of specific cell populations in the spinal cord14 and the chemoattractive effect7 that were mediated by Ptc and Smo, Shh uses a different receptor, Hip, to mediate its effect on postcommissural axons.

Figure 7 Postcommissural axons avoid high levels of Shh in vivo. (a–d) 5E1 staining of sections taken from different spinal cord levels. Shh was expressed ectopically in a spatially controlled manner in one half of the spinal cord at thoracic (a) and upper lumbosacral levels (b,c). No ectopic Shh expression was found at caudal lumbosacral levels (d). (e,f) Postcommissural axons encountering high levels of Shh stalled and did not leave the floor plate. (g,h) Axons crossing more caudally, where the ectopic Shh expression was decreasing, showed a lack of rostral bias or even turned caudally (arrows). Very few, if any, axons managed to extend rostrally (open arrow, g). (i) Summary of Shh expression and behavior of commissural axons at different spinal cord levels. At upper lumbosacral levels (top), commissural axons encountered high concentrations of ectopically expressed Shh and did not leave the floor-plate. In the transition zone between ectopic and endogenous Shh expression (middle), commissural axons encountered either no or a reversed Shh gradient and, therefore, responded with either stalling or caudal turns. At caudal lumbosacral levels (bottom), pathfinding was not affected, as axons were exposed only to the endogenous Shh gradient. (j) As a control, we traced commissural axons from the electroporated side of the spinal cord. Encountering high Shh before midline crossing, that is, before Hip is expressed, did not change their pathfinding behavior. Floor plate indicated by dashed lines in e–h and j. Rostral direction is toward the top in e–j. Bar, 100 µm in e–h,j.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

A modulatory role of Shh on axon growth during later stages of development was first described for retinal ganglion cell axons36. Further evidence for a role in axon growth was provided in a turning assay using spinal cord explants, which demonstrated that Shh can attract commis-sural axons toward the floor plate7. In both cases, the effect of Shh on axon growth was mediated by Ptc and Smo acting as coreceptors37,38.

Notably, Smo and Ptc are no longer expressed by commissural axons once they have grown into the floor-plate area (Fig. 3). Therefore, it is not surprising that we did not find an effect on commissural axon turning into the longitudinal axis using cyclopamine or in ovo RNAi to downregulate SMO (Supplementary Fig. 4 online). In search for an alternative receptor for Shh, we turned our attention to Hip34. Analysis by in situ hybridization revealed a highly dynamic temporal and spatial control of HIP expression in the developing chicken spinal cord (Fig. 4). HIP is not expressed during the initial phase of commis-sural axon growth toward and across the floor plate but is transiently upregulated when commissural axons reach the contralateral floor-plate border. After the commissural axons turn, when they extend along the contralateral floor-plate border, HIP is downregulated to barely detectable levels (Fig. 4c).

Hip is a type I transmembrane protein but lacks an intracellular domain. Its last 22 amino acids are hydrophobic and have been sug-gested to form the transmembrane domain34. Thus, it is unclear how Hip transmits the Shh signal that results in commissural axons’ turn into the longitudinal axis. In analogy to Ptc and Smo, Hip could represent a com-ponent of a receptor complex that would consist of a Shh binding unit (Hip) and a signal-transmitting unit that has not yet been identified.

Repulsion: a previously unknown activity of ShhThe graded expression of Shh, with high levels in the caudal-most region of the spinal cord, suggests a repulsive signal (Fig. 6a,b and Supplementary Fig. 6 online). Evidence for a repellent activity of Shh on postcommissural axons was obtained in gain-of-function experi-ments in vivo. Because it is unclear how far Shh can diffuse through the tissue39, we decided to express Shh locally in one half of the spinal cord at low thoracic and upper lumbosacral levels. Tracing commissural axons in embryos with a reversed Shh gradient (high levels rostrally and lower levels caudally) resulted in their expected failure to turn rostrally (Fig. 7). The repellent activity of Shh on postcommissural axons was confirmed by in vitro analysis (Fig. 8), where postcommissural axons were repelled by beads soaked in Shh but not by control beads.

Because Shh is a morphogen13 and can act as an attractant for commissural axons7, the temporal and spatial control of gene silencing that is possible with in ovo RNAi becomes extremely important. With experiments that interfered with Shh levels before commissural axons had reached the floor plate, we would not have been able to detect the involvement of Shh in commissural axon guidance. Because we blocked

SHH expression only after stage 18–19, however, either the residual Shh was sufficient to attract commissural axons to the floor plate or, alternatively, in vivo the presence of netrin was sufficient to counterbal-ance the decrease in Shh. Our in situ hybridization analysis indicated that the expression of SHH dropped transiently between stages 19 and 21 (Supplementary Fig. 7 online). Thus, the downregulation of SHH owing to gene silencing initiated at stage 18–19 would prevent the effect of Shh on axon guidance but not on axon attraction.

Morphogens act as guidance cues for postcommissural axonsThe observation that Wnt4 affects postcommissural axon guidance along the longitudinal axis of the spinal cord in rat and mouse11 is of great interest in the context of our results. Shh has been shown to regulate the expression of secreted frizzled-related proteins (Sfrps)40. Sfrps, in turn, are potent inhibitors of the effect of Wnt4 on commissural axon guid-ance11. Thus, it is tempting to speculate that Shh, Sfrps and Wnt4 coop-erate in longitudinal axon guidance. High levels of Shh would induce high levels of Sfrps in caudal segments of the spinal cord. Therefore a Wnt4 gradient with the opposite orientation (high rostral to low caudal levels) would be strengthened by inhibition through Sfrps at more cau-dal levels. Complementary expression patterns and competitive interac-tions of Wnt4 and Sfrp2 have been described in chick41,42 and mouse embryos40. Functional in vivo experiments will be required to test for a cooperation of Shh and Wnt4 in postcommissural axon guidance.

CONCLUSIONSilencing SHH by in ovo RNAi in a temporally and spatially controlled manner demonstrated the involvement of Shh in guidance of postcommissural axons along the longitudinal axis of the spinal cord. Notably, the morphogenic effects of Shh on spinal cord patterning and the chemoattractive effect of Shh on precommissural axons are medi-ated by the Ptc-Smo receptor complex, whereas the repulsive effect of Shh on postcommissural axon guidance along the longitudinal axis of the spinal cord is mediated by Hip.

METHODSSubtractive hybridization screen. To search for candidate guidance cues that would direct postcommissural axons rostrally along the longitudinal axis of the spinal cord, we set up a screen for differentially expressed floor-plate genes (stage 26 versus 20) that was based on subtractive hybridization. For this purpose, we isolated mRNA from floor-plate cells dissected from stage 25–26 embryos43. Stage

20 was the earliest time point at which floor-plate cells could be obtained without contamination by motor neurons. The subtractive hybridization was carried out according to the manufacturer’s recommendation using the PCR-Select cDNA

subtraction kit (Clontech). Using this approach we initially obtained several

hundred clones, from which 400 were randomly picked and subjected to further analysis. We used forward and reverse hybridization to eliminate false-positive

Figure 8 Shh acts as a repellent on postcommissural axons. Postcommissural axons were labeled with DiI in open-book preparation of the lumbosacral level of the spinal cord from stage 24 embryos. (a) Explants were cut as shown (blue dashed line) and cultured in collagen gels. (b–f) Beads soaked either in Shh (b–d) or in serum albumin (e,f) were positioned between 200 and 700 µm away from the explants. Postcommissural axons did not grow from the edge of explants facing Shh beads (b,c). Either axons did not leave the explant or they turned away from the bead as soon as they entered the collagen gel. The bead shown in d was positioned more than 900 µm away from the explant. In that case, axons left the explant also on the side facing the bead, but they either stalled (arrowhead) or turned away from the bead (arrow) at a distance of approximately 320 µm. The insert in d shows that the axons are, in fact, labeled dorsal postcommissural axons. Control beads coated with serum albumin did not affect the growth of postcommissural axons (e,f). Bar, 200 µm in d–f, 400 µm in b,c.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

clones. Finally, we selected 30 clones with a clear difference in expression level between the two stages. From these cDNA fragments we generated DIG-labeled

in situ probes to characterize the expression pattern in the spinal cord.

In ovo RNAi. Loss-of-function phenotypes of SHH, HIP, PTC and SMO were induced by in ovo RNAi16–18 as described earlier by injection of dsRNA followed by electroporation at stages 18–1917 (Supplementary Figs. 6 and 8 online). Embryos were injected with a solution containing dsRNA alone or mixed with a plasmid encoding yellow fluorescent protein (YFP) under the control of the ACTB (β-actin) promoter. Embryos injected and electroporated with the plasmid encoding YFP were used as controls. No difference in the overall development of the embryo or in axon growth and guidance were found between control-injected and nonin-jected embryos, indicating that the manipulation of the embryos did not induce any nonspecific changes. Numbers of embryos (n) given for control groups include only the injected and electroporated embryos; untreated control embryos that were analyzed to exclude an effect of the treatment alone were not counted.

Analysis of commissural axon growth and guidance. The analysis of commissural axon trajectories was carried out as described previously8,44. In brief, embryos were sacrificed between stages 25 and 26. The spinal cord was removed from the embryo, opened at the roof plate (open-book preparation) and fixed. The trajectories of commissural axons at the lumbosacral level of the spinal cord were visualized using the lipophilic dye Fast-DiI (5 mg ml–1 in methanol; Molecular Probes) applied to the cell bodies. Care was taken to specifically label the dorsal population of commissural neurons to avoid any confusion with more ventral populations of commissural neurons that show a different pathfinding behavior.

Cyclopamine, an alkaloid shown to interfere with Shh signaling by binding directly to Smo, was dissolved either in 45% 2-hydroxypropyl-β-cyclodextrin (Sigma; n = 11 embryos) or, because solubility was limited, in DMSO. The stock solution of cyclopamine in DMSO was diluted in PBS to limit DMSO in the injected solution to 2% (n = 5 embryos). Control embryos were injected either with 45% 2-hydroxypropyl-β-cyclodextrin (n = 9 embryos) or 2% DMSO (n = 4 embryos).

Preparation of dsRNA. dsRNA for in ovo RNAi was transcribed in vitro as described before17. The cDNA fragment of SHH obtained in the subtractive hybridization screen was extended with 5′-RACE using the FirstChoice RML-RACE kit (Ambion) according to the manufacturer’s recommendations. Primers used were 5′-TACTCAGACCCTGAAAATGGACG-3′ and 5′-GGTCAGTCATCAGAGTTACGTGC-3′. The PCR product was ligated into the pCRII-TOPO vector (Invitrogen). Sequencing of the resulting clones con-firmed their identity as chicken SHH. The fragment that was originally used to produce dsRNA and in situ probes was from the 3′-UTR of SHH and did not overlap with the open reading frame. To confirm our results we also used a second, nonoverlapping fragment covering the N-terminal sequence of SHH for dsRNA production.

DsRNA and in situ probes for PTC were obtained by RT-PCR using poly(A) RNA isolated from embryonic spinal cords (stages 25–28). The primer used for reverse transcription was 5′-AGCACCAATCCATTGAGAACTCCC-3′. For primary PCR we used 5′-TCGGGAGTTAAACTACACACGGC-3′ and 5′-CTACGGTTCTTATCTCCTATGGC-3′. Nested PCR products were obtained with 5′-ATGTACTCACAACAGAAGCACTCC-3′ and 5′-GCCAGCACCACCACAATGATCCC-3′. The final PCR product was cloned into pCRII–TOPO.

Similarly, we obtained a fragment of the cDNA encoding Smo. The prim-ers used were for RT-PCR 5′-AGTCCATGTGTGGGGACCGAAATC-3′, and for PCR the forward primers 5′-CGCGCTGCCCTACGCGCACACC-3′ and 5′ AGCTGCCCAGTCAGACCCTGTGCC-3′ with the reverse prim-ers 5′-AGCGTCCCCTTCACCCCTAAATCC-3′ and 5′-CCAGTTTCT-TCTCTCCTCCCATCC-3′. For HIP, the primers used were 5′-CAAGAATAC-CTGGCCTTGTAACTC-3′, 5′-TGCGCACACTGCTCACCTCATGCC-3′ and 5′-TCGACATGCTGGGTCACACTTTGC-3′.

Hybridoma cell grafting. The hybridoma cell lines 5E1 (producing a function-blocking anti-Shh antibody), 1E8 and 9E10 were obtained from the Developmental Studies Hybridoma Bank (Univ. Iowa). The cell lines 9E10 and

1E8, producing antibodies against c-Myc and P0, respectively, were used as controls. Cells grown in DMEM/F12 supplemented with 10% FCS were collected and resuspended in PBS for injection into the lumbar spinal cord of stage 20 embryos in ovo. A peroxidase-coupled secondary antibody (rabbit anti–mouse IgG; Cappel) was used to test for antibody production in situ. Comparable numbers of cells from the 1E8 or the 9E10 hybridoma cell lines were grafted in control embryos (n = 9).

Ectopic expression of SHH. For gain-of-function experiments the open reading frame of chicken SHH was cloned into the pMES plasmid (kindly provided by C. Krull) using EcoRI and a plasmid derived from pIRES (Clontech). The CMV promoter of the pIRES plasmid was exchanged for the chicken ACTB promoter, and the IRES sequence was removed. SHH was inserted using NheI and SalI sites. To localize transfected cells, we coinjected a plasmid encoding YFP. In the pMES plasmid, the IRES sequence is followed by EGFP, allowing for direct detection of transfected cells. To achieve a reversal of the endogenous Shh gradient, we used either shorter electrodes (2-mm rather than 4-mm) or placed them more rostrally, such that cells at caudal levels of the lumbosacral region of the spinal cord were not transfected (Fig. 7a–d,i). The density of green cells was used to determine the electroporated area of the spinal cord and to distinguish between sites with high and low Shh expression.

Staining and in situ hybridization. Antibodies recognizing Shh, Pax7, Isl1 and Nkx2.2 were obtained from the Developmental Studies Hybridoma Bank. Control and experimental embryos were sacrificed, fixed in 4% paraformaldehyde and cryoprotected in 25% sucrose. Sections 20 µm thick were stained as described previously using Cy3-conjugated goat anti–mouse IgG (Jackson Laboratories) as secondary antibody24.

The expression of SHH, HIP, SMO and PTC was analyzed by in situ hybridization in transverse spinal cord sections45.

In vitro assay. Open-book preparations of the lumbosacral spinal cord of stage 24–25 embryos43 were cultured in collagen gels essentially as described. However, we used the lipophilic dye Fast-DiI to label dorsal postcommissural axons rather than an anti–axonin-1 antibody, as axonin-1 (the ortholog of rat TAG-1) is expressed also by motoneurons of the lumbosacral spinal cord (Fig. 3a). As described previously for two-dimensional explant cultures of chicken commis-sural axons9, we used a serum-free medium to grow postcommissural axons in collagen gels. Heparin acrylic beads (Sigma) were soaked in either 0.5 mg ml–1 recombinant human Shh (R&D, with 25 mg ml–1 bovine serum albumin as carrier), or 25 mg ml–1 bovine serum albumin (Albumax, Invitrogen) for 1 h. Shh or control beads were positioned 200–700 µm from the explants. Cultures were grown for 24–36 h before inspection on an inverted microscope. Ex plants with only very few axons entering the collagen gels or explants in direct contact with the bead were not included in the analysis. A total of 19 explants with Shh and 20 explants with control beads, respectively, from three independent experiments were analyzed.


ACKNOWLEDGMENTSWe thank C. Krull for the pMES plasmid, M. Mielich for technical assistance and M. Gesemann for critical reading of the manuscript. This work was supported by grants from the Swiss National Science Foundation, the Human Frontier Science Program Organization, the Olga Mayenfisch Stiftung and the Ott Foundation.



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

2. Tessier-Lavigne, M. & Goodman, C.S. The molecular biology of axon guidance. Science 274, 1123–1133 (1996).

3. Stoeckli, E.T. & Landmesser, L.T. Axon guidance at choice points. Curr. Opin. Neurobiol. 8, 73–79 (1998).

4. Kaprielian, Z., Runko, E. & Imondi, R. Axon guidance at the midline choice point. Dev. Dyn. 221, 154–181 (2001).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

5. Leonardo, E.D. et al. Guidance of developing axons by netrin-1 and its receptors. Cold Spring Harb. Symp. Quant. Biol. 62, 467–478 (1997).

6. Livesey, F.J. Netrins and netrin receptors. Cell. Mol. Life Sci. 56, 62–68 (1999).7. Charron, F., Stein, E., Jeong, J., McMahon, A.P. & Tessier-Lavigne, M. The morphogen

sonic hedgehog is an axonal chemoattractant that collaborates with netrin-1 in midline axon guidance. Cell 113, 11–23 (2003).

8. Stoeckli, E.T. & Landmesser, L.T. Axonin-1, Nr-CAM, and Ng-CAM play different roles in the in vivo guidance of chick commissural neurons. Neuron 14, 1165–1179 (1995).

9. Stoeckli, E.T., Sonderegger, P., Pollerberg, G.E. & Landmesser, L.T. Interference with axonin-1 and NrCAM interactions unmasks a floor-plate activity inhibitory for com-missural axons. Neuron 18, 209–221 (1997).

10. Burstyn-Cohen, T. et al. F-Spondin is required for accurate pathfinding of commissural axons at the floor plate. Neuron 23, 233–246 (1999).

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

12. Placzek, M. The role of the notochord and floor plate in inductive interactions. Curr. Opin. Genet. Dev. 5, 499–506 (1995).

13. Briscoe, J. & Ericson, J. The specification of neuronal identity by graded Sonic Hedgehog signalling. Semin. Cell Dev. Biol. 10, 353–362 (1999).

14. Jessell, T.M. Neuronal specification in the spinal cord: inductive signals and trans-criptional codes. Nat. Rev. Genet. 1, 20–29 (2000).

15. Lee, S.K. & Pfaff, S.L. Transcriptional networks regulating neuronal identity in the developing spinal cord. Nat. Neurosci. 4, Suppl. 1183–1191 (2001).

16. Bourikas, D. & Stoeckli, E.T. New tools for gene manipulation in chicken embryos. Oligonucleotides 13, 411–419 (2003).

17. Pekarik, V. et al. Screening for gene function in chicken embryo using RNAi and electroporation. Nat. Biotechnol. 21, 93–96 (2003).

18. Stoeckli, E.T. RNAi in avian embryos. in RNAi: A Guide to Gene Silencing (ed. Hannon, G.J.) 297–312 (Cold Spring Harbor Laboratory Press, Cold Spring Harbor, New York, USA, 2003).

19. Bovolenta, P. & Dodd, J. Perturbation of neuronal differentiation and axon guidance in the spinal cord of mouse embryos lacking a floor plate: analysis of Danforth’s short-tail mutation. Development 113, 625–639 (1991).

20. Matise, M.P., Lustig, M., Sakurai, T., Grumet, M. & Joyner, A.L. Ventral midline cells are required for the local control of commissural axon guidance in the mouse spinal cord. Development 126, 3649–3659 (1999).

21. van Straaten, H.W. & Hekking, J.W. Development of floor plate, neurons and axonal outgrowth pattern in the early spinal cord of the notochord-deficient chick embryo. Anat. Embryol. (Berl.) 184, 55–63 (1991).

22. Greenspoon, S., Patel, C.K., Hashmi, S., Bernhardt, R.R. & Kuwada, J.Y. The noto-chord and floor plate guide growth cones in the zebrafish spinal cord. J. Neurosci. 15, 5956–5965 (1995).

23. Clarke, J.D., Holder, N., Soffe, S.R. & Storm-Mathisen, J. Neuroanatomical and func-tional analysis of neural tube formation in notochordless Xenopus embryos; laterality of the ventral spinal cord is lost. Development 112, 499–516 (1991).

24. Perrin, F.E., Rathjen, F.G. & Stoeckli, E.T. Distinct subpopulations of sensory afferents require F11 or axonin-1 for growth to their target layers within the spinal cord of the chick. Neuron 30, 707–723 (2001).

25. Briscoe, J., Pierani, A., Jessell, T.M. & Ericson, J. A homeodomain protein code

specifies progenitor cell identity and neuronal fate in the ventral neural tube. Cell 101, 435–445 (2000).

26. Wijgerde, M., McMahon, J.A., Rule, M. & McMahon, A.P. A direct requirement for Hedgehog signaling for normal specification of all ventral progenitor domains in the presumptive mammalian spinal cord. Genes Dev. 16, 2849–2864 (2002).

27. Briscoe, J. et al. Homeobox gene Nkx2.2 and specification of neuronal identity by graded Sonic hedgehog signalling. Nature 398, 622–627 (1999).

28. Goodrich, L.V., Johnson, R.L., Milenkovic, L., McMahon, J.A. & Scott, M.P. Conservation of the hedgehog/patched signaling pathway from flies to mice: induction of a mouse patched gene by Hedgehog. Genes Dev. 10, 301–312 (1996).

29. Stone, D.M. et al. The tumour-suppressor gene patched encodes a candidate receptor for Sonic hedgehog. Nature 384, 129–134 (1996).

30. Alcedo, J., Ayzenzon, M., Von Ohlen, T., Noll, M. & Hooper, J.E. The Drosophila smoothened gene encodes a seven-pass membrane protein, a putative receptor for the hedgehog signal. Cell 86, 221–232 (1996).

31. van den Heuvel, M. & Ingham, P.W. smoothened encodes a receptor-like serpentine protein required for hedgehog signalling. Nature 382, 547–551 (1996).

32. Ingham, P.W. & McMahon, A.P. Hedgehog signaling in animal development: paradigms and principles. Genes Dev. 15, 3059–3087 (2001).

33. Incardona, J.P., Gaffield, W., Kapur, R.P. & Roelink, H. The teratogenic Veratrum alkaloid cyclopamine inhibits sonic hedgehog signal transduction. Development 125, 3553–3562 (1998).

34. Chuang, P.T. & McMahon, A.P. Vertebrate Hedgehog signalling modulated by induction of a Hedgehog-binding protein. Nature 397, 617–621 (1999).

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

36. Trousse, F., Marti, E., Gruss, P., Torres, M. & Bovolenta, P. Control of retinal ganglion cell axon growth: a new role for Sonic hedgehog. Development 128, 3927–3936 (2001).

37. Marigo, V., Davey, R.A., Zuo, Y., Cunningham, J.M. & Tabin, C.J. Biochemical evidence that patched is the Hedgehog receptor. Nature 384, 176–179 (1996).

38. Murone, M., Rosenthal, A. & de Sauvage, F.J. Sonic hedgehog signaling by the pat-ched-smoothened receptor complex. Curr. Biol. 9, 76–84 (1999).

39. Chuang, P.T. & Kornberg, T.B. On the range of hedgehog signaling. Curr. Opin. Genet. Dev. 10, 515–522 (2000).

40. Lee, C.S., Buttitta, L.A., May, N.R., Kispert, A. & Fan, C.M. SHH-N upregulates Sfrp2 to mediate its competitive interaction with WNT1 and WNT4 in the somitic mesoderm. Development 127, 109–118 (2000).

41. Baranski, M., Berdougo, E., Sandler, J.S., Darnell, D.K. & Burrus, L.W. The dynamic expression pattern of frzb-1 suggests multiple roles in chick development. Dev. Biol. 217, 25–41 (2000).

42. Ladher, R.K. et al. Cloning and expression of the Wnt antagonists Sfrp-2 and Frzb during chick development. Dev. Biol. 218, 183–198 (2000).

43. Hamburger, V. & Hamilton, H.L. A series of normal stages in the development of the chick embryo. J. Morphol. 88, 49–92 (1951).

44. Perrin, F.E. & Stoeckli, E.T. Use of lipophilic dyes in studies of axonal pathfinding in vivo. Microsc. Res. Tech. 48, 25–31 (2000).

45. Schaeren-Wiemers, N. & Gerfin-Moser, A. A single protocol to detect transcripts of various types and expression levels in neural tissue and cultured cells: in situ hybridiza-tion using digoxigenin-labelled cRNA probes. Histochemistry 100, 431–440 (1993).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Local calcium transients regulate the spontaneous motility of dendritic filopodiaChristian Lohmann, Alexei Finski & Tobias Bonhoeffer

During development, dendrites, and in particular dendritic filopodia, undergo extensive structural remodeling, presumably to help establish synaptic contacts. Here, we investigated the role of calcium signaling in dendritic plasticity by simultaneously recording calcium dynamics and filopodial growth in rat hippocampal slice cultures. Local calcium transients occurred in dendritic filopodia and shafts, often at putative synaptic sites. These events were highly correlated with filopodial motility: comparatively rare when individual filopodia emerged from the dendrite, they became more frequent after filopodia started growing, finally causing them to halt. Accordingly, an experimental reduction of the frequency of local calcium transients elicited filopodial growth and, conversely, calcium uncaging reduced filopodial motility. Our observations suggest that low levels of local calcium transients facilitate filopodial outgrowth, whereas high levels inhibit the formation of filopodia and stabilize newly formed ones. This process may facilitate synapse formation and may serve as a homeostatic mechanism distributing synapses evenly along developing dendrites.

When synapses are generated during the development of the nervous system, dendrites are highly motile. In particular, dendritic filopodia that are transiently present during the formation of neuronal networks grow and retract over the course of seconds to minutes. The function of this filopodial motility is not entirely clear, but it has been suggested that it increases the probability of synaptogenic contact between den-drites and nearby axons1–3. In addition, it is thought that dendritic motility may help to increase synaptic turnover during the formation of neuronal networks by increasing the pool of temporary synaptic contacts from which the final set of synapses can be selected4.

How is the structural plasticity of dendrites regulated? Calcium signaling is important in sculpting dendrites5–9. However, whether calcium signaling also regulates the motility of individual dendritic filopodia is still a matter of active debate. Biochemical studies suggest that calcium channels and calcium-sensing proteins—such as CaMKII—are essential for the regulation of filopodial motility6,10–12. In addition, stimulating the release of calcium from internal stores induces morphological changes in spines13. Yet simultaneous recordings of spontaneously occurring spine movements and calcium signaling in hippocampal neurons have not shown any correlation between the two phenomena14.

Here, we re-addressed the question of whether calcium signals might correlate with filopodial motility in the light of new findings. A particular type of calcium transient, namely the local release of calcium from internal stores, occurs spontaneously in dendrites of developing hippocampal neurons in dissociated cultures15 as well as in ganglion cells from retinal whole mounts7. This type of calcium signaling stabilizes developing dendrites, in contrast to global calcium increases during burst activity, which have no apparent role in dendritic development7. We therefore investigated whether local calcium transients might control filopodial motility.

We found that developing hippocampal neurons in organotypic slices generated local calcium transients similar to those observed previously in dissociated hippocampal neurons and developing retinal ganglion cells7,15. Simultaneous imaging of filopodial motility and calcium dynamics showed that the frequency of local calcium transients usually was below baseline before a filopodium started growing. After the onset of growth, however, local calcium transients occurred frequently and filopodial growth subsequently halted. Accordingly, pharmacological reduction of the frequency of local calcium transients initiated filopo-dial growth, and calcium uncaging inhibited filopodial motility. Taken together, these results indicate that, somewhat counterintuitively, local calcium transients inhibit the growth of filopodia.

RESULTSDendrites generate spontaneous local calcium transientsTo measure calcium transients in the dendrites of developing neurons, we filled single neurons with a calcium indicator in organotypic slices from the hippocampus of neonatal rats. We focused on the developmental period (P0–2 + DIV1–3) when dendrites bear motile filopodia and synapses form at a high rate in vivo and in slice cultures16–18. Hippocampal neurons were labeled completely, allowing for recordings from proximal and distal dendrites.

Hippocampal neurons generated spontaneous calcium flashes that were localized to small stretches of dendrite (Fig. 1a, Supplementary Video 1). Such calcium transients occurred in putative pyrami-dal neurons as well as in interneurons, and, for both types of neu-rons, were similar in duration (interneurons: 5.4 ± 0.8 s; pyramidal cells: 4.1 ± 1.2 s) and extent along dendritic stretches (interneurons: 11.5 ± 0.7 µm; pyramidal cells: 11.0 ± 1.2 µm).In fact, the calcium transients observed were also very similar to those recently described

Max Planck Institute of Neurobiology, Am Klopferspitz 18, 82152 Planegg-Martinsried, Germany. Correspondence should be addressed to C.L. ([email protected]).


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

in retinal ganglion cells from embryonic chicks7.Hippocampal neurons also generated global calcium rises (Fig. 1a). These calcium transients occurred at mean frequencies of 0.75 ± 0.28 per min and 0.56 ± 0.18 per min in putative pyramidal neurons and interneurons, respectively. We noticed simultaneous calcium rises in neurons close to each other. These transients presumably represent network activity that has been described before19.

As in the retina, the local calcium transients in hippocampal neurons were blocked by agents that interfere with calcium release from internal stores. Both cyclopiazonic acid (CPA, 20 µM) and 2-amino-e thoxydiphenyl borate (2-APB, 100 µM)20,21 prevented the generation of local calcium transients almost completely (Fig. 1b), indicating that these events involve calcium release from internal stores. By bath appli-cation of antagonists for neurotransmitter receptors, we investigated whether neurotransmission induces local calcium transients. Neither methyl-(4-carboxyphenyl) glycine (MCPG; 1 mM), an antagonist of metabotropic glutamate receptors, nor a combination of 2-amino-

5-phosphonovaleric acid (APV; 50 µM) and 1,2,3,4-tetrahydro-6-nitro-2,3-dioxo-benzo[f]quinoxaline-7-sulfonamide (NBQX; 10 µM), which block NMDA and non-NMDA glutamate receptors, respectively, affected the frequency of local calcium transients significantly (Fig. 1b). In contrast, in the presence of SR-95531 (10 µM), a specific GABAA receptor antagonist, the frequency of local calcium transients was signifi-cantly reduced (Fig. 1b). Latrotoxin, a stimulant of vesicular exocytosis22 (100 pM, in the presence of 1 µM tetrodotoxin), increased the occur-rence of local calcium transients markedly (to 463% of that seen with tetrodotoxin alone, P < 0.05, n = 5; Fig. 1c), supporting the argument that the observed dendritic calcium transients can be caused by presyn-aptic transmitter release.

The above result that SR-95531 reduces calcium transients could still be due to network effects rather than to direct action on the imaged neu-ron. However, we also found that SR-95531 blocked 70% of latrotoxin-induced calcium transients (P < 0.05, n = 5; Fig. 1c), which showed that GABA signaling directly induces local calcium transients. Together, these

Figure 1 Local and global calcium transients occur in dendrites of developing hippocampal neurons. (a) Putative pyramidal cell in the CA3 region (P2/DIV3) filled with Calcium Green-1 dextran by ballistic delivery. Right, local (top) and global (bottom) increases of [Ca2+]i, represented as ∆F/F0 values. Arrowheads mark extent of the measured dendritic segment (traces and pseudo line scans). (b) Frequency of spontaneously occurring local calcium transients upon bath application of cyclopiazonic acid (CPA), 2-aminoethoxydiphenyl borate (2-APB), 2-amino-5-phosphonovaleric acid and 1,2,3,4-tetrahydro-6-nitro-2,3-dioxo-benzo[f]quinoxaline-7-sulfonamide (APV and NBQX), methyl-(4-carboxyphenyl) glycine (MCPG) or SR-95531 (SR) is shown as a percentage of the frequency before drug application (control). The number of experiments is shown in parentheses. Error bars, s.e.m. (*P < 0.05). (c) Latrotoxin (LTX) caused a strong increase in the frequency of local calcium transients in the presence of tetrodotoxin (TTX). LTX-mediated transients are blocked by the GABAA receptor antagonist SR-95531 (*P < 0.05), supporting the argument that the relevant transmitter is GABA.

Figure 2 Most dendritic calcium transients occur near synapses. (a) CA3 pyramidal cell loaded with Calcium Green-1 dextran and Alexa-594 dextran. (b) Higher magnification of the imaged dendritic segment. (c) The same dendrite as in b, after fixation. (d) A local calcium transient in the dendrite shown in b. (e) The dendritic segment shown in d after immunolabeling with anti-synapsin antibodies and high-resolution confocal reconstruction. The arrow marks a putative synapse and points to the same position as the arrow in d. (f) High-magnification (x-y) and orthogonal (y-z) views of this dendrite. The putative synapse on this dendrite is labeled by the arrow (yellow pixels in both views). Note that all other structures overlap with the dendrite in no more than one view (arrowheads). (g) Distribution of local calcium transients with respect to synaptic sites (average of the histograms from 5 neurons, 65 local calcium transients; ‘measured distribution’) and results from a Monte Carlo analysis assuming independent distribution of synapses and local calcium transients (‘control’). Close spatial relationships were much more frequent in the measured than in the simulated distribution.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

pharmacological experiments indicated that synaptic signaling, and in particular activation of dendritic GABAA receptors, contributed to trig-gering spontaneous local calcium transients at the investigated ages.

This observation led us to ask whether local calcium transients occur preferentially at synapses along the dendrites or at nonsynaptic sites. To address this question, we loaded neurons with both Calcium Green-1 dextran and the fixable dye Alexa-594 dextran. After recording dendritic calcium transients, we fixed the cells and immunolabeled them for syn-apsin 1 (Fig. 2; for details, see Methods). High-magnification confo-cal reconstructions in three dimensions allowed us to identify putative presynaptic structures terminating on the imaged dendrites (Fig. 2e,f). After determining the positions of local calcium transients and puta-tive synapses, we calculated the distance between the center of each cal-cium transient and the center of the nearest putative synapse (Fig. 2g). Most local calcium transients (41 of 65 from 5 cells) occurred within a distance of 1.5 µm of a putative synapse. The mean inter-synapse distance was 10.5 ± 1.4 µm. This value is in line with numbers that can be deduced from electron microscopy studies of CA1 tissue in the hip-pocampus23. Monte Carlo analyses (performed for each cell separately) showed that (in four of the five cells) the median of the distances between each calcium transient and its nearest synapse was significantly smaller (P < 0.05) than expected if the transients were distributed randomly. These results indicated that local calcium transients are generated at or close to synaptic sites. However, approximately 40% of the transients occurred at sites that were not identified as putative synapses. These sites may be genuine nonsynaptic locations (possibly sites of newly forming contacts). Alternatively, they could be locations that we could not discern as synapses in the immunolabeled preparations.

Optical recordings with high spatial and temporal resolution showed that local calcium transients also occurred spontaneously in dendritic filopodia (Fig. 3, Supplementary Video 2). These filopodial calcium transients were always accompanied by fluorescence increases in the parent dendrite. In many instances, filopodial and dendritic calcium transients occurred simultaneously. However, in a number of cases (n = 9; 6 filopodia, 5 cells), we unequivocally observed calcium

transients that occurred first in the filo-podium and only later in the dendritic shaft, with latencies of 100 ms to 1.5 s. The fluorescence increases in the dendritic shaft and the filopodium usually had similar amplitudes (Fig. 3b,c), indicating that the calcium transients propagated actively into the dendritic shaft, possibly through calcium-induced calcium release. However, in one case

the amplitude of the relative fluorescence change in the dendritic shaft was considerably smaller than in the filopodium (Fig. 3a), probably reflecting passive calcium diffusion.

These observations indicate that (i) filopodia can generate calcium transients autonomously and (ii) most (if not all) filopodial calcium transients are transmitted into the dendrite.

Local calcium dynamics correlate with filopodial motilityTo image dendritic motility and calcium dynamics simultaneously, den-drites were stained simultaneously with both the calcium indicator Calcium Green-1 dextran and the calcium-insensitive dye Alexa-594 dextran. The Alexa dye not only facilitated the visualization of fine morphological structures, but also made ratiometric measurements of calcium transients possible.

Because spontaneous filopodial growth is a relatively rare event, we needed to record for comparatively long periods. This necessitated very low-excitation light intensities to prevent bleaching and phototoxic damage. Consequently, the signal-to-noise ratio in these experiments was not sufficient to track the calcium dynamics in individual filopo-dia. Instead, we recorded the changes in the intracellular calcium con-centration ([Ca2+]i) in a dendritic segment of 10 µm centered around the insertion site of a filopodium. Because we determined that calcium transients in filopodia are accompanied by rises of [Ca2+]i in the den-dritic shaft (as discussed above), we are confident that we recorded all relevant events.

In single recordings, we observed that local calcium transients occurred in the dendritic shaft very rarely before a filopodium started growing, but frequently after it started growing (see Fig. 4a,b and Supplementary Video 3 for examples). We assessed the temporal rela-tionship between filopodial growth and local calcium flashes for all recorded filopodia (n = 22 from 12 neurons; Fig. 4c). During a 2-min period before a filopodium started growing, the frequency of local cal-cium transients in the dendritic shaft was low. Throughout the growth phase, the rate of transients rose and peaked (1.19 ± 0.28 per min) in most cells just before filopodial length reached its maximum (48–240 s

Figure 3 Local calcium transients are generated in filopodia and transmitted to the dendritic shaft. (a) Calcium measurements in a filopodium and its parent dendritic segment of a CA3 pyramidal neuron (P2/DIV3). A calcium transient was generated in the filopodium. The increase of [Ca2+]i in the dendritic shaft was small. (b) Measurement from a filopodium of an interneuron. Calcium transients were generated in the filopodium and transmitted into the dendritic shaft (delay 1–2 s). (c) Filopodium of a different pyramidal cell. Calcium transients occurred almost simultaneously in the filopodium and the dendritic shaft. The inset shows that increases in [Ca2+]i in the filopodium preceded those in the dendritic shaft. Bars: 5 s (inset: 500 ms), 20% ∆F/F0.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

after growth onset). Thereafter, the frequency dropped to intermediate levels (0.26 ± 0.12 per min).

That calcium dynamics and filopodial motility were correlated suggests calcium signals may regulate the growth of filopodia. What could this regulatory function be? Two lines of evidence from our experiments indi-cate that calcium transients restrain filopodial growth. First, we compared the calcium activity in the dendrite before a filopodium started growing with the calcium dynamics at dendritic sites where no filopodia grew. We found that the frequency of local calcium transients was significantly lower during the 2-min period before a filopodium started growing than at control sites (0.09 ± 0.05 per min versus 0.30 ± 0.05 per min, P < 0.01, sign test). Therefore, a low frequency of calcium transients may be neces-sary to initiate filopodial growth at a certain dendritic segment.

Second, we asked whether the behavior of growing filopodia changes when a calcium transient occurs. Plotting the growth rates of filopodia with respect to the occurrence of the first local calcium transient after growth onset showed that filopodia grew at high rates until the first calcium transient occurred, but growth rates dropped significantly after the first transient, often not immediately but after 1–2 min (Fig. 4d).

In summary, we found that, somewhat counterintuitively, low levels of calcium activity seemed to trigger filopodial growth, whereas high levels inhibited filopodial extension. However, this finding applies only to local calcium transients; global events were not obviously correlated with filopodial outgrowth. There was no significant difference in the rate of global calcium transients before or after the onset of growth (change: 15 ± 43%, 5 filopodia from 5 cells).

Manipulating calcium transients affects filopodial growthBecause we were able to show a clear correlation between local calcium levels and dendritic growth, we wanted to test whether the manipula-tion of local calcium transients would produce similar effects.

First, we elicited dendritic calcium rises by the uncaging of calcium and tested their effect on the motility of dendritic filopodia. We found that three consecutive light pulses induced large and sustained calcium rises lasting several seconds in the illuminated dendrite (Fig. 5a–c). One set of three light pulses sufficed to reduce the motility of den-dritic filopodia by about 50% for a period of 5 min after illumination (Fig. 5d,e). In control cells that did not contain caged calcium, the

Figure 5 Uncaging of calcium in dendrites blocks filopodial motility. (a) A CA3 pyramidal neuron loaded with calcium indicator and caged calcium. (b) High magnification of one of the dendrites of the cell shown in a. Arrowheads mark the extent of the measured dendritic segment (trace in c). (c) Calcium responses of the dendrite shown in b upon illumination (arrows). (d) Average motility of the filopodia of this cell before and after illumination. (e) Summary of the results showing motility before and during the first 5 min after illumination in neurons loaded with caged calcium (n = 5) and control cells (‘Without caged Ca2+’; n = 5) (*P < 0.05).

Figure 4 Filopodial growth and calcium dynamics are correlated. (a) Hippocampal neuron (P1/DIV1) labeled with Calcium Green-1 dextran and Alexa-594 dextran. Two filopodia grew during the recording period. Local calcium transients occurred after filopodia had started growing. Changes in [Ca2+]i are shown as ratios (green/red, high-pass filtered). Scale bars, 20 µm. (b) CA1 pyramidal neuron (P2/DIV2). Filopodial growth and calcium dynamics are illustrated as in a. Global and local calcium transients can be clearly identified. Local calcium transients did not occur before the onset of growth in the filopodium. (c) Comparison between filopodial growth (black line) and the occurrence of local calcium transients (red columns). Data from 22 filopodia from 12 cells. The hatched horizontal line indicates the frequency of local calcium transients at random locations (where no filopodial growth occurred) along dendrites of the same cells. (d) Average growth rates of filopodia with respect to the occurrence of the first local calcium transient after growth had started. Growth rates were stable before the first transient but declined afterwards.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

motility of filopodia was not affected by the light exposure (Fig. 5e). This result indicated that dendritic calcium transients are sufficient to reduce dendritic growth within minutes.

Next, we pharmacologically reduced the frequency of local calcium transients by locally applying 2-APB with a superfusion system. The superfusion system facilitates fast delivery of the blocker and a stable environment in the area of interest. After monitoring baseline calcium activity as well as filopodial motility for 4 min, 2-APB (1 mM) was applied locally via the superfusion pipettes for 2 min (Fig. 6a–e). This treatment reduced the frequency of local calcium transients by more than 50%. This reduction became apparent 2–5 min after starting the superfusion. The variation in the onset of the effect was probably due to differences in tissue penetration. During the period when the frequency of local calcium transients was reduced, some filopodia started growing (Fig. 6c). In addition, new filopodia emerged from the dendritic shaft. Another subset of filopodia did not show any change in motility.

We analyzed the rates of filopodial length changes in relation to the time-point when the frequency of local calcium transients became reduced by more than 50% (onset of 2-APB effect). We found a signifi-cant increase in growth (0.22 ± 0.04 µm per min versus –0.03 ± 0.03 µm per min, n = 5 cells, P < 0.01; Fig. 6f) within a 2-min period after the onset of the 2-APB effect (that is, 2–7 min after the beginning of the 2-APB application). This finding supports the idea that a reduction in the frequency of local calcium transients promotes the growth of den-dritic filopodia.

To specifically investigate the role of those local calcium transients that are transmitter induced, we carried out superfusion experiments using SR-95531. As expected, SR-95531 superfusion reduced the frequency of local calcium transients significantly (36 ± 16% of control; n = 5, P < 0.01). As with 2-APB, superfusion of SR-95531 also triggered sig-nificant growth of filopodia within 2 min after the reduction of local calcium signaling became apparent (0.21 ± 0.06 µm per min, n = 5 cells), as compared with the motility before superfusion (0.008 ± 0.03 µm per min, P < 0.005; Fig. 6g).

Superfusion with control solution did not cause changes in calcium activity (mean change in frequency of local calcium transients: 2 ± 28%,

n = 5) or in dendritic motility. We calculated the mean growth rates of filopodia for three 2-min periods after the beginning of superfu-

sion for each cell and found that growth in 2-APB– or SR-95531–treated cells was significantly higher (2-APB: 0.24 ± 0.02 µm per min; SR: 0.21 ± 0.06 µm per min) than in control cells (0.06 ± 0.005 µm per min; Fig. 6h). Superfusing APV and NBQX neither decreased calcium activ-ity (111 ± 64% of control, n = 4 cells) nor stimulated filopodial growth: filopodial motility (0.06 ± 0.004 µm per min) in this case was not different from that of controls, but was significantly lower than after SR-95531 treatment (P < 0.05).

DISCUSSIONWe have described spontaneously occurring local calcium transients in dendrites of developing hippocampal neurons during the period of synapse formation. Interestingly, these transients resemble those described as occurring in developing neurons from other systems and preparations. Retinal ganglion cells from embryonic chicken, for instance, generate local calcium transients with characteristics very sim-ilar to those described here (Fig. 1 and ref. 7). Moreover, a recent study reported localized [Ca2+]i rises in dendrites of hippocampal neurons from primary cultures that also last for seconds and extend for 4–8 µm in dendritic segments15. All three studies show that local calcium tran-sients require release from internal stores, suggesting that the type of local calcium transient described here might be a general phenomenon occurring in the developing neurons of different species.

As reported above, we found that in many instances local calcium transients arise in the immediate vicinity of putative synapses. In addi-tion, latrotoxin, a stimulant of vesicular release, increased the occur-rence of local transients, further supporting the idea that the calcium transients we observed are largely triggered by synaptic signaling. However, 40% of local calcium transients occurred at locations that could not be unequivocally characterized as synapses. There are two possible explanations for this finding: either (i) all local calcium tran-sients occur at synapses, but we failed to detect a certain proportion of these synapses, or (ii) local calcium transients occur at both synapses and nonsynaptic sites. It seems plausible that local calcium transients are generated not only at established synapses, but also at immature synapses or newly formed axodendritic contacts that have not accumu-

Figure 6 Blockade of local calcium transients induces filopodial growth. (a) Interneuron from the CA3 region (P1/DIV3). (b) Higher magnification of dendritic area. Red, distribution of superfusate. (c–e) Images (c) and growth curves (d) of single filopodia (1–4) and frequency of local calcium transients in both dendrites shown in b (e) before and after 2-APB superfusion. (f) Growth rates of filopodia from five cells before and after the time when 2-APB reduced frequency of local calcium transients by more than 50% (*P < 0.01). (g) Growth rates of filopodia from five cells before and after superfusion with the GABAA receptor antagonist SR-95531 (*P < 0.005). (h) Comparison of maximal filopodial growth rates after superfusion with control, 2-APB, SR-95531, or APV and NBQX solution. For each cell, mean length changes were determined for three 2-min periods, and highest growth values of each cell were averaged within each group. The number of experiments are shown in parentheses (*P < 0.05; **P < 0.002 compared with control).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

lated sufficient synapsin for immunodetection. Thus, the local calcium transients that we observed at nonsynaptic sites are most likely to be explained by their occurrence at immature synapses.

This raises the question of how these transients are generated. Among the major hippocampal neurotransmitters, we found only GABA to be involved in eliciting local calcium transients at the ages we investigated (P0–2 + DIV1–3). Blocking either ionotropic or metabotropic gluta-mate receptors did not affect the frequency of local calcium transients. This finding is consistent with the role of GABA as the main depolar-izing transmitter in the neonatal hippocampus24–26. It is also consistent with global calcium activity in the hippocampus being critically depen-dent on GABA signaling at these ages19. Yet GABA signaling clearly accounts only in part for the generation of local calcium transients, as the blockade of GABA receptors did not completely abolish the signals. Therefore, other signaling molecules must be involved in inducing local calcium transients as well. Candidate molecules known to be present in the hippocampus include receptors that recognize adhesion mol-ecules and extracellular matrix components, such as β1 integrin27,28 or neurotrophins29.

A recent study in primary cultures of hippocampal neurons found no clear-cut correlation between morphological changes and calcium transients; nevertheless, spines showing a high frequency of calcium transients over longer periods of time were generally more stable than those with intermediate activity14. The former of these results seems in direct contradiction with our findings. However, the authors did not differentiate between different types of calcium transients (for example, global versus local transients), so it is entirely possible that a high inci-dence of global transients masked a potential correlation between local transients and filopodial movements. Another study reported that a global blockade of calcium signaling in hippocampal neurons did not change filopodial motility30. The authors concluded that basal filopo-dial activity does not require calcium signaling, but may be modulated by significant elevation of the [Ca2+]i. Our findings support the idea that basal filopodial motility is independent of global calcium signaling, but that local and transient calcium signals are able to alter the motility of individual filopodia temporarily.

The calcium signaling in dendritic filopodia described here is simi-lar to that seen in filopodia of axonal growth cones31. In addition, the effect of calcium signaling on the dynamics of dendritic filopodia is comparable to that on axonal filopodia, where calcium transients also reduce filopodial motility31,32. Whereas filopodia in axons contribute to steering of growth cones, the role of dendritic filopodia is probably primarily in synapse development. Dendritic filopodia are present in many neuronal cell types during synapse formation, are highly motile and sometimes bear synapses17,18,33,34. It is therefore thought that they help to increase the probability of synaptogenic contact or to select among competing inputs1,3,4.

With respect to the role of filopodia in synaptogenesis, we propose the following model that shows how filopodial plasticity regulated by local calcium transients could help to distribute synapses evenly along the dendrite (see Supplementary Fig. 1). Local calcium transients are generated at established synapses and prevent the emergence of filopodia in the vicinity of synaptic sites. Because the half-maximal width of local calcium transients was in the range of 10 µm, we expect an inhibitory effect to have roughly this spatial extent. At locations farther from synapses, filopodia can arise and grow until calcium transients are generated, which halt further growth. These transients may be triggered by diffusible factors, such as the neurotransmitter GABA, or by interactions of adhesion molecules and their receptors upon contact with an axon. Ongoing calcium signaling may then stabilize a contact-bearing filopodium, allowing for the maturation

of a synapse at that site. Calcium signaling from the filopodium into the dendrite would prevent the growth of additional filopodia at this location. A mechanism such as this would allow the dendrite to search efficiently for synaptic partners at segments that are not yet receiving synapses, whereas no energy would be wasted at sites that receive sufficient input. Furthermore, this mechanism may serve as a homoeostatic process that prevents overload of synaptic inputs on some dendrites as well as under-representation of inputs on others.

METHODSCultures. Cultures from newborn rats (P0–2) were prepared according to a previously described method35. Rats were decapitated quickly and brains placed in ice-cold GBSS (Life Technologies) under sterile conditions. Coronal slices (300 µm) were cut using a tissue chopper (McIlwain) and incubated with serum containing medium on Millicell culture inserts (CM, Millipore) for up to 3 d.

Labeling. After 1–3 d in culture, hippocampal neurons were filled with calcium dyes with a gene gun as described recently36, using a modified barrel liner37. Briefly, slices on membranes were cut out from Millipore culture plate inserts, excess medium was removed, and slices were transferred under a culture plate insert (Falcon; for protection from damage by the pressure wave) and shot at 40–60 PSI. Tungsten particles (1.7 µm, Bio-Rad) were loaded with Calcium Green-1 dextran (mol. wt. 10,000, Molecular Probes) or double-loaded with Calcium Green-1 dextran (5 parts) and Alexa-594 dextran (mol. wt. 10,000, Molecular Probes; 1 part). At these concentrations the calcium indicator Calcium Green-1 dextran does not influence calcium signals to a noticeable extent7.

For the uncaging experiments, we electroporated the calcium indicator Oregon Green BAPTA-1 (250 µM, dissolved in water) alone or together with caged calcium (50 mM, Np-EGTA, Molecular Probes) into single neurons according to a previously described method for dye electroporation38. We used patch pipettes with ∼1-µm tip diameter and square voltage pulses (10 V, 20-ms duration). Calcium dynamics were indistinguishable between cells that were electroporated or labeled using the gene gun (data not shown). Both labeling techniques yielded completely filled neurons within a few minutes.

Imaging. After labeling, slices were immediately transferred to a record-ing chamber that was perfused with modified HBSS (plus 2 mM CaCl2 and 4.2 mM NaHCO3) and temperature controlled at 35 °C. Recordings were obtained using an Axioplan-2 microscope (Zeiss) and a cooled CCD camera (VisiCam QE, Visitron Systems) controlled with Metamorph software (Universal Imaging). Images were acquired at 0.3–30 Hz, and integration time was 0.02–1.5 s for con-tinuous recordings up to 20 min with a Zeiss 0.95–63× water immersion objec-tive. Dual-wavelength recordings were performed with a filter wheel (Visitron Systems) in the excitation pathway switching between a 492/18 nm for Calcium Green-1 dextran and a 575/25 nm for Alexa-594 dextran in combination with a multiband beam splitter and emission filter (all filters: Chroma).

Uncaging. Hippocampal neurons were loaded with calcium indicator and caged calcium as described earlier for labeling. After at least 30 min of incubation of loaded neurons, we obtained time-lapse recordings for 10 min under control conditions. The light of a flash lamp (35 S, Chadwick-Helmuth) was directed into a glass fiber (diameter ∼100 µm), the tip of which pointed to the dendritic area of the labeled neurons. Neurons were illuminated by triggering the flash lamp three times within 5–10 s, and changes in [Ca2+]i were monitored simultaneously. The illumination caused reliable increases in [Ca2+]i (Fig. 5c). Immediately after illumination, we started another 10-min time-lapse recording.

Superfusion. We used a modified version of a superfusion device39,40, which con-sists of two glass pipettes with an inter-tip distance of ∼20 µm. One pipette (tip diameter ∼20 µm) served for solution delivery and was connected to a PicoSpritzer (Parker Instrumentation) to provide a constant outflow of the superfusion solu-tion. The second glass pipette (tip diameter ∼30 µm) was connected to a peristaltic pump that removed the superfusion solution from the external medium, thereby producing a localized and stable drug concentration. The extent of the superfu-sion area was visualized by including Alexa-594 in the superfusion solution. The

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

two pipettes were mounted on a custom-made set of manipulators to adjust them toward each other before the experiment. These manipulators were themselves mounted on a manipulator, which was fixed on the microscope stage.

The superfusion solution contained pharmacological agents at concentra-tions ten times higher those used for bath application (1 mM 2-APB, 100 µM SR-95531, 500 µM APV and 100 µM NBQX). The concentration in the superfu-sion solution is significantly diluted by the time it has reached the dendrites in the tissue. We estimated that the imaged dendrites were located between 10 and 30 µm below the surface of the slice. We observed that the application of 1 mM 2-APB for 2 min reduced the frequency of local calcium transients by at least 50%. All filopodia in the focal plane were analyzed.

Image analysis. Changes in [Ca2+]i are given by ∆F/F0, where F0 is baseline fluorescence (typically taken from images during a 10-s period before the onset of a respective signal or from the beginning of a recording), or as the ratio (green/red) for two wavelength recordings. Ratiometrically determined changes in fluorescence were short-pass filtered. Spontaneous local calcium transients were defined as local rises in fluorescence of more than 5% with a half-maximal spatial extent of 3–30 µm. We defined global transients as fluorescence increases of at least 10% that were observed in all structures of the cell within the focal plane. The spatial extent (half-maximal width) of local flashes was measured from line scans of ∆F/F0 at the time of maximal intensity change.

For analysis of filopodial growth, eight high-resolution (pixel size: 0.18 µm × 0.18 µm) images from the red channel (spanning 24 s) were aver-aged to improve the signal-to-noise ratio. For comparison with the occurrence of calcium transients, the first average image of a stack (comprising 24 s of data) in which growth was observed was defined as time point zero.

Care was taken to discard all changes that were caused by focal shifts or sam-ple movement in any direction. Frequencies of calcium transients and changes in filopodial length were analyzed by a person who was blind with respect to the experimental situation.

Immunohistochemistry. After calcium imaging, slices were immediately fixed in paraformaldehyde (4%, in 0.1 M sodium phosphate buffer) and incubated in the fixative at 4 °C overnight. Slices were rinsed (for 3 h in phosphate buffer), preincubated in a blocker solution (0.4% Triton X-100, 1.5% horse serum, 0.1% bovine serum albumin in phosphate buffer, 4 °C, overnight), and incubated with the primary antibody (rabbit anti-synapsin, Chemicon, dilution 1/500 in 0.4% Triton-X, 1.5% horse serum for 7–10 d at 4 °C). After thorough rinsing, slices were incubated with the secondary antibody (anti-rabbit–FITC, 1/50 in phosphate buf-fer for 2–3 d, 4 °C), rinsed and mounted with GelMount (Biomeda). The recorded cells could be identified by their labeling with Alexa-594 dextran. We obtained stacks of high-resolution images (voxel size, xyz: 0.12 × 0.12 × 0.24 µm) from dendrites and anti-synapsin–labeled structures with a confocal microscope (Leica SP2, pinhole corresponding to 1 airy disk with respect to the green emission) in the sequential acquisition mode using a 1.25/×40 oil immersion objective (HCX PL APO CS, Leica). Putative synapses were identifie d as sites of spectral overlap of the dendrite with anti-synapsin–labeled structures (yellow pixels) in all rotational views of three-dimensional reconstructions. After identifying the positions of putative synapses and local calcium transients along each dendrite, we aligned the images of the live and fixed dendrite and determined the distance between each calcium transient and its nearest synapse. The dendrite segment was treated as a one-dimensional line and the positions of synapses and calcium transients along that line were recorded. For statistical analysis, we performed Monte Carlo simulations for each dendrite, testing the null hypothesis that calcium transients are distributed randomly along the dendrite, following an approach similar to one used previously41. The median distance between each calcium transient and its nearest synapse was compared with 999 simulations in which the position of each transient was independently positioned at random along the dendrite. We rejected the null hypothesis in cases in which the simulation produced less than 5% of median values that were smaller than the observed one.


ACKNOWLEDGMENTSWe thank F. Siegel for performing data analysis; O. Momoh and N. Stöhr for preparation and maintenance of slice cultures; S. Eglen for help with the Monte Carlo statistics; and R. Wong and M. Hübener for critically reading the manuscript.

This work was supported by the Max-Planck-Gesellschaft and the Schloessmann Foundation (C.L.).



1. Wong, W.T. & Wong, R.O. Rapid dendritic movements during synapse formation and rearrangement. Curr. Opin. Neurobiol. 10, 118–124 (2000).

2. Dunaevsky, A. & Mason, C.A. Spine motility: a means towards an end? Trends Neurosci. 26, 155–160 (2003).

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

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

5. Meberg, P.J., Kossel, A.H., Williams, C.V. & Kater, S.B. Calcium-dependent alterations in dendritic architecture of hippocampal pyramidal neurons. Neuroreport 10, 639–644 (1999).

6. Wu, G.Y., Deisseroth, K. & Tsien, R.W. Spaced stimuli stabilize MAPK pathway acti-vation and its effects on dendritic morphology. Nat. Neurosci. 4, 151–158 (2001).

7. Lohmann, C., Myhr, K.L. & Wong, R.O. Transmitter-evoked local calcium release stabilizes developing dendrites. Nature 418, 177–181 (2002).

8. Ciccolini, F. et al. Local and global spontaneous calcium events regulate neurite outgrowth and onset of GABAergic phenotype during neural precursor differentiation. J. Neurosci. 23, 103–111 (2003).

9. Brunig, I., Kaech, S., Brinkhaus, H., Oertner, T.G. & Matus, A. Influx of extracellu-lar calcium regulates actin-dependent morphological plasticity in dendritic spines. Neuropharmacology 47, 669–676 (2004).

10. Fink, C.C. et al. Selective regulation of neurite extension and synapse formation by the beta but not the alpha isoform of CaMKII. Neuron 39, 283–297 (2003).

11. Portera-Cailliau, C., Pan, D.T. & Yuste, R. Activity-regulated dynamic behavior of early dendritic protrusions: evidence for different types of dendritic filopodia. J. Neurosci. 23, 7129–7142 (2003).

12. Jourdain, P., Fukunaga, K. & Muller, D. Calcium/calmodulin-dependent protein kinase II contributes to activity-dependent filopodia growth and spine formation. J. Neurosci. 23, 10645–10649 (2003).

13. Korkotian, E. & Segal, M. Release of calcium from stores alters the morphology of dendritic spines in cultured hippocampal neurons. Proc. Natl Acad. Sci. USA 96, 12068–12072 (1999).

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

15. Koizumi, S. et al. Characterization of elementary Ca2+ release signals in NGF- diffe-rentiated PC12 cells and hippocampal neurons. Neuron 22, 125–137 (1999).

16. Dailey, M.E., Buchanan, J., Bergles, D.E. & Smith, S.J. Mossy fiber growth and synap-togenesis in rat hippocampal slices in vitro. J. Neurosci. 14, 1060–1078 (1994).

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

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

19. Garaschuk, O., Hanse, E. & Konnerth, A. Developmental profile and synaptic origin of early network oscillations in the CA1 region of rat neonatal hippocampus. J. Physiol. (Lond.) 507, 219–236 (1998).

20. Missiaen, L., Callewaert, G., De Smedt, H. & Parys, J.B. 2-Aminoethoxydiphenyl borate affects the inositol 1,4,5-trisphosphate receptor, the intracellular Ca2+ pump and the non-specific Ca2+ leak from the non-mitochondrial Ca2+ stores in permeabilized A7r5 cells. Cell Calcium 29, 111–116 (2001).

21. Bootman, M.D. et al. 2-aminoethoxydiphenyl borate (2-APB) is a reliable blocker of store-operated Ca2+ entry but an inconsistent inhibitor of InsP3-induced Ca2+ release. FASEB J. 16, 1145–1150 (2002).

22. Capogna, M., Volynski, K.E., Emptage, N.J. & Ushkaryov, Y.A. The α-latrotoxin mutant LTXN4C enhances spontaneous and evoked transmitter release in CA3 pyramidal neurons. J. Neurosci. 23, 4044–4053 (2003).

23. Steward, O. & Falk, P.M. Selective localization of polyribosomes beneath developing synapses: a quantitative analysis of the relationships between polyribosomes and developing synapses in the hippocampus and dentate gyrus. J. Comp. Neurol. 314, 545–557 (1991).

24. Ben-Ari, Y., Cherubini, E., Corradetti, R. & Galarsa, J-L. Giant synaptic potentials in immature rat CA3 hippocampal neurones. J. Physiol. (Lond.) 416, 303–325 (1989).

25. Tyzio, R. et al. The establishment of GABAergic and glutamatergic synapses on CA1 pyramidal neurons is sequential and correlates with the development of the apical dendrite. J. Neurosci. 19, 10372–10382 (1999).

26. Hennou, S., Khalilov, I., Diabira, D., Ben-Ari, Y. & Gozlan, H. Early sequential forma-tion of functional GABAA and glutamatergic synapses on CA1 interneurons of the rat foetal hippocampus. Eur. J. Neurosci. 16, 197–208 (2002).

27. Schuster, T. et al. Immunoelectron microscopic localization of the neural recognition molecules L1, NCAM, and its isoform NCAM180, the NCAM-associated polysialic acid, β1 integrin and the extracellular matrix molecule tenascin-R in synapses of the adult rat hippocampus. J. Neurobiol. 49, 142–158 (2001).

28. Spitzer, N.C. Activity-dependent neuronal differentiation prior to synapse formation: the functions of calcium transients. J. Physiol. (Paris) 96, 73–80 (2002).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

29. Kang, H. & Schuman, E.M. Intracellular Ca2+ signaling is required for neurotrophin-induced potentiation in the adult rat hippocampus. Neurosci. Lett. 282, 141–144 (2000).

30. Zhang, S. & Murphy, T.H. Ca2+-independent spine dynamics in cultured hippocampal neurons. Mol. Cell. Neurosci. 25, 334–344 (2004).

31. Gomez, T.M., Robles, E., Poo, M. & Spitzer, N.C. Filopodial calcium transients promote substrate-dependent growth cone turning. Science 291, 1983–1987 (2001).

32. Robles, E., Huttenlocher, A. & Gomez, T.M. Filopodial calcium transients regulate growth cone motility and guidance through local activation of calpain. Neuron 38, 597–609 (2003).

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

34. Niell, C.M., Meyer, M.P. & Smith, S.J. In vivo imaging of synapse formation on a growing dendritic arbor. Nat. Neurosci. 7, 254–260 (2004).

35. Stoppini, L., Buchs, P.-A. & Muller, D. A simple method for organotypic cultures of

nervous tissue. J .Neurosci.Meth. 37, 173–182 (1991).36. Kettunen, P. et al. Imaging calcium dynamics in the nervous system by means of

ballistic delivery of indicators. J. Neurosci. Meth. 119, 37 (2002).37. O’Brien, J.A., Holt, M., Whiteside, G., Lummis, S.C.R. & Hastings, M.H.

Modifications to the hand-held Gene Gun: improvements for in vitro Biolistic transfection of organotypic neuronal tissue. J. Neurosci. Methods 112, 57–64 (2001).

38. Haas, K., Sin, W.C., Javaherian, A., Li, Z. & Cline, H.T. Single-cell electroporation for gene transfer in vivo. Neuron 29, 583–591 (2001).

39. Engert, F. & Bonhoeffer, T. Synapse specificity of long-term potentiation breaks down at short distances. Nature 388, 279–284 (1997).

40. Veselovsky, N.S., Engert, F. & Lux, H.D. Fast local superfusion technique. Pflugers Arch. 432, 351–354 (1996).

41. Diggle, P.J. Displaced amacrine cells in the retina of a rabbit: analysis of a bivariate spatial point pattern. J. Neurosci. Methods 18, 115–125 (1986).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Histone deacetylase 9 couples neuronal activity to muscle chromatin acetylation and gene expressionAlexandre Méjat1, Francis Ramond1, Rhonda Bassel-Duby2, Saadi Khochbin3, Eric N Olson2 & Laurent Schaeffer1

Electrical activity arising from motor innervation influences skeletal muscle physiology by controlling the expression of many muscle genes, including those encoding acetylcholine receptor (AChR) subunits. How electrical activity is converted into a transcriptional response remains largely unknown. We show that motor innervation controls chromatin acetylation in skeletal muscle and that histone deacetylase 9 (HDAC9) is a signal-responsive transcriptional repressor which is downregulated upon denervation, with consequent upregulation of chromatin acetylation and AChR expression. Forced expression of Hdac9 in denervated muscle prevents upregulation of activity-dependent genes and chromatin acetylation by linking myocyte enhancer factor 2 (MEF2) and class I HDACs. By contrast, Hdac9-null mice are supersensitive to denervation-induced changes in gene expression and show chromatin hyperacetylation and delayed perinatal downregulation of myogenin, an activator of AChR genes. These findings show a molecular mechanism to account for the control of chromatin acetylation by presynaptic neurons and the activity-dependent regulation of skeletal muscle genes by motor innervation.

The regulation of gene expression in excitable cells, such as neurons and skeletal muscle cells, by presynaptic inputs has been well documented. More recently, it has been shown that chromatin acetylation is also influ-enced by the type of stimuli that neurons receive. In Aplysia californica sensorimotor neurons, for example, stimuli generating long-term poten-tiation (LTP) induce histone acetylation and promote the recruitment of the histone acetyl transferase CBP to the immediate early gene C/EBP, whereas cues provoking long-term depression induce HDAC recruitment and histone deacetylation on this gene1. In addition, CBP+/– mice show altered LTP, learning and memory associated with a global reduction in histone 2B acetylation2,3.

In most excitable cells, extracellular signaling is concentrated primarily at synapses; electrical activity elicited in postsynaptic cells by presynaptic neurons is a major consequence of such signaling. Neuron-induced electri-cal activity has long been known to control muscle gene expression4. The muscle nicotinic AChR, which mediates the transmission of nerve influxes to skeletal muscle fiber, is the best characterized example of gene regula-tion by electrical activity in excitable cells. The AChR is a heteropentameric cationic ion-gated channel composed of four subunits: α2βδγ/ε. Before innervation, the expression of AChR genes depends on the basic helix-loop-helix (bHLH) myogenic transcription factors4, except for the gene encoding the AChR ε subunit, which is insensitive to myogenic factors and is only expressed after birth in subsynaptic nuclei. After motor innervation, muscle electrical activity elicited by the nerve triggers calcium-dependent signals that cause a decline in expression of the myogenic bHLH transcrip-tion factor myogenin, thus preventing the expression of AchR genes in the

extrasynaptic regions of muscle fibers4. Concomitantly, the expression of AChR genes is strongly upregulated in subsynaptic nuclei in response to neural factors5–7. The combination of these two mechanisms results in the strict compartmentalization of expression AChR genes in subsynaptic nuclei. Blockade of muscle electrical activity by denervation induces the re-expression of myogenin and, shortly thereafter, of AChR subunit genes along the length of the muscle fiber. The extrasynaptic expression of AChR subunits is thus strictly coupled to the presence of myogenin. Accordingly, binding sites for the myogenic bHLH transcription factors, known as E boxes, within the promoters of AChR subunit genes are necessary for transcriptional activation in response to denervation. Moreover, myogenin expression is sufficient to induce AChR expression in vivo, as transgenic mice overexpressing myogenin in postnatal muscle express AChR genes along the entire length of the muscle fiber8. Thus, it has been proposed that electrical activity controls the transcription of AChR genes through its effects on myogenin expression8–14.

Chromatin acetylation is regulated by the balance between histone acetyl transferase (HAT) and HDAC activities, but how presynaptic inputs influence these activities has not been explored. There are two major classes of HDACs that are distinguished by their structures and expression patterns. Class I (HDAC1, HDAC2 and HDAC3) simply contain a catalytic domain and are widely expressed. In contrast, class II (HDAC4, HDAC5, HDAC7 and HDAC9) contain an N-terminal extension that mediates association with a variety of coactivators and corepressors and also confers signal responsiveness to calcium-dependent kinases. The class II HDACs show highest expression in striated muscle and brain.

1Equipe Différenciation Neuromusculaire, Institut Fédératif de Recherche 128, Unité Mixte de Recherche 5161, Centre National de la Recherche Scientifique/Ecole Normale Supérieure, Ecole Normale Supérieure 46, allée d’Italie, 69364 Lyon Cedex 07, France. 2Department of Molecular Biology, University of Texas Southwestern Medical Center, Dallas, Texas 75390-9148, USA. 3Institut National de la Santé et de la Recherche Médicale U309, Institut Albert Bonniot, 38706 La Tronche Cedex, France. Correspondence should be addressed to L.S. ([email protected]).


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Although the transcriptional repression of the bHLH myogenic f actors by electrical activity has been known for over a decade, the upstream transcription factors involved in this repression have remained unknown. Recently, regulators of histone acetylation have been shown to regulate the myogenic determination factors. MyoD and MEF2 both recruit the HAT p300/CBP, which is critical for myogenic differentiation15–18.

Conversely, HDAC1 interacts with MyoD in proliferative myoblasts, and this interaction is disrupted upon differentiation. Class II HDACs also modulate the activity of the muscle differentiation factors and regulate differentiation of myoblasts into myotubes18–20. These HDACs associate with MEF2 and are responsive to extracellular differentiation signals that control their phosphorylation and thereby their nucleocytoplasmic local-ization in muscle cells. At the onset of muscle differentiation, HDAC5 and HDAC7 are phosphorylated and translocate to the cytoplasm, thereby alleviating MEF2 repression19–22.

In the present study, we investigated the possibility that neuron-elicited electrical activity controls activity-responsive genes by modu-lating the activity of HDACs and chromatin acetylation. Our results show that chromatin is hyperacetylated in muscle myonuclei after abrogation of motor innervation and that this hyperacetylation is par-alleled by downregulation of the class II HDAC9 in skeletal muscle. In addition, preventing Hdac9 downregulation in denervated muscle abrogates both histone hyperacetylation and myogenin and AChR gene activation. The inhibition of activity-dependent genes and chromatin acetylation by electrical activity and HDAC9 is mediated by MEF2, HDAC1 and HDAC3. Our results demonstrate that HDAC9 mediates

Figure 1 Innervation-dependent repression of muscle chromatin acetylation by MITR. (a) Denervation causes muscle chromatin hyperacetylation. Histone H3 acetylation was visualized by immunofluorescence in muscle fibers isolated from innervated and 48-h-denervated tibialis anterior muscles. (b) Quantification of HDAC4, HDAC5, HDAC7 and MITR expression in tibialis anterior muscle. Gene expression was evaluated by real-time quantitative RT-PCR on total RNA extracted from tibialis anterior muscles of 5-week-old Swiss Webster male mice. Absolute quantification was obtained by comparing the results to the values obtained with dilutions of known amounts of the corresponding cDNA. Error bars, s.e.m. Each measure was performed with two dilutions of each cDNA and repeated twice. (c) MITR expression profile. MITR expression was evaluated in embryonic diaphragm muscle before and after innervation by real-time RT-PCR. For post-natal muscles, the measurements were performed on tibialis anterior muscles. ED, embryonic day of development; PN, postnatal day; D, tibialis anterior muscles of 28-d-old animals, 48 h after denervation. The RNA level was arbitrarily set to 1 in ED13 muscle. Error bars, s.e.m. (d) Forced expression of MITR prevents chromatin hyperacetylation in denervated muscle. pECFP-nuc and either a MITR expression vector or pcDNA3 (control) were coelectroporated in innervated tibialis anterior muscles. The muscles were denervated 10 d later, and after another 48 h, histone H3 acetylation was visualized by immunofluorescence in isolated fibers.

Figure 2 MITR interacts with HDAC1 and HDAC3 in skeletal muscle in vivo. (a) HDAC1, HDAC2 and HDAC3 expression is not regulated by innervation in skeletal muscle. HDAC 1, HDAC2 and HDAC3 expression was evaluated in innervated and denervated tibialis anterior muscles by real-time RT-PCR. Error bars, s.e.m. Expression levels were arbitrarily set to 1 in the innervated muscle samples. (b) HDAC1 and HDAC3 protein levels are not affected by denervation. HDAC1 and HDAC3 were detected in innervated and denervated tibialis anterior muscles by western blotting using anti-HDAC3 (α-HDAC3) and anti-HDAC1 (α-HDAC1), respectively. Gel loading was controlled with an antibody to tubulin (α-tubulin). (c) MITR interacts with HDAC1 and HDAC3 in skeletal muscle, as shown by western blotting. An HA-MITR expression vector was coelectroporated in tibialis anterior muscles with either Flag-HDAC1, Flag-HDAC2 or Flag-HDAC3 expression vectors, and the electroporated muscles were subjected to coimmunoprecipitation with anti-Flag. Bottom, the results were analyzed (α-Flag IP) by western blotting with an antibody to HA (α-HA). Top, western blots performed with anti-HA and anti-Flag to visualize the expression of the various Flag- and HA- tagged electroporated constructs. Gel loading was controlled with an antibody to tubulin (α-tubulin). Electroporated vectors are indicated above the panels. (–), control muscles electroporated with pcDNA3.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

the effect of neuron-elicited electrical activity on activity-responsive gene expression by governing the acetylation state of chromatin. These results constitute the first direct evidence that a presynaptic neuron can control chromatin acetylation in postsynaptic cells and provide the first indication of a physiological role for class II HDACs in skeletal muscle in vivo.

RESULTSInnervation triggers MITR and reduces histone acetylationTo investigate whether motor innervation regulates chromatin acetylation in muscle, histone H3 acetylation was visualized by immunofluorescence in innervated and denervated muscle fibers. Histone H3 acetylation sub-stantially increased in myonuclei upon denervation (Fig. 1a).

Recent studies have shown that the class II histone deacetylases HDAC4, HDAC5, HDAC7 and HDAC9 act as powerful repressors of skeletal mus-cle transcription that respond to calcium-dependent signals18,20,22,23. To determine whether the expression of these HDACs might be regulated by motor innervation, their transcripts in innervated and denervated muscles were measured using real-time quantitative RT-PCR (qRT-PCR).

The Hdac9 gene was recently characterized and its major product was shown to encode the splice variant MEF2-interacting transcrip-tion repressor/histone deacetylase–related protein (MITR/HDRP), which lacks the characteristic HDAC domain24,25. Although the spliced

HDAC9 isoform corresponding to MITR was abundant in skeletal mus-cle (Fig. 1b), we could not detect full-length Hdac9 mRNA in adult tibialis anterior mouse muscle.

Absolute quantification of HDAC4, HDAC5 and MITR transcripts showed that they are expressed in comparable amounts in innervated muscle (500, 90 and 100 copies per ng of total RNA, respectively), whereas Hdac7 mRNA was hardly detectable, with about 1 copy per ng of total RNA (Fig. 1b). In denervated muscle, we found that the expression of MITR was 80% lower than in innervated muscle (Fig. 1c). As the denervation experiments suggested that MITR might be regulated by innervation-induced electrical activity, it was of interest to evaluate MITR expression when muscles acquired electrical activity as a consequence of motor innervation at embryonic day (ED) 14.5. Indeed, MITR expression in embryonic muscle was upregulated by 350% between ED14 and ED19 (Fig. 1c).

MITR specifically prevents chromatin hyperacetylationTo investigate the potential role of MITR in the denervation response of adult skeletal muscle, we used an in vivo DNA electroporation tech-nique to restore high levels of MITR in denervated muscles. MITR and nuclear cyan fluorescent protein (CFP) expression vectors were coin-jected into adult tibialis anterior muscles, which were denervated by sciatic nerve section 10 d later. CFP-positive fibers were microdissected 48 h after denervation. Evaluation of histone H3 acetylation by immu-nofluorescence in electroporated fibers showed that MITR efficiently prevented denervation-induced histone hyperacetylation (Fig. 1d).

MITR recruits HDAC1 and HDAC3As MITR is the only HDAC9 isoform expressed in skeletal muscle and lacks a catalytic domain, we investigated whether it might control histone acetylation in muscle by recruiting the class I HDACs HDAC1 and HDAC3, as reported in cultured cells24,26. Using qRT-PCR, we found that HDAC 1, HDAC2 and HDAC3 were expressed in skel-etal muscle, and their expression was not modulated by denervation (Fig. 2a). HDAC1 and HDAC3 were also detectable by western blotting at comparable amounts in innervated and denervated muscle (Fig. 2b).

Figure 3 Acetylation and expression of myogenin and AChR genes in innervated and denervated muscle. (a) MITR downregulation precedes the upregulation of myogenin and AChR α expression. Gene expression in tibialis anterior muscle after denervation by sciatic nerve transection was evaluated at different times by real-time RT-PCR. Error bars, s.e.m. (b) Denervation induces hyperacetylation of the myogenin and AChR α promoters. Histone H3 acetylation on the promoters of the myogenin, AChR α and AChR ε subunit genes was quantified in innervated and denervated tibialis anterior muscles by real-time quantitative RT-PCR after ChIP with an antibody to acetyl-H3. Acetylation levels of the promoters were arbitrarily set to 1 in the innervated muscle samples. Error bars, s.e.m.

Figure 4 Suppression of the transcriptional response to denervation by MITR. (a–c) MITR blocks the response to denervation of the genes encoding myogenin and the AChR α subunit. MITR does not affect the expression of genes insensitive to or repressed by denervation (b). Transcription levels for myogenin and AChR α subunit (a), and for muscle creatine kinase (MCK), acetylcholine esterase (AChE) and AChR ε subunit (b), were analyzed in microdissected fibers by real-time RT-PCR. The black, gray and white bars correspond to the ratio between the mRNA levels measured in denervated and innervated muscles electroporated with pCDNA3, MITR and the N-terminal fragment of HDAC5, respectively (a,b). The abundance of myogenin, MITR and the N-terminal fragment of HDAC5 in microdissected fibers was visualized by western blotting with antibodies to myogenin or to HA from MITR and HDAC5 (c). Error bars, s.e.m.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

To determine whether MITR interacted with these HDACs in vivo, a hemagglutinin(HA)-MITR expression vector was co-electroporated into muscle with either Flag-HDAC1, Flag-HDAC2 or Flag-HDAC3 expres-sion vectors, and coimmunoprecipitation experiments were performed on extracts derived from the electroporated muscles. The antibody to Flag (anti-Flag) efficiently coprecipitated MITR when HDAC1 or HDAC3 was present, whereas no interaction could be detected with HDAC2 (Fig. 2c). We conclude that MITR specifically interacts with HDAC1 and HDAC3 in skeletal muscle, which is likely to account for the ability of MITR to suppress chromatin acetylation.

MITR mediates the control of activity-responsive genesAs MITR mediates the neuronal control of muscle chromatin acetylation, we next investigated whether it controlled the acetylation and expression of genes known to be responsive to electrical activity. The acetylation and responsiveness to MITR of the myogenin (Myog) and AChR subunit genes were thus evaluated. When the profiles of MITR, myogenin and AChR α subunit expression after denervation were compared, it seemed that the decline in MITR expression immediately preceded the upregula-tion of the myogenin and AChR α subunit genes (Fig. 3a).

Histone acetylation on the myogenin and AChR promoters was next measured in innervated and denervated muscles by chromatin immu-noprecipitation (ChIP), with an antibody against acetylated histone H3. Denervation caused histone H3 acetylation to increase on the AChR α subunit and myogenin promoters (Fig. 3b), whereas acetylation remained constant on the gene encoding the AChR ε subunit, which is unique among AChR subunit genes for its insensitivity to electrical activity13,14,27–29. Motor innervation thus maintains activity-responsive muscle genes in an inactive and hypoacetylated state, whereas denerva-tion induces MITR downregulation followed by activity-responsive gene hyperacetylation and expression.

To determine whether the downregulation of MITR was required for the upregulation of activity-responsive genes, we forced MITR expres-sion in denervated muscle by electroporation of MITR expression vector, and evaluated gene expression 2 d after denervation. Forced expres-sion of MITR blocked the upregulation of AChR α subunit mRNAs in denervated muscle (Fig. 4a), whereas the expression of genes repressed (acetylcholine esterase) or only slightly affected (muscle creatine kinase and the AChR ε subunit) by denervation remained unchanged (Fig. 4b). In control muscles, the amounts of myogenin and AChR α subunit mRNA increased 18- and 23-fold, respectively, in denervated muscles compared with innervated muscles (Fig. 4a). The ability of MITR to block the denervation-dependent upregulation of myogenin was also observable at the protein level by western blotting (Fig. 4c).

MITR corresponds to the N-terminal non-catalytic region of HDAC9. To determine whether its effects on activity-dependent gene expression were specific, we also expressed the N-terminal non-catalytic region of HDAC5. Among the class II HDACs, this region of HDAC5 is the most homologous to MITR, and it has been shown to efficiently block MEF2 activity in cultured cells30,31. However, overexpression of this construct in denervated muscle only slightly affected the response to denervation, thus demonstrating the specific sensitivity of the denervation response to MITR (Fig. 4a–c).

Class II HDACs are substrates of calcium-dependent kinases that control their subcellular localization. The finding that MITR, but not HDAC5, could block the denervation response suggested that the effects we observed could not be attributed to saturation of the endogenous system that controls HDAC localization. To further confirm that electro-poration of class II HDAC expression vectors did not saturate the HDAC-localization control system, and to determine whether the specificity of action of MITR correlated with a specific spatial distribution of MITR

compared to the other class II HDACs expressed in skeletal muscle, we per-formed immunofluorescence studies on muscle fibers expressing MITR, HDAC5 or HDAC4. MITR showed a punctuate nuclear distribution (14.6 ± 4.1 spots per nuclei, mean ± s.e.m.), whereas HDAC5 and HDAC4 were distributed diffusely in the nucleus and cytoplasm, respectively (Fig. 5 and Supplementary Fig. 1 online). MITR thus specifically accumu-lated at a discrete number of nuclear spots that did not seem to correspond to the heterochromatic regions visualized by Hoechst staining. Conversely, HDAC4 remained strictly cytoplasmic, although its expression was at least equivalent to that of MITR, thus demonstrating that the endogenous class II HDACs nuclear export system was not saturated.

Hdac9–/– mice have increased histone H3 acetylationOur results suggested that MITR participated in the repression of histone acetylation and electrical activity–responsive gene expression in skeletal muscle. To further test this possibility, we analyzed the expression of activity-responsive genes and their acetylation in HDAC9 knockout mice32. Visualization of global histone H3 acetylation by immunofluo-rescence or western blotting showed it to be greater in mutant innervated muscle compared with wild-type innervated muscle; it was still below the acetylation levels observed in denervated muscle, which were similar in wild-type and mutant muscles (Fig. 6a,b). Consistently, chromatin immunoprecipitation (ChIP) experiments on mutant and wild-type

Figure 5 HDAC 4, HDAC5 and MITR localization in skeletal muscle. Expression vectors for HA-tagged MITR, MEF2 interaction–defective MITR (MITR∆MEF), HDAC5 and HDAC4 were co-electroporated with pECFP-Nuc into adult tibialis anterior muscles. Localization was analyzed 6 d later in isolated muscle fibers by immunofluorescence. Scale bar, 1 µm. Red, Anti-HA labeling; green, CFP; blue, Hoechst staining.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

muscles showed that histone H3 acetylation was elevated on the myo-genin promoter in mutant muscles, but not on the AChR α promoter. As expected, acetylation on the AChR ε promoter was unaffected (Fig. 6c).

In contrast to the changes in histone acetylation, myogenin and AChR α subunit mRNA levels were not higher in Hdac9–/– innervated muscle (Supplementary Fig. 2), suggesting that alleviating MITR-mediated repression alone is insufficient to allow the expression of activity-responsive genes in innervated muscle.

Hdac9–/– mice have altered timing of myogenin regulationThe perinatal downregulation of myogenin as well as the upregulation of myogenin and AChR genes after denervation were next investigated in Hdac9–/– mice. In wild-type mice, myogenin expression started to decline at ED19, just after MITR had reached its maximal expression (compare Figures 1c and 7a). In Hdac9–/– mice, the decline of myogenin was delayed until the fourth and the fifth postnatal day (Fig. 7a). Thus, in the absence of MITR, perinatal repression of the myogenin gene is greatly delayed but ultimately occurs. Conversely, after denervation, both myogenin and AChR α subunit mRNAs appeared much sooner in Hdac9–/– than in wild-type mice, suggesting that in these Hdac9–/– mice the transcriptional upregulation triggered by denervation takes place precociously because it does not require the prior downregulation of MITR expression (Fig. 7b).

MEF2 mediates MITR action on activity-dependent genesMITR was initially identified in a yeast two-hybrid screen as a partner and repressor of MEF2 activity24–26,33. As MEF2 plays a pivotal role

in the expression of myogenin during muscle differentiation34–36, we investigated its inhibition by MITR in adult muscle and its role in the transcriptional response to denervation (Fig. 8). We first electroporated a MITR mutant lacking the MEF2-binding domain (MITR∆MEF) into muscle31 and observed that it had only a limited effect on the upregulation of myogenin or AChR α subunit mRNAs after denervation, thus suggesting that MITR-mediated repres-sion of these genes requires the interaction with MEF2 (Fig. 8a). Comparable results were obtained by western blotting for myogenin (Fig. 8e). Analysis of the subcellular localization of the MITR∆MEF mutant by immunofluorescence showed that deletion of the MEF2 binding domain of MITR induced a loss of its punctuate nuclear distribution and resulted in a diffuse nuclear distribution, similar to that of HDAC5 (Fig. 5). Together, these results indicate that the interaction with MEF2 is crucial for both the repressive action and the localization of MITR.

To confirm that MITR could indeed inhibit MEF2-dependent transcription in adult muscle, it was co-electroporated with a MEF2 reporter30. Electroporated muscles were denervated and luciferase activity measured 72 h later. As expected, MITR inhibited the expres-sion of the reporter gene as compared to the coexpression of a control plasmid, plasmids encoding the N-terminal region of HDAC5, or the MITR∆MEF mutant (Fig. 8b). This demonstrates that in adult skeletal muscle, MITR represses MEF2-dependent transcription. However, the N-terminal portion of HDAC5 or the MITR∆MEF mutant had no sub-stantial effect on the expression of the reporter gene (Fig. 8b), although they retained the ability to slightly affect the expression of endogenous

Figure 6 Chromatin acetylation is increased in Hdac9–/– muscles. (a,b) Histone H3 acetylation was visualized by immunofluorescence (a) or western blotting (b) in muscle fibers isolated from innervated and 48 h denervated wild-type tibialis anterior muscles (WT I and WT D, respectively) and from innervated Hdac9–/– tibialis anterior muscles (KO I). Nuclei were visualized by Hoechst staining. AcetylH3, antibody to acetylated histone H3; H3, total histone H3. (c) Histone H3 acetylation on the promoters of the myogenin, AChR α and AChR ε subunit genes was quantified in innervated and denervated wild-type (WT) tibialis anterior muscles as well as in innervated Hdac9–/– muscles by real-time quantitative RT-PCR after ChIP with an antibody to acetyl H3 antibody. The acetylation levels on the promoters were arbitrarily set to 1 in the innervated situation. Error bars, s.e.m.

Figure 7 Hdac9–/– mice are sensitized to denervation and have delayed myogenin repression after innervation. (a) Diaphragm muscles were collected in Hdac9–/– and Hdac9+/+ (WT) at different prenatal or postnatal days to measure myogenin expression by real-time quantitative RT-PCR. Error bars, s.e.m. (b) Five-week-old Hdac9–/– mice and Hdac9+/+ (WT) littermates were denervated by sciatic nerve section, and the expression of myogenin (top) and AChR α subunit (bottom) genes in the tibialis anterior muscle was measured at different times by real-time quantitative RT-PCR. Error bars, s.e.m.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

genes (Figs. 4a and 8a), possibly because injected plasmids remain episomal and loosely associated with chromatin37.

Motor innervation represses MEF2 activityTo investigate whether the inhibition of MEF2-dependent transcription is sufficient to suppress myogenin and AChR gene expression in the absence of electrical activity, a MEF2 dominant-negative mutant lack-ing a transactivation domain was overexpressed in denervated muscles to prevent endogenous MEF2 recruitment38. Quantitative RT-PCR and western blotting showed that this MEF2 mutant had the same effects on the denervation response as MITR (Fig. 8a). Thus, inhibition of MEF2 seems to be sufficient to prevent the upregulation of myogenin and AChR subunit genes upon denervation.

Finally, we investigated whether denervation could activate MEF2 using transgenic mice harboring a lacZ transgene controlled by three MEF2 binding sites30,39. In these animals, lacZ is not expressed in most adult muscle fibers, although MEF2 is present, suggesting that it is maintained in a repressed state. Upon denervation, the transgenic mice re-expressed lacZ to different degrees depending on the muscle type (Fig. 8c). Variability in response to denervation is likely to reflect vari-ability among individuals in their response to physiological stimuli39,40. Consistent with the fact that fast and slow muscle fibers respond simi-larly to denervation41, this variability was not correlated with muscle

fiber type. To confirm that the increase in MEF2 activity in response to denervation was not due to an increase in MEF2 expression, the mRNAs coding for the four members of the MEF2 family were quantified in innervated and denervated muscle. Their expression was unaffected by denervation (Fig. 8d). The increase in MEF2 activity induced by denervation is thus most likely to be due to a derepression of its trans-activation potential rather than an increase in its expression.

DISCUSSIONThe neuromuscular system is the best characterized example of gene regulation by neuron-induced activity in excitable cells. In skeletal mus-cle, neuron-induced electrical activity has long been known to control activity-responsive genes. In this study, we have shown that it also con-trols chromatin acetylation. Calcium flux has been shown to be an early mediator of the signals generated by membrane receptors in response to stimulation, but the downstream factors involved in the control of chromatin acetylation by HAT and HDACs have been poorly defined. In this study, we have investigated the possibility that class II HDACs could be involved in the control of both activity-dependent gene expression and chromatin acetylation in response to electrical activity in skeletal muscle. We have shown that HDAC9/MITR is regulated by electrical activity and that it regulates both chromatin acetylation and AChR subunit gene expression in muscle fibers. The action of MITR on the

Figure 8 MITR-mediated repression by innervation is caused by the inhibition of MEF2. (a) pECFP-nuc was co-electroporated with expression vectors for MITR, MEF2 dominant-negative mutant (MEF2dn), or MITR lacking its MEF2 interaction domain (MITR∆MEF) in innervated tibialis anterior muscles as described in Methods. The left tibialis anterior muscle was denervated 10 d later, and after another 48 h, the CFP-positive fibers were microdissected to measure the expression of endogenous myogenin and AChR α subunit by real-time RT-PCR. Error bars, s.e.m. Expression levels correspond to the ratio between the mRNA levels measured in denervated and innervated muscles. (b) A MEF2 reporter plasmid (p2xMEF2RE-Luc; schematic shown at top) and pCMVsportβgal were co-electroporated with expression vectors for MITR, MEF2dn or MITR∆MEF in innervated tibialis anterior muscles. The left tibialis anterior muscle was denervated 10 d later, and after another 48 h, whole-muscle luciferase and β-galactosidase activities were measured. To avoid the bias of electroporation efficacy, for each muscle, the luciferase measurement was divided by the β-galactosidase activity. Error bars, s.e.m. (c) Adult transgenic mice expressing lacZ under the control of the MEF2 binding sites from the desmin promoter were denervated. Their hindlimbs were dissected 3 d later and stained to detect β-galactosidase expression. The plantaris and part of the gastrocnemius muscle are shown. (d) Expression levels of MEF2 family members in innervated and denervated muscle (log scale). Gene expression was evaluated by real-time quantitative RT-PCR on RNA isolated from tibialis anterior muscle either innervated or 48 h after denervation. Error bars, s.e.m. (e) Expression of myogenin, MITR, MITR∆MEF and MEF2dn measured by western blot.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

expression of activity-sensitive genes and on chromatin acetylation is mediated by MEF2, HDAC1 and HDAC3.

MITR regulates myogenin expression through MEF2MITR was first identified as a MEF2-interacting factor capable of repressing activity of MEF224–26,31,33. MEF2 cooperates with bHLH myogenic factors to control skeletal muscle gene expression35,36, includ-ing expression of the myogenin gene, which is a direct transcriptional target of MEF234. The function of MITR in skeletal muscle differentia-tion has been investigated before, but a physiological role in this process has not been demonstrated in vivo. Indeed, myoblast differentiation was not inhibited by MITR overexpression but could nevertheless be blocked by the expression of mutant MITR deprived of its phosphory-latable serines 218 and 448, thus suggesting that MITR is inactivated by phosphorylation during muscle differentiation31. Our in vivo electro-poration experiments demonstrate that in adult skeletal muscle, MITR is active and inhibits MEF2 activity.

Myogenin is one of the transcription factors with the strongest electri-cal activity dependence identified to date12,13. Experiments with MEF2-dependent lacZ transgenic mice show that MEF2 is also very tightly regulated by motor innervation, but at the level of its activity rather than the level of its expression. Altogether, these observations suggest that myogenin expression is regulated by electrical activity through MEF2, which is itself regulated by MITR, which acts as the more upstream tran-scription factor in the regulatory cascade initiated by electrical activity.

MITR sequentially regulates activity-responsive genesOur results show that neuron-elicited electrical activity can coordinately control chromatin acetylation and gene expression through MITR. The ChIP assays suggest that chromatin acetylation specifically affects activity-responsive genes. In the Hdac9–/– mice, the myogenin promoter, but not the AChR α promoter, is hyperacetylated; this suggests that myogenin is a primary target of MITR, and that AChR subunits are regulated indi-rectly through myogenin. In innervated Hdac9–/– muscles, histone H3 acetylation levels are intermediary between innervated and denervated wild-type muscles; this suggests that the acetylation of a first set of genes that includes myogenin is directly regulated by MITR, and the products of these genes then affect the acetylation of terminal acceptor genes, such as AChR subunit genes. This correlates with the fact that bHLH myogenic factors are required for the upregulation of AChR subunit genes after denervation9–13,28,29,42. In addition, we have shown that the effect of MITR on the expression of activity-responsive genes involves MEF2, which is known to regulate the expression of myogenin35,36, whereas AChR genes do not contain MEF2 binding sites.

Altogether, these observations suggest that MITR coordinately regu-lates AChR gene expression and acetylation in a multistep process, ini-tially affecting the acetylation and expression of primary target genes, through MEF2, and HDAC1 and HDAC3. The expression of these genes, including myogenin, then finally triggers the acetylation and expression of AChR subunit genes.

MITR has several target genes beside myogeninImmunofluorescence with anti–acetyl histone H3 has shown many acetylated loci in the genome, thus suggesting that the myogenin gene is not the only target of MITR. Notably, MITR is located at discrete nuclear dots that are much larger than the acetyl-H3 dots. This could suggest that the genes controlled by MITR are clustered in transcrip-tionally inactive regions of the nucleus and that upon activation they are redistributed in the nucleus, as suggested by the histone H3 acetyla-tion pattern. Such gene clustering has already been described in several cases of gene silencing43–45.

The ChIP assays also shows that in innervated Hdac9–/– muscles chromatin is acetylated on the myogenin promoter, although myogenin is not expressed. This demonstrates that removing HDACs from the myogenin promoter is required but not sufficient to allow its expression and suggests that it is possible to uncouple chromatin acetylation from gene expression. It is therefore likely that in addition to MITR, HDAC1 and HDAC3, non-HDAC inhibitors repress myogenin expression in the presence of electrical activity. Thus, the expression of myogenin is probably blocked at several levels, and removing a single inhibitor is not sufficient to allow its expression. The fact that in Hdac9–/– mice, the perinatal downregulation of myogenin is delayed but finally occurs favors this hypothesis.

MITR action involves other factors beside MEF2That the MITR∆MEF mutant retained some inhibitory activity on the expression of endogenous activity-dependent genes indicates that MITR could have additional targets besides MEF2. In addition, MITR∆MEF inactivity on episomal reporter genes suggests that these additional MITR targets are chromatin-specific regulatory factors. Accordingly, our results show that MITR recruits HDAC1 and HDAC3; class II HDACs, including MITR, have been shown to interact with the transcriptional repressor BCL6 and the heterochromatin protein HP1 (refs. 46,47).

A new regulatory mechanism for class II HDACsCalcium has a critical role in mediating the effect of electrical activity on gene expression27,48, and the regulation of MEF2 activity by calcium has been extensively documented49. Calcium has been shown to regu-late MEF2 activity during myocyte hypertrophy as well as fiber type specification by modulating class II HDAC phosphorylation. It thereby regulates the nucleocytoplasmic localization of the class I HDACs and thus controls the amount of nuclear class II HDACs available to inter-act with (and repress) MEF2 (refs. 19–21,23,31). Our results suggest that, depending on the physiological stimulus, calcium can regulate MEF2 through another mechanism involving the modulation of MITR at the level of its expression rather than the level of its phosphoryla-tion. The 350% increase of MITR expression after innervation and the 80% decrease of MITR expression upon blockade of the electrical activity evoked by the nerve constitute the first demonstration of the regulation of a class II HDAC family member at the level of its mRNA rather than at the level of its subcellular localization.

In vivo MITR has a specific action among class II HDACsThe cytoplasmic localization of HDAC4 and the nuclear localization of HDAC5 and MITR in muscle fibers are reminiscent of what was previously observed in cultured myoblasts19,20,23. However, the highly divergent nuclear distribution of HDAC5 and MITR was not reported in cultured myoblasts. These distinct distributions correlate with the fact that MITR (not corresponding N-terminal region of HDAC5) can inhibit MEF2 in skeletal muscle, whereas they both inhibit MEF2 in cultured cells25,26,30,31. Consistently, the inactive MITR mutant lacking a MEF2 interaction domain was localized similarly to HDAC5 in muscle fibers. Altogether, these results suggest that the functional redundancy observed in vitro among class II HDACs does not exist in vivo.

MITR: a stress-response molecular integrator?The regulation of muscle chromatin acetylation and electrical activ-ity–responsive genes in innervated muscle is the first physiologi-cal role that can be attributed to MITR in skeletal muscle. In the Hdac9–/– mice, skeletal muscles presented no obvious perturbation under normal conditions but were sensitized to denervation. These observations are reminiscent of the cardiac phenotype of these

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

mice. Indeed, no evident cardiac phenotype could be observed in Hdac9–/– or Hdac5–/– embryos of young animals, whereas the adult mice were sensitized to hypertrophic stimuli and showed profound stress-dependent cardiomegaly. This suggested that these HDACs act as suppressors of the signaling mechanism that conveys a stress signal to hypertrophic growth of the heart32,50.

Altogether, the observations made in the heart and skeletal muscle suggest that the Hdac9 gene is preferentially involved in the coordinated regulation of chromatin acetylation and gene expression in physiologi-cal and stress responses specific to postnatal life.

METHODSAnimal care. The animals were housed individually and provided with mouse chow and water ad libitum in a restricted-access, specific-pathogen-free animal care facility at the Ecole Normale Superieure of Lyon.

For operative procedures, mice were anaesthetized with an intraperitoneal injection of ketamine (100 mg kg–1) and xylazine (10 mg kg–1) to obtain a deep state of general anesthesia. All operative procedures were performed using asep-tic techniques and according to the local ethical committee recommendations (Comité Rhône Alpes d’Ethique pour l’Expérimentation Animale).

Denervation. The tibialis anterior muscles were denervated by sciatic nerve transection. Briefly, after the mice were anesthetized, the sciatic nerve was exposed in the thigh and doubly cut (5–10 mm apart) just distal to the sciatic notch. Contralateral muscles that were not operated on were used as controls.

Constructs. pECFP-nuc (Clontech) and pCMVsportβgal (Invitrogen) encode a nuclear cyan fluorescent protein and β-galactosidase, respectively, placed down-stream of the CMV promoter. pHDAC4, pHDAC5, pHDAC5 N-ter (which cor-responds to amino acids 123–673 of HDAC5) and p2xMEF2RE-Luci have been previously described30. pMITR, pMITR∆MEF and pMEF2Adn were generated in pcDNA3 (Invitrogen). As previously described31,38, MITR∆MEF lacks the MEF2 binding domain between amino acids 135 and 152, and the MEF2A dominant-negative mutant corresponds to amino acids 1–131 of MEF2A.

Immunodetection. The various MITR, MEF2 and HDAC5 proteins contained an N-terminal HA tag that allowed their detection by immunofluorescence or western blotting (HA 11, Babco).

When immunostaining had to be compared, the different samples were pro-cessed in parallel. The images were acquired on an Axioplan fluorescence micro-scope (Zeiss) with a Coolsnap digital camera (Nikon), and to avoid saturation, the optimal acquisition time was determined for the samples with the greatest staining, and the same acquisition time was used for the other samples.

In vivo electroporation. Expression vectors were injected into tibialis anterior muscles (5 µg of total DNA in 30 µl of 0.9% NaCl) of 5-week-old Swiss Webster male mice. Injected muscles were then electroporated with 1-cm2 plaque elec-trodes placed on each side of the leg and eight 200 V cm–1 pulses of 20 ms applied at 2 Hz. The electroporated DNA mixtures contained 2 µg of p2XMEF2RE-Luci and 1 µg of pCMVsportβgal reporter plasmid for subsequent quantification of luciferase activity and normalization to β-galactosidase activity on tibialis ante-rior muscle homogenates (FastPrep, Bio 101, in passive lysis buffer, Promega), or the mixture contained 2 µg of pECFP-Nuc to allow the visualization and microdissection of electroporated fibers for further gene expression analysis by real-time quantitative RT-PCR or by western blot.

Specifically, electroporated tibialis anterior muscles were quickly collected and placed in cold 1×. Groups of CFP-positive fibers consisting of about ten fibers were isolated under a fluorescence microscope (SZX 12, Olympus) using ultrafine forceps (Moria, No. 5). Non-GFP-positive extremities of the fibers were eliminated, and CFP-positive portions were placed in RLT buffer (RNeasy mini Kit, Qiagen) to prevent RNA degradation or in passive lysis buffer (Promega) supplemented with protease inhibitor cocktail (Complete, Roche).

Electroporations were performed in quintuplicate (when amount of endo-genous mRNA was quantified) or in duplicate (when reporter gene expression was measured). All experiments were repeated at least three times. The results were subjected to ANOVA and all P values were below 0.01.

Real-time quantitative RT-PCR. Gene expression was evaluated by real-time quantitative RT-PCR (LightCycler, Roche) using the LightCycler FastStart DNA Master SYBR Green 1 RT-PCR kit (Roche) on total RNA extracted (RNeasy mini, Qiagen) from homogenized (FastPrep, Bio 101, in RLT buffer, Qiagen) whole tibialis anterior muscles or microdissected fibers. The measures were normalized to β-actin (Actb) RNA levels. For RNA extraction from embryos, the diaphragm muscle was used.

The sequences of the primers were as follows [AU: Because these refer to genes, we have added the MGI-approved gene symbols; please confirm these are the genes from which the primers were derived.]: β-actin (Actb) forward 5′-CCCTGTATGCCTCTGGTCGT-3′, reverse 5′-ATGGCGTGAGGGAGAGCAT-3′; MITR forward 5′-GCGGTCCAGGTTAAAACAGA-3′, reverse 5′-GAGCTGAAGCCTCATTTTCG-3′; myogenin (Myog) forward 5′-CTACAGGCCTTGCTCAGCTC-3′, reverse 5′-AGATTGTGGGCGTCTGTAGG-3′; AChR α (Chrna1) forward 5′-ACCTGGACCTATGACGGCTCT-3′, reverse 5′-AGTTACTCAGGTCGGGCTGGT-3′; AChR ε (Chrne) forward 5′-CTTGGTGCTGCTCGCTTACTT-3′, reverse 5′-CGTTGATAGAGACCGTGCATT-3′; muscle creatine kinase (Ckm) forward 5′-GACGAAGGCGAGTGAGAATC-3′, reverse 5′-CATGGAGAAGGGAGGCAATA-3′; acetylcholinester-ase (Ache) forward 5′-GGTTCCCACTCGGTAGTTCA-3′, reverse 5′-CCTGGGTTTGAGGGTACTGA-3′; MEF2A (Mef2a) forward 5′-GAATGCCCAAAGGATAAGCA-3′, reverse 5′-CAGCATTCCAGGGGAAGTAA-3′; MEF2B (Mef2b) forward 5′-ACTTTCACCAAGCGCAAGTT-3′, reverse 5′-CTCGGTGTATTTGAGCAGCA-3′; MEF2C (Mef2c) forward 5′-ATCTGCCCTCAGTCAGTTGG-3′, reverse 5′-CAGCTGCTCAAGCTGTCAAC-3′; MEF2D (Mef2d) forward 5′-CCGTTTCTCTCAGCAACCTC-3′, reverse 5′-CGGTCTCATAGGATCCTCCA-3′.

ChIP assay. Tibialis anterior muscles from wild-type or Hdac9–/– mice were dis-sected and fixed (1% formalin, 1× PBS) for 20 min at 4 °C. After fixation, glycine was added to a final concentration of 0.125 M, and the samples were washed three times with 1× PBS. Muscles were resuspended in lysis buffer (10 mM EDTA, 1% SDS, 50 mM Tris, pH 8.0, and 1 mM PMSF) and homogenized using Fast Prep (Bio 101). Nuclei were collected and resuspended in lysis buffer and sonicated. The average size of DNA fragments was approximately 1,200 bp. The same amount of chromatin was used to perform immunoprecipitations with anti–acetylated histone H3 (Upstate). Acetylated histone-DNA complexes were purified using a ChIP assay kit (Upstate). The presence of promoters was analyzed by quantitative PCR (LightCycler FastStart DNA Master SYBR Green 1 PCR kit, Roche).

Primer sequences were as follows: myogenin promoter forward 5′-AGAGGGAAGGGGAATCACAT-3′, reverse 5′-AAGGCTTGTTCCTGCCACT-3′; AChR α promoter forward 5′-TTTGATGGGAACACAGGACA-3′, reverse 5′-GCGCTACAATGATCACATGG-3′; AChR ε promoter forward 5′-GATGACAGGCCTTGTGGATT-3′, reverse 5′-ACAAGCTTGAGGGAACAGG-3′; H4 promoter forward 5′-GACACCGCATGCAAAGAATAGCTG-3′, reverse 5′-CTTTCCCAAGGCCTTTACCACC-3′.


ACKNOWLEDGMENTSWe thank E. Chopin for technical assistance in chromatin immunoprecipitation (ChIP) assay; C. Antos for kind and helpful assistance; and A. Ravel-Chapuis, M. Vandromme, K. Ancelin, A. de Kerchove d'Exaerde, Y.-G. Gangloff and A. Sergeant for fruitful discussions and critical reading of the manuscript. This work was supported by the Association Française contre les Myopathies, the Centre National de la Recherche Scientifique, the Ministere de l’Education Nationale, de la Recherche et de la Technologie, and grants from the U.S. National Institutes of Health, the Texas Advanced Technology Program, The Robert A. Welch Foundation and the Donald W. Reynolds Foundation to E.N.O. and R.B.D.



1. Guan, Z. et al. Integration of long-term-memory-related synaptic plasticity involves bidirectional regulation of gene expression and chromatin structure. Cell 111, 483–493 (2002).

2. Alarcon, J.M. et al. Chromatin acetylation, memory, and LTP are impaired in CBP+/−

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

mice: a model for the cognitive deficit in Rubinstein-Taybi syndrome and its amelioration. Neuron 42, 947–959 (2004).

3. Korzus, E., Rosenfeld, M.G. & Mayford, M. CBP histone acetyltransferase activity is a critical component of memory consolidation. Neuron 42, 961–972 (2004).

4. Sanes, J.R. & Lichtman, J.W. Induction, assembly, maturation and maintenance of a postsynaptic apparatus. Nat. Rev. Neurosci. 2, 791–805 (2001).

5. Schaeffer, L., Duclert, N., Huchet-Dymanus, M. & Changeux, J.P. Implication of a mul-tisubunit Ets-related transcription factor in synaptic expression of the nicotinic acetyl-choline receptor. EMBO J. 17, 3078–3090 (1998).

6. Fromm, L. & Burden, S.J. Synapse-specific and neuregulin-induced transcription require an ets site that binds GABPα/GABPβ. Genes Dev. 12, 3074–3083 (1998).

7. Schaeffer, L. de Kerchove d’Exaerde, A. & Changeux, J. P. Targeting transcription to the neuromuscular synapse. Neuron 31, 15–22 (2001).

8. Gundersen, K., Rabben, I., Klocke, B.J. & Merlie, J.P. Overexpression of myogenin in muscles of transgenic mice: interaction with Id-1, negative crossregulation of myogenic factors, and induction of extrasynaptic acetylcholine receptor expression. Mol. Cell. Biol. 15, 7127–7134 (1995).

9. Piette, J., Bessereau, J.L., Huchet, M. & Changeux, J.P. Two adjacent MyoD1-binding sites regulate expression of the acetylcholine receptor alpha-subunit gene. Nature 345, 353–355 (1990).

10. Duclert, A., Piette, J. & Changeux, J.P. Influence of innervation of myogenic factors and acetylcholine receptor alpha-subunit mRNAs. Neuroreport 2, 25–28 (1991).

11. Witzemann, V. & Sakmann, B. Differential regulation of MyoD and myogenin mRNA levels by nerve induced muscle activity. FEBS Lett. 282, 259–264 (1991).

12. Eftimie, R., Brenner, H.R. & Buonanno, A. Myogenin and MyoD join a family of skeletal muscle genes regulated by electrical activity. Proc. Natl. Acad. Sci. USA 88, 1349–1353 (1991).

13. Merlie, J.P., Mudd, J., Cheng, T.C. & Olson, E.N. Myogenin and acetylcholine receptor alpha gene promoters mediate transcriptional regulation in response to motor innervation. J. Biol. Chem. 269, 2461–2467 (1994).

14. Bessereau, J.L., Laudenbach, V., Le Poupon, C. & Changeux, J.P. Nonmyogenic factors bind nicotinic acetylcholine receptor promoter elements required for response to denerva-tion. J. Biol. Chem. 273, 12786–12793 (1998).

15. Sartorelli, V., Huang, J., Hamamori, Y. & Kedes, L. Molecular mechanisms of myogenic coactivation by p300: direct interaction with the activation domain of MyoD and with the MADS box of MEF2C. Mol. Cell. Biol. 17, 1010–1026 (1997).

16. Puri, P.L. et al. Differential roles of p300 and PCAF acetyltransferases in muscle dif-ferentiation. Mol. Cell 1, 35–45 (1997).

17. Polesskaya, A. et al. Interaction between acetylated MyoD and the bromodomain of CBP and/or p300. Mol. Cell. Biol. 21, 5312–5320 (2001).

18. McKinsey, T.A., Zhang, C.L. & Olson, E.N. Control of muscle development by dueling HATs and HDACs. Curr. Opin. Genet. Dev. 11, 497–504 (2001).

19. Lu, J., McKinsey, T.A., Zhang, C.L. & Olson, E.N. Regulation of skeletal myogenesis by association of the MEF2 transcription factor with class II histone deacetylases. Mol. Cell 6, 233–244 (2000).

20. McKinsey, T.A., Zhang, C.L., Lu, J. & Olson, E.N. Signal-dependent nuclear export of a histone deacetylase regulates muscle differentiation. Nature 408, 106–111 (2000).

21. McKinsey, T.A., Zhang, C.L. & Olson, E.N. Activation of the myocyte enhancer factor-2 transcription factor by calcium/calmodulin-dependent protein kinase-stimulated bind-ing of 14–3-3 to histone deacetylase 5. Proc. Natl. Acad. Sci. USA 97, 14400–14405 (2000).

22. Dressel, U. et al. A dynamic role for HDAC7 in MEF2-mediated muscle differentiation. J. Biol. Chem. 276, 17007–17013 (2001).

23. McKinsey, T.A., Zhang, C.L. & Olson, E.N. Identification of a signal-responsive nuclear export sequence in class II histone deacetylases. Mol. Cell. Biol. 21, 6312–6321 (2001).

24. Zhou, X., Richon, V.M., Rifkind, R.A. & Marks, P.A. Identification of a transcriptional repressor related to the noncatalytic domain of histone deacetylases 4 and 5. Proc. Natl. Acad. Sci. USA 97, 1056–1061 (2000).

25. Zhou, X., Marks, P.A., Rifkind, R.A. & Richon, V.M. Cloning and characterization of a histone deacetylase, HDAC9. Proc. Natl. Acad. Sci. USA 98, 10572–10577 (2001).

26. Sparrow, D.B. et al. MEF-2 function is modified by a novel co-repressor, MITR. EMBO

J. 18, 5085–5098 (1999).27. Klarsfeld, A. et al. Regulation of muscle AChR alpha subunit gene expression by electrical

activity: involvement of protein kinase C and Ca2+. Neuron 2, 1229–1236 (1989).28. Bessereau, J.L., Stratford-Perricaudet, L.D., Piette, J., Le Poupon, C. & Changeux, J.P.

In vivo and in vitro analysis of electrical activity-dependent expression of muscle acetyl-choline receptor genes using adenovirus. Proc. Natl. Acad. Sci. USA 91, 1304–1308 (1994).

29. Walke, W., Xiao, G. & Goldman, D. Identification and characterization of a 47 base pair activity-dependent enhancer of the rat nicotinic acetylcholine receptor delta-subunit promoter. J. Neurosci. 16, 3641–3651 (1996).

30. Lemercier, C. et al. mHDA1/HDAC5 histone deacetylase interacts with and represses MEF2A transcriptional activity. J. Biol. Chem. 275, 15594–15599 (2000).

31. Zhang, C.L., McKinsey, T.A. & Olson, E.N. The transcriptional corepressor MITR is a signal-responsive inhibitor of myogenesis. Proc. Natl. Acad. Sci. USA 98, 7354–7359 (2001).

32. Zhang, C.L. et al. Class II histone deacetylases act as signal-responsive repressors of cardiac hypertrophy. Cell 110, 479–488 (2002).

33. Zhang, C.L., McKinsey, T.A., Lu, J.R. & Olson, E.N. Association of COOH-terminal-bind-ing protein (CtBP) and MEF2-interacting transcription repressor (MITR) contributes to transcriptional repression of the MEF2 transcription factor. J. Biol. Chem. 276, 35–39 (2001).

34. Cheng, T.C., Wallace, M.C., Merlie, J.P. & Olson, E.N. Separable regulatory elements governing myogenin transcription in mouse embryogenesis. Science 261, 215–218 (1993).

35. Molkentin, J.D., Black, B.L., Martin, J.F. & Olson, E.N. Cooperative activation of muscle gene expression by MEF2 and myogenic bHLH proteins. Cell 83, 1125–1136 (1995).

36. Black, B.L. & Olson, E.N. Transcriptional control of muscle development by myocyte enhancer factor-2 (MEF2) proteins. Annu. Rev. Cell Dev. Biol. 14, 167–196 (1998).

37. Wolff, J.A. et al. Direct gene transfer into mouse muscle in vivo. Science 247, 1465–1468 (1990).

38. Ornatsky, O.I., Andreucci, J.J. & McDermott, J.C. A dominant-negative form of transcrip-tion factor MEF2 inhibits myogenesis. J. Biol. Chem. 272, 33271–33278 (1997).

39. Naya, F.J., Wu, C., Richardson, J.A., Overbeek, P. & Olson, E.N. Transcriptional activity of MEF2 during mouse embryogenesis monitored with a MEF2-dependent transgene. Development 126, 2045–2052 (1999).

40. Wu, H. et al. MEF2 responds to multiple calcium-regulated signals in the control of skeletal muscle fiber type. EMBO J. 19, 1963–1973 (2000).

41. Hyatt, J.P., Roy, R.R., Baldwin, K.M. & Edgerton, V.R. Nerve activity-independent regula-tion of skeletal muscle atrophy: role of MyoD and myogenin in satellite cells and myonu-clei. Am. J. Physiol. Cell Physiol. 285, C1161–C1173 (2003).

42. Tang, J., Jo, S.A. & Burden, S.J. Separate pathways for synapse-specific and electrical activity-dependent gene expression in skeletal muscle. Development 120, 1799–1804 (1994).

43. Hediger, F. & Burden, S.J. Nuclear organization and silencing: putting things in their place. Nat. Cell Biol. 4, E53–E55 (2002).

44. Fisher, A.G. & Merkenschlager, M. Gene silencing, cell fate and nuclear organization. Curr. Opin. Genet. Dev. 12, 193–197 (2002).

45. Baxter, J., Merkenschlager, M. & Fisher, A.G. Nuclear organization and gene expression. Curr. Opin. Cell Biol. 14, 372–376 (2002).

46. Lemercier, C. et al. Class II histone deacetylases are directly recruited by BCL6 transcrip-tional repressor. J. Biol. Chem. 277, 22045–22052 (2002).

47. Zhang, C.L., McKinsey, T.A. & Olson, E.N. Association of class II histone deacetylases with heterochromatin protein 1: potential role for histone methylation in control of muscle differentiation. Mol. Cell. Biol. 22, 7302–7312 (2002).

48. Xu, Q. et al. p38 Mitogen-activated protein kinase-, calcium-calmodulin-dependent protein kinase-, and calcineurin-mediated signaling pathways transcriptionally regulate myogenin expression. Mol. Biol. Cell 13, 1940–1952 (2002).

49. McKinsey, T.A., Zhang, C.L. & Olson, E.N. MEF2: a calcium-dependent regulator of cell division, differentiation and death. Trends Biochem. Sci. 27, 40–47 (2002).

50. Chang, S. et al. Histone deacetylase 5 and 9 govern responsiveness of the heart to a subset of stress signals and play redundant roles in heart development. Mol. Cell. Biol. 24, 8467–8476 (2004).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Soluble CPG15 expressed during early development rescues cortical progenitors from apoptosisUlrich Putz1,2, Corey Harwell1,2 & Elly Nedivi1

The balance between proliferation and apoptosis is critical for proper development of the nervous system. Yet, little is known about molecules that regulate apoptosis of proliferative neurons. Here we identify a soluble, secreted form of CPG15 expressed in embryonic rat brain regions undergoing rapid proliferation and apoptosis, and show that it protects cultured cortical neurons from apoptosis by preventing activation of caspase 3. Using a lentivirus-delivered small hairpin RNA, we demonstrate that endogenous CPG15 is essential for the survival of undifferentiated cortical progenitors in vitro and in vivo. We further show that CPG15 overexpression in vivo expands the progenitor pool by preventing apoptosis, resulting in an enlarged, indented cortical plate and cellular heterotopias within the ventricular zone, similar to the phenotypes of mutant mice with supernumerary forebrain progenitors. CPG15 expressed during mammalian forebrain morphogenesis may help balance neuronal number by countering apoptosis in specific neuroblasts subpopulations, thus influencing final brain size and shape.

During mammalian evolution, the cerebral cortex has greatly expanded through a tremendous increase in the number of cortical neurons. The surface of the cortical plate has extended and become indented and con-voluted as a result of neuron addition in columnar radial units1. At the onset of cortical neurogenesis, the proliferative population of founder cells is confined to the ventricular zone of the embryonic cerebral wall2. Even modest alterations in the size of this progenitor population during its early exponential growth phase can markedly affect final neuronal numbers1,3. Thus, it has been proposed that cellular mechanisms that influence founder cell number may underlie the telencephalic expansion and sculpting that are characteristic of mammalian forebrain develop-ment and evolution1. Apoptosis within the founder population is one putative mechanism for influencing eventual brain size and shape4,5.

It has recently been recognized that the role of apoptosis in brain development extends beyond matching of neuronal populations with their appropriate target fields, as specified in the ‘neurotrophic hypoth-esis’5,6. Caspase 3, a key enzyme in the mammalian apoptotic pathway, is expressed at high levels in the mouse cerebral wall around embryonic day 12 (E12)7, when dying cells are prevalent in proliferative zones of the cerebral cortex8–10. Consistent with these observations, mutant mice deficient in the pro-apoptotic genes Casp3, Casp9 and Apaf1 show gross nervous system malformations resulting from improper expan-sion of specific neural progenitor populations11–15. The excess neurons in some of these mutants are added as extra radial units, expanding the surface of the cortical plate, rather than influencing its thickness. The cortical plate, with increased size, forms convolutions resembling the gyri and sulci of the primate brain. In addition, later generated cells accumulate below the cortical plate, forming heterotopic cell masses within the ventricles11,12. Despite the essential role of the core

apoptotic pathway in brain morphogenesis, little is known about the signals regulating apoptosis of proliferative neurons. Identification of the molecules involved is vital to understanding the complex morpho-genetic processes that shape the mammalian brain.

cpg15 (also known as Nrn1) was identified in a screen for activity- regulated genes involved in synaptic plasticity16,17 and encodes a small, highly conserved protein18-19 (also termed neuritin-1). In a membrane-bound form attached by a glycosylphosphatidylinositol (GPI) link, CPG15 has been shown to function non–cell autonomously to coordinately regulate growth of apposing dendritic and axonal arbors, and to promote synaptic maturation19,20. As cpg15 is an activity- regulated gene, late cpg15 expression is contemporaneous with critical periods for activity-dependent plasticity and requires action potential activity. However, cpg15 is also expressed in an activity-independent manner during early brain development before circuit formation and maturation21,22, suggesting that it may have a different role at this stage. We hypothesized that, like the neurotrophic factors, CPG15 has multiple roles during nervous system development. In addition to its previously characterized role as a growth and differentiation factor that affects process outgrowth and synaptic maturation, CPG15 may also function as a survival factor during early brain development. Here we describe the identification of a soluble CPG15 expressed in the embryonic brain that regulates survival of cortical progenitors by preventing caspase-mediated apoptosis.

RESULTScpg15 is expressed in embryonic proliferative zonesTo examine localization of early, activity-independent cpg15 expression, we performed in situ hybridizations on sections from embryonic rat

1The Picower Center for Learning and Memory, Departments of Brain and Cognitive Sciences and Biology, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA. 2These authors contributed equally to this work. Correspondence should be addressed to E.N. ([email protected]).

Published online 13 February 2005 2005; doi:10.1038/nn1407

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

brains. At the earliest times tested, embryonic days 14 (E14) and 15 (E15), cpg15 mRNA is present in the cortical plate, in the ventricular zone of the dorsal thalamus and in retinal ganglion cells (Fig. 1a–f). At E17–E19, cpg15 is expressed in the telencephalic and dorsal diencephalic subventricular zones (Fig. 1g–l), is expressed in the hippocampal primordia (Fig. 1i–j), and at postnatal day 7 (P7) appears in the external granular layer of the cerebellum (Fig. 1m). In all these regions early cpg15 expression is temporally correlated with expansion of the progenitor pool and apoptotic elimination of superfluous neuroblasts10. cpg15 is not expressed in all proliferative zones and is markedly absent from the olfactory epithelium and gan-glionic eminence (Fig. 1a,b,e,f,i,j), suggesting that its function may be cell type specific. cpg15 is also expressed when target-derived tro-phic support is crucial for protection from apoptosis used to match neuron number with target size. From E19 to P7, cpg15 mRNA is present in the trigeminal ganglia, sensory thalamus and various brainstem nuclei (Fig. 1i,j,m), at times of afferent ingrowth, tar-get selection and synaptogenesis in these structures. From P0, cpg15 expression in the cerebral cortex is downregulated to undetectable levels22, coincident with cessation of apoptosis in this region8,9. At P14, cpg15 mRNA re-appears, not in the ventricular or subventricular zones, but in the differentiated cortical layers (Fig. 1n), where activ-ity-dependent plasticity is thought to occur postnatally. cpg15 mRNA patterns are thus consistent with an early role as a survival factor

during brain morphogenesis, and a later role in structural remodeling and synaptic maturation associated with developmental and adult plasticity.

CPG15 is primarily expressed in a soluble, secreted formTo generate CPG15 for testing in a survival assay, we cloned full-length CPG15 tagged with a Flag epitope into a vector containing an inter-nal ribosomal entry site (IRES) for enhanced green fluorescent protein (EGFP) coexpression (pIRES-EGFP-CPG15-FLAG, Fig. 2a) and expressed it in HEK293T cells. The Flag tag allowed detection and subsequent affinity purification of the CPG15 protein, whereas EGFP marked transfected cells. Consistent with CPG15’s GPI link to the cell surface18,19, immuno-histochemistry with a monoclonal antibody to Flag (anti-Flag) showed membrane staining of transfected cells (data not shown). Notably, we observed CPG15 staining in untransfected EGFP- negative cells, suggesting intercellular transfer of CPG15 from transfected cells to their untransfected neighbors. To test whether cell-to-cell contact is necessary for intercellular transfer of CPG15, we cocultured transfected and untransfected HEK293T cells on the same coverslip but in distinct locations without physi-cal contact (Fig. 2b). Anti-Flag immunohisto-chemistry showed CPG15 membrane staining of all untransfected cells (Fig. 2c), suggest-ing that a soluble form of CPG15 can diffuse between isolated cells. Cells cocultured with cells expressing vector alone (data not shown) or a control Flag-tagged cytoplasmic protein showed no membrane staining (Fig. 2d).

We verified the presence of soluble CPG15 in supernatants from CPG15-transfected HEK293T cells by western blot analysis (Fig. 2e). Anti-Flag staining showed two CPG15-specific bands of distinct molec-ular weights in whole-cell extracts. The lower-molecular-weight protein was also present in the supernatant fraction. Treatment of CPG15-transfected cells with phospholipase C (to promote cleavage of GPI anchors and release GPI-anchored proteins from the cell surface23) resulted in a disappearance of the higher molecular weight protein from the cell extracts and its concurrent appearance in the superna-tant fraction (Fig. 2e). The lower-molecular-weight protein remained unaffected by phospholipase C treatment. These results suggest that the higher-molecular-weight protein represents membrane-bound GPI-linked CPG15 and that the smaller protein is a soluble form of CPG15.

To determine whether both CPG15 forms are expressed in vivo and at what developmental times, we prepared membrane and soluble protein fractions from brains of E14, E18 and adult rats and examined them by western blot analysis using an antibody against CPG1519. In both embry-onic brains and adult cortex, CPG15 was detected predominantly in the soluble protein fractions, with low levels of the membrane-bound protein detected only in the adult (Fig. 2f). As the prevalent form of CPG15 in vivo is soluble, this form is likely the primary mediator of CPG15’s early role during embryonic brain development and possibly of its later role as an activity-regulated growth and differentiation factor19,20.

Figure 1 cpg15 mRNA expression is biphasic. (a–n) cpg15 in situ hybridizations on (a–l) coronal sections through the telencephalon during prenatal development, and (m,n) on saggital sections through postnatal brains (ages designated at left). Dark-field photomicrographs of embryonic times are shown in left column; on the right are bright-field photomicrographs of the same sections counterstained with toluidine blue and overlaid with their dark-field views. Marked are the third ventricle (3V), dorsal thalamus (DT), lateral ventricle (LV), retina (RET), olfactory epithelium (OE), cortical plate (CP), subventricular zone (SVZ), ganglionic eminence (GE), trigeminal ganglion (TG), hippocampus (HI), neocortex (CTX), sensory thalamus (sTH), external granular layer (EGL), inferior colliculus (IC) and superior colliculus (SC). Scale bars: a,b,e–j, 0.5 mm; c,d,k,l, 150 µm; m,n, 1 mm.

E14

P7

E19

E17

E19

E15

E14

P14

i

CTXEGL

sTH

CP

IZ

SVZ

VZ

3V CP

a b

c d

e

i

f

j

g

k

n

m

h

l

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

CPG15 rescues cultured cortical neurons from apoptosisThe presence of soluble CPG15 in brain extracts at early developmental times and the localization of its mRNA to specific proliferative populations and to neurons at their target selection phase led us to test whether it may serve as a survival factor that protects against apoptosis. We examined whether solu-ble CPG15, affinity purified from supernatants of CPG15-Flag–transfected HEK293T cells, was capable of preventing cultured cortical neurons from undergoing apoptosis. Using Hoechst 33324 staining to identify cells with fragmented nuclei, we counted the number of apoptotic neurons in untreated cortical cultures and in cultures after growth factor deprivation (starvation), with or without addition of purified CPG15 (Fig. 3a–c). Growth factor depriva-tion more than doubled the percentage of apoptotic neurons in the cultures, from approximately 15% to 40%. The increased apoptosis could be completely prevented by addition of soluble CPG15, but not by addition of affinity column elution buffer (Fig. 3d). To confirm that CPG15 was rescuing neurons from apoptotic rather than necrotic cell death, the treated and control cultures were immunostained with an antibody specific to the p17 subunit of activated caspase 3, a key component of the apoptotic path-way in brain development11. All neurons contain-ing pyknotic nuclei visualized by Hoechst staining also stained positive for the p17 cleavage product of activated caspase 3 (Fig. 3e–j). Independent quanti-

fication of immunostained neurons expressing cleaved caspase 3 showed that starvation more than doubled their number. CPG15 provision completely pre-vented the increased caspase 3 activation induced by starvation (Fig. 3k). We conclude that soluble CPG15 protects cortical neurons from apoptosis by pre-venting activation of caspase pathways induced by growth factor deprivation.

Figure 2 CPG15 is predominantly expressed in a soluble secreted form. (a) Schematic of CPG15-FLAG-IRES-EGFP construct. The Flag peptide sequence (gray) was inserted between the secretion signal (SS) and the CPG15 core domain. Tagged, full-length CPG15 was then cloned upstream of an internal ribosome entry site (IRES) and EGFP. (b) HEK293T cells transfected with CPG15-FLAG-IRES-EGFP (green) were cocultured with untransfected cells (white) for 2 d. (c) Untransfected cells (outlined by square in b) show CPG15 immunoreactivity (white arrow). Scale bar, 10 µm. (d) HEK293T cells cocultured with cells transfected with a Flag-tagged control protein show no membrane staining. (e) Western blots of cell extracts and supernatants from CPG15-FLAG-IRES-EGFP transfected HEK293T cells either untreated or treated with phospholipase C (PLC). (f) Western blots of membrane and soluble fractions from E14 and E18 brains, and adult cortices, probed with antibody against CPG15. Staining for the transferrin receptor (Tf-rec.) and the enzyme Akt serve as membrane- and soluble-fraction controls, respectively. Ponceau staining (Pon.) serves as a loading control.

G

Starvation – + + Protein – – CPG15

20

40

60

80

Cle

aved

cas

p. 3

(p

erce

ntag

e of

tota

l)

* *

Ant

i-cle

aved

cas

pase

3

Starvation+ CPG15Starvation

Hoe

chst

Ove

rlay

Control

Starvation – + + +Protein – – CPG15 TE

20

40

60

80

Apo

ptot

ic c

ells

(per

cent

age

of to

tal)

* * *

*

a b c

e f g

h i j k

d

Figure 3 Soluble CPG15 rescues cortical neurons from apoptosis induced by growth factor deprivation (starvation). (a,b) Starvation increased the number of fragmented nuclei seen with Hoechst staining (marked by arrows). (c) CPG15 addition prevented this increase. (d) Quantification of a–c. Starvation significantly increased the percentage of apoptotic neurons (*P < 0.002). CPG15 application prevented this increase (** P < 0.003). (e,f) Starvation induces apoptosis, as seen by increased numbers of neurons immunopositive for cleaved caspase 3. (g) Purified CPG15 prevented the starvation-induced increase in cleaved caspase 3 immunoreactivity. (h–j) Overlay of a and e, b and f, and c and g shows that cells with fragmented nuclei scored by Hoechst staining contain activated caspase 3. (k) Quantification of e–g. Starvation significantly increased the percentage of cleaved caspase 3–immunopositive neurons (* P < 0.001). CPG15 application prevented this increase (**P < 0.002). Scale bar, 10 µm.

SS CPG15 EGFPGPI IRES

FlagCPG15-Flag

CPG15-Flag

Mem

.Sol.

Mem

.Sol.

Mem

.Sol.

E14 E18 Adult

Pon

.

CP

G15

Tf-

rec.

Akt

2 d

– PLC +PLC

kD

25

15

10

Cells

Sup.

Cells

Sup.

a

b

e f

c d

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Lentivirus-delivered cpg15 shRNA reduces CPG15 levelsTo test whether CPG15 also promotes survival of cortical progeni-tors, we used lentivirus delivery24,25 combined with RNA interference (RNAi)26 to knock down CPG15 expression in vitro and in vivo. We generated lentiviruses expressing CPG15-Flag, a cpg15 shRNA (small hairpin RNA), and a control scrambled cpg15 shRNA with four of the hairpin nucleotides inverted (Fig. 4a). shRNAs are processed to small interfering RNAs (siRNAs) that guide the specific cleavage and elimi-nation of their cognate mRNAs. When primary cortical cultures were coinfected with the cpg15-FLAG and cpg15 shRNA lentiviruses, both the exogenous cpg15-FLAG and the endogenous cpg15 mRNAs were

severely reduced as detected by northern blotting (Fig. 4b), but they were unaffected by coinfection with the scrambled cpg15 shRNA control virus. The cpg15 shRNA lentivirus was also effective in knocking down cellular levels of the CPG15 protein as detected by immunocytochem-istry (Fig. 4c–f) and by western blotting (Fig. 4g).

CPG15 knockdown increases cortical progenitor apoptosisTo address the role of CPG15 in progenitor cell survival, we first examined the effect of CPG15 knockdown on cultured cortical progenitors that were isolated from E14–E15 embryonic rat cortex and plated in the presence of basic fibroblast growth factor (bFGF). At plating, cells were infected with cpg15 shRNA, scrambled cpg15 shRNA or EGFP lentiviruses. After 4 d in culture the vast majority of cells in uninfected, EGFP-infected or scrambled hairpin–infected control cultures were positive for nestin, a marker of neural progeni-tors (Fig. 5a,b red staining). In cpg15 shRNA–infected cultures, the number of progenitors was greatly reduced, whereas the number of differentiated neurons (marked by neurofilament-M (Nf-M) staining, blue) was unchanged (Fig. 5c). Quantification of these results shows that the decrease in progenitor numbers in cpg15 shRNA–infected cultures is similar to that seen in cultures deprived of bFGF from plating (Fig. 5d,e). Progenitor loss in the cpg15 shRNA–infected cul-tures was accompanied by a marked increase in apoptotic cell death identified by Hoechst staining (Fig. 5f). These results demonstrate that depletion of endogenous CPG15 results in increased apoptosis of neuronal progenitors, suggesting that CPG15 is required for their

GFPCPG15

cpg15-sh

+ + – – – – – – + + + + – – – – + +

1510

kD

Cells

Sup.

Cells

Sup.

Cells

Sup.

Scrambled

cpg15-sh

CAAGA T G T A GGGCTTTTCAGACTGTTTG CAAACAGTCTGAAAAGCCC

CAAGA T G

T A GGGCTTGACTTACTGTTTG CAAACAGTAAGTCAAGCCC

CMV R U5 HIV-1 Ubiquitin EGFP WRE ∆U3 R U5enhancer flap promoter

Viru

s – – Scram.

OverlayFlag

CP

G15

CP

G15

cpg1

5 sh

RN

Acpg15shRNA

CPG15 – + + +

Exo.

Endo.

a b

c d

e f

g

5S RNA

Figure 4 A lentivirus-delivered cpg15 small hairpin RNA (shRNA) knocks down CPG15 expression. (a) Schematic representation of short hairpin sequences and lentivirus vector used for CPG15 knockdown. The shRNA or scrambled shRNA sequences connected to the U6 promoter were inserted upstream of the ubiquitin-C promoter driving EGFP expression (see Methods for details). Red boxes mark the inverted region in the scrambled shRNA. (b–g) In cultured cortical neurons, lentivirus-delivered cpg15 shRNA, but not the scrambled version, markedly reduces cpg15 mRNA assayed by northern blotting (b) and reduces protein expression as assayed by immunocytochemistry (c–f) or by western blotting (g). Neurons were infected with CPG15-FLAG-IRES-EGFP lentivirus, alone (b, lane 2; c,d; and g, lanes 3–4), together with a scrambled cpg15 shRNA control lentivirus (b, lane 3) or together with the cpg15 shRNA lentivirus (b, lane 4; e,f; and g, lanes 5–6). CPG15 immunostaining is red, and EGFP (green) marks infected cells. In g, Sup., supernatant; cpg15-sh, cpg15 shRNA. Scale bar, 25 µm.

Figure 5 Endogenous CPG15 is required for cortical progenitor survival in vitro. (a–d) Cortical progenitor cultures stained with the progenitor marker nestin (red), the neuronal marker neurofilament-M (Nf-M) (blue) and Hoechst nuclear staining (pseudocolored green). (e) Quantification of a–d and an additional EGFP lentivirus control. Comparison of total number of cells per field staining positive for nestin or Nf-M (*P < 0.001). Infection with the cpg15 shRNA lentivirus (cpg15-sh), but not the scrambled control, leads to a marked decrease in the number of neural progenitors. (f) Quantification of apoptotic cells in a–d and an additional EGFP lentivirus control. Knockdown of CPG15 is accompanied by a significant increase in number of apoptotic cells (*P < 0.001). Scale bar, 100 µm.Uninf. EGFP Scram. cpg15-sh No FGF

Apo

ptot

ic c

ells

(per

cent

age

of to

tal)

Uninf. EGFP Scram.cpg15-sh No FGF0

5

10

15

20

25

Uninfected Scrambled

cpg15-sh No FGF

Nf-M Nestin

0

20

40

60

80

100

Cel

ls p

er fi

eld

*

*

* *

a b

c d

e f

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

in vitro survival. As acute CPG15 loss has no immediate affect on the number of Nf-M positive neurons, our results further suggest that CPG15 primarily affects progenitor survival.

Our finding that CPG15 is required for survival of cortical progenitors in culture led us to predict that depletion of endogenous CPG15 at early developmental times would increase apoptosis and reduce survival within the cortical progenitor population in vivo. To test this hypothesis, we delivered cpg15 shRNA lentiviruses or control lentiviruses into E15 embryonic brains by direct ventricular injection. Embryos were harvested at E22 and their brains sectioned. Nissl staining showed moderate shrinkage in the size and ventricular volume of cpg15 shRNA lentivirus–infected brains when compared with uninfected (data not shown), EGFP lentivirus–infected or scrambled shRNA lentivirus–infected brains (Fig. 6a–e), suggest-ing a decrease in neuronal number. We stained alternate sections by TUNEL and quantified apoptotic neurons in the neocortex, where CPG15 is highly expressed during embryonic development, and in the diencephalon, where CPG15 expression is low in early develop-ment (Fig. 1). When compared to uninfected (Fig. 6f), EGFP lenti-virus–infected (Fig. 6g) and scrambled shRNA lentivirus–infected brains (Fig. 6h), the cpg15 shRNA brains (Fig. 6i) showed an increase in apoptotic cells in the neocortex. Overlay of TUNEL staining and EGFP staining on the same sections (Fig. 6j–m) demonstrated com-parable infection levels by the different lentiviruses. Quantification of TUNEL-stained cells showed that cpg15 shRNA increases apopto-sis in the neocortex (Fig. 6n) but not in the diencephalon (Fig. 6o). These results demonstrate that decreasing endogenous CPG15 levels during embryonic development results in increased apoptosis and

diminished survival of cortical neurons. Reduced neuronal number is likely to cause shrinkage of the cortical plate and its contraction around the lateral ventricles, as seen in the deformed cpg15 shRNA lentivirus–infected brains (Fig. 6a–e). The specific effect of CPG15 depletion on neocortical neurons suggests that it is not essential for survival in all progenitor populations. Alternatively, the RNAi intervention may be past the critical time when CPG15 is necessary for survival of the diencephalic progenitors. In any case, the lack of effect of the cpg15 shRNA lentivirus on apoptosis in the developing diencephalon demonstrates that the lentivirus-delivered RNAi has no deleterious effect in brain regions surrounding the ventricles.

CPG15 overexpression results in an enlarged cortical plateIn a complementary study, we further examined the role of CPG15 in vivo by overexpressing CPG15-Flag in the developing brain. CPG15-Flag lentivirus was injected into the ventricles of E15 embryonic brains harvested as described above. When compared with control EGFP lentivirus–infected brains, brains overexpressing CPG15 were significantly larger in diameter, with enlarged ventricles but with no alteration in cortical thickness (Fig. 7a–e). Closer examination of the enlarged brains showed sulcus-like indentations (Fig. 7f–k; boxes in a–c are shown at higher magnification in f,h,i, and two additional examples from different CPG15-Flag lentivirus–infected brains are shown in j,k), consistent with a larger surface area resulting from addition of super-numerary neurons in radial units4. Sections from control EGFP lenti-virus–infected brains (Fig. 7l) and from brains overexpressing CPG15 (Fig. 7m–p) were double-labeled with nestin (blue) and Nf-M (red), showing heterotopic cell masses within the proliferative ventricular zone.

Uninfected EGFP Scrambled cpg15-sh



Neocortex

Diencephalon

TU

NE

L-po

sitiv

e ce

llsT

UN

EL-

posi

tive

cells

EGFP Scrambled cpg15-sh cpg15-sh EGFP

Scram

.

shRNA

Ven

tric

le a

rea

(mm

2 )

1.2

0.8

0.4

0

a

f g h i

j k l m

b c d e

n

o

Figure 6 In vivo knockdown of endogenous CPG15 causes shrinkage of the cortical plate and increased apoptosis of cortical neurons. (a–e) In vivo knockdown of endogenous CPG15 results in cortical plate shrinkage. Coronal hemi-sections from (a) brains infected with EGFP lentivirus, (b) scrambled cpg15-shRNA or (c,d) cpg15-shRNA lentivirus. Low to high infection levels shown from left to right. (e) Scatter plot summarizing ventricular area of infected brains. Each symbol represents one hemisphere. (f–i) TUNEL staining of control and cpg15-shRNA lentivirus–infected brains. (j–m) Overlay of TUNEL with EGFP staining shows similar infection levels with control and cpg15-shRNA viruses. (n,o) Quantification of TUNEL-positive cells in the neocortex (n) or diencephalon (o) of control and infected brains (*P < 0.003). Scale bars: a, 1 mm; j, 50 µm.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Thus, CPG15 overexpression results in the expansion and involution of the cortical plate, and in heterotopias and discontinuities of the ven-tricular zone. Malformations of this type are typically seen in mutant mice with an increase in forebrain progenitor numbers11,12,27 and are consistent with an expansion of the progenitor pool.

CPG15 expands the progenitor pool by reducing apoptosisTo directly measure an affect of CPG15 on progenitor number, we used BrdU to label proliferating neurons. Pregnant dams were injected with BrdU 3 d after intraventricular injection of the CPG15-Flag lentivirus into embryonic brains. Embryos were harvested 2 or 24 h after BrdU injection (Fig. 8a,b), earlier than for previous experiments to avoid confounding secondary effects of the deformed cortical plate. Low-magnification views showed that 2 h after BrdU injection, the CPG15-Flag lentivirus–infected brains are indistinguishable from uninfected control brains, with normal cellular lamination and cellular distri-bution within the ventricular and subventricular zones (Fig. 8c,d). However, a higher-magnification view showed a small but significant increase in BrdU-labeled cells within the ventricular zone of CPG15-Flag lentivirus–infected brains, suggesting an increase in progenitor number (Fig. 8e–g).

Expansion of the progenitor pool can result from increased mitotic rates, a decrease in cell cycle re-entry, decreased cell death or any combination of these factors. To test for changes in mitotic rates of dividing progenitors, embryos injected with the CPG15-Flag lentivirus and harvested 24 h after BrdU injection at E19 (Fig. 8b) were double stained for BrdU, to mark cells in S phase of the cell cycle, and for

phospho-histone-H3 (p-H3), to mark cells in M phase undergoing mitosis (Fig. 8h,i). We found no significant difference between un-infected control and CPG15-Flag lentivirus–infected brains in the ratio of dividing progenitors at M phase to progenitors in S phase (Fig. 8j), indicating that CPG15 does not increase mitotic rate. We next tested whether CPG15 affects the number of progenitors that re-enter the cell cycle rather than progressing towards terminal differentiation. E15 embryos were injected with the CPG15-Flag lentivirus, followed by BrdU injection at E18 (Fig. 8b). Twenty-four hours later, embryos were harvested and double-stained for BrdU, to mark cells that were dividing at the time of injection, and Ki67, to mark progenitors. We identified cells that had exited the cell cycle within the 24 h of BrdU labeling as BrdU+ and Ki67– (red) and divided progenitors in the cell cycle as BrdU+ and Ki67+ (yellow) (Fig. 8k,l). We found that in the CPG15-Flag lentivirus–infected embryos, there was no significant change in the proportion of cortical progenitors exiting the cell cycle as compared to uninfected controls (Fig. 8m).

Finally, to examine whether CPG15 overexpression decreases progenitor apoptosis, we performed TUNEL staining on E18 brains from uninfected control (Fig. 8n) and CPG15-Flag lentivirus–infected (Fig. 8o) embryos 3 d after viral delivery. Counts of TUNEL-stained cells in the cortex of CPG15-Flag lentivirus–infected brains showed a significant decrease in the numbers of apoptotic cells when compared to uninfected control brains (Fig. 8n–p). Together, these studies indicate that the observed increase in progenitor pool size and the expanded cortices seen in CPG15-Flag lentivirus–infected brains are not due to increased progenitor mitotic rates or decreased cell

*

Bra

in d

iam

eter

(m

m)

Ven

tric

le a

rea

(mm

)

a b c d e

f

l m n o p

g h i j k

Figure 7 In vivo CPG15 overexpression in the embryonic brain results in an expanded cortical plate and heterotopic cell masses in the ventricular zone. (a–e) CPG15 lentivirus–infected brains are larger in diameter and ventricular area. (a–c) Nissl stained coronal sections from EGFP and CPG15 lentivirus–infected brains. (d) Measurements of brain diameter comparing control and CPG15-overexpressing brains (* P < 0.02). (e) Scatter plot summarizing ventricular area of the same brains. Each symbol represents one hemisphere. (f–k) Nissl-stained coronal sections of brains infected with the indicated viruses. Boxes in (a–c) are shown at higher magnification in f, h and i, respectively. (l–p) Double labeling for nestin in blue and neurofilament in red on EGFP lentivirus–injected (l) and CPG15 lentivirus–injected brains (m–p). Scale bars: a, 1 mm; f, 50 µm; l, 100 µm.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

cycle exit, but rather are likely to be a result of decreased apoptosis within the progenitor population.

DISCUSSIONcpg15 expression patterns in the embryonic brain concomitant with periods of rapid cell proliferation and apoptosis, and during cir-cuit formation and target selection, led us to test whether early in development CPG15 might function as a survival factor, similarly to neurotrophins. We found that CPG15 is able to rescue cortical neurons from starvation-induced apoptosis by preventing caspase 3 activation and is also crucial for survival of cortical progenitors. Consequently, manipulating in vivo CPG15 levels in utero had a profound effect on the size and shape of the neocortical plate. Decreasing CPG15 levels caused increased apoptosis of cortical pro-genitors and shrinkage of the cortical plate, whereas overexpression resulted in decreased progenitor apoptosis and an expanded and convoluted cortical plate.

It is particularly informative to compare the CPG15 overexpression phenotype to developmental mutants with similar presentations. In the case of mutants in the cell-death cascade, expansion of the cortical plate occurs only when the apoptotic pathway is affected in proliferating

neurons rather than in postmitotic neurons and is a direct result of an enlarged progenitor pool4,5. Gene-targeting studies have identified Bax, Bcl-XL (Bcl2l1), Apaf1, Casp9 and Casp3 as key elements in neuronal apoptosis occurring during brain development11–15,28,29. By anal-ogy to their counterparts in Caenorhabditis

elegans, the proteins are likely to form a linear cell-death cascade, with Apaf1 bound to caspase 9, activating caspase 3, and Bcl-XL acting as an upstream anti-apoptotic regulator of Apaf1 that can be blocked by Bax12,29. Epistatic analysis shows that the upstream components of this pathway, Bax and Bcl-XL, are obligatory only in postmitotic neu-rons, whereas caspase 3 is unique in its effect on apoptosis of neuronal founder cells29,30. Global formation of the nervous system is unaffected in Bax-deficient embryos31,32, whereas null mutants of Casp3, Casp9 and Apaf1 show severe forebrain malformations that result from hyper-plasia and an enlarged neocortical plate11–15. This is consistent with the idea that even modest alterations in the size of the progenitor pool during its exponential growth phase can drastically effect final cortical size and shape1,3. Hyperplasia and an enlarged neocortical plate can also be seen when the cortical progenitor pool is expanded for reasons other than a decrease in cell death: for example, in β-catenin transgenic mice, where cortical progenitors fail to exit the cell cycle after mito-sis33. The similar phenotype seen with in vivo CPG15 overexpression suggests that elevated CPG15 expands the progenitor pool, likely by reducing apoptosis. This is supported by results showing that increasing amounts of CPG15 during corticogenesis reduce apoptosis and enlarge the progenitor pool but have no effect on mitotic rates or cell cycle exit.

Figure 8 CPG15 overexpression reduces apoptosis in the progenitor pool but does not affect mitotic index or cell cycle exit. (a) Schematic of BrdU labeling experiments shown in c–j. Embryos were injected with virus at E15 and then harvested at E18 2 h after BrdU injection. (b) Schematic of cell cycle exit experiments shown in k–m. Embryos were injected with virus at E15 then harvested at E19 24 h after BrdU injection. (c,d) Low-magnification view of BrdU labeling (green) and propidium iodide nuclear staining (red) in uninfected (c) and CPG15 lentivirus–infected (d) brains. (e,f) Areas outlined in white in c,d, respectively shown at higher magnification. (g) Quantification of BrdU staining shows a significant increase in the percentage of BrdU-labeled cells in CPG15 lentivirus–infected brains (*P = 0.008) as compared to uninfected controls. (h,i) Double labeling for BrdU (red) and the mitosis marker p-H3 (green). (j) Quantification of mitotic rates (ratio of S-phase BrdU-labeled cells to M-phase p-H3–labeled cells) showed no significant difference between CPG15 lentivirus–infected and uninfected brains. (k,l) Double labeling for BrdU (red) and the progenitor cell marker Ki-67 (green). (m) Quantification showed no significant difference in the rate of cell cycle exit between CPG15 lentivirus–infected and control brains. (n,o) TUNEL staining in the ventricular zone of uninfected (n) and CPG15 lentivirus–infected (o) brains. (p) Quantification of TUNEL-positive cells shows a significant decrease in ventricular zone apoptosis in CPG15 lentivirus–infected brains as compared to uninfected controls (*P < 0.001). Scale bars: c, 100 µm; e,h,k,n, 50 µm.

Uninfected CPG1505

10152025303540

Brd

U-p

ositi

ve c

ells

(%

)

Harvest2 hBrdUCPG15

E15

3 d

Harvest24 hBrdUCPG15

E15

3 d

Uninfected CPG15

Uninfected CPG150

10203040506070

Cel

l cyc

le e

xit (

%)

VZ

IZ

VZ

IZ

Uninfected CPG15

Uninfected CPG15

E18

E18 E19

BrdU + PI stainBrdU + p-H3 stain

BrdU + Ki-67 stain

*

Uninfected CPG150.00

0.04

0.08

0.12

0.16

0.20

Mito

tic r

ate

(M/S

)T

UN

EL-

posi

tive

cells

Uninfected CPG15

10

20

0

30

Uninfected CPG15

5

15

25

35

*

Uninfected CPG15

Brd

U +

PI

Brd

U +

p-H

3B

rdU

+ K

i-67

TU

NE

La

b

e f

h i

k l

n o

g

j

m

p

c d

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Furthermore, an increased number of apoptotic neurons can be seen in brains with RNAi-mediated knockdown of endogenous CPG15, and in vitro data demonstrates CPG15 function as an anti-apoptotic factor for cortical neurons as well as cortical progenitors.

Other extracellular signaling molecules previously shown to regulate cerebral cortical size through their effect on the progenitor pool are basic fibroblast growth factor34–36, pituitary adenylate cyclase-activating polypeptide37 and, more recently, lysophosphatidic acid38. Basic fibro-blast growth factor and pituitary adenylate cyclase–activating polypep-tide, respectively, expand or shrink the progenitor pool by acting as mitogenic or anti-mitogenic signals. Lysophosphatidic acid expands the progenitor pool by increasing terminal mitosis and reducing cas-pase-mediated cell death, and is thus the only factor other than CPG15 that has been shown to decrease apoptosis of cortical progenitors38. Studies of cultured cortical progenitors suggest that BDNF and NT-3 can mediate progenitor cell survival in vitro39. Yet direct intrauterine, intraventricular application of neurotrophins into the embryonic brain does not affect proliferation or apoptosis of cortical progeni-tors40. Furthermore, classic neurotrophins are absent in most parts of the embryonic brain41, and prenatal CNS defects related to neu-ronal number have not been seen in neurotrophin single- or double-knockout mice or in animals lacking their receptors42. Thus, CPG15 is one of few molecules shown to be essential for in vivo survival of undifferentiated cortical progenitors. CPG15 expression is specific to a subset of progenitor populations in the developing brain and seems to be required only in these populations. We propose that by counter-ing early apoptosis in specific progenitor subpopulations, CPG15 has a role in regulating brain size and shape during morphogenesis of the mammalian forebrain.

As CPG15 is also expressed in some differentiated neurons during target selection and circuit formation and is able to rescue cortical neu-rons from starvation-induced apoptosis, it is possible that CPG15 may also function as a target-derived survival factor for differentiated neu-rons analogous to neurotrophins. It remains to be examined whether CPG15 is required for survival in various populations of differenti-ated neurons and whether this function of CPG15, as well as its later activity-dependent function as a growth and differentiation factor, are mediated by the soluble form described here. Another possible future direction will be analysis of a cpg15 knockout mouse. As it is techni-cally difficult to intervene by intrauterine delivery of an shRNA earlier than E14–E15, the CPG15 knockdown using RNAi can only affect the very last rounds of cortical progenitor division. In the knockout, cpg15 deletion at an earlier step may result in a more extreme phenotype. However, deleting a gene from the onset of embryogenesis could also lead to compensation and lack of a discernible phenotype, something that can be avoided with an acute knockdown using RNAi at a criti-cal developmental step43. For example, mice with targeted deletion of the doublecortin gene (Dcx) seem to develop a normal neocor-tex44, whereas electroporation of plasmids encoding shRNA against the doublecortin protein in utero disrupts radial migration in the rat neocortex, resulting in a malformed cerebral cortex45. In this case, acute intervention using RNAi results in a phenotype more similar to the double cortex syndrome seen in humans with mutations in the Dcx gene46 than is seen with the complete Dcx knockout. Thus, RNAi approaches may be complementary to gene knockouts in functional studies of genes important for brain development43.

METHODSAll animal work was approved by the Massachusetts Institute of Technology Committee on Animal Care and conforms to US National Institutes of Health guidelines for the use and care of vertebrate animals.

In situ hybridization. In situ hybridizations were performed as previously described22.

CPG15 immunocytochemistry and protein purification. HEK293T cells were grown to 80–90% confluence in 100-mm culture dishes containing 15 ml medium (10% calf serum, 50 U penicillin, 50 µg streptomycin, 4 mM L-glutamate in Dulbecco’s Modified Eagles medium (BioWhittaker)) and then transfected with 8 µg of the pIRES-EGFP-CPG15-FLAG plasmid (Fig. 2a) using Lipofectamine 2000 (Invitrogen). For immunocytochemistry, cells were fixed in 4% paraformaldehyde and stained with anti-Flag (Sigma). For protein purifica-tion, the medium was harvested 4 d after transfection and debris removed by centrifugation (3,000 rpm, 15 min, 4 °C). The medium was then incubated with 40 µl anti-Flag coupled to agarose (EZview Red ANTI-FLAG M2 Affinity Gel, Sigma) for 12 h at 4 °C on a rotator. The agarose was pelleted by centrifugation for 10 min at 2,000 rpm and washed three times with TBS. The tagged CPG15 protein was eluted from the anti-Flag by incubation for 4 h at 4 °C with 30 µg of 3× Flag peptide (Sigma) diluted in 200 µl TBS; it was recovered by centrifu-gation for 5 min at 2,000 rpm. Protein concentration in the supernatant was determined by the Bradford assay.

Preparation of membrane and soluble fractions from brain, and western blot-ting. Membrane and soluble fractions from brains were prepared as detailed in Supplementary Methods online. Western blots were incubated with a polyclonal anti-CPG1519, followed by staining for the transferrin receptor (1:1,000, Zymed) detected by a goat anti-mouse horseradish peroxidase (HRP)-conjugated sec-ondary antibody (1:5,000, Sigma), and by staining for protein Akt (1:1,000, Cell Signaling Technology) detected by a goat anti-rabbit–HRP secondary antibody (1:2,500, Jackson ImmunoResearch).

Growth factor deprivation and apoptosis assays. Primary cortical cultures were done essentially as previously described47. After 6 d in vitro, cortical neu-rons were washed three times with Neurobasal medium without supplements and then incubated for 12 h in the unsupplemented medium with or without 50 ng ml–1 purified CPG15 protein. After an additional 12 h in feeding medium, cells were fixed in 4% formaldehyde/PBS for 30 min at 4 °C. Hoechst staining and immunocytochemistry were done as described in Supplementary Methods. For quantification, fragmented apoptotic nuclei as well as healthy nuclei were counted blind to experimental treatment using a fluorescence microscope with a UV filter setting for the Hoechst staining (excitation 330–380; emission 420) and rhodamine settings for visualizing the antibody to cleaved caspase 3 (excitation 528–553; emission 600–660). Treatments were repeated in three independent experiments with two coverslips per treatment in each experiment. Each data point represents the mean of 500–600 cells, counted in 40–50 different fields per coverslip. The percentage of apoptotic cells was calculated based on the num-ber of fragmented nuclei divided by the total number of nuclei. Comparisons between groups were analyzed using a Student’s t-test.

Immunocytochemistry of neural progenitor cultures. Cortical progenitor cul-tures were prepared as previously described in detail48. Immunocytochemistry for nestin (1:2,000; Chemicon) and neurofilament M (1:3,000; Chemicon) is described in Supplementary Methods. Quantifications were from three inde-pendent experiments (five random images per treatment). Groups were com-pared using ANOVA post-hoc analysis with the Bonferroni/Dunn method.

Lentivirus generation and infections. Five different cDNA sequences span-ning the cpg15 core domain were synthesized, fused to a loop region and then annealed to their antisense sequences and cloned separately into pSilencer1.0-U6 plasmid (Ambion) downstream of the U6 promoter. To test the effectiveness of the cpg15 shRNAs in reducing CPG15 levels, HEK293T cells were separately cotransfected with each one of the pSilencer-cpg15-shRNA plasmids together with the pIRES-EGFP-CPG15-FLAG plasmid at a 40:1 ratio using Lipofectamine 2000 (Invitrogen). CPG15-IRES-EGFP mRNA knockdown was determined by reduced expression of EGFP. The most effective small hairpin sequence (5′-GGGCTTTTCAGACTGTTTG-3′) was then amplified with its upstream U6 promoter by PCR and subcloned into the pFUGW lentivirus transfer vector24. To generate the scrambled shRNA construct, the small hairpin sequence was modified to 5′-GGGCTTGACTTACTGTTTG-3′ (the inverted sequence is underlined) and cloned into the pFUGW lentivirus transfer vector as described

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

above. The CPG15-FLAG-IRES-EGFP cDNA was subcloned into pFUGW down-stream of the ubiquitin promoter. pFUGW lentivirus was used as the EGFP control. Lentivirus production, concentration and titer determination were done as described24. Typical titers for in vitro experiments were 1 × 106 and for in vivo injections, 5 × 106 to 20 × 106.

For viral infection of primary cortical cultures, culture medium was reduced to 0.5 ml at 3 d in vitro, and viruses were added for >8 h incubation. The medium was then removed and replaced with 0.5 ml saved plating medium and 0.5 ml new feeding medium. Cells were fixed and mounted 4 d later. For infection of progenitor cultures, the different lentiviruses were added to the medium at plating.

In vivo lentivirus injections were performed on timed-pregnant Sprague-Dawley rats (Taconic). Pregnancies were timed with day of plug detection as E1 and birth usually occurred on E23. Neurons and progenitors were infected by injecting 2–2.5 µl of lentivirus into the lateral ventricle of E15 brains. Surgery and injection procedures were as previously described49.

Northern blot hybridization. Northern blot hybridization was performed as previously described50.

Histological and immunohistochemical analysis of embryonic brains. Animals were euthanized at E18, E19 or E22: 3, 4 or 7 d, respectively, after viral injection. Brains were removed and submerged in 4% paraformaldehyde in PBS overnight at 4 °C, transferred to 30% sucrose in PBS at 4 °C until they sank, and then frozen and sectioned at 20 µm using a cryostat. Every sixth section was Nissl stained and used to match sections from different brains. Sections at equivalent levels on the anterior-posterior axis were then processed for TUNEL. Cryosections were blocked in 10% goat serum, 0.3% Triton X-100 for 1 h. Primary antibodies against nestin (1:2,000, Chemicon, mouse mono-clonal) and neurofilament (1:1,000, Chemicon, rabbit polyclonal) were diluted in blocking solution. Sections were counterstained with Hoechst 33342 in PBS for 10 min to highlight nuclear DNA. Fluorescent secondary antibodies Alexa-555-conjugated goat anti-mouse (1:200, Molecular Probes) and Alexa-647-conjugated anti-rabbit (1:200, Molecular Probes) were used for visualization as described above.

TUNEL staining on frozen brain sections was performed as described by the manufacturer using the Roche In Situ Cell Death Detection Kit, TMR red. The total number of TUNEL-positive cells present in the neocortex and the diencephalon of uninfected (three each at E18 and E22), EGFP-infected (three at E22) and cpg15 shRNA–infected (four at E22) brains was quantified. Six to twelve sections were analyzed from each brain. Statistical significance was deter-mined by Student’s t-test.

To label proliferating cells in the cortex of E18 embryos, BrdU was injected intraperitoneally into the mother at 50 mg kg–1 body weight. BrdU and Ki-67 or p-H3 double labeling was done as previously described for BrdU and Ki6733. Primary antibodies used were BrdU (1:200; Harlan, rat monoclonal), Ki-67 (1:500; Novocastra, rabbit polyclonal), and p-H3 (1:1,000; Upstate, rabbit polyclonal). BrdU-labeled and p-H3–labeled cells were counted from matching sections and percentages were compared by unpaired Student’s t-tests. For cell cycle exit studies, a similar approach was used: Ki-67 expression was scored in 50 BrdU-positive cells from each of five randomly chosen fields of view from at least three sections per brain (two brains per each experimental group). Statistical significance was determined by Student’s t-test.


ACKNOWLEDGMENTSWe thank members of the Nedivi laboratory and P. Garrity, J. Hoch, and J. Mintern for helpful comments on the manuscript, J. Cottrell for initiating and help with shRNA cloning and testing, C. Lois for advice on construction and use of lentivirus vectors, C. Walsh and E. Olson for guidance on embryonic injections, and J. Pungor for help with cell counts. This work was sponsored by grants from National Eye Institute and the Ellison Medical Foundation to E. Nedivi. U. Putz was supported by a postdoctoral fellowship from the Deutsche Forschungsgemeinschaft, and C. Harwell by a Ford Foundation predoctoral fellowship.



1. Rakic, P. A small step for the cell, a giant leap for mankind: a hypothesis of neocortical expansion during evolution. Trends Neurosci. 18, 383–388 (1995).

2. Takahashi, T., Nowakowski, R.S. & Caviness, V.S.J. The mathematics of neocortical neuronogenesis. Dev. Neurosci. 19, 17–22 (1997).

3. Caviness, V.S.J., Takahashi, T. & Nowakowski, R.S. Numbers, time and neocortical neuronogenesis: a general developmental and evolutionary model. Trends Neurosci. 18, 379–383 (1995).

4. Haydar, T.F., Kuan, C-Y., Flavell, R.A. & Rakic, P. The role of cell death in regulating the size and shape of the mammalian forebrain. Cereb. Cortex 9, 621–626 (1999).

5. Kuan, C.-H., Roth, K.A., Flavell, R.A. & Rakic, P. Mechanisms of programmed cell death in the developing brain. Trends Neurosci. 23, 291–297 (2000).

6. de la Rosa, E.J. & de Pablo, F. Cell death in early neural development: beyond the neurotrophic theory. Trends Neurosci. 23, 454–458 (2000).

7. Pompeiano, M., Blaschke, A.J., Flavell, R.A., Srinivasan, A. & Chun, J. Decreased apoptosis in proliferative and postmitotic regions of the caspase 3-deficient embryonic central nervous system. J. Comp. Neurol. 423, 1–12 (2000).

8. Blaschke, A.J., Staley, K. & Chun, J. Widespread programmed cell death in proliferative and postmitotic regions of the fetal cerebral cortex. Development 122, 1165–1174 (1996).

9. Thomaidou, D., Mione, M.C., Cavanagh, J.F.R. & Parnavelas, J.G. Apoptosis and its relation to the cell cycle in the developing cerebral cortex. J. Neurosci. 17, 1075–1085 (1997).

10. Blaschke, A.J., Weiner, J.A. & Chun, J. Programmed cell death is a universal feature of embryonic and postnatal neuroproliferative regions throughout the central nervous system. J. Comp. Neurol. 396, 39–50 (1998).

11. Kuida, K. et al. Decreased apoptosis in the brain and premature lethality in CPP32-deficient mice. Nature 384, 368–372 (1996).

12. Kuida, K. et al. Reduced apoptosis and cytochrome c-mediated caspase activation in mice lacking caspase 9. Cell 94, 325–337 (1998).

13. Hakem, R. et al. Differential requirement for caspase 9 in apoptotic pathways in vivo. Cell 94, 339–352 (1998).

14. Cecconi, F., Alvarez-Bolado, G., Meyer, B.I., Roth, K.A. & Gruss, P. Apaf1 (CED-4 Homolog) regulates programmed cell death in mammalian development. Cell 94, 727–737 (1998).

15. Yoshida, H. et al. Apaf1 is required for mitochondrial pathways of apoptosis and brain development. Cell 94, 739–750 (1998).

16. Nedivi, E., Hevroni, D., Naot, D., Israeli, D. & Citri, Y. Numerous candidate plasticity-related genes revealed by differential cDNA cloning. Nature 363, 718–722 (1993).

17. Hevroni, D. et al. Hippocampal plasticity involves extensive gene induction and multiple cellular mechanisms. J. Mol. Neurosci. 10, 75–98 (1998).

18. Naeve, G.S. et al. Neuritin: a gene induced by neural activity and neurotrophins that promotes neuritogenesis. Proc. Natl Acad. Sci. USA 94, 2648–2653 (1997).

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

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

21. Corriveau, R., Shatz, C.J. & Nedivi, E. Dynamic regulation of cpg15 during activity-dependent synaptic development in the mammalian visual system. J. Neurosci. 19, 7999–8008 (1999).

22. Lee, W.C.A. & Nedivi, E. Extended plasticity of visual cortex in dark-reared animals may result from prolonged expression of genes like cpg15. J. Neurosci. 22, 1807–1815 (2002).

23. Hooper, N.M. Determination of glycosyl-phosphatidylinositol membrane protein anchor-age. Proteomics 1, 748–755 (2001).

24. Lois, C., Hong, E.J., Pease, S., Brown, E.J. & Baltimore, D. Germline transmission and tissue-specific expression of transgenes delivered by lentiviral vectors. Science 295, 868–872 (2002).

25. Rubinson, D.A. et al. A lentivirus-based system to functionally silence genes in primary mammalian cells, stem cells and transgenic mice by RNA interference. Nat. Genet. 33, 401–406 (2003).

26. McManus, M.T. & Sharp, P.A. Gene silencing in mammals by small interfering RNAs. Nat. Rev. Genet. 3, 737–747 (2002).

27. Chenn, A. & Walsh, C.A. Increased neuronal production, enlarged forebrains and cyto-architectural distortions in β-catenin overexpressing transgenic mice. Cereb. Cortex 13, 599–606 (2003).

28. Motoyama, N. et al. Massive cell death of immature hematopoietic cells and neurons in Bcl-x-deficient mice. Science 267, 1506–1510 (1995).

29. Roth, K.A. et al. Epistatic and independent functions of Caspase-3 and Bcl-XL in developmental programmed cell death. Proc. Natl Acad. Sci. USA 97, 466–471 (2000).

30. Shindler, K.S., Latham, C.B. & Roth, K.A. bax deficiency prevents the increased cell death of immature neurons in bcl-x-deficient mice. J. Neurosci. 17, 3112–3119 (1997).

31. Knudson, C.M., Tung, K.S.K., Tourtellotte, W.G., Brown, G.A.J. & Korsmeyer, S.J. Bax-deficient mice with lymphoid hyperplasia and male germ cell death. Science 270, 96–99 (1995).

32. White, F.A., Keller-Peck, C.R., Knudson, C.M., Korsmeyer, S.J. & Snider, W.D. Widespread elimination of naturally occurring neuronal death in Bax-deficient mice. J. Neurosci. 18, 1428–1439 (1998).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

33. Chenn, A. & Walsh, C.A. Regulation of cerebral cortical size by control of cell cycle exit in neural precursors. Science 297, 365–369 (2002).

34. Ortega, S., Ittmann, M., Tsang, S.H., Ehrlich, M. & Basilico, C. Neuronal defects and delayed wound healing in mice lacking fibroblast growth factor 2. Proc. Natl Acad. Sci. USA 95, 5672–5677 (1998).

35. Dono, R., Texido, G., Dussel, R., Ehmke, H. & Zeller, R. Impaired cerebral cortex development and blood pressure regulation in FGF-2-deficient mice. EMBO J. 17, 4213–4225 (1998).

36. Vaccarino, F.M. et al. Changes in cerebral cortex size are governed by fibroblast growth factor during embryogenesis. Nat. Neurosci. 2, 246–253 (1999).

37. Suh, J., Lu, N., Nicot, A., Tatsuno, I. & DiCicco-Bloom, E. PACAP is an anti-mitogenic signal in developing cerebral cortex. Nat. Neurosci. 4, 123–124 (2001).

38. Kingsbury, M.A., Rehen, S.K. & Contos, J.J.A. Higgins, C.M.a. & Chun, J. Non-pro-liferative effects of lysophosphatidic acid enhance cortical growth and folding. Nat. Neurosci. 6, 1292–1299 (2003).

39. Barnabé-Heider, F. & Miller, F.D. Endogenously produced neurotrophins regulate survival and differentiation of cortical progenitors via distinct signaling pathways. J. Neurosci. 23, 5149–5160 (2003).

40. Brunstrom, J.E., Gray-Swain, M.R., Osborne, P.A. & Pearlman, A.L. Neuronal het-erotopias in the developing cerebral cortex produced by neurotrophin-4. Neuron 18, 505–517 (1997).

41. Ernfors, P., Merlio, J.-P. & Persson, H. Cells expressing mRNA for neurotrophins and their receptors during embryonic rat development. Eur. J. Neurosci. 4, 1140–1158 (1992).

42. Conover, J.C. & Yancopoulos, G.D. Neurotrophin regulation of the developing nervous system: analyses of knockout mice. Rev. Neurosci. 8, 13–27 (1997).

43. Götz, M. Doublecortin finds its place. Nat. Neurosci. 6, 1245–1247 (2003).44. Corbo, J.C. et al. Doublecortin is required in mice for lamination of the hippocampus

but not the neocortex. J. Neurosci. 22, 7548–7557 (2002).45. Bai, J. et al. RNAi reveals doublecortin is required for radial migration in rat neocortex.

Nat. Neurosci. 6, 1277–1282 (2003).46. Gleeson, J.G. et al. Doublecortin, a brain-specific gene mutated in human X-linked

lissencephaly and double cortex syndrome, encodes a putative signaling protein. Cell 92, 63–72 (1998).

47. Zhou, J. & Tang, X.C. Huperzine A attenuates apoptosis and mitochondria-dependent caspase-3 in rat cortical neurons. FEBS Lett. 526, 21–25 (2002).

48. Ghosh, A. & Greenberg, M.E. Distinct roles for bFGF and NT-3 in the regulation of cortical neurogenesis. Neuron 15, 89–103 (1995).

49. Walsh, C. & Cepko, C.L. Widespread dispersion of neuronal clones across functional regions of the cerebral cortex. Science 255, 434–440 (1992).

50. Sambrook, J., Fritsch, E.F. & Maniatis, T. Molecular Cloning: A Laboratory Manual. (Cold Spring Harbor Laboratory Press, Cold Spring Harbor, New York, 1989).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

332 ADVANCE ONLINE PUBLICATION NATURE NEUROSCIENCE

A R T I C L E S

Inhibitory synapses in the developing auditory system are glutamatergicDeda C Gillespie1, Gunsoo Kim1,2 & Karl Kandler1,2

Activity-dependent synapse refinement is crucial for the formation of precise excitatory and inhibitory neuronal circuits. Whereas the mechanisms that guide refinement of excitatory circuits are becoming increasingly clear, the mechanisms guiding inhibitory circuits have remained obscure. In the lateral superior olive (LSO), a nucleus in the mammalian sound localization system that receives inhibitory input from the medial nucleus of the trapezoid body (MNTB), specific elimination and strengthening of synapses that are both GABAergic and glycinergic (GABA/glycinergic synapses) is essential for the formation of a precise tonotopic map. We provide evidence that immature GABA/glycinergic synapses in the rat LSO also release the excitatory neurotransmitter glutamate, which activates postsynaptic NMDA receptors (NMDARs). Immunohistochemical studies demonstrate synaptic colocalization of the vesicular glutamate transporter 3 with the vesicular GABA transporter, indicating that GABA, glycine and glutamate are released from single MNTB terminals. Glutamatergic transmission at MNTB-LSO synapses is most prominent during the period of synapse elimination. Synapse-specific activation of NMDARs by glutamate release at GABAergic and glycinergic synapses could be important in activity-dependent refinement of inhibitory circuits.

In the developing brain, activity-dependent refinement has a central role in establishing precise neuronal circuits1. Although the refinement of excitatory circuits has been extensively studied, the rules and mechanisms by which inhibitory circuits are refined remain poorly understood. Numerous studies have shown developmental and activity-dependent modifications of inhibitory circuitry that include changes in receptor and synaptic properties and in overall expression of GABAergic markers2–7. However, the complexity of the majority of inhibitory networks in the vertebrate brain has hindered progress in understanding how these molecular and synaptic changes translate into plasticity at the level of functionally-defined inhibitory circuits.

Reorganization of specific inhibitory circuits has been well documented in the developing auditory system8. The LSO is a binaural auditory brainstem nucleus involved in sound localization. In order to compute interaural intensity differences, the LSO integrates excitatory input from the cochlear nucleus with inhibitory input from the MNTB, whose neurons are both GABAergic and glycinergic at the ini-tial stages in development but become glycinergic in early postnatal life9–12. Activity-dependent refinement of MNTB-LSO connections is necessary for the formation of a precise inhibitory tonotopic map. In the MNTB-LSO pathway, tonotopic precision is achieved through an early phase of functional refinement that occurs before the onset of hearing13 and a later phase of structural reorganization that occurs after hearing onset14. The pre-hearing phase of functional refinement is characterized by elimination of most of the initial GABA/glycin-ergic inputs and by strengthening of the remaining inputs13. These processes take place at an age when MNTB-LSO synapses are primarily

GABAergic rather than glycinergic11,12 and when GABA and glycine are depolarizing rather than hyperpolarizing15. Although functional refinement occurs before hearing onset and thus without sound-evoked neuronal activity, it clearly depends on cochlea-generated spontaneous activity, as tonotopic precision is impaired by neonatal cochlea ablation or by pharmacological blockade of glycine receptors14.

Here we show that activation of the GABA/glycinergic MNTB-LSO pathway in slices from neonatal rats elicits a glutamate response in postsynaptic LSO neurons. This current is not due to glycine spillover16 but instead reflects glutamate release from MNTB terminals. Our mini-mal stimulation experiments indicated that glutamate, GABA and gly-cine are released by single MNTB axons, and immunohistochemical evidence suggested that all three neurotransmitters are released from single synaptic terminals. Glutamate transmission was age dependent and was most prominent during the early period of functional refine-ment. The transient glutamatergic phenotype of the immature GABA/glycinergic MNTB-LSO pathway may provide synapse-specific activa-tion of MNTB-type glutamate receptors and thus could represent a pre-viously unknown mechanism for the developmental reorganization of this inhibitory circuit.

RESULTSMNTB stimulation elicits glutamatergic responses in LSOWhole-cell voltage-clamp recordings were made from LSO neurons in acute brain slices from postnatal day 1–12 (P1–12) rats in Mg2+-free solution. Electrical stimulation of the MNTB produced synaptic currents that were inwardly directed owing to the high

1Department of Neurobiology and 2Center for Neurological Basis of Cognition, University of Pittsburgh School of Medicine, W1412 Biomedical Science Tower, 3500 Terrace St., Pittsburgh, Pennsylvania 15261, USA. Correspondence should be addressed to K.K. ([email protected]).


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

NATURE NEUROSCIENCE ADVANCE ONLINE PUBLICATION 333

A R T I C L E S

internal chloride concentration of the pipette solution (Fig. 1a). Despite the well-documented GABA/glycinergic nature of MNTB neurons17, these currents were only partially blocked by the GABAA and glycine receptor antagonists bicuculline (10–20 µM) and strych-nine (1–30 µM) (31/42 cells, P1–P12). The current remaining after application of bicuculline or strychnine reversed at positive membrane potentials (Vrev = +5.0 ± 4.6 mV, n = 4), in contrast to pure GABA/glycine currents, which reversed at –14.8 ± 4.8 mV (n = 4 cells), close to the calculated reversal potential for chloride (–14 mV) (Fig. 1b). In all cases (n = 31), the bicuculline- and strychnine-insensitive current was blocked by the glutamate receptor antagonists APV (50 µM D-APV or 100 µM D,L-APV), CNQX (5 µM), or both, indicating that it was mediated by glutamate receptors.

Glutamate acts on NMDA receptorsMNTB-elicited glutamate responses had a long decay time course suggestive of an NMDAR-mediated component. Consistent with this, the NMDAR antagonist D-APV (50 µM, n = 6) abolished a large portion of the current, leaving behind a fast, rapidly decaying component that was sensitive to the AMPA and kainate receptor (AMPA/KA-R) antagonist CNQX (5 µM; n = 5 cells) (Fig. 1c). On average, NMDARs contributed over three-quarters of the glutama-tergic synaptic charge and over half of the peak current (80.8 ± 7.9% charge, or 64.2 ± 10.9% peak current, n = 7) (Fig. 1d,e).

In the spinal cord, glycinergic synapses can activate NMDARs through the spillover of glycine onto the glycine site of the NMDAR16, a mechanism that could also be responsible for the apparent glutamatergic response we observed in the MNTB-LSO pathway. To test this possibility, we saturated the NMDAR glycine site with high c oncentrations of the agonist D-serine (100–500 µM) in an attempt to occlude the response to synaptically released glycine. D-Serine had little effect on MNTB-elicited glutamate currents (Fig. 1f; 7.0 ± 9.0% reduction in peak amplitude, n = 8 cells), indi-cating that the majority of NMDARs were acti-vated by glutamate, not by glycine. This result, together with the fact that most responses included an AMPAR-mediated component that could not have been activated by glycine (Fig. 1c–f), strongly suggests that MNTB fibers release glutamate or another excitatory neu-rotransmitter such as aspartate that is capable of activating ionotropic glutamate receptors. The sensitivity of the glutamate response to APV (Fig. 1c–f), together with the failure of D-serine to block the response, further excludes the possibility that the current resulted from activation of NR3-containing excitatory gly-cine receptors18, lending support to the idea that glutamate is released by MNTB fibers.

Glutamate is released from MNTB neuronsIt is possible that this glutamate release stems from activation of a disynaptic pathway between the MNTB and the LSO, as GABA and glycine are depolarizing and excitatory at this age19 and thus could activate glutama-tergic neurons that project to the LSO. This

would have to be a pathway whose glutamatergic neurons were excited independently of GABAA and glycine receptor activation, as these recep-tors were blocked here. In addition, if there were a disynaptic pathway, we would expect a latency difference between monosynaptic GABA/glycine currents and disynaptic glutamate currents. However, the small difference in response latencies for the isolated glutamate current and the mixed current (Fig. 1f,g) (0.6 ± 0.5 ms at room temperature, P = 0.3, paired t-test, n = 6) argues against a disynaptic MNTB-LSO pathway. This is in agreement with previous anatomical and physiological stud-ies13,20 that did not find disynaptic connections from the MNTB to the LSO. The small latency differences we observed at room temperature may instead reflect release of GABA/glycine and glutamate from distinct presynaptic vesicular pools or differences in the spatial distribution of postsynaptic receptors (for example, extra- versus perisynaptic) .

Previous studies in vivo and in vitro have demonstrated that in the adult, the MNTB-LSO pathway is purely glycinergic, and that during development it is GABAergic5,9–12 as well. We were therefore concerned that the glutamatergic responses might have resulted from electrical stimulation of unknown glutamatergic fibers running near or through the MNTB. To test this possibility, we used focal photolysis of caged glu-tamate (n = 9 cells in 7 slices, P2–5) to activate specifically the somata and dendrites of MNTB neurons and to avoid stimulation of en passant axons (Fig. 2). Glutamate uncaging in the MNTB in the presence of bicucul-line (10 µM) and strychnine (1–10 µM) still elicited a synaptic current

Figure 1 Electrical stimulation of the MNTB causes release of glutamate at synapses in the LSO. (a) In Mg2+-free ACSF, MNTB stimulation caused a large, slowly decaying response that was reduced but not abolished by the GABAAR antagonist bicuculline (10 µM) and the glycine receptor antagonist strychnine (1 µM). The residual response was abolished by the addition of ionotropic glutamate receptor antagonists D,L-APV (100 µM) and CNQX (5 µM). P5 slice; traces shown are average of 20 responses. Scale bars: 20 pA, 50 ms. Black arrow: stimulus artifact. Unless otherwise indicated, holding potentials were approximately –60 mV. (b) Isolated GABA/glycine and glutamate currents in LSO neuron in Mg2+-free ACSF, in P2 slice. Application of D-APV (50 µM) and CNQX (5 µM) isolated a GABA/glycine current that reversed near the calculated Cl– reversal potential of –14 mV (left). Washout of APV and CNQX isolated a glutamate current that reversed at +10 mV (right). Scale bars: 50 pA, 50 ms. (c) Recordings from LSO neuron in Mg2+-free ACSF with bicuculline (10 µM) and strychnine (10 µM). APV removed the slowly decaying component, leaving a CNQX-sensitive component. Scale bars: 20 pA, 10 ms. (d) For six cells in which AMPA and NMDA currents were pharmacologically separated, amount of glutamate current charge (left) and amplitude (right) mediated by NMDARs. (e) Percentage of glutamate peak amplitude (left) or charge (right) mediated by NMDARs. Filled circles: averages ± s.e.m. (f) Recording in Mg2+-free ACSF in a P7 slice. Saturation of the glycine site by perfusion of D-serine (200 µM) for 5 min did not occlude the glutamatergic MNTB response, which is composed of both NMDAR and AMPAR components. Scale bars: 20 pA, 50 ms. (g) Response latencies for mixed and isolated glutamatergic currents recorded in Mg2+-free ACSF were not significantly different (P = 0.3, paired t-test, n = 6).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

in LSO neurons that reversed at positive membrane potentials and was blocked by APV (Fig. 2d). These glutamatergic synaptic responses were very sensitive to the mediolateral location of the uncaging site within the MNTB (Fig. 2a–c). This reflects the topographic organization of the MNTB-LSO pathway and the very focal activation of the MNTB neurons around the uncaging site, making it very unlikely that neurons outside the MNTB were activated. These results indicate that immature MNTB neurons release glutamate in the LSO.

We next addressed the question of whether glutamate, on the one hand, and GABA and glycine, on the other, are released from distinct populations of MNTB neurons or from the same MNTB fibers (Fig. 3a). In these experiments, we used minimal stimulation techniques13,21 and recorded in physiological Mg2+ conditions to minimize background noise caused by spontaneous NMDAR activation. In brain slices from P4–P7 rats, one-third of all presumptive single fibers (4 of 12 single-fiber recordings made on 12 cells in response to minimal stimula-tion) elicited a synaptic current (21.47 ± 6.55 pA) that persisted in the combined presence of bicuculline (10 µM) and strychnine (1 µM) (Fig. 3b,c) but was blocked by CNQX (5 µM) and D,L-APV (100 µM) (n = 3; data not shown). As was the case with multifiber stimula-tion, single-fiber glutamate currents had slightly, but statistically significantly, longer response latencies than mixed currents (0.175 ± 0.025 ms later, P ≤ 0.01). These data, together with previous results11, support the idea that individual MNTB fibers can release three neu-rotransmitters: glutamate, GABA and glycine.

Developmental profileBecause of the importance of NMDARs in the development and plasticity of excitatory and inhibitory neuronal circuits22–24, we determined the developmental profile of glutamatergic transmis-sion at MNTB-LSO synapses. In brain slices from P1–P8 rats, 96% of LSO neurons (25 of 26 cells) showed MNTB-elicited glutamater-gic responses, whereas in slices from P9–P12 rats, this fraction was only 31% (4 of 13 cells) (Fig. 4). Notably, the period during which glutamatergic transmission was encountered most frequently coincides with the period during which GABA and glycine are also depolarizing in the LSO and during which MNTB-LSO synapses are undergoing functional refinement13–15,25.

MNTB terminals contain both GABA and glutamate vesiclesWe next asked whether markers of both glutamate and GABA/gly-cine terminals were expressed together in terminals in the LSO, as would be expected if glutamate release indeed is involved in synapse-specific refinement. We addressed this question by using the gluta-mate and GABA/glycine vesicular transporters, which are specific for their respective neurotransmitters and thus determine syn-aptic vesicle content26, as markers. In the LSO, immunolabeling for the vesicular glutamate transporters VGLUT1 and VGLUT3 (n = 6 rats, P4–14) was so intense that the characteristic S-shaped LSO was instantly recognizable (Fig. 5a,b). VGLUT3 label in the LSO formed clusters (<1 µm) presumably representing synaptic termi-nals, as indicated by their immunoreactivity for the synaptic protein SV2 (n = 3 rats, Fig. 5c). Many of these VGLUT3-immunopositive clusters were also immunopositive for VGAT (Fig. 5d, n = 3 rats, Pearson’s correlation coefficient = 0.22, control = 0.02). In triple labeling experiments, clusters labeled for both VGLUT3 and VGAT were also positive for SV2 (Fig. 5e,f), indicating that VGLUT3 and VGAT were present together in the same presynaptic terminals. In contrast, VGLUT1, which also was strongly expressed in the LSO, did not colocalize with VGAT (Fig. 5g), although it did colocalize with VGLUT2 (Fig. 5h).

Although the MNTB provides the principal inhibitory input to the LSO, other potential GABA/glycinergic inputs from the ventral nucleus of the trapezoid body have been described5,17. The clear labeling of most MNTB cell bodies not only for VGAT, as is expected from their GABA/glycinergic phenotype27,28, but also for VGLUT3 makes it likely that most VGAT- and VGLUT3-positive terminals in the LSO are axon termi-nals from the MNTB (Fig. 6a, n = 6 rats, P4–14). To pursue this further, we filled individual MNTB cells (Fig. 6b,c) in acute brain slices (8 cells, 4 rats) and subsequently processed for SV2 and VGLUT3 immunore-activity. SV2-positive presynaptic MNTB terminals were also VGLUT3 immunopositive (Fig. 6d,e), suggesting that individual MNTB terminals can contain glutamatergic as well as GABA/ glycinergic vesicles, or per-haps even mixed glutamate/GABA/glycinergic synaptic vesicles.

DISCUSSIONHere we have presented physiological and anatomical evidence that developing GABA/glycinergic synapses also release glutamate. To our

Figure 2 The glutamatergic response arises from cell bodies of the MNTB. (a–d) Position of optical fiber for glutamate uncaging is shown in left column (MNTB outlined in solid white; initial position of optical fiber outlined with dashed black line) and corresponding postsynaptic response in LSO neuron is shown in right column. Traces (average of three stimulations) are presented in order of acquisition. Focal application of glutamate in the MNTB produced an inward current (a) in an LSO neuron ∼500 µm away. Uncaging at a nearby site (b) within the MNTB did not elicit a response, though it could subsequently elicit a response at the original site (c). Finally, D-APV (50 µM) bath application abolished the LSO response (d). Bicuculline (10 µM) and strychnine (10 µM) were present throughout. P4 slice. Scale bars: 20 pA, 100 ms.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

knowledge, this constitutes the first description of glutamate release at a GABA/glycinergic synapse. Previous studies have reported release of two fast-acting neurotransmitters by individual neurons, including GABA release from glutamatergic mossy fibers in young or kindled hippocampus29–33. The release of the three classical small amino acid neurotransmitters, with seemingly opposing postsynaptic functions, represents the most extreme example of multiple neurotransmitter release. Our results also provide a functional explanation for the long puzzling but neglected findings of high levels of glutamate immuno-reactivity in MNTB cells34 and of immunoreactivity for both glycine and glutamate in axon terminals in the LSO35.

Why has glutamatergic transmission not been previously described in this well-studied pathway? In addition to inputs from the MNTB, LSO neurons receive prominent glutamatergic inputs from other sources17. Consequently, the glutamatergic currents that have occa-sionally been observed after MNTB stimulation have been seen as resulting from activity in these other glutamatergic fibers and hence have been ignored or pharmacologically blocked11,36. In addition, most of the MNTB-elicited glutamate response is mediated by NMDARs, which, as elsewhere, are magnesium sensitive in the LSO37. Therefore, in voltage-clamp recordings at normal resting membrane potentials and at physiological Mg2+ concentrations, NMDARs will remain blocked. In current-clamp recordings, relief of Mg2+ block is likely to depend on depolarizations by GABA or glycine15,36,38, so pharma-cological blockade of GABA and glycine channels would also prevent activation of NMDARs.

Our immunolabeling results demonstrate that VGAT-positive MNTB neurons and terminals also express the vesicular glutamate transporter VGLUT3 but do not express VGLUT1. This suggests that glutamate release from GABA/glycinergic MNTB terminals is medi-ated by VGLUT3, whereas glutamate release from cochlea nucleus axon terminals is mediated by VGLUT1 and VGLUT2 or by VGLUT2 alone. VGLUT3 was recently identified by homology to VGLUT1 and VGLUT2, the vesicular glutamate transporters most commonly asso-ciated with glutamatergic terminals39–42. The role of VGLUT3 in the nervous system is still unknown, as this transporter is structurally and functionally similar to VGLUT1 and VGLUT2 yet is expressed in termi-nals of cholinergic, serotonergic, GABAergic and glycinergic neurons42. It has been proposed that VGLUT3 supports glutamate release from these ‘non-glutamatergic’ neurons, and VGLUT3 has been reported to mediate release of glutamate from dendrites43. Dendritic glutamate release from LSO neurons is unlikely here, as it would have to be trig-gered by a signal other than activation of GABAA and glycine receptors, which were blocked in our experiments. In addition, dendritic release would have added an apparent synaptic delay, which was not observed, to the glutamate response. Glutamate release from VGLUT3-positive GABA/glycinergic terminals thus provides the first functional support, albeit only correlative, for the hypothesis that VGLUT3 enables release of glutamate from ‘non-glutamatergic’ presynaptic terminals.

Glutamatergic neurotransmission in this pathway was most preva-lent during the first postnatal week. It is unlikely that a switch in neonatal glycine receptor subtypes44, accompanied by decreased strychnine affinity, could alone account for our results, as (i) we increased the strychnine concentration tenfold in some cases to ensure that glycine receptors were blocked and (ii) the residual cur-rent showed the reversal potential and pharmacology expected of a glutamate, not a glycine, current. We do not currently know what signals and mechanisms initiate and mediate the closure of this glutamatergic period. However, it is intriguing that downregulation of glutamatergic transmission correlates with other fundamental changes at this synapse, some of which could be involved in closing the glutamatergic period. For example, at the end of the first postnatal week, the quality of MNTB-LSO synapses switches from depolarizing to hyperpolarizing11,12,15,25. As a consequence of this switch, these synapses lose their ability to increase postsynaptic calcium con-centration, which in turn may influence the number or activity of postsynaptic glutamate receptors45. At the same time, the trans-mitter phenotype of MNTB neurons changes from predominantly

Figure 3 Minimal stimulation in a P7 LSO neuron in normal ACSF elicits a mixed GABA/glycine and glutamate response. (a) Input/output relationship in the absence of GABAA/glycine receptor antagonists. (b) Current amplitudes recorded in response to minimal stimulation (failure rate 39% for this cell). Insets: individual (gray) and average (black) successful responses in no-drug and bicuculline + strychnine conditions. Scale bars: 20 pA, 5 ms. (c) Average mixed current (black trace, 25.9 pA) and isolated glutamate current (gray trace, 15.6 pA) responses to minimal stimulation. Scale bars: 10 pA, 5 ms. (d) Average single-fiber mixed response (top trace) with no drugs, isolated glutamate response (middle trace), and remaining response after application of CNQX (bottom trace). Responses were recorded in normal ACSF at holding potentials of –60 mV. a–c, P7 neuron; d, P6 neuron.

Figure 4 Developmental profile of the glutamatergic response. For each of three age groups, the percentage of LSO cells recorded that showed a gluta-matergic response to MNTB stimulation is shown. (n = 9, 19 and 13 cells for groups P1–4, P5–8 and P9–12, respectively, χ2 P < 0.001). Age groups match published chloride reversal potentials in LSO neurons of –47, –58 and –82 mV (ref. 25).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

GABAergic to predominantly glycinergic, and MNTB inputs conse-quently activate different classes of postsynaptic receptors along with their associated downstream effectors. Finally, the apparent down-regulation of functional glutamate transmission could also reflect changes in presynaptic glutamate release or in glutamate uptake46. Future experiments addressing these possibilities will be necessary to understand potential interactions between early glutamatergic trans-mission and these changes in basic synaptic properties.

What is the role of glutamate release at GABA/glycinergic synapses? Glutamatergic transmission coincides with the period when MNTB-LSO connections are either eliminated or strengthened, which is cru-cial for tonotopic sharpening. NMDARs are pivotal in the development and plasticity of not only excitatory but also inhibitory synapses23. For example, at several GABAergic and glycinergic synapses, NMDAR-mediated calcium influx is required for the induction of long-term plasticity47,48. In immature hippocampal neurons, GABAergic syn-apses can activate NMDARs through the depolarizing action of GABA, which removes the Mg2+ block of NMDARs and allows activation of NMDARs by glutamate38. However, the source of glutamate at these GABAergic synapses has been an open question. Glutamate release at depolarizing GABA/ glycinergic synapses provides a straightforward mechanism by which GABA/glycinergic inputs could gain access to activated NMDARs in a synapse-specific manner. This previously unknown mechanism for NMDAR activation by GABA/glycinergic

synapses might mediate synapse-specific refinement and tonotopic sharpening during the development of this glycinergic map. It remains to be shown whether glutamate release at developing inhibitory syn-apses also occurs at other GABAergic or glycinergic synapses. However, in light of the VGLUT3 expression observed in a variety of GABA and glycinergic neurons42, this seems a likely possibility.

METHODSElectrophysiology. Experimental procedures were in accordance with US National Institutes of Health guidelines and were approved by the Institutional Animal Care and Use Committee at the University of Pittsburgh. Rats aged postnatal day 1–12 (P1–12) were deeply anesthetized with hypothermia or isofluorane, brains were removed in ice-cold artificial CSF (ACSF) with 1 mM kynurenic acid, and 300-µm-thick coronal slices were cut with a vibratome. Slices were allowed to recover for at least 1 h in an interface chamber before recording. Slices were trans-ferred to a submersion-type chamber mounted on an upright microscope and perfused continuously with Mg2+-free ACSF, containing (in mM), NaCl, 126; K2SO4, 1; KCl, 3; KH2PO4, 1.25; dextrose, 10; NaHCO3, 26; and CaCl2, 2. A 1-MΩ patch electrode filled with ACSF was placed in the MNTB for electrical stimulation. For photostimulation, p-hydroxyphenacylglutamate49 (200 µM) was added to the ACSF. UV light pulses (50–100 ms) were delivered via an optical fiber (40 µm) placed directly above the MNTB to uncage glutamate focally. Electrodes (borosili-cate glass, A-M Systems) were pulled to 1–4 MΩ and filled with a cesium gluconate solution containing (in mM) D-gluconic acid, 64; CsOH, 64; EGTA 11; CsCl, 56; MgCl2, 1; CaCl2, 1; HEPES, 10; and QX-314, 5). Stock solutions of bicuculline (Tocris), strychnine (Sigma), CNQX (Tocris) and/or D- or D,L-APV (Tocris) were diluted in ACSF for bath application. Whole-cell voltage clamp recordings were obtained from principal cells of the LSO visualized with gradient contrast optics (Luigs-Neumann). Signals were filtered at 2 kHz (Axon Instruments) and digitized at 5 kHz (custom LabView acquisition program). Offline analysis of acquired traces was performed with custom LabView and Matlab programs. Latencies,

Figure 5 Immunostaining for vesicular transporters in the LSO. (a) VGLUT3 expression clearly identifies the S-shaped LSO of P5 rat brain, and also brightly labels the superior paraolivary nucleus just medial to the LSO. (b) VGLUT1 expression in the LSO of P5 rat clearly identifies the S-shaped LSO and the medial superior olive. Scale bars: 100 µm. Dorsal is up and lateral to the right in both frames. (c) VGLUT3 (red) and SV2 (green) immunoreactivity colocalize (yellow) in the LSO. (d) VGLUT3 clusters (red) and VGAT clusters (green) colocalize (yellow) in presumptive terminals in the LSO. (e) Two-channel image of triple-labeled section shows colocalized (yellow) VGLUT3 clusters (red) and VGAT clusters (green). (f) Addition of the SV2 channel (blue) to the image shows that these VGLUT3-VGAT colocalized clusters are in presynaptic terminals, as indicated by their colocalization (white) with SV2. Scale bars in e,f: 1 µm. (g) VGLUT1 does not colocalize with VGAT, as indicated by the distinct green and red, and the absence of yellow, label. Scale bars in c,d,g: 2 µm. Images d and g are from adjacent sections of the same P5 LSO. (h) Overlap coefficient (Pearson’s r) for different immunolabeled protein pairs (black bars). As a random colocalization control using two images with the same spatial statistics, correlation coefficients were recalculated for the same image pairs after reflecting the green channel across the vertical or horizontal axis (white bars indicate average correlation coefficients for image pairs with the green channel reflected across each axis).

Figure 6 Immunostaining for vesicular transporters VGAT and VGLUT3 in cells of the MNTB. (a) MNTB cell bodies contain both VGAT (green) and VGLUT3 (red) immunoreactivity. Scale bar: 10 µm. (b) MNTB neuron filled with Alexa 568. Scale bar: 10µm. (c) Dye-filled MNTB axon collateral (red) in the LSO, compressed z-stack. Scale bar: 10 µm. (d) Two-channel image shows SV2-labeled (green) presynaptic terminals of identified MNTB axon (red; presynaptic terminals of this axon are thus yellow). Scale bar: 10 µm. (e) Three-channel image with VGLUT3 (blue) indicates the presence of VGLUT3 in identified MNTB presynaptic terminal. Scale bar: 1 µm.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

determined from averaged responses, were taken to be the point at which responses reached 20% of their peak amplitude. For minimal stimulation experiments, which were performed in normal Mg2+ ACSF (containing, in mM: NaCl, 126; MgSO4, 1.3; KCl, 3; KH2PO4, 1.25; dextrose, 10; NaHCO3, 26; and CaCl2, 2), a stimulus-response relation was obtained and the first plateau was determined at stimulus intensities for which failure rate decreased without increasing the mean amplitude of successful responses (Fig. 3a). Within this (plateau) range of stimulus intensities, a stimulus intensity was chosen that resulted in failures 30–40% of the time. Under these conditions, it is likely that only one fiber is activated. At these stimulus intensities (10–40 µA), 50–100 responses were evoked at 0.2 Hz before drug application, and another ∼50 responses were evoked in the presence of bicu-culline and strychnine to unmask glutamate components. Successful responses were identified offline by eye. Only those responses whose latencies fell within a 1-ms window were accepted.

Immunohistochemistry. Coronal brainstem slices (400 µm) were prepared as described above and fixed in 4% paraformaldehyde (PFA) for 2 h. Alternatively, rats were deeply anesthetized with isofluorane and then perfused transcardially with phosphate-buffered saline followed by 4% PFA and brains were removed and postfixed overnight. Tissue was cryoprotected in 30% sucrose and then sectioned at 10 µm with a cryostat. For double and triple immunofluorescence labeling, the following primary antibodies were used: rabbit anti–VGLUT1-2 and VGAT (gift of R. Edwards), goat anti–VGLUT1-3 (Chemicon), and mouse anti-SV2 (Iowa Developmental Studies Hybridoma Bank, gift of W. Halfter). Secondary antibod-ies were conjugated to Cy2, Alexa 488, Cy3 or Cy5 (Jackson; Molecular Probes). Control sections incubated with secondary antibodies alone showed only weak, diffuse background staining. To identify MNTB-LSO axons, MNTB principal cells in 400 µm acute slices (P5–8) were patched with electrodes (3–6 MΩ) containing a 5% solution of Alexa 568 KCl in the standard pipette solution and were held for 20–30 min; slices were then removed to an interface chamber and allowed to recover for 2–3 h before fixation in 4% PFA for 2 h. These slices were resectioned on the cryostat and immunostained for VGLUT3 and SV2. Fluorescence images were acquired in z-stacks using a confocal microscope (60× 1.4-NA lens, FV-500, Olympus), with sequential imaging of each channel to minimize bleedthrough between channels. All images shown are of single confocal planes unless stated otherwise. Quantification was performed on images from single confocal planes, with each channel thresholded separately to save the top 10–15% of signal pixels in that channel, using custom Matlab functions. To estimate colocalization across rats and LSOs, the Pearson’s correlation coefficient was calculated on the thresh-olded images for multiple sites in LSOs from different rats and staining runs, and these coefficients were averaged for each pair of labeled proteins. To compare the acquired images with ‘random’ images with the same spatial statistics, one channel was reflected relative to the other, thus preserving intrachannel spatial statistics while gaining relatively random interchannel correlations. One channel from each image pair was flipped vertically and horizontally, then the correlation coefficient was determined for each flipped image pair; these two values were averaged to obtain the pseudorandom correlation coefficient shown in the white bars.

ACKNOWLEDGMENTSThe authors thank K. Cihil for technical assistance, R. Edwards for VGAT and VGLUT1/2 antibodies, R. Givens for caged glutamate, W. Halfter for SV2 antibodies, and P. Land, S. Shand and S. Watkins for lending histology advice and equipment. We are grateful to E. Aizenman, S. Amara, N.K. Baba, J. Johnson and L. Lillien for comments on an earlier version of the manuscript. This work was supported by grants from the National Institute on Deafness and Other Communication Disorders (K.K., D.C.G.) and the National Institute of Neurological Disorders and Stroke (D.C.G.).


Received 7 December; accepted 29 December 2004Published online at http://www.nature.com/natureneuroscience/

1. Katz, L.C. & Shatz, C.J. Synaptic activity and the construction of cortical circuits. Science 274, 1133–1138 (1996).

2. Hendry, S.H. & Jones, E.G. Reduction in number of immunostained GABAergic neurones in deprived-eye dominance columns of monkey area 17. Nature 320, 750–753 (1986).

3. Otis, T.S., DeKoninck, Y. & Mody, I. Lasting potentiation of inhibition is associated

with an increased number of gamma-aminobutyric acid type A receptors activated during miniature inhibitory postsynaptic currents. Proc. Natl. Acad. Sci. USA 91, 7698–7702 (1994).

4. Micheva, K.D. & Beaulieu, C. An anatomical substrate for experience-dependent plas-ticity of the rat barrel field cortex. Proc. Natl. Acad. Sci. USA 92, 11834–11838 (1995).

5. Korada, S. & Schwartz, I.R. Development of GABA, glycine, and their receptors in the auditory brainstem of gerbil: a light and electron microscopic study. J. Comp. Neurol. 409, 664–681 (1999).

6. Kilman, V., van Rossum, M.C. & Turrigiano, G.G. Activity deprivation reduces minia-ture IPSC amplitude by decreasing the number of postsynaptic GABA(A) receptors clustered at neurocortical synapses. J. Neurosci. 22, 1328–1337 (2002).

7. Woodin, M.A., Ganguly, K. & Poo, M.M. Coincident pre- and postsynaptic activity modifies GABAergic synapses by postsynaptic changes in Cl– transporter activity. Neuron 39, 807–820 (2003).

8. Kandler, K. Activity-dependent organization of inhibitory circuits: lessons from the auditory system. Curr. Opin. Neurobiol. 14, 96–104 (2004).

9. Boudreau, J.C. & Tsuchitani, C. Binaural interaction in the cat superior olive S seg-ment. J. Neurophysiol. 31, 442–454 (1968).

10. Moore, M.J. & Caspary, D.M. Strychnine blocks binaural inhibition in lateral superior olivary neurons. J. Neurosci. 3, 237–242 (1983).

11. Kotak, V.C., Korada, S., Schwartz, I.R. & Sanes, D.H. A developmental shift from GABAergic to glycinergic transmission in the central auditory system. J. Neurosci. 18, 4646–4655 (1998).

12. Nabekura, J. et al. Developmental switch from GABA to glycine release in single central synaptic terminals. Nat. Neurosci. 7, 17–23 (2004).

13. Kim, G. & Kandler, K. Elimination and strengthening of glycinergic/GABAergic con-nections during tonotopic map formation. Nat. Neurosci. 6, 282–290 (2003).

14. Sanes, D.H. & Friauf, E. Development and influence of inhibition in the lateral superior olivary nucleus. Hear. Res. 147, 46–58 (2000).

15. Kandler, K. & Friauf, E. Development of glycinergic and glutamatergic synaptic transmis-sion in the auditory brainstem of perinatal rats. J. Neurosci. 15, 6890–6904 (1995).

16. Ahmadi, S. et al. Facilitation of spinal NMDA receptor currents by spillover of synapti-cally released glycine. Science 300, 2094–2097 (2003).

17. Schwartz, I. The superior olivary complex and lateral lemniscal nuclei. in The Mammalian Auditory Pathway: Neuroanatomy (eds. Popper, A.N., Fay, R.R. & Webster, D.B.) 117–167 (Springer, New York, 1992).

18. Chatterton, J.E. et al. Excitatory glycine receptors containing the NR3 family of NMDA receptor subunits. Nature 415, 793–798 (2002).

19. Kullmann, P.H. & Kandler, K. Glycinergic/GABAergic synapses in the lateral superior olive are excitatory in neonatal C57B1/6J mice. Brain Res. Dev. Brain Res. 131, 143–147 (2001).

20. Lohmann, C., Ehrlich, I. & Friauf, E. Axon regeneration in organotypic slice cultures from the mammalian auditory system is topographic and functional. J. Neurobiol. 41, 596–611 (1999).

21. Stevens, C.F. & Wang, Y. Facilitation and depression at single central synapses. Neuron 14, 795–802 (1995).

22. Aamodt, S.M., Shi, J., Colonnese, M.T., Veras, W. & Constantine-Paton, M. Chronic NMDA exposure accelerates development of GABAergic inhibition in the superior colliculus. J. Neurophysiol. 83, 1580–1591 (2000).

23. Gaiarsa, J.L., Caillard, O. & Ben-Ari, Y. Long-term plasticity at GABAergic and gly-cinergic synapses: mechanisms and functional significance. Trends Neurosci. 25, 564–570 (2002).

24. Dickson, K.S. & Kind, P.C. NMDA receptors: neural map designers and refiners? Curr. Biol. 13, R920–R922 (2003).

25. Ehrlich, I., Lohrke, S. & Friauf, E. Shift from depolarizing to hyperpolarizing glycine action in rat auditory neurons is due to age-dependent Cl– regulation. J. Physiol. (Lond.) 520, 121–137 (1999).

26. Fon, E.A. & Edwards, R.H. Molecular mechanisms of neurotransmitter release. Muscle Nerve 24, 581–601 (2001).

27. Helfert, R.H., Bonneau, J.M., Wenthold, R.J. & Altschuler, R.A. GABA and glycine immunoreactivity in the guinea pig superior olivary complex. Brain Res. 501, 269–286 (1989).

28. Bledsoe, S.C. Jr. et al. Immunocytochemical and lesion studies support the hypothesis that the projection from the medial nucleus of the trapezoid body to the lateral superior olive is glycinergic. Brain Res. 517, 189–194 (1990).

29. Docherty, M., Bradford, H.F. & Wu, J.Y. Co-release of glutamate and aspartate from cholinergic and GABAergic synaptosomes. Nature 330, 64–66 (1987).

30. O’Malley, D.M. & Masland, R.H. Co-release of acetylcholine and gamma-a2inobutyric acid by a retinal neuron. Proc. Natl. Acad. Sci. USA 86, 3414–3418 (1989).

31. Jonas, P., Bischofberger, J. & Sandkuhler, J. Corelease of two fast neurotransmitters at a central synapse. Science 281, 419–424 (1998).

32. Walker, M.C., Ruiz, A. & Kullmann, D.M. Monosynaptic GABAergic signaling from dentate to CA3 with a pharmacological and physiological profile typical of mossy fiber synapses. Neuron 29, 703–715 (2001).

33. Gutierrez, R. et al. Plasticity of the GABAergic phenotype of the ‘glutamatergic’ granule cells of the rat dentate gyrus. J. Neurosci. 23, 5594–5598 (2003).

34. Glendenning, K.K., Masterton, R.B., Baker, B.N. & Wenthold, R.J. Acoustic chiasm. III: Nature, distribution, and sources of afferents to the lateral superior olive in the cat. J. Comp. Neurol. 310, 377–400 (1991).

35. Helfert, R.H. et al. Patterns of glutamate, glycine, and GABA immunolabeling in four synaptic terminal classes in the lateral superior olive of the guinea pig. J. Comp. Neurol. 323, 305–325 (1992).

36. Kullmann, P.H., Ene, F.A. & Kandler, K. Glycinergic and GABAergic calcium responses

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

in the developing lateral superior olive. Eur. J. Neurosci. 15, 1093–1104 (2002).37. Ene, F.A., Kullmann, P.H., Gillespie, D.C. & Kandler, K. Glutamatergic calcium

responses in the developing lateral superior olive: receptor types and their spe-cific activation by synaptic activity patterns. J. Neurophysiol. 90, 2581–2591 (2003).

38. Ben-Ari, Y., Khazipov, R., Leinekugel, X., Caillard, O. & Gaiarsa, J.L. GABAA, NMDA and AMPA receptors: a developmentally regulated ‘menage a trois’. Trends Neurosci. 20, 523–529 (1997).

39. Bellocchio, E.E., Reimer, R.J., Fremeau, R.T. & Edwards, R.H. Uptake of glutamate into synaptic vesicles by an inorganic phosphate transporter. Science 289, 957–960 (2000).

40. Wojcik, S.M. et al. An essential role for vesicular glutamate transporter 1 (VGLUT1) in postnatal development and control of quantal size. Proc. Natl. Acad. Sci. USA 101, 7158–7163 (2004).

41. Fremeau, R.T. et al. Vesicular glutamate transporters 1 and 2 target to functionally distinct synaptic release sites. Science 304, 1815–1819 (2004).

42. Fremeau, R.T. Jr., Voglmaier, S., Seal, R.P. & Edwards, R.H. VGLUTs define subsets of excitatory neurons and suggest novel roles for glutamate. Trends Neurosci. 27, 98–103 (2004).

43. Harkany, T. et al. Endocannabinoid-independent retrograde signaling at inhibitory synapses in layer 2/3 of neocortex: involvement of vesicular glutamate transporter 3. J. Neurosci. 24, 4978–4988 (2004).

44. Lynch, J.W. Molecular structure and function of the glycine receptor chloride channel. Physiol. Rev. 84, 1051–1095 (2004).

45. Malinow, R. AMPA receptor trafficking and long-term potentiation. Philos. Trans. R. Soc. Lond. B Biol. Sci. 358, 707–714 (2003).

46. Furuta, A., Rothstein, J.D. & Martin, L.J. Glutamate transporter protein subtypes are expressed differentially during rat CNS development. J. Neurosci. 17, 8363–8375 (1997).

47. McLean, H.A., Caillard, O., Ben-Ari, Y. & Gaiarsa, J.L. Bidirectional plasticity expressed by GABAergic synapses in the neonatal rat hippocampus. J. Physiol. (Lond.) 496, 471–477 (1996).

48. Morishita, W. & Sastry, B.R. Postsynaptic mechanisms underlying long-term depression of GABAergic transmission in neurons of the deep cerebellar nuclei. J. Neurophysiol. 76, 59–68 (1996).

49. Givens, R.S., Weber, J.F., Jung, A.H. & Park, C.H. New photoprotecting groups: desyl and p-hydroxyphenacyl phosphate and carboxylate esters. Methods Enzymol. 291, 1–29 (1998).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S


Alcohol-induced motor impairment caused by increased extrasynaptic GABAA receptor activityH Jacob Hanchar1,3, Paul D Dodson2,3, Richard W Olsen1, Thomas S Otis2 & Martin Wallner1

Neuronal mechanisms underlying alcohol intoxication are unclear. We find that alcohol impairs motor coordination by enhancing tonic inhibition mediated by a specific subtype of extrasynaptic GABAA receptor (GABAR), α6β3δ, expressed exclusively in cerebellar granule cells. In recombinant studies, we characterize a naturally occurring single-nucleotide polymorphism that causes a single amino acid change (R100Q) in α6 (encoded in rats by the Gabra6 gene). We show that this change selectively increases alcohol sensitivity of α6β3δ GABARs. Behavioral and electrophysiological comparisons of Gabra6100R/100R and Gabra6100Q/100Q rats strongly suggest that alcohol impairs motor coordination by enhancing granule cell tonic inhibition. These findings identify extrasynaptic GABARs as critical targets underlying low-dose alcohol intoxication and demonstrate that subtle changes in tonic inhibition in one class of neurons can alter behavior.

Humans have been consuming alcohol for thousands of years, and the use of alcoholic beverages pervades human culture and society and can have substantial health effects1. Different mechanisms by which ethanol might depress brain function have been proposed based on ethanol’s ability to modulate a wide variety of ion channels2–4, neu-rotransmitter receptors5–10 and transporters11. Among these diverse targets, however, GABARs are arguably the most attractive candidates. This is in part because other classes of known GABAR modulators such as benzodiazepines, barbiturates and certain anesthetics lead to behavioral effects that closely resemble ethanol intoxication. Yet despite strong evidence implicating GABARs in ethanol’s action, critical details remain unclear. For instance, although it is known that native GABARs are heteropentamers assembled from 19 possible subunits12,13, it has not been possible to link the activity of particular GABAR subunits to changes in behavioral sensitivity to ethanol.

Recent studies suggest that specific combinations of GABAR subunits (those containing α4β3δ and α6β3δ) are uniquely sensitive to ethanol, showing dose dependencies that mirror blood alcohol levels associ-ated with intoxication in humans9,10. GABARs containing α4 and δ subunits are expressed in many brain regions14,15, but α6 is found in only two types of neurons (cerebellar granule cells and granule cells in the cochlear nucleus) and is expressed together with δ only in cerebel-lar granule cells14,16,17. In granule cells α6 and δ combine with β sub-units to give rise to high-affinity18 extrasynaptic GABARs19–21. These GABARs generate a tonic inhibitory conductance20–22 that exerts strong control over granule cell firing patterns in vivo23.

We set out to examine whether such extrasynaptic GABARs con-taining α6 and δ subunits account for behavioral effects of ethanol at moderately intoxicating doses. To link these particular GABARs to

behavioral sensitivity, we first characterized a naturally occurring sin-gle-nucleotide polymorphism in the gene encoding rat α6 (Gabra6). This polymorphism is of interest because it has been reported to be present in alcohol-nontolerant (ANT) rats24 and enriched in Sardinian alcohol-nonpreferring rats25, two lines of animals selec-tively bred either for heightened ethanol-induced motor impairment (ANT rats) or for aversion to ethanol (Sardinian nonpreferring rats). A single nucleotide change (from a guanine to an adenine) leads to a single amino acid substitution, from arginine (R) to glutamine (Q), at amino acid position 100 in α6. The α6-R100Q polymorphism causes a marked increase in benzodiazepine impairment of motor coordi-nation in vivo and has been shown to convert recombinant GABARs containing α6 and γ subunits from benzodiazepine insensitive to benzodiazepine sensitive24. However, because it has not been shown that this polymorphism affects ethanol sensitivity in recombinant or native GABARs, it has been suggested that other co-segregating polymorphisms might be responsible for the increased ethanol sen-sitivity in ANT rats26,27.

We report here that the α6-R100Q polymorphism further enhances the ethanol sensitivity of a specific subtype of GABAR composed of α6, β3 and δ subunits. Notably, we found that the α6-R100Q poly-morphism is common in outbred strains of Sprague-Dawley rats. This allowed us to obtain rats homozygous for each of the two alleles and then to test ethanol sensitivity in situ. We showed that behaviorally-relevant concentrations of ethanol enhance tonic inhibition mediated by extrasynaptic GABARs in cerebellar granule cells from Gabra6100R/100R rats and cause an even larger increase in tonic inhibition in granule cells from Gabra6100Q/100Q rats. Finally, in a cerebellum-dependent behav-ioral task we found that ethanol more severely impairs coordination in

1Department of Molecular and Medical Pharmacology and 2Department of Neurobiology, David Geffen School of Medicine, University of California, Los Angeles, California 90095, USA. 3These authors contributed equally to this work. Correspondence should be addressed to M.W. ([email protected]) or T.S.O. ([email protected]).


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S


Gabra6100Q/100Q rats. Together these findings imply that small increases in a tonic GABA conductance in a single class of cerebellar neurons can account for the adverse effects of ethanol on motor coordination. The results strongly imply that similar, more widely expressed isoforms of extrasynaptic GABARs are critical components of ethanol intoxication.

RESULTSα6-R100Q polymorphism enhances ethanol sensitivityThe ethanol sensitivities of heteromeric GABARs containing the α6-100R and α6-100Q polymorphisms were tested by expressing each variant in Xenopus laevis oocytes together with a β subunit and either δ or a γ2 sub-unit. We evaluated ethanol dose-response curves for several combina-tions of GABAR subunits thought to exist in granule cells (α6β2δ, α6β3δ, α6β2γ2 and α6β3γ2)16,28,29 as well as other combinations that may be pres-ent in these cells (Fig. 1 and Table 1). Consistent with published results24, we found that introduction of the α6-R100Q polymorphism into GABARs composed of α6, β3 and γ2 gave rise to benzodiazepine-sensitive receptors (data not shown). However, the low ethanol sensitivity of α6(100R)β3γ2 GABARs was unchanged in α6(100Q)β3γ2 receptors (Table 1).

In contrast, for the highly ethanol-sensitive α6β3δ GABARs10, we found that changing amino acid position 100 in α6 from R to Q leads to a further increase in ethanol sensitivity (Fig. 1). Notably, α6β3δ receptors were the only GABARs that showed a significant change in ethanol sensitivity in response to the α6-R100Q polymorphism (Fig. 1 and Table 1). In par-ticular, α6β2δ receptors, another species of GABAR that may contribute to tonic GABA currents in granule cells, were unaffected.

Ethanol enhances tonic inhibition mediated by α6β3δ GABARsBased on these results, we hypothesized that the presence of this poly-morphism in α6 should enhance ethanol sensitivity of tonic GABA con-ductances in cerebellar granule cells, the only class of neurons in the brain that express α6 together with β3 and δ subunits. We fortuitously discovered that the Gabra6100Q allele is present in outbred Sprague-Dawley rats obtained from Charles River (Fig. 2). Genotyping 35 rats showed that 10 were Gabra6100R/100R, 11 were Gabra6100Q/100Q and 14 were heterozygotes. Whole-cell recordings were made from granule cells in cerebellar slices prepared from rats of the two homozygous genotypes (Fig. 3). We tested ethanol concentrations that would result in moderate to severe intoxication (30 and 100 mM) and found that they enhanced tonic GABA currents in granule cells from rats of both genotypes. However, the enhancement was significantly larger in granule cells from Gabra6100Q/100Q rats (Fig. 3b,c). Consistent with a recent report30, etha-nol also enhanced the frequency of spontaneous inhibitory postsynaptic currents (sIPSCs) in granule cells, and the mean increases in frequency were larger in recordings from the mutant rats (Fig. 3d). We did not find any ethanol-induced changes in the mean amplitudes of sIPSCs nor in their decay rates. These findings demonstrate that tonic GABA current is sensitive to low concentrations of ethanol in wild-type rats and that the presence of R100Q in α6 subunits renders tonic GABA current even more sensitive to ethanol. Furthermore, the increased sIPSC frequency in Gabra6100Q/100Q compared with Gabra6100R/100R rats is likely to result indirectly from changes in granule cell excitability, as α6 is not expressed in any other cerebellar cell type14,16.

Low ethanol doses act directly on extrasynaptic GABARsIn recombinant studies, we have shown that wild-type GABARs con-taining α4 or α6 along with β3 and δ subunits are enhanced by 10 mM

a

b

Ethanol (mM)

Figure 1 The α6-R100Q polymorphism leads to a marked increase in ethanol sensitivity when expressed with β3 and δ subunits. (a) GABARs of the indicated subunit compositions were expressed in X. laevis oocytes and activated by steady-state GABA (300 nM ∼EC30). Brief applications of ethanol at the indicated concentrations (in mM) enhance the current in a dose-dependent manner. (b) Dose-response curves for ethanol showing the peak enhancement of GABA current. Shown are wild-type and mutant versions of subunit combinations that are likely to be responsible for tonic GABA current in granule cells: α6β2δ (cross, n = 6), α6β3δ (asterisk, n = 10), α6(R100Q)β2δ (inverted triangles, n = 7) and α6(R100Q)β3δ (upright triangles, n = 8). Note that the β3 and δ subunits are required for the R100Q mutation to exert an effect and that replacement of β2 with β3 leads to an almost tenfold increase in ethanol sensitivity. Other combinations of subunits expressed in granule cells are summarized in Table 1.

ATTCCGAAAT TTCCGAAAT TTCCAAAATF R N F R N F Q N Q

Gabra6 Gabra6 Gabra6100R/100R 100R/100Q 100Q/100Q

Figure 2 A single-nucleotide polymorphism (a guanine-to-adenine substitution) in the gene encoding the rat GABAA receptor α6 subunit (Gabra6) is common in Sprague-Dawley rats obtained from Charles River Laboratories. Shown from left to right are the sequence chromatograms from a homozygous Gabra6100R/100R, a heterozygous Gabra6100R/100Q and a homozygous Gabra6100Q/100Q rat. The lower panel shows the sequence of each Gabra6 genotype along with the corresponding amino acid translation (bottom). Gray boxes denote the codon for amino acid position 100.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 3 |MARCH 2005 341

ethanol, a concentration below the legal maximally permitted blood alcohol level for motor vehicle operation in many countries (typically 0.08% (wt/vol) or 17.4 mM). To examine postsynaptic effects of such low doses of ethanol on tonic GABA currents, we performed experiments in the presence of 0.5 µM tetrodotoxin, 2 µM NBQX, 300 nM added GABA and 10 µM NO-711, a GABA transporter antagonist. Significant enhancement by 10 mM ethanol was observed for Gabra6100R/100R gran-ule cells, and even larger increases were observed for Gabra6100Q/100Q granule cells (Fig. 4). Together, these results indicate that increases in tonic GABA current could be detected independent of changes in GABA

release and in response to concentrations of ethanol associated with minimal intoxication.

Ethanol more severely impairs behavior in Gabra6100Q/100Q ratsTo explore whether increases in granule cell tonic inhibition in response to alcohol could account for motor behaviors associated with intoxication, we compared Gabra6100R/100R and Gabra6100Q/100Q rats in cerebellar and non-cerebellar behavioral tests. In the accelerat-ing rotarod test, a cerebellum-dependent behavioral assay, we found that low ethanol concentrations impaired Gabra6100R/100R rats in

Table 1 Ethanol and GABA sensitivity for GABARs of different subunit combinationsa

Percent enhancement by ethanol

Receptor GABA EC50 (n) n 10 mM ethanol 30 mM ethanol 100 mM ethanol 300 mM ethanol

α6(R100Q)β3δ 0.68 ± 0.1 (5) 8 99.3 ± 15.0 180.1 + 28.2 275.3 ± 32.4 389.2 ± 65.0

α6β3δ 0.70 ± 0.4 (6) 10 41.2 ± 4.3 92.5 ± 9.0 125.3 ± 20.5 245.0 ± 33.6

α6(R100Q)β2δ 0.51 ± 0.09 (5) 7 0 24.5 ± 10.7 97.0 ± 11.2 199.0 ± 38.1

α6β2δ 0.50 ± 0.03 (5) 6 0 23.1 ± 7.9 88.4 ± 15.6 175.0 ± 35.8

α6(R100Q)β1δ 0.62 + 0.04 8 0 24.1 ± 4.0 50.3 ± 7.8 185.2 ± 9.4

α6β1δ 0.56 + 0.07 9 0 21.2 ± 3.3 52.0 ± 5.6 167.9 ± 10.0

α6(R100Q)β2/3γ2L/S 19 ± 3.5 (6) 16 0 0 40.3 ± 7.8 167.5 ± 14.2

α6β2/3γ2L/S 19 ± 0.5 (6) 10 0 0 39.0 ± 7.5 182.6 ± 11.3

α1β3δ 0.56 ± 0.05 (4) 8 32.5 ± 5.4 88.2 ± 7.0 117.5 ± 10.6 295.3 ± 19.7

α1β2/3γ2L 6.8 ± 0.8 (5) 9 0 0 34.4 ± 9.9 147.3 ± 12.9

α1β2/3γ2S 8.8 ± 1.1 (6)aFor each indicated subunit combination, GABA dose-response relationships were determined by fitting the Hill equation (see Methods). Current enhancement by ethanol was measured by co-application of ethanol and GABA (∼EC30). Reported values are mean (±s.d.) percent increases above responses to GABA alone, zeroes represent measurements of no change in current, and n indi-cates the number of oocytes used to determine the average ethanol dose response curves. In cases where the identities of β or γ2 subunits did not lead to differences, results have been pooled.

Figure 3 Ethanol enhances granule cell tonic GABA current, and the enhancement is larger in Gabra6100Q/100Q rats. (a,b) Tonic GABA currents in granule cells recorded from Gabra6100R/100R (a) or Gabra6100Q/100Q (b) slices in the presence of indicated concentrations of ethanol or in a saturating concentration (10µM) of the GABAR antagonist SR95531. To the right are histograms of all points in each segment. Gaussian functions have been fit to each condition and are superimposed. The dashed lines indicate the mean current from these fits. (c) Summary of the mean ± s.e.m. percentage change in tonic current amplitude caused by 30 or 100 mM ethanol in the two genotypes (wild type, n = 5; mutant, n = 7 granule cells). Ethanol caused a significantly larger enhancement of mutant versus wild-type currents. Mean tonic GABA current under control conditions was –9.7 ± 1.7 pA and did not differ significantly between the two genotypes (P > 0.4). (d) Ethanol-induced increases in sIPSC frequency are significantly larger in mutant granule cells (Gabra6100R/100R, n = 5; Gabra6100Q/100Q, n = 7 granule cells). No significant changes were observed in either the amplitude (Gabra6100R/100R, 112.43 ± 12.57% of control, P > 0.49, n = 5; Gabra6100Q/100Q, 101.55 ± 2.42% of control, P > 0.7, n = 7) or the decay of sIPSCs (Gabra6100R/100R, 90 ± 12% of control, P > 0.42, n = 4; Gabra6100Q/100Q, 96 ± 18% of control, P > 0.77, n = 6) in ethanol (data not shown).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S


a dose- and time-dependent manner (Fig. 5a–d). Administration of identical doses of ethanol to Gabra6100Q/100Q rats led to signifi-cantly larger impairment on the rotarod at the three doses shown (0.75, 1 and 1.25 g kg–1). Blood ethanol concentrations obtained in these behavioral tests 25–30 min after administration of 1g kg–1 etha-nol (15.2 ± 2 mM in Gabra6100R/100R, 16.4 ± 2 mM in Gabra6100Q/100Q, P = 0.34, n = 4 and 5, respectively) confirmed that plasma concentra-tions of ethanol were not affected by genotype and showed that the concentrations achieved were the equivalent of mildly intoxicating doses in humans. In contrast, hypersensitivity to ethanol was not found in the loss of righting reflex (LORR) assay, a common behav-ioral test for sedative effects that arise principally from non-cerebellar brain areas (Fig. 5e,f).

Our finding that this single amino acid difference is sufficient for increased ethanol sensitivity of motor coordination suggests that it makes an important contribution to the ANT phenotype. Moreover, the α6-R100Q polymorphism enhances ethanol sensitivity only in GABARs composed of α6, β3 and δ subunits. As receptors of this molecular makeup are found only in extrasynaptic membrane of granule cells, the tonic GABA current carried they carry must reduce the excitability of this specific class of cerebellar neurons to account for the hypersensitive phenotype.

DISCUSSIONMechanisms underlying the effects of ethanolTwo general hypotheses have been put forward to explain how ethanol depresses neuronal activity. One proposes an indirect influence on ion channels and receptors through nonspecific interactions with mem-brane lipids. An alternative hypothesis is that ethanol interacts with specific, saturable binding sites on receptor or ion channel proteins. Many targets have been proposed including potassium channels2–4, glutamate-gated channels6, glycine-gated7 channels and GABAA recep-tors5,8–10. However, not all of these channels respond to the low concen-trations of ethanol associated with mild intoxication, and it is unclear how ethanol acting at these various sites alters behavior. Our results suggest that extrasynaptic GABARs are important targets for mildly intoxicating concentrations of ethanol. This is dramatically demon-strated by the Gabra6100Q/100Q rats, in which a single amino acid change in an extrasynaptic GABA receptor subtype expressed in one class of cerebellar neurons (granule cells) leads to the predicted enhancement of ethanol sensitivity of granule cell tonic inhibition and of cerebel-lum-dependent behavior.

Link between tonic inhibition and behaviorTonic inhibition mediated by extrasynaptic, δ subunit–contain-ing GABARs has been identified in a number of neurons31–35 and is thought to regulate neuronal excitability; however, the role of tonic inhibition in shaping behavior is unclear. The results reported here were obtained from cerebellar granule cells, a cell type for which there is strong evidence that receptors containing α6 and δ subunits are extrasynaptic16,19,36 and are required for tonic GABA current20,21. We demonstrate that tonic current is facilitated by low ethanol concentra-tions and, in addition, identify a variant of α6 (α6-100Q) that ren-ders tonic current carried by α6β3δ GABARs much more sensitive to ethanol. Behavioral analysis suggests that enhancement of this specific extrasynaptic GABAR subtype in granule cells results in motor impair-ment. How do modest changes in tonic inhibition of these neurons lead to such pronounced behavioral differences? First, granule cells have extremely high input resistance; only small changes in conductance are required to affect excitability. Second, tonic inhibition seems to influ-ence the gain of the input-output relationship in addition to shifting the amount of excitation required for granule cell firing37. Ethanol is not the only compound reported to selectively increase tonic inhibition in granule cells. Tonic currents have also been shown to be highly sensitive to endogenous neuroactive steroids21, raising the possibility that these compounds could similarly affect motor behavior. Future studies using such specific endogenous and exogenous modulators should provide insight into the role of extrasynaptic GABARs in influencing informa-tion processing in the granule cell layer.

Gabra6100Q allele and behavioral sensitivity to ethanolThrough selective breeding, several lines of rodents that show etha-nol hypersensitivity have been generated. In two such lines, the ANT and the Sardinian nonpreferring rats, the Gabra6100R and Gabra6100Q

Figure 4 Tonic GABA current in granule cells is enhanced by low concentrations of ethanol in Gabra6100R/100R and Gabra6100Q/100Q rats. (a,b) Tonic currents recorded from Gabra6100R/100R (a) or Gabra6100Q/100Q (b) granule cells showing that 10 mM ethanol enhances the tonic current in the continuous presence of 300 nM GABA, TTX, NBQX and NO-711. (c) Summary data indicating that in response to 10 mM ethanol, significant increases in tonic current are observed in Gabra6100R/100R granule cells (P < 0.05, n = 4) and larger enhancements are found in Gabra6100Q/100Q granule cells (P < 0.05, n = 4).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S


alleles have been found to segregate into ethanol-hyposensitive and -hypersensitive groups, respectively. For the Sardinian nonpreferring rats, it is a partial segregation25, whereas for the alcohol tolerant (AT)/alcohol non-tolerant (ANT) line, almost all of the ANT rats are homo-zygous for the Gabra6100Q allele24. However, because the α6-R100Q polymorphism had not been shown to affect ethanol sensitivity of recombinant receptors24 and because backcrosses of ANT/AT rats had led researchers to question the correlation between this polymorphism and the ethanol-hypersensitive phenotype, it has been concluded that genetic differences other than the one causing the α6-R100Q polymor-phism were important27,38.

It remains to be determined whether the α6 polymorphism accounts for all aspects of the ANT phenotype and whether it contributes to complex behaviors that do not obviously involve the cerebellum (such as alcohol preference in the Sardinian line or anxiety in ANT rats39). In these selectively bred lines, control and experimental groups of rats are typically isolated by more than 40 generations of breeding, a strategy likely to segregate multiple genetic loci. In the present study, control (Gabra6100R/100R) and experimental (Gabra6100Q/100Q) groups were iden-tified only by genotype with respect to this one gene. As is the case for all knockout and ‘knock-in’ animal studies, we cannot completely exclude the possibility that other alleles near the genetic locus of interest coseg-regate and contribute to cellular and behavioral phenotypes. However, considered with the positive evidence provided here that α6-100Q markedly increases ethanol sensitivity of recombinant α6β3δ receptors and tonic current in cerebellar granule cells, it seems very likely that the α6-100Q polymorphism is responsible for the additional ethanol-induced motor impairment.

Our findings suggest that mice deficient in the Gabra6 gene prod-uct might show less ethanol-induced motor impairment. High doses of ethanol (2 g kg–1) have been reported to similarly impair wild-type and Gabra6−/− mice in rotarod tests40, and at ethanol doses of 3.5 g kg–1, no significant differences have been observed between wild type and Gabra6−/− in a LORR test41,42. At such high ethanol doses, we also find no change for Gabra6100R/100R and Gabra6100Q/100Q rats in rotarod per-formance (data not shown) or in LORR (Fig 5e,f). It is possible that

Gabra6−/− mice will show less impairment on the rotarod in a lower range of ethanol doses, but there are reasons to believe that ethanol sensitivity of knockout mice may be complicated. First, the expression of GABAR subunits α1, β2, γ2 and δ is changed markedly in Gabra6−/− mice19,36,43, and other potential ethanol targets such as two-pore domain K+ chan-nels20 show increased function in granule cells, making it difficult to rule out compensatory changes in Gabra6−/− mice. Second, differences in cer-ebellum-dependent behavior at moderate ethanol doses may be obscured in the Gabra6−/− mouse by other more abundant and widely expressed ethanol targets. These and other high-affinity ethanol targets may make it more difficult to detect behavioral changes in animals lacking the α6 gene than in animals having the ethanol-hypersensitive polymorphism, in which very low doses of ethanol selectively impair motor behavior (as in Fig. 5b).

The observation that a specific combination of GABAR subunits forms an important ethanol target in cerebellum is similar to recent results showing that certain anesthetics and benzodiazepines act on heteromeric GABARs composed of particular subunits. Some of these studies make use of mice engineered to carry single amino acid changes in positions homologous to the R100Q site in α6. Point mutations at these sites, which are known to form part of a high-affinity bind-ing pocket for benzodiazepines at the α-γ2 interface in the pentam-eric receptor, render GABARs containing those particular subunits insensitive to benzodiazepines. The resulting 'knock-in' mice show specific behavioral insensitivities in response to diazepam13, implying that important clinical properties of diazepam such as sedation44,45, amnesia44 and anxiolysis46 are attributable to GABARs containing par-ticular α subunits. A similar strategy involving mice carrying a point mutation in β3 shows that this subunit is necessary for the anesthetic actions of etomidate47. The naturally occurring α6-R100Q mutation (polymorphism) in rats allowed us to examine the roles of α6 subunits and of granule cells in ethanol-induced impairment of cerebellar func-tion. Like the mice with experimentally introduced ('knock-in') point mutations discussed above, these rats should prove to be useful tools for examining the role of α6 in certain features of ethanol and benzo-diazepine intoxication.

Figure 5 Rats homozygous for the α6-100Q polymorphism show increased alcohol-induced motor impairment as compared with Gabra6100R/100R rats. Motor coordination was assessed by testing Gabra6100R/100R (open circles) and Gabra6100Q/100Q (filled circles) rats in an accelerating rotarod test. (a–d) Performance was measured as the latency to fall from the rotating rod before (‘Pre’) and 20, 40 and 60 min after intraperitoneal injection of saline (a; n = 5 rats each group), 0.75 g kg–1 (b, n = 7 and 8 rats for Gabra6100R/100R and Gabra6100Q/100Q, respectively), 1 g kg–1 (c; n = 6 and 7 rats for Gabra6100R/100R and Gabra6100Q/100Q, respectively) or 1.25 g kg–1 ethanol (d; n = 8 rats each group). Tests on the three postinjection data points yielded the following statistics: saline controls, Gabra6100R/100R versus Gabra6100Q/100Q: F2,7 = 0.50, P = 0.63; 0.75 g kg–1 ethanol, Gabra6100R/100R versus Gabra6100Q/100Q: F2,12 = 6.66, P = 0.011; 1 g kg–1 ethanol, Gabra6100R/100R versus Gabra6100Q/100Q: F2,10 = 5.41, P = 0.026; 1.25 g kg–1 ethanol, Gabra6100R/100R versus Gabra6100Q/100Q: F2,13 = 135.6, P < 0.001. (e,f) Latency (e) and duration (f) of LORR was determined after intraperitoneal injection of 3 g kg–1 ethanol (n = 9 for each group) and did not differ in Gabra6100Q/100Q versus Gabra6100R/100R rats (P = 0.51 for LORR latency, P = 0.81 for LORR duration).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S


Ethanol sensitivity of other extrasynaptic GABARsExtrasynaptic GABARs composed of α4, β3 and δ subunits are likely to be found in neurons located in the thalamus3, dentate gyrus21, stria-tum14,15 and cerebral cortex15. As such GABARs are also sensitive to low ethanol concentrations9,10,48, we predict that ethanol-induced increases in tonic inhibition in these neuronal populations may contribute to sedative-hypnotic and anxiolytic effects and to the depressant actions of this drug on higher cognitive functions.

METHODSElectrophysiology. Standard methods were used for isolation, injection and recordings from X. laevis oocytes and for preparation of cRNA10. For brain slice experiments, 300 µm parasagittal slices of cerebellum were prepared from 24- to 42-day-old Sprague-Dawley rats using standard techniques20,21 with the exceptions that slicing solution consisted of (in mM) 250 sucrose, 26 NaHCO3, 10 glucose, 4 MgCl2, 3 myoinositol, 2.5 KCl, 2 sodium pyruvate, 1.25 NaH2PO4, 0.5 ascorbic acid, 0.1 CaCl2, and 0.001 D,L-APV. Slice storage and recording solutions were saturated with 95% O2/5% CO2 and consisted of (in mM) 119 NaCl, 26 NaHCO3, 11 glucose, 2.5 KCl, 2.5 CaCl2, 1.3 MgCl2, and 1 NaH2PO4; in addition, storage solution contained 0.001 D,L-APV. All procedures were in accordance with a protocol approved by the University of California at Los Angeles (UCLA) Chancellor’s Animal Research Committee. For voltage-clamp recordings (holding potential –70 mV, 20–23 °C), whole-cell pipettes contained (in mM) 140 CsCl, 10 HEPES, 1 EGTA, 4 magnesium ATP, 0.4 GTP, titrated to pH 7.3 with CsOH. Recording pipettes had a bath resistance of 6–10 MΩ.

Whole-cell data were filtered at 5 kHz and acquired at a sampling rate of 20 kHz. Analysis was conducted using customized routines written in IGOR Pro 4.0 (Wavemetrics). Tonic GABAR-mediated current was defined as the steady-state current blocked by 10 µM SR95531; its magnitude was calculated by plot-ting all-point histograms of relevant 30-s segments of data (as in Figs. 3,4). These data were fit to the Gaussian equation, constraining fits to values 2 bins more negative than the peak. This ensured that the tail of higher amplitude values (representing sIPSCs) did not influence the fit. The effects of 10 mM ethanol on tonic current (Fig. 4) were compared with changes in tonic current observed over otherwise identical sham perfusion periods.

Genotyping. After isolation of genomic DNA from ear snips, the exon coding for the α6-100 position was amplified with primers designed to be located in introns flanking the region of interest (to avoid amplification of mRNA). The PCR fragment was sequenced using standard fluorescent dye sequencing.

Behavior. Rats were housed with food and water ad libitum in a 12 h/12 h light/dark cycle. Homozygous male and female rats (Gabra6100R/100R and Gabra6100Q/100Q, > P55) were used for the rotarod (MedAssociates) and sleep time (LORR) studies. These rats were either obtained directly from a breeding colony at Charles River Laboratories or bred at UCLA. In the accelerating rotarod test, the speed of rotation increases at a constant rate from 4 to 40 r.p.m. over 5 min. All rats used in the rotarod tests were naive to ethanol and were used to test only one condition. Blood samples (20–50 µl) were taken from the tail and serum ethanol concentration was determined with an Analox enzymatic blood alcohol analyzer.

Statistics. Values are reported as mean ± s.e.m. unless noted otherwise. To evalu-ate the effect of ethanol dose and time in the rotarod experiments, we used a general linear model (GLM) with repeated measures. The Wilk's lambda multi-variate test was used to test the effect of the ethanol dose–time interaction. These statistical analyses were conducted using SPSS for Windows version 12. All other statistical comparisons were conducted using paired and unpaired Student's t-tests as appropriate.

ACKNOWLEDGMENTSWe thank C. Gundersen and the UCLA Anesthesiology Department for providing X. laevis oocytes, A. Taylor and D. Tio for help with blood alcohol analysis, and K. Olofsdotter-Otis for helpful comments on the manuscript. The work was supported by a Human Frontiers Science Program Long Term Fellowship to P.D.D. and by US National Institutes of Health grants AA015460 to H.J.H, NS41651 to T.S.O., and NS35985 and AA07680 to R.W.O.



1. Vallee, B.L. Alcohol in the western world. Sci. Am. 278, 80–85 (1998).2. 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).3. Kobayashi, T. et al. Ethanol opens G-protein-activated inwardly rectifying K+ channels.

Nat. Neurosci. 2, 1091–1097 (1999).4. Lewohl, J.M. et al. G-protein-coupled inwardly rectifying potassium channels are

targets of alcohol action. Nat. Neurosci. 2, 1084–1090 (1999).5. Hodge, C.W. et al. Supersensitivity to allosteric GABAA receptor modulators and alcohol

in mice lacking PKCε. Nat. Neurosci. 2, 997–1002 (1999).6. Lovinger, D.M., White, G. & Weight, F.F. Ethanol inhibits NMDA-activated ion current

in hippocampal neurons. Science 243, 1721–1724 (1989).7. Mihic, S.J. et al. Sites of alcohol and volatile anaesthetic action on GABAA and glycine

receptors. Nature 389, 385–389 (1997).8. Suzdak, P.D. et al. A selective imidazobenzodiazepine antagonist of ethanol in the

rat. Science 234, 1243–1247 (1986).9. Sundstrom-Poromaa, I. et al. Hormonally regulated α4β2δ GABAA receptors are a

target for alcohol. Nat. Neurosci. 5, 721–722 (2002).10. Wallner, M., Hanchar, H.J. & Olsen, R.W. Ethanol enhances α4β3δ and α6β3δ GABAA

receptors at low concentrations known to affect humans. Proc. Natl. Acad. Sci. USA 100, 15218–15223 (2003).

11. Choi, D.S. et al. The type 1 equilibrative nucleoside transporter regulates ethanol intoxication and preference. Nat. Neurosci. 7, 855–861 (2004).

12. Macdonald, R.L. & Olsen, R.W. GABAA receptor channels. Annu. Rev. Neurosci. 17, 569–602 (1994).

13. Mohler, H., Crestani, F. & Rudolph, U. GABAA-receptor subtypes: a new pharmacology. Curr. Opin. Pharmacol. 1, 22–25 (2001).

14. Pirker, S., Schwarzer, C., Wieselthaler, A., Sieghart, W. & Sperk, G. GABAA receptors: immunocytochemical distribution of 13 subunits in the adult rat brain. Neuroscience 101, 815–850 (2000).

15. Peng, Z. et al. GABAA receptor changes in δ subunit-deficient mice: altered expression of α4 and γ2 subunits in the forebrain. J. Comp. Neurol. 446, 179–197 (2002).

16. 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).

17. Campos, M.L., de Cabo, C., Wisden, W., Juiz, J.M. & Merlo, D. Expression of GABAA receptor subunits in rat brainstem auditory pathways: cochlear nuclei, superior olivary complex and nucleus of the lateral lemniscus. Neuroscience 102, 625–638 (2001).

18. Saxena, N.C. & Macdonald, R.L. Properties of putative cerebellar GABAA receptor isoforms. Mol. Pharmacol. 49, 567–579 (1996).

19. Nusser, Z. et al. Alterations in the expression of GABAA receptor subunits in cerebel-lar granule cells after the disruption of the α6 subunit gene. Eur. J. Neurosci. 11, 1685–1697 (1999).

20. Brickley, S.G., Revilla, V., Cull-Candy, S.G., Wisden, W. & Farrant, M. Adaptive regula-tion of neuronal excitability by a voltage-independent potassium conductance. Nature 409, 88–92 (2001).

21. 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 δ subunit-containing GABAA receptors. Proc. Natl. Acad. Sci. USA 100, 14439–14444 (2003).

22. Rossi, D.J. & Hamann, M. Spillover-mediated transmission at inhibitory synapses promoted by high affinity α6 subunit GABAA receptors and glomerular geometry. Neuron 20, 783–795 (1998).

23. Chadderton, P., Margrie, T.W. & Hausser, M. Integration of quanta in cerebellar granule cells during sensory processing. Nature 428, 856–860 (2004).

24. 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).

25. Saba, L. et al. The R100Q mutation of the GABAA α6 receptor subunit may contribute to voluntary aversion to ethanol in the sNP rat line. Brain Res. Mol. Brain Res. 87, 263–270 (2001).

26. Farrant, M. & Cull-Candy, S. GABA receptors, granule cells and genes. Nature 361, 302–303 (1993).

27. Radcliffe, R.A. et al. Behavioral characterization of alcohol-tolerant and alcohol-non-tolerant rat lines and an f(2) generation. Behav. Genet. 34, 453–463 (2004).

28. Wisden, W., Korpi, E.R. & Bahn, S. The cerebellum: a model system for studying GABAA receptor diversity. Neuropharmacology 35, 1139–1160 (1996).

29. Poltl, A., Hauer, B., Fuchs, K., Tretter, V. & Sieghart, W. Subunit composition and quantitative importance of GABAA receptor subtypes in the cerebellum of mouse and rat. J. Neurochem. 87, 1444–1455 (2003).

30. 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).

31. Kaneda, M., Farrant, M. & Cull-Candy, S.G. Whole-cell and single-channel currents activated by GABA and glycine in granule cells of the rat cerebellum. J. Physiol. (Lond.) 485, 419–435 (1995).

32. Overstreet, L.S. & Westbrook, G.L. Paradoxical reduction of synaptic inhibition by vigabatrin. J. Neurophysiol. 86, 596–603 (2001).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S


33. Nusser, Z. & Mody, I. Selective modulation of tonic and phasic inhibitions in dentate gyrus granule cells. J. Neurophysiol. 87, 2624–2628 (2002).

34. Semyanov, A., Walker, M.C., Kullmann, D.M. & Silver, R.A. Tonically active GABAA receptors: modulating gain and maintaining the tone. Trends Neurosci. 27, 262–269 (2004).

35. Semyanov, A., Walker, M.C. & Kullmann, D.M. GABA uptake regulates cortical excitability via cell type-specific tonic inhibition. Nat. Neurosci. 6, 484–490 (2003).

36. Jones, A. et al. Ligand-gated ion channel subunit partnerships: GABAA receptor α6 subunit gene inactivation inhibits δ subunit expression. J. Neurosci. 17, 1350–1362 (1997).

37. Mitchell, S.J. & Silver, R.A. Shunting inhibition modulates neuronal gain during synaptic excitation. Neuron 38, 433–445 (2003).

38. Korpi, E.R. et al. Cerebellar GABAA receptors in two rat lines selected for high and low sensitivity to moderate alcohol doses: pharmacological and genetic studies. Alcohol 9, 225–231 (1992).

39. Vekovischeva, O.Y., Haapalinna, A., Sarviharju, M., Honkanen, A. & Korpi, E.R. Cerebellar GABAA receptors and anxiolytic action of diazepam. Brain Res. Mol. Brain Res. 837, 184–187 (1999).

40. Korpi, E.R. et al. Cerebellar granule-cell-specific GABAA receptors attenuate benzo-diazepine-induced ataxia: evidence from α6-subunit-deficient mice. Eur. J. Neurosci. 11, 233–240 (1999).

41. Homanics, G.E. et al. Mice devoid of GABAA receptor β3 subunit have epilepsy, cleft palate, and hypersensitive behavior. Proc. Natl. Acad. Sci. USA 94, 4143–4148 (1997).

42. Homanics, G.E. et al. Gene knockout of the α6 subunit of the GABAA receptor: lack of effect on responses to ethanol, pentobarbital, and general anesthetics. Mol. Pharmacol. 51, 588–596 (1997).

43. Uusi-Oukari, M. et al. Long-range interactions in neuronal gene expression: evidence from gene targeting in the GABAA receptor β2-α6-α1-γ2 subunit gene cluster. Mol. Cell. Neurosci. 16, 34–41 (2000).

44. Rudolph, U. et al. Benzodiazepine actions mediated by specific GABAA receptor subtypes. Nature 401, 796–800 (1999).

45. McKernan, R.M. et al. Sedative but not anxiolytic properties of benzodiazepines are mediated by the GABAA receptor α1 subtype. Nat. Neurosci. 3, 587–592 (2000).

46. Low, K. et al. Molecular and neuronal substrate for the selective attenuation of anxiety. Science 290, 131–134 (2000).

47. Jurd, R. et al. General anesthetic actions in vivo strongly attenuated by a point muta-tion in the GABAA receptor β3 subunit. FASEB J. 17, 250–252 (2003).

48. 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).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Allodynia limits the usefulness of intraspinal neural stem cell grafts; directed differentiation improves outcomeChristoph P Hofstetter1, Niklas A V Holmström2, Johan A Lilja1, Petra Schweinhardt1,4, Jinxia Hao3, Christian Spenger1, Zsuzsanna Wiesenfeld-Hallin3, Shekar N Kurpad5, Jonas Frisén2 & Lars Olson1

Several studies have reported functional improvement after transplantation of neural stem cells into injured spinal cord. We now provide evidence that grafting of adult neural stem cells into a rat thoracic spinal cord weight-drop injury improves motor recovery but also causes aberrant axonal sprouting associated with allodynia-like hypersensitivity of forepaws. Transduction of neural stem cells with neurogenin-2 before transplantation suppressed astrocytic differentiation of engrafted cells and prevented graft-induced sprouting and allodynia. Transduction with neurogenin-2 also improved the positive effects of engrafted stem cells, including increased amounts of myelin in the injured area, recovery of hindlimb locomotor function and hindlimb sensory responses, as determined by functional magnetic resonance imaging. These findings show that stem cell transplantation into injured spinal cord can cause severe side effects and call for caution in the consideration of clinical trials.

Spontaneous recovery after spinal cord injury is hindered by the limited ability of the mammalian central nervous system to re- establish functional neural connections, remyelinate spared nerve fibers and replace lost cells1. Complete spinal cord injury leads to total and permanent sensorimotor loss and disruption of autonomic nervous system control caudal to the level of injury. Most spinal cord injury victims also develop chronic pain conditions2 that severely reduce quality of life3. Although most forms of spinal cord injuries in humans are incomplete and are characterized by variable degrees of tissue sparing across the lesion4, surviving axons are often compromised with respect to the propagation of electrical impulses because of loss of myelin5.

Grafting of embryonic stem cells6 and marrow stromal cells7, as well as neural stem cells (NSCs)8, into the injured spinal cord improves functio-nal recovery. Transplanted neural stem cells give rise almost exclusively to astrocytes and to only relatively few oligodendrocytes and occasional neurons8,9, suggesting that the beneficial effects are mediated either by trophic support provided by astrocytes10 or by remyelination of spared axons by graft-derived oligodendrocytes11. Neurotrophic factors have also been reported to cause severe side effects. Studies in rodents using several different trophic factors12,13 as well as a clinical trial using nerve growth factor for Alzheimer disease14 have reported neuropathic pain such as allodynia: that is, pain from stimuli that normally are not noxious.

To further explore the potential function of graft-derived astrocytes and oligodendrocytes in the treatment effects, we transplanted naive NSCs and NSCs in which astrocytic differentiation was suppressed by ectopic expression of neurogenin-2. Neurogenin-2, a member of

the basic helix-loop-helix family of transcription factors, is involved in the determination and differentiation of multiple neural lineages during development15. Neurogenins inhibit gliogenesis by sequestering CBP-Smad1 away from astrocyte differentiation genes and by blocking activation of STAT1/3 (ref. 16).

Here we report that although transplantation of naive NSCs improved motor function, it also caused aberrant host fiber sprouting associated with allodynia-like hypersensitivity of non-affected forepaws. Suppression of astroglial differentiation effectively reduced both s prouting and allo-dynia and allowed for further sensory and motor recovery.

RESULTSNeurogenin-2 modulates the fate of engrafted NSCsWe studied the differentiation of adult spinal cord–derived NSCs grafted to the site of a 1-week-old low-thoracic spinal cord weight-drop injury. The fate of naive NSCs was compared with that of NSCs that had been transduced to express neurogenin-2 (Ngn2-NSCs).

Two weeks after transplantation, NSC-derived cells were present mainly in the area of injury, close to the injection site. Engrafted cells had round cell bodies and some extended processes (Fig. 1a). Ngn2-NSCs were also located mainly near the injection site at this time. However, many cells had migrated beyond the astroglial barrier into the white matter, where they extended processes oriented along spared nerve fibers (Fig. 1a). The total number of engrafted 5-bromodeoxyuridine (BrdU)-immunoreactive cells at all four depositions was determined using stereological cell counting. A total of 79,496 ± 3,350 NSC- derived cells

1Department of Neuroscience, 2Department of Cell and Molecular Biology and 3Division of Clinical Neurophysiology, Karolinska Institutet, 17177 Stockholm, Sweden. 4Department of Human Anatomy, and Genetics, University of Oxford, OX1 1QX Oxford, UK. 5Department of Neurosurgery, Medical College of Wisconsin, Milwaukee, Wisconsin 53226, USA. Correspondence should be addressed to ([email protected]).


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

and 80,963 ± 17,660 Ngn2-NSC-derived cells were detected per rat 2 weeks after transplantation. Both types of grafts extended approximately 8 mm in a rostrocaudal direction and were centered on the injury center (Table 1). The majority (74%) of engrafted NSCs were glial fibrillary acidic protein (GFAP) immunoreactive, characteristic of astrocytes (Table 2). Ectopic expression of neurogenin-2, verified by immunohis-tochemical detection of the c-Myc tag (Supplementary Fig. 1 online), suppressed astroglial differentiation to 3% and allowed for oligoden-droglial and limited neuronal differentiation (Table 2). However, not all engrafted Ngn2-NSCs expressed neurogenin-2, given their in vitro transduction efficiency of approximately 60%. Consequently, the effect

on cell fate, although still very pronounced, was not as substantial when all transplanted BrdU-labeled cells were examined (Table 2). Retroviral expression was downregulated with time and only approximately 5% of all BrdU-positive cells still expressed green fluorescent protein (GFP) 2 weeks after transplantation9.

At 9 weeks after transplantation, gross examination showed that cysts had for-med around the injury center. Estimation of cyst size using the Cavalieri principle demonstrated no significant differen-ces between treatment groups (data not

shown). NSCs extended radiating processes and were preferenti-ally located in areas filled with debris and macrophages (Fig. 1a). Of 43,140 ± 14,326 BrdU- positive NSCs detected per rat, 80% had become astrocytes and 17% had differentiated toward an oligo-dendroglial fate (Table 2). At the same time, Ngn2-NSCs extended processes of varying thickness forming a network in white mat-ter tracts (Fig. 1a,b). Of all engrafted BrdU-positive Ngn2-NSCs (40,560 ± 5,067), significantly fewer cells showed GFAP immuno-reactivity (32%) and significantly more cells showed RIP immu-noreactivity (31%) than did naive NSCs (P < 0.01 and P < 0.05, respectively, Mann-Whitney U-test; Table 2). Most of the RIP-

Figure 1 Histological characterization of engrafted cells. (a) NSCs show a rather immature cell morphology 2 weeks after transplantation and acquire stellate cell morphology at 9 weeks. At 2 weeks after injection, Ngn2-NSCs extend thickened and less structured processes that develop into either long, fine arborizing processes or short, thick ones at 9 weeks. (b) Engrafted Ngn2-NSCs (green) are intermingled between spared neurofilament-immunoreactive nerve fibers (red). The asterisk indicates a cyst. (c) Examples of astroglial (GFAP) and oligodendroglial (RIP) differentiation of engrafted cells. Some engrafted cells show NG2 immunoreactivity in their processes, which is characteristic of oligodendrocyte precursor cells. More mature oligodendroglial differentiation is indicated by CNP immunoreactivity. Other engrafted cells express the early neuronal markers NeuN and Tuj1. (d) Osmium tetroxide staining of myelin. Photomicrographs of coronal spinal cord sections at the level of the lesion center of rats treated with vehicle, NSCs or Ngn2-NSCs. Bottom row, high-magnification images of the regions outlined above with red and green rectangles show myelin profiles in superficial and deeper layers of the injured cord, respectively. (e) Quantification of myelin at the level of the injury center as well as 1 mm above and below the injury center. *P < 0.05; **P < 0.01 (ANOVA followed by a Fisher’s post-hoc test). Data represent mean ± s.e.m. (f) A BrdU-immunoreactive Ngn2-NSC–derived cell shows MBP immunoreactivity and forms myelin sheaths 9 weeks after transplantation. Scale bars, 100 µm (a,b), 20 µm (c and high magnification in d), 200 µm (top row of d) and 5 µm (f).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

immunoreactive Ngn2-NSCs had fine arborizing processes that expressed the chondroitin sulfate proteoglycan NG2 in their distal parts (Fig. 1c). NG2 immunoreactivity suggests that these cells were oligodendrocyte precursor cells. Other RIP-expressing cells had thicker processes and were immunoreactive for the myelin-specific protein cyclic nucleotide 3-phosphohydrolase (CNP; Fig. 1c), indicating that they had acquired a more mature oligodendroglial phenotype. At 9 weeks, 3% of BrdU-positive Ngn2-NSCs had neuron-specific nuclear protein (NeuN) immunoreactivity. These cells extended a main process with a length of several cell body dia-meters. The processes were βIII- tubulin immunoreactive, indicating that these engrafted Ngn2-NSCs had differentiated toward an early neuronal phenotype (Fig. 1c).

Ngn2-NSCs increase myelinization at the injury siteOsmium tetroxide myelin staining showed a rim of myelinated axons surrounding the lesion cavity–gray matter remnants in the injured area, except at the center of the injured area, where such a dorsome-dial rim of myelin was lacking (Fig. 1d). Axons with well developed myelin sheaths were present in the spared rim beneath the pial sur-face, whereas areas closer to the cavity contained axons with thin myelin sheaths. The total amount of white matter was quantified at the injury center as well as 1 mm rostral and 1 mm caudal to the injury center. Grafting of Ngn2-NSCs led to a significantly greater amount of white matter at the injury center than that of rats given vehicle alone (Fig. 1e). In the caudal aspect of the injury zone, Ngn2-NSCs led to significantly more white matter than did either NSC or vehicle treatment. We identified engrafted cells as one source of myelin basic protein (MBP)–positive cells. In good accordance with the differentiation pattern observed with the early oligodendroglial marker RIP, Ngn2-NSCs frequently gave rise to MBP-immunoreac-tive cells that in turn formed myelin sheaths (Fig. 1f). Notably, MBP was exclusively detected in engrafted (BrdU-positive) cells that had downregulated expression of neurogenin-2.

Transplantation improves functional outcomeWe assessed locomotor recovery using Basso, Beattie and Bresnahan (BBB) scoring of open field walking (Fig. 2a). Three weeks after

transplantation, NSC-engrafted rats had significantly higher BBB scores than did vehicle-treated rats. At 5 weeks after transplantation, Ngn2-NSC–engrafted rats also performed significantly better than vehicle-treated rats. At 9 weeks after transplantation, motor recovery of Ngn2-NSC–engrafted rats significantly exceeded that of vehicle-treated and of NSC-engrafted rats (Fig. 2a). We found that 62% of the rats that received Ngn2-NSCs showed frequent toe clearance during limb advancement, compared with 35% of NSC-engrafted rats and 7% of vehicle-treated rats. Recovery of hindpaw locomotor function in individual rats was significantly correlated with the amount of white matter at the injury center (P < 0.001; correlation coefficient, 0.73; χ2 statistics).

Assessment of sensorimotor integration by gridway testing showed that rats grafted with either Ngn2-NSCs or NSCs commit-ted significantly fewer foot misplacements than did vehicle-treated rats from 5 weeks after transplantation onward (Fig. 2b). At 9 weeks after transplantation, rats that had received Ngn2-NSCs performed significantly better than rats in the two other groups. Improved per-formance in the gridway task of individual rats was significantly correlated to the amount of myelin at the injury center (P < 0.05; correlation coefficient, –0.53; χ2 statistics) as well as to the number of RIP-immunoreactive graft-derived cells (P < 0.05; correlation coefficient, –0.65; χ2 statistics). Sensory function of the hindpaws was examined on a hotplate, which elicited licking of hindpaws in all intact rats with a latency of 22 ± 3 s (Fig. 2c). Grafting of Ngn2-NSCs increased the number of rats that responded to the thermal stimulation by licking their hindpaws. It also significantly decreased the latency of response to the hot stimulus compared with that of vehicle- and NSC-treated rats (52 ± 2 s, 59 ± 1 s and 59 ± 1 s, respec-tively; P < 0.01, ANOVA followed by a Fisher’s post-hoc test).

Table 1 Number of grafted cells, volume and location of engrafted area

Time after grafting

Number of

BrdU+ cells

Graft volume(mm3)

Total longitudinalextent (mm)

Extent rostral toinjury center (mm)

2 weeks NSCs (n = 3) 79,496 ± 3,350 8.1 ± 0.5 7.9 ± 0.1 3.6 ± 0.3

Ngn2-NSCs (n = 3) 80,963 ± 17,660 8.6 ± 1.0 8.3 ± 0.7 4.2 ± 0.3

9 weeks NSCs (n = 6) 43,140 ± 14,326 6.0 ± 0.3 7.1 ± 0.2 2.6 ± 0.5

Ngn2-NSCs (n = 6) 40,560 ± 5,067 6.3 ± 0.4 6.6 ± 0.5 2.5 ± 0.3

n = number of rats. Data represent mean ± s.e.m.

Figure 2 Functional recovery of rats treated with vehicle, NSCs or Ngn2-NSCs. (a) Open field walking as determined by the BBB rating scale. Upward arrow indicates the impact injury and Tx indicates time of transplantation. Hindlimb function was assigned scores from 0 to 21 (flaccid paralysis to normal gait). (b) Sensorimotor integration, as assessed by a gridway task. Dashed line indicates the performance of intact rats. (c) Hotplate testing shows sensory function of hindlimbs. Black bars indicate the percentage of animals licking their hindpaws during testing and gray bars indicate the percentage of animals not showing a response. BBB and gridway scores were compared by ANOVA followed by a Fisher’s post-hoc test. Ngn2-NSC (n = 29) or NSC (n = 20) compared to vehicle (n = 28): *P < 0.05, **P < 0.01 and ***P < 0.001. Ngn2-NSC compared to NSC: #P < 0.05.

0

25

50

75

100

NSC

Ngn2-

NSC

Vehicl

eIn

tact

0

5

10

15

20

2 3 4 5 6 7 8 9

* * ****

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

*

#

#

0

2

4

6

8

10

12

14

16

TX 1 2 3 4 5 6 7 8 9

****

*****

*****

***

***

#

VehicleNSCNgn2-NSC

Weeks after transplantation Weeks after transplantation

Loco

mot

or p

erfo

rman

ce (

BB

B r

atin

g sc

ale)

Foo

t mis

plac

emen

ts

Res

pond

ers

(%)

a

b c

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Ngn2-NSCs partially restore hindpaw fMRI responsesWe recorded cerebral blood oxygen level–dependent (BOLD) signal changes in response to hindpaw stimulation in intact rats and in spinal cord–injured rats treated with vehicle, NSCs or Ngn2-NSCs. Unilateral electric hindpaw stimulation of intact rats led to activation of the contralateral pri-mary somatosensory cortex17. Additionally, ipsilateral activation was occasionally observed (Fig. 3a). At 9 weeks after injury, vehicle-treated rats showed minimal bilateral cortical BOLD signals18(Fig. 3b). Transplantation of NSCs did not result in substantial recov-ery of the cortical hindpaw representation (Fig. 3c). Conversely, grafting of Ngn2-NSCs led to significant recov-ery of contralateral cortical BOLD signals in response to hindlimb stimulation as compared to vehicle-treated and NSC-grafted rats (P < 0.001and P < 0.01, respectively, ANOVA followed by a Fisher’s post-hoc test; Fig. 3d,e). Grafting of Ngn2-NSCs partially restored activation of the contralateral somatosensory cortex. In addition, activation was detected bilaterally in cortical areas adjacent to the primary somato-sensory cortex. These areas were not activated in intact rats.

Naive NSCs cause allodynia-like responses in forepawsThe withdrawal threshold for mechanical stimulation of the forepaws was not significantly altered by spinal cord injury (Fig. 4a). However, an

Contralateral cortical activation

Ipsilateral cortical activation

0

5

10

15

20

Cor

tical

act

ivat

ion

(vox

els/

anim

al)

*

*****

Ngn2-NSCsNSCsVehicleIntact

Ngn2-NSCs

NSCs

Vehicle

Intact

a

b

c

d

e

increased sensitivity to cold stimulation of the forepaws was observed in rats subjected to weight-drop injury compared with that of intact rats (Fig. 4b). Rats with spinal cord injury that were transplanted with naive NSCs showed a considerably decreased withdrawal threshold in response to mechanical forepaw stimulation compared with that of vehicle-treated rats (Fig. 4a). The withdrawal threshold of individual rats correlated significantly with the number of GFAP-immunoreactive graft-derived cells (P < 0.01; correlation coefficient, –0.83; χ2 statis-tics). Additionally, NSC-engrafted rats showed significantly increased pain-like responses to cold stimulation compared with vehicle-treated rats (Fig. 4b). Directed differentiation of NSCs with neuro-genin-2 significantly alleviated the allodynia-like responses observed

in NSC-treated rats and led to responses similar to those of vehicle-treated rats (Fig. 4a,b). Moreover, rats engrafted with naive NSCs, but not rats given Ngn2-NSCs, showed significantly enhanced pain-like responses to heat stimulation of the forepaw and reduced response latency in the hotplate test (Fig. 4c,d).

Transplantation of naive NSCs causes axonal sproutingWe quantified calcitonin gene–related peptide (CGRP)–immunoreactive sensory axons and biotinylated dextran amine (BDA)–labeled corticospinal tract fibers using stereology in 6-mm-long spinal cord segments above the injury. Transplantation of naive NSCs gave rise to sprouting of thin CGRP-immunoreac-tive fibers into Rexed’s lamina III (Fig. 5a). Stereological quantification showed twice as many CGRP-positive nerve fibers in lamina III in rats that had received naive NSCs as in rats treated with either vehicle or Ngn2-NSCs (Fig. 5b). The amount of CGRP-posi-tive fibers in individual rats was positively correlated with the number of graft-derived GFAP-immunoreactive cells (P < 0.01; correlation coefficient, 0.78; χ2 statistics). Moreover, there was a reciprocal relationship between the withdrawal threshold to von Frey hair stimulation (increased hypersensitivity) of individual rats and the amount of CGRP-immunoreactive fibers (P < 0.01; correlation

Table 2 Histological characterization of engrafted neural stem cells

Time after grafting % GFP+% of all GFP+ cells % of all BrdU+ cells

GFAP RIP NeuN GFAP RIP NeuN

2 weeks NSCs (n = 3) ∼5

N = 269

74 ± 9

N = 204

17 ± 5

N = 224

4 ± 2

N = 237

69 ± 4

N = 221

13 ± 2

N = 241

1 ± 1

N = 242

Ngn2-NSCs (n = 3) ∼5

N = 214

3 ± 1

N = 201

32 ± 6

N = 217

37 ± 6

N = 293

16 ± 4

N = 221

37 ± 5

N = 209

11 ± 1

N = 244

9 weeks NSCs (n = 6) ∼1.5

N = 204

84 ± 4

N = 235

18 ± 3

N = 203

4 ± 4

N = 203

80 ± 3

N = 224

17 ± 2

N = 273

1 ± 1

N = 231

Ngn2-NSCs (n = 6) ∼1.5

N = 213

6 ± 2

N = 200

32 ± 2

N = 261

44 ± 4

N = 228

32 ±3

N = 205

31 ± 4

N = 249

3 ± 2

N = 219

n = number of rats; N = number of cells analyzed. Data represent mean ± s.e.m.

Figure 3 Functional MRI during stimulation of hindpaws. (a–d) Sensory representation of hindpaws in intact rats (a) and rats treated with vehicle (b), NSCs (c) or Ngn2-NSCs (d). Bars (bottom right) indicate the T value of the obtained functional MRI signal, ranging from 0 (dark red) to 8 (white; a) or to 4.5 (b–d). (e) Ipsilateral and contralateral cortical activation in response to hindpaw stimulation in intact rats (n = 7) and rats treated with vehicle (n = 11) or engraftment of NSCs (n = 8) or Ngn2-NSCs (n = 11). The amount of cortical activation was compared by ANOVA followed by Fisher’s post-hoc test: *P < 0.05, **P < 0.01 and ***P < 0.001. Data represent mean ± s.e.m.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

coefficient, –0.72; χ2 statistics; Fig. 5c). Immediately rostral to the injury, the corticospinal tract was abolished or considerably displaced by a cyst, with some fibers traveling either through the gray matter of the dorsal horn or along gray–white matter interfaces. At 9 weeks after injury, there was still complete interruption of all BDA-labeled corticospinal tract fibers at the site of injury in all groups, and thus no fibers were seen caudal to the lesion. Cranial to the injury site, engrafted cells were intermingled with the proximal stumps of the corticospinal tract (Fig. 5d). Transplantation of naive NSCs caused excessive sprouting of corticospinal tract fibers within the dorsal horns rostral to the injury (Fig. 5e). Stereological fiber quantification detected significantly more corticospinal tract fibers within super-ficial as well as deep dorsal horn layers of NSC-engrafted rats than in vehicle-treated or Ngn2-NSC–engrafted rats (Fig. 5f,g). Although there was excessive sprouting at the rostral edge of the lesion site, there was no evidence of rostrocaudal regeneration and consequently no sprouting of corticospinal tract fibers caudal to the lesion (Fig. 5h). There were no statistically significant differences with regard to the most distant extent of the corticospinal tract for rats given vehicle, NSCs or Ngn2-NSCs (distance of corticospinal tract fibers to injury center, 1.7 ± 0.3 mm, 1.1 ± 0.2 mm and 1.6 ± 0.1 mm, respectively). The amount of corticospinal tract sprouting was not correlated to the degree of recovery of motor function in NSC-treated rats.

DISCUSSIONHere we have shown that treatment of spinal cord injury with naive NSCs improves recovery of motor function but causes allodynia in unaffected forepaws. Transduction of NSCs with neurogenin-2 before transplantation depressed astroglial differentiation and increased oligodendroglial differentiation of engrafted cells. We have provided evidence that this treatment alleviates allodynia and allows for better sensory and motor recovery of the hind limbs than is seen with naive NSCs.

Promotion of host axonal regeneration is regarded as a key component of a successful treatment strategy for spinal cord injury. Delivery of growth factors can enhance and direct such outgrowth of severed nerve fibers19. Trophic support by engrafted fetal NSCs induces local sprouting of different nerve fiber systems10. Using a new stereological method, we detected excessive sprouting of the corticos-pinal tract and of CGRP-immunoreactive fibers after transplantation of naive adult NSCs. Despite prominent local sprouting of interrup-ted corticospinal tract fibers, there was no regeneration beyond the site of injury within the time interval we investigated. Thus, trophic support does not seem sufficient for fiber tract regeneration. Instead, overcoming the growth-inhibitory properties of the injured spinal cord milieu20 and providing a permissive, growth-promoting environment for advancing axons is necessary21. Another strategy for overcoming growth inhibition is to increase the intrinsic capability of neurons to extend fibers across inhibitory substrates by increasing intracellular cyclic adenosine monophosphate levels22.

The formation of new intraspinal circuits relaying cortical input from the injured, sprouting corticospinal tract above the injury to centers below injury through other remaining descending pathways has been brought forward as an explanation for spontaneous functional recovery after spinal cord injury23. Excessive local sprouting in our present study might have facilitated the formation of such pathways, although we could not detect a correlation between the amount of local sprouting and motor recovery. However, the number of graft-derived astrocytes was correlated to both sprouting of CGRP-immunoreactive fibers and the development of allodynia-like responses. Astrocytes constitute a known source of multiple neurotrophic factors, especially in areas of central nervous system trauma24. CGRP-immunoreactive dorsal root neurons express the nerve growth factor receptor TrkA25 and increase both neuropeptide expression and sprouting26 in response to increased nerve growth factor levels. CGRP is expressed in small- to medium-diameter dorsal root ganglion neurons27, which are involved in the transmission of pain, temperature, and noxious and non-noxious mechanical stimuli28. Notably, blocking of CGRP attenuates allodynia in the forelimbs in spinal cord models of chronic pain29.

NSCs differentiate almost exclusively into astrocytes after transplanta-tion into the injured spinal cord. Here we have shown that transduction of NSCs with neurogenin-2 enhanced oligodendroglial differentiation. The low grade of neuronal differentiation was unexpected, as neuro-genins, independent of their inhibition of an astrocytic fate, are potent stimulators of neuronal differentiation in vitro30,31. However, neuro-genin-2 does not unconditionally commit NSCs to a neuronal fate; environmental cues also influence the fate of neurogenin-2-expressing cells32. Our data suggest that a non-neurogenic environment in the spi-nal cord largely overrides the neurogenic effect of neurogenin, leading instead to a notably increased number of oligodendroglial cells.

In an incomplete spinal cord injury, loss of myelin impairs the con-duction properties of spared axons5. Some degree of spontaneous

Figure 4 Rats engrafted with NSCs develop allodynia-like responses in unaffected forepaws. (a,b) NSC-treated rats show a significantly enhanced pain-like response to mechanical (a) and cold (b) stimulation compared with that of both vehicle-treated and Ngn2-NSC–engrafted rats. (c,d) NSC-engrafted rats also show significantly shortened withdrawal latency (c) and significantly decreased time to forepaw licking in the hotplate test (d) compared with vehicle-treated rats. n = 20 for intact rats; n = 8 for other groups. Responses to cold stimulation were tested by a Kruskal-Wallis test followed by a Dunn’s multiple comparison test; the remaining data were compared by ANOVA followed by a Fisher’s post-hoc test: *P < 0.05; **P < 0.01; ***P < 0.001. Data represent mean ± s.e.m.

0

20

40

60

tcatnIelciheV

sCSN-2ngN

sCSN

)s( emit n oi tcae

R

0

1

2

3

4

5

6

tcatnIelciheV

sCSN-2ngN

sCSN

With

draw

al th

resh

old

(s)

0

0.5

1.0

1.5

2.0

2.5

tcatnIelciheV

sCSN-2ngN

sCSNC

old

scor

e (1

– 3)

0

10

20

30

tcatnIelciheV

sCSN-2ngN

sCSN

With

draw

al th

resh

old

(g)

***** **

*

*

*

a b

c d

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

remyelination is seen and may be the result of invading Schwann cells33 and/or oligodendrocyte precursor cells34. This remyelination is often abnormal, with some areas remaining partially demyelinated and other areas having thin myelin sheaths35. Consequently, incom-plete spinal cord injury can be associated with abnormal conduction

properties36. In the lesion periphery, oligo-dendrocyte precursor cells proliferate and are able to remyelinate injured axons34. However, the center of the injury is notably depleted of both oligodendrocyte precursor cells and oligodendrocytes, and repopulation is ineffi-cient because of the limited migratory capa-city of these cells37. The increased amount of myelin at the injury center detected in rats engrafted with Ngn2-NSCs may be attribu-ted to remyelination and/or the prevention of dysmyelination. The latter possibility is supported by the mostly immature status of oligodendroglial cells derived from engraf-ted Ngn2-NSCs as well as by the relatively rapid time course of functional improvement after grafting in our study. Similar findings consistent with the prevention of dysmyeli-nation have been observed after grafting of oligodendrocyte precursors into a contusion injury38. However, we also obtained evidence

of remyelination by engrafted cells. Some engrafted cells gave rise to MBP-immunoreactive cells forming myelin sheaths. Neural stem cells differentiate into mature myelin-producing cells after transplantation into demyelinated rat spinal cord11. Replacement of oligodendrocytes, which restores signal transmission39, is thus likely to have contribu-

Figure 5 Grafting of NSCs induces sprouting of nerve fibers rostral to the injury. (a) Dense CGRP-immunoreactive innervation of lamina I and the outer part of lamina II. Grafting of NSCs leads to more CGRP-positive fibers in lamina III than does treatment with vehicle or grafting of Ngn2-NSCs. (b) Stereological quantification of CGRP-positive fibers in lamina III. (c) The amount of CGRP-positive fibers in lamina III correlates with the withdrawal threshold of rats that developed allodynia-like behavior of the forepaws (P < 0.01; correlation coefficient, –0.72; χ2 statistics). (d) BDA-labeled corticospinal tract fibers (red) at the rostral edge of the lesion, which contains engrafted BrdU-immunoreactive NSCs (green). Retraction bulbs are formed by proximal stumps of the corticospinal tract fibers. (e) Coronal spinal cord sections show extensive sprouting of corticospinal tract fibers into the dorsal horns of rats engrafted with NSCs. Few BDA-labeled fibers sprout beyond Rexed’s lamina III into more superficial laminae. (f,g) Transplantation of NSCs increases corticospinal tract sprouting into the superficial and deep dorsal horn layers. (h) Rostrocaudal distribution of corticospinal tract fibers rostral to the injury zone. Excessive local sprouting of the corticospinal tract is not associated with improved regeneration of this tract beyond the injury center (IC). , NSCs; , Ngn2-NSCs; , vehicle. n ≥ 6 per group. Fiber length is compared by ANOVA followed by a Fisher’s post-hoc test: ** P < 0.01; and *** P < 0.001 (for bracketed comparisons). Data represent mean ± s.e.m. Bottom row (a,e), higher magnification of areas outlined in images above. Scale bars, 50 µm (top row of a), 100 µm (d and top row of e) and 10 µm in high magnification (a,e).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

ted to the improved behavioral recovery seen after transplantation of NSCs and Ngn2-NSCs. This is further supported by the positive cor-relation between the number of graft-derived oligodendrocytes and the performance on the gridway task of individual rats. The additional behavioral improvement obtained with Ngn2-NSCs compared with that obtained with naive NSCs was a delayed event, compatible with the time needed for newly formed oligodendrocytes to remyelinate axons. Functional MRI recordings demonstrated significant recovery of BOLD signals in the contralateral somatosensory cortex in response to hindpaw stimulation in rats engrafted with Ngn2-NSCs. Ngn2-NSCs partially restored a topographically appropriate cortical activation in response to hindpaw stimulation, a phenomenon that is less evident in animals that recover spontaneously from mild spinal cord injury18. The topographically appropriate recovery of hindpaw representation seen after grafting of Ngn2-NSCs is compatible with recovery due to remyelination of existing fiber tracts.

We conclude that the grafting of naive NSCs into injured spinal cord improves motor function but also causes aberrant fiber sprouting associated with allodynia of unaffected forepaws. The suppression of astroglial differentiation and enhancement of oligodendroglial diffe-rentiation by neurogenin-2 prevents adverse side effects and allows for robust recovery of motor and sensory function. Controlled differenti-ation of transplanted stem cells may be required to avoid serious side effects and to achieve optimal functional improvement. These findings should facilitate the development of safe stem cell–based treatment strategies for patients suffering from spinal cord injury.

METHODSPreparation of NSCs. Spinal cords of 134 adult female Sprague-Dawley rats (B&K Universal) were used to prepare primary neural stem cells40. After removal of the meninges, spinal cord tissue was dissociated at 37 °C for 30 min in a solution of 0.7 mg hyaluronic acid, 200 U DNase and 1.33 mg trypsin per ml. After filtration (70-µm mesh) and centrifugation (200g for 5 min), cells were resuspended in 0.5 M sucrose in 0.5× Hank’s balanced salt solution. After a second centrifugation (750g for 10 min), the cell pellet was resuspended in neurosphere culture medium based on DMEM F12 (Life Technologies) with Glutamax and supplemented with 20 ng extracellular growth factor, 20 ng basic fibroblast growth factor, 20 µl B27, 100 U penicillin and 100 µg streptomycin per ml. Neurospheres that had formed after 7 d were dissociated in 1.33 mg/ml of trypsin and single cells were further cultured for 3 d into secondary neuro-spheres in 50% neurosphere-conditioned medium and 50% fresh medium. In some experiments, the medium was supplemented with BrdU (10 µM; Sigma) 48 h before collection to facilitate detection of engrafted cells.

Viral transduction of NSCs. Vesicular stomatitis virus G protein–pseudotyped retroviral particles were prepared with two retroviral constructs41. One construct encoded c-Myc-tagged neurogenin-2 followed by an internal ribosomal entry site and enhanced GFP. The control construct was identical except that it did not contain the neurogenin-2 cDNA41. Secondary neurospheres were infected 24 h after passage at a multiplicity of infection of 3.5. The efficiency of the transduc-tion protocol is approximately 60% (ref. 41). Cells were propagated in complete neurosphere culture medium for another 48 h and then were resuspended to a concentration of approximately 10,000 viable cells per microliter of DMEM-F12, as determined by trypan blue dye exclusion.

Impact injury, transplantation and tracing. A total of 104 female Sprague-Dawley rats (250 g) were subjected to a contusion injury of the spinal cord. In rats sedated by halothane anesthesia, the spinal cord was exposed by a lami-nectomy of T8–9 and was subjected to impact by a weight dropped from a height of 12.5 mm. One week later, rats were randomly assigned to groups receiving vehicle (n = 37), NSCs (n = 29) or Ngn2-NSCs (n = 38). Four injec-tions were made at 1 mm cranial to, caudal to and left and right of the lesion. At each site, 2.5 µl of NSC suspension or vehicle was infused at 0.5 µl/min (UltraMicroPump-II, World Precision Instruments). Thus, a total of 100,000 NSCs collected from two donors per recipient was delivered to the spinal cord.

For anterograde labeling of the corticospinal tract, five injections of 0.5 µl of a 10% BDA solution (10,000 molecular weight; Sigma) were made per senso-rimotor cortex 4 weeks before rats were killed42. Experiments were approved by the Animal Research Committee of Stockholm.

Behavioral testing. Open field walking was evaluated according to the BBB rat-ing scale43. In a gridway task, rats crossed a 1.2-m gridway three times, and the number of hindlimb misplacements was counted44. For hotplate testing, rats were placed on a hotplate and the latency to licking of paws was measured. Non-responders were removed from the plate after 60 s (ref. 45). Mechanical sensitivity was assessed with von Frey hairs (Stoelting), which tests the withdrawal threshold to graded mechanical touch and/or pressure. Von Frey hairs were applied in ascending order on the palmar surface at a frequency of 0.5 per second. The lowest force leading to at least three withdrawals of five trials was defined as the mechanical threshold. The response rate to cold was tested by the application of ethyl chloride spray to the palmar surface. The response ranged from 1 (no observable response) to 2 (brief withdrawal and licking) to 3 (vocalization, pro-longed withdrawal, licking and aversive reactions). Heat response was tested with a modified Hargreaves method46. Radiant heat was applied to the palmar surface and the latency to withdrawal of the stimulated paw was measured. Behavior was assessed by two observers ‘blinded’ to experimental conditions.

Functional MRI. Functional MRI was done with the BOLD technique. Five sequential MRI slices 1 mm in thickness were acquired and centered 5 mm posterior to the rhinal fissure17. A 540-image ‘package’ was acquired with a block design with ten alternating stimulus and rest periods. Stimulation pulses (1 mA and 3 Hz) were applied to the hindpaws of all tested rats. Data sets were evalu-ated with SPM99. Voxels that showed a significant increase (P < 0.05) in signal intensity during stimulus compared with rest periods were color coded. Active brain areas were defined as positive voxel clusters consisting of a minimum of two clustered voxels and were located with the aid of structural magnetic reso-nance images at the same levels.

Histology. Indirect immunocytochemistry was done with antisera raised in goat to GFP (Rockland) or in rabbit to GFP (Molecular Probes), GFAP (Sigma), NG2 (Chemicon), MBP (Chemicon), CGRP (a gift from Tomas Hökfelt, Department of Neuroscience, Karolinska Institutet, Stockholm, Sweden) and neurofilament (Dahl), and mouse monoclonal antibodies to BrdU (Dako), c-Myc (9E10, DSHB), NeuN (Chemicon), βIII-tubulin (BabCO), RIP (Developmental Studies Hybridoma Bank) and CNP (Chemicon) were used. Secondary antisera were conjugated with carbocyanine, indocarbocyanine and indodicarbocyanine (Jackson ImmunoResearch) for analysis with confocal microscopy (LSM 510 META, Zeiss). For detection of BrdU, slides were pretreated with 2 mM HCl at 37 °C. For immunohistochemical characterization, ‘number-weighted’ samples of engrafted cells were analyzed with confocal microscopy. Z-series of a total of 6,385 individual cells were assessed for colocalization at high magnification (100×, 1.45–numerical aperture, oil-immersion objective). Bleedthrough was minimized by using multitrack scanning. For stereology, primary antibodies to BrdU and CGRP were detected with biotinylated secondary antibodies (Vector). Biotinylated secondary antibodies and BDA were visualized with an avidin–biotin–horseradish peroxidase complex (Vectastain Elite; Vector Laboratories). For myelin staining, histological specimens were incubated for 1 h at 4 °C in a 1% osmium tetroxide solution. Slides were then dehydrated and mounted with Pertex (CellPath).

Stereology. BrdU-positive engrafted cells as well as CGRP- and BDA-positive nerve fibers were quantified using unbiased stereology. A three-dimensional probe, the optical dissector47, was used to count cells and a two-dimensional probe, the surface of a sphere48, was used to quantify nerve fiber length. Stereological estimates were done in a uniform and systematic sample, consti-tuting a known fraction of the region containing either engrafted cells or labeled nerve fibers47. Stereological probes were superimposed on the field of view with appropriate software (Stereologer; SPA). The area of myelination was measured with a stereological point grid, where every point was associated with 20,000 µm2. The average myelin area of two slides separated by 0.3 mm was acquired in the injury center as well as 1 mm above and below the injury center.


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

ACKNOWLEDGMENTSWe thank E. Lindqvist, K. Pernold and K. Lundströmer for outstanding technical assistance. This study was supported by grants from AMF, the Swedish Research Council, the Karolinska Institute, the Swedish Cancer Society, the Tobias Foundation, the Göran Gustafsson Foundation, the US Public Health Service, the European Union and the Foundation for Strategic Research. S.K. was supported by a Van Wagenen fellowship from the AANS.



1. Cajal, S.R.Y. Degeneration and Regeneration of the Nervous System. (Oxford University Press, London, 1928).

2. Christensen, M.D. & Hulsebosch, C.E. Chronic central pain after spinal cord injury. J. Neurotrauma 14, 517–537 (1997).

3. Cairns, D.M., Adkins, R.H. & Scott, M.D. Pain and depression in acute traumatic spinal cord injury: origins of chronic problematic pain? Arch. Phys. Med. Rehabil. 77, 329–35 (1996).

4. Bunge, R.P., Puckett, W.R. & Hiester, E.D. Observations on the pathology of several types of human spinal cord injury, with emphasis on the astrocyte response to pen-etrating injuries. Adv. Neurol. 72, 305–315 (1997).

5. Waxman, S.G. Demyelination in spinal cord injury. J. Neurol. Sci. 91, 1–14 (1989).

6. McDonald, J.W. et al. Transplanted embryonic stem cells survive, differentiate and promote recovery in injured rat spinal cord. Nat. Med. 5, 1410–1412 (1999).

7. Hofstetter, C.P. et al. Marrow stromal cells form guiding strands in the injured spinal cord and promote recovery. Proc. Natl. Acad. Sci. USA 99, 2199–2204 (2002).

8. Ogawa, Y. et al. Transplantation of in vitro-expanded fetal neural progenitor cells results in neurogenesis and functional recovery after spinal cord contusion injury in adult rats. J. Neurosci. Res. 69, 925–933 (2002).

9. Vroemen, M., Aigner, L., Winkler, J. & Weidner, N. Adult neural progenitor cell grafts survive after acute spinal cord injury and integrate along axonal pathways. Eur. J. Neurosci. 18, 743–751 (2003).

10. Lu, P., Jones, L.L., Snyder, E.Y. & Tuszynski, M.H. Neural stem cells constitutively secrete neurotrophic factors and promote extensive host axonal growth after spinal cord injury. Exp. Neurol. 181, 115–129 (2003).

11. Akiyama, Y. et al. Transplantation of clonal neural precursor cells derived from adult human brain establishes functional peripheral myelin in the rat spinal cord. Exp. Neurol. 167, 27–39 (2001).

12. Hao, J., Ebendal, T., Xu, X., Wiesenfeld-Hallin, Z. & Eriksdotter Jonhagen, M. Intracerebroventricular infusion of nerve growth factor induces pain-like response in rats. Neurosci. Lett. 286, 208–212 (2000).

13. Jubran, M. & Widenfalk, J. Repair of peripheral nerve transections with fibrin sealant containing neurotrophic factors. Exp. Neurol. 181, 204–212 (2003).

14. Eriksdotter Jonhagen, M. et al. Intracerebroventricular infusion of nerve growth factor in three patients with Alzheimer’s disease. Dement. Geriatr. Cogn. Disord. 9, 246–257 (1998).

15. Nieto, M., Schuurmans, C., Britz, O. & Guillemot, F. Neural bHLH genes control the neuronal versus glial fate decision in cortical progenitors. Neuron 29, 401–413 (2001).

16. Sun, Y. et al. Neurogenin promotes neurogenesis and inhibits glial differentiation by independent mechanisms. Cell 104, 365–376 (2001).

17. Spenger, C. et al. Functional MRI at 4.7 tesla of the rat brain during electric stimula-tion of forepaw, hindpaw, or tail in single- and multislice experiments. Exp. Neurol. 166, 246–253 (2000).

18. Hofstetter, C.P., Schweinhardt, P., Klason, T., Olson, L. & Spenger, C. Numb rats walk - a behavioural and fMRI comparison of mild and moderate spinal cord injury. Eur. J. Neurosci. 18, 3061–3068 (2003).

19. Coumans, J.V. et al. Axonal regeneration and functional recovery after complete spi-nal cord transection in rats by delayed treatment with transplants and neurotrophins. J. Neurosci. 21, 9334–9344 (2001).

20. Bradbury, E.J. et al. Chondroitinase ABC promotes functional recovery after spinal cord injury. Nature 416, 636–640 (2002).

21. Olson, L. Regeneration in the adult central nervous system: experimental repair strate-gies. Nat. Med. 3, 1329–1335 (1997).

22. Pearse, D.D. et al. cAMP and Schwann cells promote axonal growth and functional recovery after spinal cord injury. Nat. Med. 10, 610–616 (2004).

23. Bareyre, F.M. et al. The injured spinal cord spontaneously forms a new intraspinal circuit in adult rats. Nat. Neurosci. 7, 269–277 (2004).

24. Brown, A., Ricci, M.J. & Weaver, L.C. NGF message and protein distribution in the injured rat spinal cord. Exp. Neurol. 188, 115–127 (2004).

25. Averill, S., McMahon, S.B., Clary, D.O., Reichardt, L.F. & Priestley, J.V. Immunocytochemical localization of trkA receptors in chemically identified subgroups of adult rat sensory neurons. Eur. J. Neurosci. 7, 1484–1494 (1995).

26. Christensen, M.D. & Hulsebosch, C.E. Spinal cord injury and anti-NGF treatment results in changes in CGRP density and distribution in the dorsal horn in the rat. Exp. Neurol. 147, 463–475 (1997).

27. Hokfelt, T. et al. Calcitonin gene-related peptide in the brain, spinal cord, and some peripheral systems. Ann. NY Acad. Sci. 657, 119–134 (1992).

28. Willis, W.D. & Coggeshall, R.E. Sensory Mechanisms of the Spinal Cord. (Plenum, New York, 1991).

29. Bennett, A.D., Chastain, K.M. & Hulsebosch, C.E. Alleviation of mechanical and thermal allodynia by CGRP(8–37) in a rodent model of chronic central pain. Pain 86, 163–175 (2000).

30. Tomita, K., Moriyoshi, K., Nakanishi, S., Guillemot, F. & Kageyama, R. Mammalian achaete-scute and atonal homologs regulate neuronal versus glial fate determination in the central nervous system. EMBO J. 19, 5460–5472 (2000).

31. Guillemot, F. Vertebrate bHLH genes and the determination of neuronal fates. Exp. Cell Res. 253, 357–364 (1999).

32. Zirlinger, M., Lo, L., McMahon, J., McMahon, A.P. & Anderson, D.J. Transient expres-sion of the bHLH factor neurogenin-2 marks a subpopulation of neural crest cells biased for a sensory but not a neuronal fate. Proc. Natl. Acad. Sci. USA 99, 8084–8089 (2002).

33. Blight, A.R. & Young, W. Central axons in injured cat spinal cord recover electrophysi-ological function following remyelination by Schwann cells. J. Neurol. Sci. 91, 15–34 (1989).

34. Gensert, J.M. & Goldman, J.E. Endogenous progenitors remyelinate demyelinated axons in the adult CNS. Neuron 19, 197–203 (1997).

35. Griffiths, I.R. & McCulloch, M.C. Nerve fibres in spinal cord impact injuries. Part 1. Changes in the myelin sheath during the initial 5 weeks. J. Neurol. Sci. 58, 335–349 (1983).

36. Nashmi, R. & Fehlings, M.G. Changes in axonal physiology and morphology after chronic compressive injury of the rat thoracic spinal cord. Neuroscience 104, 235–251 (2001).

37. Franklin, R.J., Gilson, J.M. & Blakemore, W.F. Local recruitment of remyelinating cells in the repair of demyelination in the central nervous system. J. Neurosci. Res. 50, 337–344 (1997).

38. Bambakidis, N.C. & Miller, R.H. Transplantation of oligodendrocyte precursors and sonic hedgehog results in improved function and white matter sparing in the spinal cords of adult rats after contusion. Spine J. 4, 16–26 (2004).

39. Utzschneider, D.A., Archer, D.R., Kocsis, J.D., Waxman, S.G. & Duncan, I.D. Transplantation of glial cells enhances action potential conduction of amyelinated spinal cord axons in the myelin-deficient rat. Proc. Natl. Acad. Sci. USA 91, 53–57 (1994).

40. Johansson, C.B. et al. Identification of a neural stem cell in the adult mammalian central nervous system. Cell 96, 25–34 (1999).

41. Falk, A. et al. Gene delivery to adult neural stem cells. Exp. Cell Res. 279, 34–39 (2002).

42. Bregman, B.S. et al. Recovery from spinal cord injury mediated by antibodies to neurite growth inhibitors. Nature 378, 498–501 (1995).

43. Basso, D.M., Beattie, M.S. & Bresnahan, J.C. A sensitive and reliable locomotor rating scale for open field testing in rats. J. Neurotrauma 12, 1–21 (1995).

44. Behrmann, D.L., Bresnahan, J.C., Beattie, M.S. & Shah, B.R. Spinal cord injury produced by consistent mechanical displacement of the cord in rats: behavioral and histologic analysis. J. Neurotrauma 9, 197–217 (1992).

45. Gale, K., Kerasidis, H. & Wrathall, J.R. Spinal cord contusion in the rat: behavioral analysis of functional neurologic impairment. Exp. Neurol. 88, 123–134 (1985).

46. Hargreaves, K., Dubner, R., Brown, F., Flores, C. & Joris, J. A new and sensitive method for measuring thermal nociception in cutaneous hyperalgesia. Pain 32, 77–88 (1988).

47. Gundersen, H.J. Stereology of arbitrary particles. A review of unbiased number and size estimators and the presentation of some new ones, in memory of William R. Thompson. J. Microsc. 143, 3–45 (1986).

48. Mouton, P.R., Gokhale, A.M., Ward, N.L. & West, M.J. Stereological length estimation using spherical probes. J. Microsc. 206, 54–64 (2002).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Intraglomerular inhibition: signaling mechanisms of an olfactory microcircuitGabe J Murphy, Daniel P Darcy & Jeffry S Isaacson

Microcircuits composed of principal neuron and interneuron dendrites have an important role in shaping the representation of sensory information in the olfactory bulb. Here we establish the physiological features governing synaptic signaling in dendrodendritic microcircuits of olfactory bulb glomeruli. We show that dendritic γ-aminobutyric acid (GABA) release from periglomerular neurons mediates inhibition of principal tufted cells, retrograde inhibition of sensory input and lateral signaling onto neighboring periglomerular cells. We find that L-type dendritic Ca2+ spikes in periglomerular cells underlie dendrodendritic transmission by depolarizing periglomerular dendrites and activating P/Q type channels that trigger GABA release. Ca2+ spikes in periglomerular cells are evoked by powerful excitatory inputs from a single principal cell, and glutamate release from the dendrites of single principal neurons activates a large ensemble of periglomerular cells.

The olfactory bulb is a brain region rich in microcircuits that is involved in the early processing of sensory information. Olfactory information is conveyed directly to the brain through conventional synapses in olfactory bulb glomeruli. Here, the axons of olfactory receptor neurons make glutamatergic synaptic contacts with the dendrites of principal neurons, mitral and tufted (M/T) cells and local interneurons called periglomerular neurons1–3.

Most synaptic interactions within the bulb occur at dendrodendritic synapses. M/T neurons make dendrodendritic synaptic contacts with GABAergic periglomerular neurons in glomeruli (Fig. 1a), whereas recip-rocal dendrodendritic synapses link M/T lateral dendrites and granule cells in the external plexiform layer4. Dendrodendritic synapses between M/T and granule cells provide a basis for self- and lateral inhibition, which is thought to contribute to odor discrimination5,6. Dendrodendritic syn-apses between M/T primary dendrites and periglomerular cells are poised to mediate intraglomerular feedback inhibition and ‘gate’ M/T cell activ-ity at the first site of olfactory sensory input. In addition, dendrodendritic self-inhibition is thought to have an important role in the temporal pat-terning of odorant-evoked activity in M/T neurons7–9.

The mechanisms underlying dendrodendritic signaling between mitral and granule cells in the bulb are beginning to be explored. N-Methyl-D-aspartate receptors (NMDARs) are believed to have an important role in triggering GABA release from granule dendrites10–13. Ca2+ influx through NMDARs has been suggested to trigger GABA exocytosis directly11,13. Alternatively, it has been proposed that the slow kinetics of NMDARs are especially good for bringing dendrites to threshold for activating voltage-gated Ca2+ channels that trigger release10,12,14. Evidence supports a role for high voltage– activated (HVA) N- or P/Q-type channels in dendritic GABA release14, whereas low voltage–activated (LVA) T-type currents have also been implicated15.

It is thus unclear whether subthreshold events in interneurons are sufficient to trigger dendrodendritic transmission, or whether strong depolarization provided by Na+ and/or Ca2+ spikes are required for dendritic microcircuits to function under physiological condi-tions. It is also unclear why brief activation of mitral neurons elic-its GABAergic feedback that persists for hundreds of milliseconds. Answering these questions is a critical step toward understanding how the bulb processes and represents information about odorant stimuli. Unfortunately, progress has been hampered by the difficulty in making paired recordings between synaptically coupled cells of the mitral and granule cell layers14,15.

Several factors suggest that the mechanisms governing signaling between M/T and granule cells are likely to be closely related to those at M/T and periglomerular dendrite contacts. Anatomically, reciprocal synapses formed between glomerular dendritic tufts of M/T cells and dendritic spines of periglomerular cells1–3 seem virtually identical to dendrodendritic contacts formed between M/T lateral dendrites and granule cells16. In addition, NMDAR- mediated excitation of periglomerular cells drives prolonged inhibition of external tufted cells at dendrodendritic synapses in glomeruli, as it does between mitral cells and granule cells in the external plexiform layer17.

In this study, we take advantage of the convergence of tufted cell and periglomerular cell dendrites within individual glomeruli to identify the fundamental processes that govern synaptic signaling in reciprocal dendrodendritic microcircuits.

RESULTSMultiple modes of glomerular GABA signalingWe first examined classical reciprocal dendrodendritic inhibition (DDI) in external tufted cells, a class of principal neurons lacking

Neuroscience Graduate Program and Department of Neuroscience, University of California, San Diego, La Jolla, California 92093-0608, USA. Correspondence should be addressed to J.S.I. ([email protected]).

Published online 6 February 2005; corrected online 7 February 2005; doi:10.1038/nn1403

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

lateral dendrites that project a single primary dendrite into a nearby glomerulus. To study dendritic transmission, tetrodotoxin (TTX, 1 µM) was added to the artificial cerebrospinal fluid (aCSF) to block Na+- dependent action potentials. This abolishes the ability of axons to convey action potentials and thus isolates dendritic transmis-sion. Under these conditions, brief (20 ms) activation of tufted cell calcium channels (+70 mV voltage step, CsCl internal) generates a prolonged barrage of synaptic currents mediated through the GABAA receptor (GABAAR; Fig. 1b).

Olfactory nerve terminals express metabotropic GABAB receptors (GABABRs), and GABABR agonists reduce evoked glutamate release from olfactory nerve terminals onto M/T and periglomerular neurons18,19. In addition, application of GABABR antagonists facilitates olfactory nerve–evoked responses, suggesting that GABA release from periglomerular neurons can mediate diffuse presynaptic inhibition of olfactory nerve transmission18. We tested directly whether GABA released from periglomerular neurons acts in a retrograde manner to inhibit glutamate release from olfactory nerve terminals. Periglomerular cells were voltage-clamped using a Cs+-based internal solution to block any postsynaptic action of GABABRs. In the presence of the GABAAR antagonist bicuculline methiodide (BMI, 10–20 µM), olfactory nerve stimulation evoked an excitatory postsynaptic current (EPSC) mediated by AMPA (α-amino-3-hydroxyl-5-methyl-4- isoxazole propionate) receptors in periglomerular neurons19. To examine the effect of

GABA release from periglomerular neurons on glutamate release from olfactory nerve terminals, we interleaved trials of olfactory nerve stimulation alone with trials in which

a depolarizing step (+70 mV) to the periglomerular neuron preceded olfactory nerve stimulation by 75–100 ms (Fig. 1c). On average, olfac-tory nerve–evoked EPSCs in periglomerular neurons were reduced in amplitude by ∼30% when olfactory nerve stimulation was preceded by a postsynaptic voltage step; this inhibition was abolished by the GABABR antagonist CGP 55845 (10 µM; Fig. 1c). These data indicate that GABA released from periglomerular cells can act in a retrograde fashion to inhibit glutamate release from olfactory nerve terminals by means of presynaptic GABABRs.

GABA release from individual periglomerular neurons can generate self-inhibition by activating GABAARs on the same cell20. It has also been suggested that dendritically released GABA might generate a GABAAR response in neighboring periglomerular cells through GABA spillover20. To examine GABAergic signaling directly between periglomerular cell dendrites, we made simultaneous recordings from pairs of periglomerular neurons that seemed to extend dendrites to a common glomerulus. In the presence of TTX (0.5–1 µM) and glutamate receptor blockers (NBQX, 10 µM, and D-APV, 50 µM, or MK-801, 10 µM), a brief voltage step from –70 to 0 mV in one cell (Fig. 1d, cell 1, gray traces) generated a Ca2+ current followed by a long-lasting tail current; the tail current reflects dendritically released GABA activating GABAARs on the cell from which it was released (self-inhibition21).

In many cases, a voltage step that produced self-inhibition in one cell also generated responses in a neighboring unstimulated periglomerular

Figure 1 Multiple types of GABAergic signaling in olfactory glomeruli. (a) Schematic representation of an olfactory bulb glomerulus. ON, olfactory nerve. (b) Tufted cell (TC) dendrodendritic self-inhibition mediated by periglomerular (PG) cells. In the presence of TTX (1 µM), a voltage step (+70 mV, 20 ms) to a TC is followed by a long-lasting barrage of IPSCs that are abolished by the GABAAR antagonist bicuculline (+BMI, 20 µM). (c) Retrograde GABABR-mediated inhibition of sensory input. Top, representative experiment showing that an olfactory nerve–evoked EPSC in a PG neuron is smaller when olfactory nerve stimulation is preceded by a voltage step to cause GABA release (Control, gray trace, dashed line). After CGP 55845 application, the voltage step no longer influences olfactory nerve–evoked EPSC amplitude (+CGP). Bottom, summary of effect of CGP 55845 on the ratio of EPSC amplitudes evoked by olfactory nerve stimulation after a voltage step and olfactory nerve stimulation alone (EPSCstep/EPSCcon; n = 6). (d–f) Lateral GABAAR-mediated signaling between PG cells. (d) A brief (10 ms) voltage step (shown schematically above) in PG cell 1 generates self-inhibition and postsynaptic currents in a neighboring PG neuron (PG cell 2) held at a variety of membrane potentials. Inset, peak postsynaptic current amplitude versus holding potential. (e) In another PG pair, self-inhibitory currents in the cell releasing GABA (gray traces) and currents in a neighboring PG neuron (black traces; –70 mV) are abolished by BMI (10 µM). (f) Reciprocal signaling between a pair of GABAergic PG neurons.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

neuron (Fig. 1d, cell 2). GABA release from one periglomerular cell generated a current in a neighboring periglomerular cell in 25 of 105 paired recordings. In all (25/25) paired recordings where a voltage step in cell 1 generated a current in cell 2, cell 1 showed self-inhibitory currents. The evoked currents in peri-glomerular cell 2 reversed near the predicted Cl– reversal potential (0 mV; Fig. 1d), suggesting they were due to activation of GABAARs. Consistent with this idea, both self-inhibitory currents in cell 1 and the current in cell 2 were blocked by BMI (10–20 µM, n = 7; Fig. 1e). In approximately one-third of paired recordings where GABA release from cell 1 produced a response in cell 2, a voltage step in cell 2 gener-ated a reciprocal GABAAR-mediated current in cell 1 (Fig. 1f). These data provide direct evidence that GABA released from the dendrites of one periglomerular neuron can activate GABAARs on neighboring periglomerular neurons.

Is GABA signaling between periglomerular neurons a consequence of GABA spillover or is it due to synaptic signaling? To address this issue, we examined the kinetics of currents produced by signaling between periglomerular cells. We found that currents evoked in one periglomerular neuron after GABA release from another could show very rapid rise times (∼1 ms; Fig. 2a). Indeed, in most (16/23) cases (Fig. 2b), unitary currents evoked in one periglomerular cell by GABA released from another periglomerular cell showed rise times resembling those at GABAergic synapses12,21. These data suggest that

periglomerular neurons can communicate with each other through dendrodendritic synaptic contacts because the rise times of inhibitory postsynaptic currents (IPSCs) evoked by dendritic GABA release from one periglomerular neuron onto another are too fast to be accounted for by anything other than direct synaptic signaling.

Signaling between most periglomerular neurons showed kinetics reminiscent of synaptic communication, but GABA release from one periglomerular neuron onto another did not always produce fast, unitary, GABAAR-mediated currents. In some pairs, the rise times of evoked responses were >5 ms, and there seemed to be a correlation between the rise and decay time of the events (Fig. 2b). To exam-ine whether electrotonic filtering of synapses at distant dendritic sites slowed IPSC kinetics, we used a voltage jump protocol22,23 to determine the time course of the underlying conductance in cases where currents had slow (3–7 ms) rise times (Fig. 2b, filled circles). In every instance (n = 4), the time course of the current recovered from voltage jumps matched that obtained at –70 mV, indicating that the slow kinetics of the response reflect a slow underlying conduc-tance (Fig. 2c). These data indicate that the slow kinetics of some GABAAR-mediated currents in periglomerular cells cannot be due to dendritic filtering of fast synaptic conductances.

Figure 2 Factors governing the time course of periglomerular signaling. (a) Traces from a pair of periglomerular cells in which a voltage step in cell 1 produces self-inhibition and a slow IPSC in a neighboring periglomerular cell; a voltage step in cell 2 produces self-inhibitory currents and fast synaptic currents in cell 1. Five episodes are overlaid in each panel. (b) Scatter plot of IPSC decay time (time to 1/e of peak) versus 10–90% rise time in 23 paired recordings. The dotted line represents the least-squares linear fit to data from 22 of 23 experiments. Filled circles represent responses tested with voltage jumps. (c) Slow GABAA responses reflect a slow conductance. Top, slow current evoked by a voltage pulse to a neighboring periglomerular cell. Bottom, postsynaptic voltage jumps from 0 to –70 mV at different time points after the presynaptic voltage step. Current is recovered throughout the slow IPSC. (d) Slow currents consist of summated asynchronous unitary events. Top, individual sweep showing a slowly rising IPSC. Bottom, blowup of top trace bounded region showing discrete, step-like inflections on the rising phase of the IPSC. (e) Increasing the duration of the presynaptic voltage step prolongs the time course of GABA release under control buffering conditions (0.5 mM EGTA). Left, representative experiment showing response to a brief and long depolarizing step. Right, summary of release time course during brief and long presynaptic depolarizations (n = 6 pairs). Ratio represents the time course of IPSC charge transfer evoked by 25 ms voltage steps divided by that evoked by a 10 ms voltage step. (f) Dialyzing the presynaptic neuron with an internal solution containing 10 mM EGTA prevents prolonged release (n = 6 pairs).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Buffer saturation governs the duration of GABA releaseSlow GABAAR synaptic responses between periglomerular cells may indicate a diffuse, spillover form of transmission, or a postsynaptic feature such as GABAARs with intrinsically slow kinetics. However, inspection of the rising phase of individual slow currents revealed discrete, step-like transitions (Fig. 2d). These current steps suggest that the slow rise of some responses may reflect the summation of asynchronous unitary events with fast rise times. We next took advantage of the good success rate of recording from synaptically coupled periglomerular cell pairs to explore further the factors shaping the time course of dendritic release.

A puzzling feature of DDI between M/T cells and local interneurons is the prolonged duration of asynchronous IPSCs after M/T cell glutamate release10–12,17. In periglomerular paired recordings, we found that the time course of release was dependent on the dura-tion of the presynaptic stimulus (Fig. 2e). Increasing the duration of the presynaptic voltage step from ≤10 ms to ≥20 ms greatly pro-longed the duration of dendritic GABA release. We quantified the time course of release by calculating the time it took for half of the postsynaptic charge transfer to occur after the end of the presynap-tic voltage step. Increasing the duration of the presynaptic voltage step nearly doubled the time it took for half of the charge transfer to occur after cessation of the presynaptic depolarization (Fig. 2e). Prolonged release evoked by a ≥20 ms step was abolished when the presynaptic neuron was dialyzed with an internal solution containing a high concentration of the Ca2+ chelator EGTA (10 mM; Fig. 2f). These results indicate that prolonged, asynchronous, dendritic GABA release can, in part, reflect the saturation of endogenous Ca2+ buffers in the presynaptic cell.

Dominant role for dendritic Ca2+ spikes in GABA releaseTo explore dendritic GABA release under more physiological conditions, we carried out experi-ments in which the presynaptic periglomerular

neuron was held in a current clamp. To our surprise, single action poten-tials generated by a brief presynaptic depolarizing current step (20–50 pA, 10 ms) generally failed to evoke an IPSC in postsynaptic periglomerular cells or tufted cells (Fig. 3a). In these same experiments, slightly longer current steps often triggered a burst of action potentials and a long-lasting, subthreshold depolarization that generated a barrage of IPSCs in the post-synaptic cell (Fig. 3b). Most GABA release generally occurred well after the last Na+ spike and coincided with the onset of the slow depolarization (Fig. 3c). Although single action potentials were relatively ineffective trig-gers of GABA release, IPSCs were more routinely observed during trains of action potentials (Fig. 3d). In these same cells in which action poten-tials were ineffective triggers of release, IPSCs were readily evoked in the presence of TTX by long-lasting (>100 ms) subthreshold depolarizations that followed a hyperpolarizing current step (n = 5; Fig. 3d). The marked differences in transmission evoked by action potentials as compared to slow depolarizations were consistently observed whether the postsyn-aptic cell was a tufted cell (n = 7) or another periglomerular neuron (n = 16; Supplementary Fig. 1 online).

What underlies these slow depolarizations, and why are they such effective triggers of dendritic GABA release? We first studied the rebound depolarizations triggered by hyperpolarizing current injec-tion. In the presence of TTX, rebound depolarizations were insensitive to nickel (100 µM; Fig. 3e), an inorganic blocker of LVA T-type Ca2+ channels24. In agreement with previous studies, we saw a hyperpolariza-tion-activated cation conductance in a large fraction of periglomerular cells25,26. The Ih channel blocker ZD7288 (25–50 µM) blocked the depolarizing sag during hyperpolarizing current injection, but did not abolish the rebound depolarization (Fig. 3e). In the presence of Ni2+ and ZD7288, the nonselective Ca2+ channel blocker Cd2+ (100 µM)

Figure 3 Ca2+ spikes mediated by dihydropyridine-sensitive channels are more effective triggers of periglomerular GABA release than fast Na+ spikes. (a) A brief current step in the presynaptic neuron (gray traces, KCl internal) evoked Na+ spikes but no IPSC in the postsynaptic cell (black traces, CsCl internal). (b) A slightly longer current step generates Na+ spikes and a subthreshold depolarization in the presynaptic neuron, and a barrage of IPSCs in the postsynaptic cell. Three consecutive trials are superimposed in each panel. (c) Blowup of a single trial in b. (d) Another paired recording where a train of action potentials (gray traces, KCH3SO4 internal) occasionally evoked an IPSC. In contrast, the subthreshold depolarization after a hyperpolarizing current step evoked robust release in the presence of TTX. Five traces are superimposed. (e) The subthreshold depolarization after a hyperpolarizing current step is insensitive to Ni2+ (100 µM) and ZD7288 (25–50 µM), but is blocked by 100 µM Cd2+. (f) Nimodipine blocks Ca2+ spikes evoked by a small (15 pA) depolarization from rest (∼–70 mV) in the presence of TTX, Ni2+ and ZD7288. Application of Bay K 8664 (10 µM; Bay K) recovered the spike. The time course of the experiment is shown below. Scale bars in e and f: 25 mV, 500 ms.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

abolished the rebound potential (n = 5). These data suggest that non-T-type Ca2+ channels mediate subthreshold depolarizations in periglomerular cells.

In a variety of central neurons, subthreshold depolarizations reflect dendritic Ca2+ spikes that propagate passively to the soma27–30. Similar to these neurons with active dendrites, periglomerular neurons seem to generate dendritic Ca2+ spikes that we record in the soma as an attenuated subthreshold potential. We next explored the Ca2+ channels underly-ing dendritic Ca2+ spikes in periglomerular neurons. Ca2+ spikes persisted in the presence of Ω-conotoxin MVIIC (5 µM, n = 3) or Ω-agatoxin IVA (250 nM, n = 3), ruling out a role for N-, P- or Q-type Ca2+ channels. However, the L-type Ca2+ channel antagonist nimodipine (20 µM) largely abolished Ca2+ spikes evoked by small (5–15 pA) depolariza-tions (n = 5; Fig. 3f) or hyperpolarizing current injection (n = 3; data not shown). Application of the dihydropyridine agonist Bay K 8644 (5–10 µM) rescued the Ca2+ spike after appli-cation of nimodipine (n = 4; Fig. 3f). These data indicate that L-type channels are most likely to underlie dendritic Ca2+ spikes in periglomerular cells.

Although periglomerular neurons are elec-trotonically compact, it is possible that Na+-dependent action potentials do not propagate well enough through periglomerular dendrites to cause appreciable Ca2+ influx to trigger release. To address this possibility, we filled cells with fluorescent Ca2+-indicator dye (Oregon Green BAPTA-1, 100 µM) to examine den-dritic Ca2+ transients (Fig. 4). Action potentials evoked Ca2+ transients throughout the den-dritic tree (>20 µm from the soma) of every periglomerular neuron examined. In a segment of dendrite that is ∼25 µm from the soma of a periglomerular neuron, a train of five action potentials at 40 Hz generated a Ca2+ tran-sient throughout the dendritic compartment (Fig. 4a). This transient was much smaller than that evoked by a rebound Ca2+ spike (Fig. 4b). Although the Ca2+ transients evoked by trains of action potentials were abolished by TTX (data not shown), the large transient evoked by the Ca2+ spike persisted (Fig. 4c).

On average, single action potentials generated a peak change in the fluorescence over baseline fluorescence (dF/F) of 2.1 ± 0.3% (n = 16 cells). Trains of five action potentials at 10–40 Hz produced a peak transient of 9.2 ± 1.6% (n = 11 cells). The transient evoked by five action potentials was nearly the linear sum of that produced by a single action potential in the same cells (Fig. 4d). The peak transient evoked by a Ca2+ spike (dF/F = 16.8 ± 2.6%, n = 15) always greatly exceeded that generated by a single action potential (Fig. 4e), and generally was larger than that evoked by trains of five action potentials (Fig. 4f). These results indicate that Ca2+ spikes produce much greater Ca2+ influx into periglomerular dendrites than action potentials.

LVA L-type current in periglomerular cellsWhat channels underlie the generation of low-threshold Ca2+ spikes in periglomerular dendrites? We found that low-threshold spikes are sensitive to dihydropyridines and can be evoked in the pres-ence of ZD7288 and 100 µM nickel. We next considered the possi-bility that L-type channels may be active near the resting membrane potential of periglomerular cells. In current clamp recordings, we observed that long (400 ms) hyperpolarizing current steps to –100 mV reduced dendritic Ca2+ before the generation of a rebound spike (Fig. 5a,b), and the decrease was correlated with the resting potential (n = 10 cells; Fig. 5c). This voltage-sensitive reduction in resting

Figure 4 Ca2+ spikes generate more Ca2+ influx than action potentials in periglomerular dendrites. (a–c) Pseudocolor images in the top row reflect peak Ca2+ in a segment of periglomerular dendrite in response to a brief train of action potentials (a), and a rebound Ca2+ spike under control conditions (b) and in the presence of TTX (c). Images are mapped on the same intensity scale. Colored boxes in c represent approximate regions of interest used to calculate dF/F shown in the second row. (d) Comparison of Ca2+ transient amplitudes produced by trains of five action potentials at 10–40 Hz versus single action potentials in the same cells. Solid line is a least-squares fit with slope = 4.1. AP, action potential. (e) Ca2+ influx produced by Ca2+ spikes under control conditions (black circles, n = 15) and in the presence of TTX (red circles, n = 5) is much larger than that produced by single action potentials in the same cells. (f) Dendritic Ca2+ spikes generally produce more Ca2+ influx than trains of five action potentials.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Ca2+ suggests that voltage-gated Ca2+ channels can be active near the resting membrane potential.

We next studied directly Ca2+ currents in voltage-clamped peri-glomerular cells (Vhold = –100 mV). Voltage steps evoked sustained Ca2+ currents that activated near the resting membrane potential of periglomerular cells (∼–60 mV; Fig. 5d,e). We did not find substantial T-type currents in periglomerular cells. The sustained LVA current was blocked by nimodipine (–44.1 ± 5.7% change, n = 7) and potentiated

by Bay K 8644 (+95 ± 50% change, n = 4; Fig. 5f,g). These properties are consistent with reports of LVA L-type channels in suprachi-asmatic neurons31 and pyramidal neurons of the hippocampus32,33 and amygdala34. These findings suggest that L-type Ca2+ channels can be active near the resting potential of periglo-merular neurons.

The effects of nimodipine on dendritic Ca2+ spikes and Ca2+ currents in periglomerular neurons prompted us to examine the role of L-type channels in DDI between tufted cell and periglomerular neurons. In the presence of TTX, nimodipine (20 µM) caused a reduc-tion in DDI (53 ± 15% of control) that was restored by the subsequent application of Bay K 8644 (5 µM, n = 5; Fig. 6a). We next carried out the identical experiment in cells bathed with a low Mg2+ (100 µM) aCSF. This facilitates DDI by enhancing the activa-tion of interneuron NMDARs10–12,17. Under these conditions, nimodipine also caused a reduction in DDI (69 ± 7% of control) that was recovered by Bay K 8644 (n = 4; Fig. 6a). These experiments indicate that L-type cal-cium channels contribute to DDI under physi-ological conditions as well as under conditions in which the contribution of NMDARs is greatly enhanced.

Ca2+ channels coupled to GABA exocytosisDo dihydropyridines act on the channels directly coupled to release from tufted cell or periglomerular cells? We first explored the possibility that the action of nimodipine on DDI reflected a block of the calcium channels governing glutamate release from tufted cells. We measured tufted cell glutamate release by recording self-excitation in low Mg2+ aCSF supplemented with picrotoxin (100 µM) to block GABAARs. Under these conditions, the cell is held at –60 mV and a voltage step to 0 mV is followed by a tail current owing to activation of NMDARs on the same cell35–37. Nimodipine (20 µM) had no effect on self- excitation (n = 6; Fig. 6b), ruling out a role for L-type channels in tufted cell glutamate release.

We next focused on the properties of Ca2+ channels governing dendritic GABA release by recording from synaptically coupled periglomerular cell pairs. We studied the volt-age dependence of dendritic transmission by stepping the presynaptic neuron to membrane

potentials between –70 and +50 mV from a holding potential of –90 mV. Only steps that were sufficiently large to activate HVA Ca2+ channels (>–40 mV) evoked an IPSC in synaptically coupled neurons (Fig. 6c). These data demonstrate that, like conventional nerve termi-nals, Ca2+ influx through HVA Ca2+ channels triggers dendritic GABA release from periglomerular neurons.

We used GABAAR-mediated self-inhibition to characterize the calcium channels directly coupled to GABA release from periglomerular

Figure 5 Low voltage–activated dihydropyridine sensitive Ca2+ currents in periglomerular cells. (a–c) Hyperpolarization reduces resting Ca2+ in periglomerular dendrites. (a) Image of a periglomerular dendrite. (b) Top, a long hyperpolarizing current step from rest (–60 mV) generates a rebound Ca2+ spike. Middle, changes in calcium in the three regions outlined in a. Bottom, average of the three regions shown above on an expanded scale. (c) Plot of the change in calcium level (dF) evoked at the end of a 400 ms hyperpolarizing step to approximately –100 mV versus membrane potential for ten cells. (d–g) Sustained low voltage–activated Ca2+ currents in periglomerular cells. (d) Response of a typical cell to voltage steps (Vhold = –100 mV) from –70 mV through –40 mV. Symbols represent the time points used to measure the transient (open circle) and sustained (filled circle) components of Ca2+ currents. (e) I/V relationship for LVA sustained and transient current in periglomerular cells (n = 8 cells). Inset, full I/V of the sustained current (n = 4 cells). (f) Single experiment showing calcium currents under control conditions, in the presence of nimodipine (20 µM) and after washout of nimodipine in the presence of Bay K 8644 (5 µM; Bay K). (g) Summary of effects of nimodipine (n = 7) and Bay K 8644 (n = 4) on currents evoked by voltage steps to –40 and –50 mV, respectively.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

dendrites. Nimodipine (20 µM) did not affect self-inhibition in periglomerular cells (n = 9; Fig. 6d), indicating that Ca2+ influx through L-type channels is not normally coupled to GABA release. Curiously, application of Bay K 8644 (5 µM) led to an increase in self-inhibition (160 ± 21% of control, n = 8; Fig. 6e). This indicates that although L-type channels do not normally support exocytosis, facilitating L-type channels allows them to act as a source of calcium for GABA release. The selective N-type blocker Ω-conotoxin GVIA (5 µM) did not affect self-inhibition (n = 5), suggesting that N-type channels are not involved in GABA release. However, Ω-conotoxin MVIIC (5 µM), a blocker of N- and P/Q type channels, caused a reduction in self-inhibition (Fig. 6f, 58 ± 6% of control, n = 8). In separate experiments, the P/Q antagonist agatoxin IVA (1 µM) reduced self-inhibition by an almost identical amount (Fig. 6f, 53 ± 7% of control, n = 6). Pretreatment of slices with nimodipine (20 µM) did not augment the action of the toxins, and results with and without the antagonist were pooled. These results suggest that P/Q-type calcium channels have a major role in triggering dendritic GABA release.

To confirm the role of P/Q-type channels in triggering GABA release from periglomerular dendrites, we used caged calcium to evoke DDI

in tufted cells. This approach triggers glutamate release from tufted cell dendrites, but bypasses the need for tufted cell dendrite Ca2+ chan-nels during DDI14. Photolysis of caged Ca2+ in tufted cells evoked a barrage of IPSCs similar to those evoked by depolarizing voltage steps in the same cell (Supplementary Fig. 2 online). Recordings from slices treated with Ω-conotoxin MVIIC (5 µM) and agatoxin IVA (500 nM) were interleaved with naive slices as a control. In control slices, DDI evoked by voltage steps averaged 431 ± 84 pA•s (n = 7), whereas in toxin-treated slices the charge transfer was markedly less (84 ± 42 pA•s, n = 10). DDI evoked by photolysis of caged Ca2+ in the same control slices was 145 ± 53 pA•s, whereas the response in toxin-treated slices was much smaller (33 ± 19 pA•s; Supplementary Fig. 2 online). These results indi-cate the important role of P/Q-type channels in periglomerular GABA release during dendrodendritic synaptic signaling.

Strong impact of single principal cells within glomeruliWhat is the impact of a single principal cell on interneurons in glomerular microcircuits? To address this question, we loaded slices with the cell-permeable calcium indicator Oregon Green BAPTA-1 AM (10 µM). We recorded from tufted cells in a voltage clamp using a

Figure 6 L-type Ca2+ channels are required for tufted cell–periglomerular cell (TC-PG) dendrodendritic inhibition but are not normally coupled to exocytosis. (a) DDI in a TC is reduced by nimodipine (20 µM) and recovered by Bay K 8644 (5 µM; Bay K). Summary plot shows dendrodendritic inhibition in normal aCSF (filled circles) and low Mg2+ aCSF (open circles). Top traces, representative TC recording showing reduction and recovery of dendrodendritic inhibition. (b) L-type Ca2+ channels do not contribute to TC glutamate release. Top traces, nimodipine has no effect on TC self-excitation in a representative cell. Subsequent application of APV (50 µM) abolishes the response. (c) Left, presynaptic PG Ca2+ current generated by voltage steps from –90 to –50, –30 and –10 mV and postsynaptic responses in a neighboring cell. Right, normalized Ca2+ current (filled circles) and IPSC charge (open circles) versus the presynaptic membrane potential from eight PG-PG pairs. Each point represents the average of three to eight experiments. Ca2+ currents were normalized to the response at –10 mV; IPSC charge was normalized to the response at +10 mV. IPSC charge was measured only during the step to exclude responses resulting from Ca2+ tail currents. (d) Nimodipine (20 µM) has no effect on PG self-inhibition. Top traces, bicuculline-subtracted records from a representative PG. (e) Bay K 8644 (5 µM) enhances PG self-inhibition. Top traces, bicuculline-subtracted traces from a representative cell. (f) In separate experiments, both agatoxin IVA (1 µM) and conotoxin MVIIC (5 µM) reduce PG self-inhibition. Top traces, representative PGs showing bicuculline-subtracted responses under control conditions (Con) and in the presence of agatoxin IVA (aga) or conotoxin MVIIC (ctx).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

KCl-based internal solution. Trains of brief depolarizations (+80 mV, 3 ms, 10 ms ISI) were used to evoke action currents (action potentials) to drive self-inhibition. Activation of a tufted cell generated robust calcium transients in small (<10 µm diameter), round compartments surrounding the glomerulus to which the tufted cell projected a den-dritic tuft (Fig. 7a). The size and location of these fluorescent puncta are consistent with the cell bodies of periglomerular cells. Indeed, in targeted whole-cell recordings (9/9 cells; Supplementary Fig. 3 online), the calcium transients always occurred in cells that showed hallmarks of periglomerular neurons (that is, input resistance ≥1 GΩ, self- inhibition, calcium spikes).

Activation of periglomerular cells coincided with the onset of tufted cell DDI (Fig. 7b). Single tufted cells activated multiple periglomeru-lar cells in every glomerulus examined (5 ± 1 periglomerular cells, range = 2–7, n = 8 glomeruli). Although the number of periglomeru-lar cells activated on individual trials varied with the magnitude of DDI, tufted cells reliably recruited the same ensemble of interneurons. An example of this is shown (Fig. 7c–e). We first stimulated a tufted cell with a short pulse train (five action potentials) that was just at threshold for eliciting DDI. On one trial, this stimulus activated seven periglomerular cells during and immediately after the pulse train (Fig. 7c). The DDI in this trial showed both an early, relatively synchronous component and delayed IPSCs. The next trial using the same brief stimulus activated only one periglomerular cell and did not elicit substantial DDI (Fig. 7d). Increasing the stimulus to ten action

potentials activated the same ensemble of periglomerular cells observed in the first trial and produced a prolonged barrage of IPSCs in the tufted cell (Fig. 7e). These results indicate that single tufted cells can reliably recruit the same ensemble of periglomerular cells. In addition, the kinetics of periglomerular cell activation (Fig. 7b–e) reveal considerable temporal variability (jitter) that overlaps with the time course of DDI. Similar results were obtained when DDI was evoked in the presence of TTX (Supplementary Fig. 3 online).

The somatic Ca2+ transients evoked in periglomerular cells suggest that the dendrite of a single principal cell has a powerful impact on individual interneurons. To explore the strength of this microcircuit, we made recordings from pairs of synaptically coupled periglomerular and external tufted cells in the presence of TTX. In a typical experiment, a brief depolarizing step in the voltage-clamped tufted cell generated a large excitatory postsynaptic potential (EPSP) in the periglomerular neuron (Fig. 8a). On average, tufted cell glutamate release produced a 10.8 ± 1.5 mV EPSP in periglomerular neurons (Vm ∼–65 mV, n = 14). In 7 of 14 tufted cell–periglomerular cell pairs, the EPSP was followed by an all-or-none Ca2+ spike that showed kinetics nearly identical to the rebound Ca2+ spike after a hyperpolarizing step in the same cell (Fig. 8b). APV (50 µM) had no effect on the rebound Ca2+ spike, but abolished the spike driven by glutamate release from the tufted cell (Fig. 8a, n = 3) and greatly reduced dendrodendritic self-inhibition. These data demonstrate that glutamate release from a single principal cell can trigger a dendritic Ca2+ spike in periglomerular cells. Moreover,

these data suggest that NMDARs are critical for dendritic Ca2+ spike generation.

In vivo, respiration generates rhythmic sen-sory input to the olfactory bulb8,9,38 and M/T cells are entrained by theta frequency olfactory nerve stimulation in vitro39,40. We next tested whether rhythmic activity in a tufted cell was sufficient to generate spikes in synaptically coupled periglomerular neurons. In cur-rent clamp, tufted cells fired bursts of action potentials in response to sinusoidal current injection (2 Hz; 100–200 pA). Single action potentials in a tufted cell generated a large (8.3 ± 1.7 mV, n = 12 pairs) EPSP in syn-aptically coupled periglomerular neurons, and theta-patterned bursts in tufted cells led to a complex Na+/Ca2+ spike in peri-glomerular neurons in 6 of 12 paired recordings (Fig. 8c). Tufted cell activity was most effective at bringing periglomerular neurons to thresh-old during the first theta cycle. The dynamic nature of tufted cell input was apparent from analyzing tufted cell–evoked depolarizations in periglomerular neurons as a function of the theta cycle; the second to fourth theta cycles in the tufted cell produced a much smaller depolarization in periglomerular neurons than the first theta cycle (n = 9; Fig. 8c, inset). These results demonstrate that physiologically relevant patterns of activity in tufted cells can produce large depolarizations in periglomeru-lar cells that are sufficient to evoke Ca2+ spikes. In addition, these data indicate that the impact of tufted cells on periglomerular neurons is dynamic during rhythmic patterns of activity like those observed in vivo.

Figure 7 Individual tufted cells (TCs) activate an ensemble of periglomerular (PG) cells. (a) Images of an intracellularly recorded TC (white) and Ca2+ indicator–loaded PG cells (colored circles) before (left) and after (right) a train of ten action currents (APs). (b) Ca2+ transients in PG cells shown in a and dendrodendritic inhibition evoked in the TC. Arrows indicate frames shown in a. (c–e) PG cell activation covaries with dendrodendritic inhibition. Top, images of peak PG cell activity evoked by TC stimulation at threshold for eliciting dendrodendritic inhibition. Bottom, time course of TC responses and PG calcium transients. A DIC image was used to estimate the position of the glomerulus (dashed circle). Displayed images were median filtered and level adjusted.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

DISCUSSIONHere we have established key physiologi-cal features of the microcircuit underlying intraglomerular inhibition. First, we found that, in addition to mediating dendroden-dritic GABAAR inhibition onto principal neurons, periglomerular cells make dendro-dendritic synaptic contacts onto each other. Periglomerular dendrite GABA release also produces retrograde inhibition of sensory input through presynaptic GABABRs. Second, we exploited the favorable electrotonic prop-erties of periglomerular neurons to study the relationship between Ca2+ influx and dendritic GABA release in detail. We found that transmitter release from periglomeru-lar dendrites requires Ca2+ influx through P/Q-type Ca2+ channels. Under physiologi-cal conditions, dendritic release is triggered much more effectively by Ca2+ spikes than Na+-dependent action potentials. We showed that L-type Ca2+ channels mediate dendritic Ca2+ spikes in periglomerular neurons but do not normally couple to exocytosis. Third, we examined the network properties under-lying signaling between principal cells and glomerular interneurons. We showed that glutamate release from a single principal cell is sufficiently powerful to generate den-dritic Ca2+ spikes in periglomerular cells. In addition, excitation from one principal cell dendrite activates a large ensemble of periglomerular cells.

Paired recordings revealed that dendritic release from one periglo-merular neuron can mediate fast and slow GABAAR-mediated responses in neighboring periglomerular neurons. Although GABA spillover or receptors with intrinsically slow kinetics may contribute to slow trans-mission, the kinetics of signaling between most periglomerular neurons was consistent with synaptic transmission. Anatomical studies have not reported evidence for dendrodendritic synapses formed between periglomerular neurons in mammals1–3. However, as pointed out previ-ously3, the complexity of dendritic processes in glomeruli can make it difficult to establish unequivocally the identity of pre- and postsynaptic cell types. Indeed, recent anatomical studies raise the possibility that calbindin-positive periglomerular neurons receive dendritic GABAergic input41,42. Our data provide evidence for synaptic signaling between periglomerular neurons in the mammalian olfactory bulb, and suggest a greater degree of synaptic complexity in glomerular circuits than was previously appreciated.

Pharmacological experiments have suggested that T-type Ca2+ chan-nels may underlie GABA release from the dendrites of granule cells in the olfactory bulb15. We examined directly the nature of the Ca2+ channels governing transmitter release from periglomerular dendrites in paired recordings. Our results indicate that HVA Ca2+ currents are required for dendritic GABA release. Given the more extensive den-dritic arbors of granule cells compared with periglomerular cells, it may be that granule cells use different mechanisms to trigger dendritic GABA release.

We found that single action potentials are far less effective at trig-gering transmitter release than dendritic Ca2+ spikes. Although some periglomerular cells have axons, these typically project into other

glomeruli, where they may contribute to lateral inhibition43. We there-fore think that the action potential–evoked release that we observed in periglomerular cell pairs in the same glomerulus reflects dendroden-dritic transmission.

We found that low-threshold Ca2+ spikes in periglomerular cells do not seem to require T-type Ca2+ channels and that nimodipine, an antagonist of L-type channels, inhibits Ca2+ spikes. Although it has been reported that nimodipine can block T-type channels32,44, the L-channel agonist Bay K 8644 recovered Ca2+ spikes in periglomerular cells. In agreement with a previous study45, we found that that Ca2+ spikes are relatively insensitive to low concentrations (100 µM) of Ni2+ that typi-cally block low-threshold Ca2+ spikes in other neurons24. At 1 mM, Ni2+ blocked Ca2+ spikes in periglomerular cells45; however, at this concentra-tion Ni2+ acts as a broad-spectrum calcium channel blocker32.

The simplest interpretation of our data is that LVA L-type channels have a critical role in generating periglomerular cell Ca2+ spikes. In sup-port of this idea, we found that hyperpolarization caused a decrease in resting [Ca2+] in periglomerular dendrites. It is unlikely that T-type chan-nels underlie a resting calcium conductance because they should be inac-tivated near the resting potential of periglomerular cells24. Although our voltage-clamp recordings did not reveal evidence for substantial T-type current, we did find rapidly activating, sustained LVA Ca2+ currents in periglomerular cells. Sustained LVA currents were partially blocked by nimodipine and enhanced by Bay K 8644. These results are consistent with other studies of neuronal LVA L-type currents31,33,34. Indeed, the expression of CaV1.3 channel subunits generates LVA L-type currents with similar properties46,47. Although we cannot rule out a contribution of T-type channels in some periglomerular cells, LVA L-channels seem sufficient for the generation of low-threshold Ca2+ spikes.

Figure 8 Dendritic glutamate release from a tufted cell generates an all-or-none Ca2+ spike in synaptically coupled periglomerular neurons. (a) In TTX, a brief voltage step to cause glutamate release from a tufted cell (gray traces) generates a large EPSP in the periglomerular neuron (black traces), which in turn leads to a Ca2+ spike in about half of the trials. The spike did not depend on the small (∼3 mV) fluctuations in postsynaptic membrane potential. In the presence of APV (50 µM), glutamate release from the tufted cell still generated a large average EPSP in the periglomerular neuron, but no Ca2+ spike. Four consecutive trials of EPSPs are superimposed. (b) The rebound Ca2+ spike, which had a similar time course to the glutamate-evoked Ca2+ spike, was not affected by APV. (c) In current clamp, 2 Hz sinusoidal current injection evokes bursts of action potentials in a tufted cell (gray traces) that generate an all-or-none complex spike in a synaptically coupled periglomerular neuron (black traces) in 4 of 10 trials. The inset plots the normalized EPSP amplitude in periglomerular neurons versus the tufted cell theta cycle number (n = 9). Periglomerular traces are shown on an expanded timescale below.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

As reported previously45, we found that prolonged hyperpolarization leads to a rebound Ca2+ spike in periglomerular cells. The deactivation and subsequent reactivation of LVA L-type channels at the resting membrane potential provides a simple mechanism for generation of rebound Ca2+ spikes in periglomerular cells. The high input resistance of periglomerular cells20,39,45,48 makes it plausible that relatively small currents can have large effects on membrane potential. A high density of dendritic Ca2+ channels in periglomerular cells would be ideal for the initiation of a regenerative Ca2+ spike not unlike the regenerative nature of conventional action potentials.

Although L-type channels are required for tufted cell– periglomerular cell DDI, they do not directly contribute to glutamate or GABA release under normal conditions. We showed that blockers of P/Q-type Ca2+ channels inhibit GABA release from periglomeru-lar dendrites. This is consistent with the high expression of P-type channels in periglomerular cells49. Residual release in the presence of toxins may reflect the contribution of R-type Ca2+ channels or dif-ficulty in achieving saturating toxin concentrations in glomerular compartments. These data demonstrate separate roles (depolarization versus exocytosis) for different Ca2+ channel types that act in concert in periglomerular dendrites.

Why do Ca2+ spikes trigger dendritic GABA release more effectively than fast Na+ spikes? We found that single action potentials cause Ca2+ transients throughout periglomerular cell dendrites, making it unlikely that action potentials do not propagate to dendritic release sites. One possibility is that the brief depolarization provided by action potentials is not coupled as effectively to the P/Q channels that underlie release. Indeed, the much slower Ca2+ spikes produce larger dendritic Ca2+ transients than trains of Na+-dependent action potentials. In olfactory bulb granule cells, A-type K+ channels specifically attenuate rapid dendritic excitation50. Similarly, we think it likely that active conduc-tances in periglomerular dendrites limit the ability of short-duration (action potential–like) inputs to activate dendritic Ca2+ spikes and trigger release.

Glutamate release from a single tufted cell can have a large impact on synaptically coupled periglomerular neurons. We observed activation of approximately five periglomerular cells by single tufted cells. This suggests that the synchronous activation of many M/T neurons is not necessary to produce intraglomerular inhibition. Given that our slice experiments are carried out on cut segments of spherical glomeruli, the number of periglomerular cells activated by single tufted cells is likely to be much higher in intact glomeruli. The activation of periglomerular cell ensembles occurred with substantial temporal jitter. Our results support the idea that the slow time course of DDI reflects both the temporal variability in onset of periglomerular cell Ca2+ spikes and calcium buffer saturation in periglomerular dendrites.

Finally, our paired tufted cell–periglomerular cell recordings indicate that single tufted cells generate a powerful EPSP in glomerular interneurons. We showed that NMDARs facilitate dendritic Ca2+ spike generation in GABAergic interneurons. This is consistent with the idea that the crucial role of NMDARs in DDI reflects their intrinsi-cally slow kinetics14,50 rather than their Ca2+ permeability. We also found that activation of periglomerular neurons by principal neurons is dynamic and adapts rapidly during physiologically relevant (theta) activity. The marked depression of tufted cell EPSPs in periglomerular neurons after the first theta cycle may reflect the depletion of available transmitter from tufted cell dendritic release sites. This adaptation of periglomerular input is similar to that of odor-evoked lateral inhibition in M/T neurons and granule cell excitation in vivo8, suggesting that adaptation may be a common feature of dendrodendritic transmission between principal neurons and interneurons in the bulb.

METHODSElectrophysiology. Olfactory bulb slices (∼250–300 µm) were prepared from 2–4-week-old Sprague-Dawley rats in accordance with institutional and national guidelines using standard procedures. Slices were prepared and maintained in aCSF containing 83 mM NaCl, 2.5 mM KCl, 3.3 mM MgSO4, 1 mM NaH2PO4, 26.2 mM NaHCO3, 22 mM glucose, 72 mM sucrose and 0.5 mM CaCl2, and equilibrated with 95% O2/5% CO2 at 34 °C for 15–30 min and at room temperature thereaf-ter. In the recording chamber, slices were viewed by means of infrared-DIC optics (BX-51W1, Olympus) and superfused with aCSF containing 119 mM NaCl, 2.5 mM KCl, 1.3 mM MgSO4, 1 mM NaH2PO4, 26.2 mM NaHCO3, 22 mM glucose and 2.5 mM CaCl2.

Whole-cell electrodes (∼3–4 MΩ) for voltage-clamp recordings were filled with a solution containing 120 mM CsCl, 10 mM TEA-Cl, 10 mM HEPES, 12 mM phosphocreatine, 0.25–10 mM EGTA, 3 mM Mg-ATP, 0.2 mM Na-GTP and 10 mM GABA (pH ∼7.35, 300 mOsm). For isolating calcium currents, Cs-gluconate replaced CsCl in the internal solution and the aCSF included 100 µM picrotoxin, 1 µM TTX, 1 mM 4-aminopyridine, 4 mM TEA and 5 mM CsCl. The holding potential was –70 mV unless otherwise noted. For tufted cell recordings, glutamate replaced GABA in the internal solution. Current-clamp recordings were carried out with a KCH3SO4-based internal solution contain-ing 115.5 KCH3SO4, 17.5 mM KCl, 10 mM HEPES, 10 mM phosphocreatine, 3 mM magnesium ATP, 0.2 mM sodium GTP, 0.5 mM EGTA and 10 mM GABA (pH ∼7.35, 300 mOsm). This internal solution mimics the Cl– reversal potential (–50 mV) previously measured in periglomerular cells20.

Series resistance was typically <10 MΩ and compensated by ≥90%. Experiments using cesium gluconate– or KCH3SO4-based internal solutions were corrected for a 10 mV junction potential. Responses were recorded with Axopatch 200B amplifiers (Axon Instruments), filtered at 2–5 KHz and digitized at 10–20 KHz (ITC-18; Instrutech). Data acquisition and analysis were per-formed with Axograph 4.8 (Axon) and IGOR Pro 4 (Wavemetrics). Experiments examining periglomerular cell–periglomerular cell signaling were carried out in the presence of CGP 55845 or CGP 54626 (2.5–5 µM) and 1,2,3,4-tetrahydro-6-nitro-2,3-dioxo-benzo[f]quinoxaline-7-sulfonamide (NBQX) to block GABAB and AMPA receptors, respectively. Experiments were carried out at 31–33 °C unless otherwise noted. Ca2+ currents were determined using leak subtraction (P/4). Experiments examining Ca2+ currents and the voltage dependence of release were carried out at room temperature.

Imaging. For Ca2+ imaging of periglomerular dendrites Oregon Green-1 (100 µM; Molecular Probes) replaced EGTA in the internal solution. Image acquisition (494 nm excitation, 2 × 2 binning, 25–50 Hz) and analysis were carried out with a cooled-CCD camera system (T.I.L.L. Photonics). Regions of interest for analysis were restricted to portions of dendrite in the focal plane.

To examine network properties of tufted cell and periglomerular cells, slices were loaded with Oregon Green-1 AM (10 µM, with 0.02% Pluronic F-127) at 34 °C for 30 min. We observed preferential labeling of periglomerular cells relative to principal neurons in the glomerular layer. Tufted cells were filled intracellularly with Alexa 568 (10 µM; Molecular Probes). Calcium transients were imaged at 494 nm excitation (30 Hz, 4 × 4 binning) and tufted cell images were acquired at 568 nm. Regions of interest (∼10 µm diameter) centered over periglomerular somata were used for kinetic analysis.

Representative traces are the average of five or more consecutive episodes, except where noted. Data are presented as mean ± s.e.m. Student’s t-test was used to determine statistical significance.


ACKNOWLEDGMENTSWe thank M. Scanziani, P. Sah and K. Franks for helpful discussions. G.J.M. received support from an NRSA predoctoral fellowship (NIDCD; DC005679). J.S.I. received support from a McKnight Scholar Award, Klingenstein Award, Burroughs-Wellcome Career Award and the NIH (RO1 DC04682).



©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

1. White, E.L. Synaptic organization of the mammalian olfactory glomerulus: new findings including an intraspecific variation. Brain Res. 60, 299–313 (1973).

2. White, E.L. Synaptic organization in the olfactory glomerulus of the mouse. Brain Res. 37, 69–80 (1972).

3. Pinching, A.J. & Powell, T.P. The neuropil of the glomeruli of the olfactory bulb. J. Cell Sci. 9, 347–377 (1971).

4. Shepherd, G.M. & Greer, C.A. Olfactory bulb. in The Synaptic Organization of the Brain (ed. Shepherd, G.M.) 159–203 (Oxford Univ. Press, Oxford, 1998).

5. Brennan, P.A. & Keverne, E.B. Neural mechanisms of mammalian olfactory learning. Prog. Neurobiol. 51, 457–481 (1997).

6. Yokoi, M., Mori, K. & Nakanishi, S. Refinement of odor molecule tuning by dendro-dendritic synaptic inhibition in the olfactory bulb. Proc. Natl Acad. Sci. USA 92, 3371–3375 (1995).

7. Spors, H. & Grinvald, A. Spatio-temporal dynamics of odor representations in the mammalian olfactory bulb. Neuron 34, 301–315 (2002).

8. Cang, J. & Isaacson, J.S. In vivo whole-cell recording of odor-evoked synaptic trans-mission in the rat olfactory bulb. J. Neurosci. 23, 4108–4116 (2003).

9. Margrie, T.W. & Schaefer, A.T. Theta oscillation coupled spike latencies yield com-putational vigour in a mammalian sensory system. J. Physiol. (Lond.) 546, 363–374 (2003).

10. Isaacson, J.S. & Strowbridge, B.W. Olfactory reciprocal synapses: dendritic signaling in the CNS. Neuron 20, 749–761 (1998).

11. Chen, W.R., Xiong, W. & Shepherd, G.M. Analysis of relations between NMDA recep-tors and GABA release at olfactory bulb reciprocal synapses. Neuron 25, 625–633 (2000).

12. Schoppa, N.E., Kinzie, J.M., Sahara, Y., Segerson, T.P. & Westbrook, G.L. Dendrodendritic inhibition in the olfactory bulb is driven by NMDA receptors. J. Neurosci. 18, 6790–6802 (1998).

13. Halabisky, B., Friedman, D., Radojicic, M. & Strowbridge, B.W. Calcium influx through NMDA receptors directly evokes GABA release in olfactory bulb granule cells. J. Neurosci. 20, 5124–5134 (2000).

14. Isaacson, J.S. Mechanisms governing dendritic γ-aminobutyric acid (GABA) release in the rat olfactory bulb. Proc. Natl Acad. Sci. USA 98, 337–342 (2001).

15. Egger, V., Svoboda, K. & Mainen, Z.F. Mechanisms of lateral inhibition in the olfactory bulb: efficiency and modulation of spike-evoked calcium influx into granule cells. J. Neurosci. 23, 7551–7558 (2003).

16. Price, J.L. & Powell, T.P. The synaptology of the granule cells of the olfactory bulb. J. Cell Sci. 7, 125–155 (1970).

17. Christie, J.M., Schoppa, N.E. & Westbrook, G.L. Tufted cell dendrodendritic inhibition in the olfactory bulb is dependent on NMDA receptor activity. J. Neurophysiol. 85, 169–173 (2001).

18. Aroniadou-Anderjaska, V., Zhou, F.M., Priest, C.A., Ennis, M. & Shipley, M.T. Tonic and synaptically evoked presynaptic inhibition of sensory input to the rat olfactory bulb via GABAB heteroreceptors. J. Neurophysiol. 84, 1194–1203 (2000).

19. Murphy, G.J., Glickfeld, L.L., Balsen, Z. & Isaacson, J.S. Sensory neuron signaling to the brain: properties of transmitter release from olfactory nerve terminals. J. Neurosci. 24, 3023–3030 (2004).

20. Smith, T.C. & Jahr, C.E. Self-inhibition of olfactory bulb neurons. Nat. Neurosci. 5, 760–766 (2002).

21. Salin, P.A. & Prince, D.A. Spontaneous GABAA receptor-mediated inhibitory currents in adult rat somatosensory cortex. J. Neurophysiol. 75, 1573–1588 (1996).

22. Hausser, M. & Roth, A. Estimating the time course of the excitatory synaptic conduc-tance in neocortical pyramidal cells using a novel voltage jump method. J. Neurosci. 17, 7606–7625 (1997).

23. Pearce, R.A. Physiological evidence for two distinct GABAA responses in rat hippo-campus. Neuron 10, 189–200 (1993).

24. Perez-Reyes, E. Molecular physiology of low-voltage-activated t-type calcium channels. Physiol. Rev. 83, 117–161 (2003).

25. Holderith, N.B., Shigemoto, R. & Nusser, Z. Cell type-dependent expression of HCN1 in the main olfactory bulb. Eur. J. Neurosci. 18, 344–354 (2003).

26. Cadetti, L. & Belluzzi, O. Hyperpolarisation-activated current in glomerular cells of the rat olfactory bulb. Neuroreport 12, 3117–3120 (2001).

27. Regehr, W., Kehoe, J.S., Ascher, P. & Armstrong, C. Synaptically triggered action potentials in dendrites. Neuron 11, 145–151 (1993).

28. Golding, N.L. & Spruston, N. Dendritic sodium spikes are variable triggers of axonal action potentials in hippocampal CA1 pyramidal neurons. Neuron 21, 1189–1200 (1998).

29. Kim, H.G. & Connors, B.W. Apical dendrites of the neocortex: correlation between sodium- and calcium-dependent spiking and pyramidal cell morphology. J. Neurosci. 13, 5301–5311 (1993).

30. Schwindt, P. & Crill, W. Equivalence of amplified current flowing from dendrite to soma measured by alteration of repetitive firing and by voltage clamp in layer 5 pyramidal neurons. J. Neurophysiol. 76, 3731–3739 (1996).

31. Pennartz, C.M., de Jeu, M.T., Bos, N.P., Schaap, J. & Geurtsen, A.M. Diurnal modula-tion of pacemaker potentials and calcium current in the mammalian circadian clock. Nature 416, 286–290 (2002).

32. Avery, R.B. & Johnston, D. Multiple channel types contribute to the low-voltage-activated calcium current in hippocampal CA3 pyramidal neurons. J. Neurosci. 16, 5567–5582 (1996).

33. Magee, J.C., Avery, R.B., Christie, B.R. & Johnston, D. Dihydropyridine-sensitive, voltage-gated Ca2+ channels contribute to the resting intracellular Ca2+ concentration of hippocampal CA1 pyramidal neurons. J. Neurophysiol. 76, 3460–3470 (1996).

34. Power, J.M. & Sah, P. Intracellular calcium store filling by an L-type calcium current in the basolateral amygdala at subthreshold membrane potentials. J. Physiol. 562, 439–453 (2005).

35. Salin, P.A., Lledo, P.M., Vincent, J.D. & Charpak, S. Dendritic glutamate autorecep-tors modulate signal processing in rat mitral cells. J. Neurophysiol. 85, 1275–1282 (2001).

36. Friedman, D. & Strowbridge, B.W. Functional role of NMDA autoreceptors in olfactory mitral cells. J. Neurophysiol. 84, 39–50 (2000).

37. Isaacson, J.S. Glutamate spillover mediates excitatory transmission in the rat olfactory bulb. Neuron 23, 377–384 (1999).

38. Macrides, F. & Chorover, S.L. Olfactory bulb units: activity correlated with inhalation cycles and odor quality. Science 175, 84–87 (1972).

39. Hayar, A., Karnup, S., Shipley, M.T. & Ennis, M. Olfactory bulb glomeruli: external tufted cells intrinsically burst at theta frequency and are entrained by patterned olfactory input. J. Neurosci. 24, 1190–1199 (2004).

40. Schoppa, N.E. & Westbrook, G.L. Glomerulus-specific synchronization of mitral cells in the olfactory bulb. Neuron 31, 639–651 (2001).

41. Toida, K., Kosaka, K., Heizmann, C.W. & Kosaka, T. Chemically defined neuron groups and their subpopulations in the glomerular layer of the rat main olfactory bulb: III. Structural features of calbindin D28K-immunoreactive neurons. J. Comp. Neurol. 392, 179–198 (1998).

42. Kosaka, K., Toida, K., Aika, Y. & Kosaka, T. How simple is the organization of the olfactory glomerulus? The heterogeneity of so-called periglomerular cells. Neurosci. Res. 30, 101–110 (1998).

43. Aungst, J.L. et al. Centre-surround inhibition among olfactory bulb glomeruli. Nature 426, 623–629 (2003).

44. Stengel, W., Jainz, M. & Andreas, K. Different potencies of dihydropyridine derivatives in blocking T-type but not L-type Ca2+ channels in neuroblastoma-glioma hybrid cells. Eur. J. Pharmacol. 342, 339–345 (1998).

45. McQuiston, A.R. & Katz, L.C. Electrophysiology of interneurons in the glomerular layer of the rat olfactory bulb. J. Neurophysiol. 86, 1899–1907 (2001).

46. Lipscombe, D., Helton, T.D. & Xu, W. L-type calcium channels: the low down. J. Neurophysiol. 92, 2633–2641 (2004).

47. Xu, W. & Lipscombe, D. Neuronal CaV1.3α1 L-type channels activate at relatively hyperpolarized membrane potentials and are incompletely inhibited by dihydropyri-dines. J. Neurosci. 21, 5944–5951 (2001).

48. Puopolo, M. & Belluzzi, O. Functional heterogeneity of periglomerular cells in the rat olfactory bulb. Eur. J. Neurosci. 10, 1073–1083 (1998).

49. Hillman, D. et al. Localization of P-type calcium channels in the central nervous system. Proc. Natl Acad. Sci. USA 88, 7076–7080 (1991).

50. Schoppa, N.E. & Westbrook, G.L. Regulation of synaptic timing in the olfactory bulb by an A-type potassium current. Nat. Neurosci. 2, 1106–1113 (1999).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

A R T I C L E S


Medial prefrontal cortex determines how stressor controllability affects behavior and dorsal raphe nucleusJ Amat1,2, M V Baratta1,2, E Paul1,2, S T Bland1,2, L R Watkins1,2 & S F Maier1,2

The degree of behavioral control that an organism has over a stressor is a potent modulator of the stressor’s impact; uncontrollable stressors produce numerous outcomes that do not occur if the stressor is controllable. Research on controllability has focused on brainstem nuclei such as the dorsal raphe nucleus (DRN). Here we find that the infralimbic and prelimbic regions of the ventral medial prefrontal cortex (mPFCv) in rats detect whether a stressor is under the organism’s control. When a stressor is controllable, stress-induced activation of the DRN is inhibited by the mPFCv, and the behavioral sequelae of uncontrollable stress are blocked. This suggests a new function for the mPFCv and implies that the presence of control inhibits stress-induced neural activity in brainstem nuclei, in contrast to the prevalent view that such activity is induced by a lack of control.

It is generally recognized that the emotional and health-related consequences associated with aversive experiences are much less severe when organisms have control over the aversive event. Lack of control has been proposed to contribute to mood and anxiety disorders such as depression1 and post-traumatic stress disorder (PTSD)2. Much of what is known about the effect of control comes from studies of what is called learned helplessness or behavioral depression. The learned helplessness paradigm uses a triadic design which compares animals in two conditions that experience a series of stressors with animals in a third condition that do not experience the stressor. Rats in one of the stress conditions (escape condition) have control over the duration of the stressors: they can learn a response that terminates or escapes from each stressor. Rats in the other stress condition have no control over any aspect of the stress experience (inescapable condition). For these rats, the duration of each stressor is determined by the animal in the escape condition to which the animal in the inescapable condition is yoked.

The importance of control was first noted in the mid 1960s, with reports that animals exposed to a series of inescapable (uncontrollable) shocks later failed to learn to escape in a different environment3 and developed ulcers4, whereas identical escapable (controllable) shocks produced neither consequence. Stressor controllability effects have since been demonstrated in a broad range of species, including humans, and have been shown to extend to a wide spectrum of behavioral and neurochemical consequences of exposure to stressors5.

Research directed at understanding the neural mechanisms that mediate how stressor controllability influences the sequelae of stress-ors has focused on aminergic nuclei in the brainstem. Both the seroto-

nergic (5-HT) DRN and the noradrenergic (NE) locus coeruleus (LC) have received intensive study. These nuclei are particularly interesting because the DRN provides the majority of the 5-HT with innervation of cortical and limbic structures, as does the LC for the majority of the NE. Both 5-HT and NE have been implicated in depression and anxiety.

Work in our laboratory has focused on the DRN. Our most salient findings are that (i) uncontrollable stress activates DRN 5-HT neurons far more than equal controllable stress does, as indicated both by c-Fos expression in 5-HT neurons6 and by extracellular 5-HT within the DRN7 and projection regions of the DRN8–10; (ii) intense activation of DRN 5-HT neurons by uncontrollable stress sensitizes these neurons for a period of time8; and (iii) this activation and subsequent sensitization are both necessary and sufficient11 to produce the behavioral changes characteristic of learned helplessness (behavioral depression). Similar results have been reported for the LC1,12, and the changes in the DRN and LC may be related13.

The DRN is a small brainstem structure (as is the LC) and is unlikely to be capable of performing the complex computations necessary to process or detect whether there exists a contingency between the organ-ism’s behavior and outcomes, and thus whether a stressor is, or is not, controllable. In addition, the DRN does not receive direct inputs from either primary sensory or motor regions, so it may not even receive the information required for such a computation to be made. The prefrontal cortex (PFC) is a likely site to process contingency/control-lability and then to regulate DRN function accordingly. The PFC has been implicated in executive function, the aspect of cognition afford-ing the ability to act flexibly in a way not locked to environmental stimuli14, as well as in affective processing15. Within the PFC, the ventral

1Department of Psychology and 2Center for Neuroscience, Campus Box 345, University of Colorado, Boulder, Colorado 80903-0345, USA. Correspondence should be addressed to S.F.M. ([email protected]).


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

medial prefrontal cortex (mPFCv) is of special interest. Anatomically, the DRN receives virtually all of its cortical input from infralimbic (IL) and prelimbic (PL) regions within the mPFCv16,17. Glutamatergic projections from these areas synapse onto predominantly GABAergic neurons within the DRN18, which in turn inhibit 5-HT neurons. Thus, activation of neurons within IL and PL regions might be expected to inhibit DRN 5-HT neurons, and indeed, electrical stimulation in the mPFCv inhibits DRN unit activity19. However, to our knowledge there are no data that indicate what sorts of environmental conditions might activate pathways that descend from the mPFCv to the DRN or other brainstem nuclei. In addition, although it has been noted20 that PFC executive functions might extend to the control of aminergic brainstem nuclei, as yet there has been no evidence for this.

These observations suggested to us a set of experiments to determine the effects of temporarily inactivating the mPFCv during exposure to equated (yoked) controllable and uncontrollable stressors on (i) activa-tion of DRN 5-HT neurons and 5-HT release in the DRN during the stressor and (ii) later behaviors typical of learned helplessness (behav-ioral depression). If the mPFCv is critical to the detection of whether a stressor is controllable and modulates the brainstem DRN accordingly,

controllable and uncontrollable stress should not differ with regard to either immediate neurochemical or delayed behavioral consequences. Here, we used microinjections of the GABAA receptor agonist muscimol in the mPFCv to temporarily inhibit mPFCv activity. We then exposed rats to either controllable or yoked uncontrollable stress. In separate experiments, we measured expression of c-Fos, the protein product of the immediate early gene c-fos, a marker of neuronal activity21, in 5-HT–labeled neurons within the DRN after the stressor. We assessed extracellular levels of 5-HT in the DRN before, during and after the stressor. 5-HT efflux in the DRN is a measure of DRN activation22. We also measured fear conditioning and escape learning 24 h after the stressor. These are behaviors that are altered differentially by stressors differing in controllability23.

We observed that the ability to control the stressor prevented the increase in DRN activation as well as subsequent behavioral changes produced by uncontrollable stress. However, mPFC inhibition with muscimol during controllable stress, while not affecting the ability to learn the wheel-turn response during stressor exposure, completely blocked the impact of stressor controllability on these outcomes.

RESULTSWheel-turn escape learningThe critical experiments are only possible if animals with mPFCv inactivation can learn to escape (control) the initial stressors. Thus, rats received a series of 100 escapable tailshocks while restrained in boxes with a wheel mounted in the front that could be turned to terminate each shock. Either muscimol or vehicle was microinjected through indwelling cannulae aimed at the juncture between IL and PL regions of the mPFCv (see Methods) 60 min before the start of the session. As in all of our previous research (for example, ref. 23), the number of wheel turns required to terminate each shock (in quarter turns of the wheel) was increased as the rats became more proficient at escape (see below). Thus, to measure the quality of escape performance, both the response requirement attained across trials (that is, the number of quarter turns to terminate the shock; Fig. 1a) and the time taken to terminate each tailshock (Fig. 1b) must be assessed. As is evident, animals injected with muscimol learned to escape quite well. Repeated-measures ANOVA did not indicate a difference between muscimol- and vehicle-treated groups on either time to terminate (F1,12 = 0.76) or response requirement (F1,12 = 1.49). In addition, there was no interaction between Groups and Trials on either measure (F19,228 = 1.11 and 1.08, respectively). The locations of the cannula placements within the mPFCv for this and the subsequent experiments are shown in Figure 2.

c-Fos expression in 5-HT–labeled neurons in the DRNWe have previously reported6 that inescapable shock (IS) produces greater activation of DRN 5-HT neurons (as assessed by c-Fos protein expression in 5-HT–labeled neurons) than is produced by exactly equal escapable shock (ES). Moreover, this effect of controllability was espe-cially prominent in middle and caudal regions of the DRN. In the pres-ent experiment, rats received ES, received yoked IS or remained as home cage controls (HCC). Half of each group received microinjections of

Figure 1 Efficiency of wheel-turn escape behavior during exposure to controllable tailshock in rats that had received mPFCv muscimol or vehicle 60 min before the shock session. (a) The mean (± s.e.m) number of quarter turns of the wheel attained as the escape requirement on each trial. (b) The mean (± s.e.m) time to reach escape criterion per trial. Filled symbols represent the group given muscimol and open symbols the group given vehicle.

Figure 2 Placements of microinjection cannula and microdialysis probe. Numerals indicate distance from bregma (mm). (a) Summary of microinjection cannula placements in the mPFCv for all experiments. Not all cannulae are shown owing to overlapping placements. (b) Summary of microdialysis probe placements in the DRN.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

muscimol as in the above experiment before treatment, whereas the remaining half received vehicle. Thus, the design was 3 (stress condition) × 2 (muscimol versus vehicle) factorial. Rats were sacrificed at 2 h after the session, the time of maximal c-Fos expression after IS6.

As in our prior studies, neither IS nor ES treatment altered the number of 5-HT–labeled neurons (data not shown). We determined the percentage of 5-HT–labeled neurons expressing c-Fos for the caudal (Fig. 3a) and rostral (Fig. 3b) DRN (coordinates in Methods) by immunohistochemistry. Again, as in our earlier studies, no 5-HT–labeled cells expressed c-Fos in HCC rats. This was true for both vehicle- and muscimol-injected HCC rats. Thus, HCC rats are not shown, as they are at zero with zero standard error, nor are they included in the statistical analysis. IS led to much greater c-Fos expression in caudal 5-HT neurons than did ES: a difference of approximately 300%. Muscimol abolished the difference between IS and ES, largely by increasing c-Fos expression in 5-HT neurons in the ES rats. ANOVA showed significant effects of stress condition (F1, 22 = 4.27, P = 0.05) and an interaction between stress condition and muscimol (F1,22 = 10.41, P < 0.004). Fisher’s PLSD post-hoc comparisons showed that the ES-vehicle group differed from the other groups, which did not differ from each other. Thus, muscimol sig-nificantly increased c-Fos expression in caudal 5-HT neurons in ES rats, whereas the small reduction in IS rats was not statistically reliable. There was no effect of stressor controllability or muscimol in rostral DRN.

Extracellular 5-HT in the caudal DRNWe have previously found7 that IS produces much greater 5-HT efflux than does equal ES within the caudal DRN, as measured by in vivo microdialysis. This 5-HT is released by axon collaterals and perhaps from dendrites themselves and is likely to reflect the activity of the DRN 5-HT neurons22. Therefore, we used the same experimental design as in the above c-Fos experiment and measured extracellular levels of 5-HT in the DRN before, during and after the stress session. Muscimol injected into the mPFCv had no detectable effect on 5-HT efflux within the DRN (Fig. 4, insert), and so the HCC groups with

and without muscimol were pooled for simplicity. Measuring extracel-lular 5-HT levels for the controls and for the groups given IS showed that IS produced a sustained increase in 5-HT that persisted during the IS session and for the period measured after the session (Fig. 4a). As in prior studies, ES (Fig. 4b) produced only a transient increase in 5-HT, with 5-HT returning to baseline levels by 40 min after the onset of the stress session. This rapid reduction in 5-HT if the stressor is controllable is notable, as the stressor exposure continued for another 80 min, but 5-HT now remained at baseline values. Muscimol, which had no effect in the IS rats, produced a marked increase in 5-HT efflux in the ES rats and elevated 5-HT to levels comparable to those observed in the IS rats. ANOVA on the baseline samples taken before the start of the stressor did not show any differences (all F values <1.0). During the stressor there were significant effects of stress condition (F2, 27 = 190.0, P < 0.00001), the interaction between stress condition and muscimol (F2, 27 = 4.31, P < 0.03), Time (F4,108 = 10.17, P < 0.0001), and the interaction between time, stress condition and muscimol (F8,108 = 2.34, P < 0.05). Fisher’s PLSD indicated that the IS-vehicle, IS, muscimol, and ES-muscimol groups differed from the other groups but did not differ from each other. ANOVA conducted on post-stress samples indicated significant effects of shock condition (F2,25 = 4.65, P < 0.02). Fisher’s PLSD indicated the same group differences as occurred during the stress treatment. Probe placements within the DRN are shown in Figure 2b.

Fear conditioning and escape learningIS potentiates subsequent fear conditioning and interferes with escape learning, whereas ES does not 24. Thus, we used the same 3 × 2 factorial design as above, with behavioral testing conducted 24 h after the ES/IS session using our typical procedures. However, we added two additional site-specificity control groups. One group was injected 2.0 mm rostral (ventral orbital cortex (VO), n = 4) and the other 2.0 mm caudal (cingulate cortex area 2 (Cg2), n = 4) relative to the usual

Figure 3 Percentage of neurons double-labeled for 5-HT and c-Fos (mean ± s.e.m). (a) Caudal DRN. (b) Rostral DRN. Gray bars represent rats that had received escapable stress; white bars represent rats that had received inescapable stress.

Figure 4 5-HT as a percentage of baseline in the DRN. The insert shows nonshocked home cage controls that received either muscimol or vehicle in the mPFCv (mean ± s.e.m). (a) Groups that received inescapable stress (IS) and controls. The open circles represent rats that had received vehicle before IS, the filled circles represent rats that had received muscimol before IS, and the dotted line represents the controls. (b) Groups that received escapable stress (ES) and controls. The open circles represent rats that had received vehicle before ES, the closed circles represent rats that had received muscimol before ES, and the dotted line represents the controls. The gray bar represents the time of stressor exposure in the wheel-turn boxes.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

mPFCv injection site. Both groups received muscimol and were subjected to ES, as pilot data indicated that muscimol altered functioning only in the ES rats. All rats received two shocks (0.6 mA) in a shuttlebox, and then fear conditioning to the environmental context was examined by measuring freezing in that environment for 20 min. It has been previously shown that this freezing is not a reaction to the shock per se, but rather to fear that has been conditioned to the context25. After this fear assessment, all rats received escape training in which two crossings of the shuttlebox (FR-2) were required to terminate each footshock. Freezing was not observed before the two footshocks in the shuttlebox but was observed after foot-shocks (Fig. 5). The ES muscimol control rats injected rostral and caudal to the critical site were pooled, as they did not differ (F < 1). As is typical, IS potentiated fear conditioning, whereas ES did not. Muscimol had no effect on IS rats but increased fear conditioning in the ES rats to the level in the IS rats. Notably, muscimol injected at the control sites had no effect on the behavior of the ES rats. ANOVA showed significant main effects of group (F6,48 = 3.49, P < 0.007) and trial block (F9,432 = 171.20, P < 0.00001), and a significant trial block × group interaction (F54, 432 = 3.03, P < 0.0001). Post-hoc Fisher’s PLSD tests (P < 0.05) indicated that the HCC-muscimol, HCC-vehicle, ES–muscimol at con-trol sites and ES-vehicle groups did not differ from each other, but did differ from the IS-muscimol, IS-vehicle and ES-muscimol groups.

Shuttlebox escape followed a similar pattern (Fig. 6). IS potently interfered with escape learning, whereas ES had no effect relative to controls. Again, muscimol had little effect on IS-treated rats, but led to poor escape in ES rats, although not when administered at the control sites. ANOVA indicated significant effects of groups (F6,48 = 7.62,

P < 0.0001), trial blocks (F4,192 = 4.36, P < 0.003) and the interaction between groups and trial blocks (F24,192 = 2.51, P < 0.003). Post-hoc Fisher’s PLSD comparisons (P < 0.05) indicated that the HCC- muscimol, HCC-vehicle, ES–muscimol at the control sites and ES-vehicle groups did not differ from each other but did differ from the IS-muscimol, IS-vehicle, and ES-muscimol groups.

DISCUSSIONThe results of this series of experiments are clear. Inactivation of regions within the mPFCv during exposure to stress eliminated the impact of behavioral control over the stressor on DRN 5-HT activation and subsequent behavior. Thus, the mPFCv is indeed involved in the processing of contingency information and regulation of DRN 5-HT activity. Notably, inactivation of the mPFCv eliminated the difference between controllable stress and uncontrollable stress by substantially altering the neural and behavioral impact of only one of these conditions: namely, controllable stress. Thus, muscimol injected into the mPFCv increased the controllable shock–induced c-Fos expression in 5-HT neurons in the DRN and extracellular 5-HT in the DRN to the levels normally produced by uncontrollable shock, and had no significant effect on either c-Fos expression or 5-HT efflux produced by the uncontrollable condition. Furthermore, inactivation of the mPFCv altered the behavioral consequences of stressor exposure measured 24 h later. Again, this manipulation had no effect in uncontrollably shocked rats, as they later showed potentiated fear conditioning and escape deficits. However, controllably stressed rats now behaved as if the stressor had been uncontrollable 24 h earlier: they showed potentiated fear conditioning and poor escape behavior. It would seem reasonable to conclude that aversive events per se drive the DRN26, and that if the events are controllable, the mPFCv inhibits that aversive stimulus–driven DRN 5-HT activity and prevents the cascade of events that follow prolonged activation of DRN 5-HT neurons. In addition, the DRN sends 5-HT projections to the mPFCv, and uncontrollable stress produces greater 5-HT efflux within the mPFCv than does controllable stress9. Because 5-HT tends to inhibit pyramidal neurons in the mPFC27, this large 5-HT efflux produced by uncontrollable stress may further contribute to differential activation of the mPFCv by stressors of differ-ing controllability. Thus, processes within the DRN alter mPFCv func-tion, and these two structures, at very different levels of the brain, are likely to interact closely to regulate stressor controllability–related phe-nomena. The present results also strengthen the tie between DRN 5-HT activation and learned helplessness and behavioral depression. mPFCv inactivation during controllable stress increased DRN 5-HT activation to the level observed with uncontrollable stress, and behaviors charac-teristic of learned helplessness and behavioral depression followed.

A potential difficulty with part of these conclusions is that our data here do not preclude the possibility that the mPFCv regulates the DRN not directly, but through projections to other structures that then in turn regulate the DRN. For example, the mPFCv projects to the central nucleus of the amygdala28, which in turn projects to the DRN16. Although the particular possibility that the central nucleus of the amyg-dala is a necessary relay is unlikely, as large lesions of this structure have no effect on the escape deficits produced by uncontrollable stress29, the general point still holds. However, even if this type of indirect regulation proves to exist, the critical role of the mPFCv in controllability effects and in brainstem regulation remains.

Notably, the mPFCv has not been implicated in escape learning per se, and the periaqueductal gray has generally been regarded as the key mediator of this type of learning30. Indeed, acquisition of the wheel-turn escape response occurred normally during mPFCv inactivation. Thus, the mPFCv is likely not to be critical to the acquisition of the

Figure 5 The mean (± s.e.m) number of 8-s intervals spent freezing, in 2-min blocks, after two footshocks 24 h after stressor exposure. Blue symbols represent rats that had received inescapable stress on day 1, red symbols escapable stress, and green symbols are home cage controls. Squares represent rats that received vehicle before day 1 treatment, whereas circles represent rats that had received muscimol. Site-specificity controls are represented by the dashed lines and orange stars.

Figure 6 Mean (± s.e.m) shuttlebox escape latencies across blocks of five shuttlebox FR-2 escape trails 24 h after stressor exposure. Blue symbols represent rats that had received inescapable stress on day 1, red symbols escapable stress, and green symbols are home cage controls. Squares represent rats that received vehicle before day 1 treatment, whereas circles represent rats that had received muscimol. Site specificity controls are represented by the dashed lines and orange stars.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

escape (controlling) response, but rather is the location of circuitry that detects that a contingency is present and then regulates other structures (here, the DRN) accordingly. Motor responses can be learned (that is, habits formed) in a manner separated from higher cognitive processing and ‘expectational-like’ phenomena31. Other circuitry then uses this information to regulate other neural and behavioral processes. It would seem that the mPFCv is the structure that performs this ‘utilizing’ func-tion for responses that escape aversive events. The escape response can be acquired without the involvement of the mPFCv, but the mPFCv is necessary to use the existence of successful escape to regulate other cir-cuitries that respond to aversive events and that influence later emotion and behavior. Thus, this interpretation of the data implies that when stressor termination is made contingent upon a particular response, at least two types of learning occur: (i) low-level processes are engaged that result in strengthening of the escape response (that is, a habit may be learned) and (ii) higher processes are engaged enabling the organism to detect, and/or use the knowledge, that there is a contingency between behavior and outcome and that its responses control the stressor. It is the latter process that mitigates the consequences associated with prolonged uncontrollable stress.

A related argument has been made with regard to the role of the PL in the learning of instrumental responses that obtain appetitive rewards32,33. Rats with PL lesions learned to make instrumental responses such as pressing a lever for food, but were sensitive only to the pairing or the temporal contiguity between the response and food, and not to the contingency between response and food. Thus, when the probability of food in the absence of the instrumental response was increased to be the same as the probability of food after responses (that is, no contingency, but still pairing between response and food), control rats responded less, but PL-lesioned rats continued to respond.

The present data and conclusions fit well with recent studies exploring the role of the mPFCv in fear-related processes. mPFCv lesions have proven to have no effect on the acquisition of conditioned fear34 or on the extinction of fear under some circumstances34. However, mPFCv lesions, in the same regions as inactivated here, interfere with the memory35 or the use of extinction to guide behavior36. Similarly, in the present experiments, the effects of mPFCv inactivation can be described as a failure to use the presence of an escape response to modulate the impact of the stressor on neural activation within the brainstem and on later behavior. At a neural level, it has been argued37 that the mPFCv mediates the memory of extinction, and perhaps the more general use of extinction information, by inhibiting amygdala function. Glutamatergic neurons from the IL region of the mPFCv project to GABAergic inhibitory interneurons within the capsular division of the central nucleus of the amygdala, which inhibit output from this latter region38. Consistent with this sort of arrangement, combat veterans with PTSD show enhanced amygdala and reduced mPFCv activation when shown combat scenes, relative to combat veterans that do not suffer from PTSD39. The suggestion has been that amygdala function is exaggerated because mPFCv inhibition of the structure is reduced. Notably, controllable stress produces less conditioned fear and anxiety than does uncontrollable stress5, findings that would follow from the activation of the mPFCv by the presence of control. It may be that the absence of a perception of control contributes to the occurrence of intense anxiety, as occurs in PTSD. Highly aversive events are likely to drive limbic and brainstem structures that induce negative affective and motivational experiences; control, perceived control or, more gener-ally, the ability to cope may activate mPFCv inhibition of these limbic and brainstem processes. From an evolutionary perspective, it may be sensible that activation of ‘lower’ centers by strongly aversive events came first, and that as species developed the ability to cope with such

events by behavioral means, inhibition from ‘higher’ centers under conditions of behavioral coping then developed.

The present findings also fit with more general conceptions of the functions of the mPFC. The mPFC is generally believed to mediate ‘executive control’, a set of processes that involves the scheduling and optimizing of subsidiary processes40. These subsidiary processes are typically viewed as residing in other cortical and perhaps limbic regions, although, as noted above, the possibility of brainstem locations has been suggested20. Here we extend to brainstem nuclei the general concept of mPFC control, with controllability as the regulator and the DRN as the target.

METHODSRats. In all experiments, rats were male Sprague-Dawley rats (Harlan Labs) weighing 275–325 g, housed four per cage on a 12-h light/12-h dark cycle (on at 0700 and off at 1900). Experiments were conducted between 0800 and 1200 h. All procedures were approved by the Institutional Animal Care and Use Committee of the University of Colorado at Boulder.

Surgery and cannulation. Surgery was carried out under anesthesia with either 60 mg kg–1 sodium pentobarbital i.p. or halothane. All rats were implanted with dual cannula guides for microinjections (26 gauge) 1 mm center-to-cen-ter distance (Plastics One). The tips of the cannulae were at the PL/IL junc-tion region of the medial prefrontal cortex (mPFC): 2.2 mm rostral to bregma, 3.5 mm ventral from the dura matter and ±0.5 mm relative to midline. For the microdialysis experiments, rats were implanted with a second cannula guide for microdialysis probes (CMA 12), with the tip terminating within the DRN: 8.3 mm caudal to bregma and 5 mm ventral from the dura matter at the mid-line41. A screw cap of a 15-ml conical centrifuge tube, whose central lid portion was removed, was also affixed to the skull so that its threads were exposed and it encircled the cannulae guide. This was done so that the skull assembly could be protected during microdialysis. Rats were allowed to recover for 1–2 weeks after surgery before experimentation.

Muscimol microinjection. Rats were injected bilaterally with 0.5 µl of either 50 ng of muscimol (Sigma) or vehicle in the mPFC. Dual 33-gauge microinjec-tors (Plastics One) attached to PE 50 tubing were inserted through the guides, from which they protruded 1 mm. The other end of the tubing was connected to a 25-µl Hamilton syringe that was attached to a Kopf microinjection unit (Model 5000). The volumes were injected over a period of about 30 s, and the injector was left in place for 2 min to allow diffusion.

Wheel-turn escape/yoked inescapable shock procedure. Each rat was placed in a Plexiglas box (14 × 11 × 17 cm) with a wheel mounted in the front and a Plexiglas rod extending from the back. The rats' tails were taped to the Plexiglas rod and affixed with copper electrodes. Rats received shocks in yoked pairs (ES and IS). The treatment consisted of 100 trials with an average intertrial interval of 60 s. Shocks began simultaneously for both rats in a pair and terminated for both whenever the ES rat met a response criterion. Initially, the shock was terminated by a quarter turn of the wheel. The response requirement was increased by one quarter turn when each of three consecutive trials was completed in less than 5 s. Subsequent latencies under 5 s increased the requirement by 50% up to a maximum of four full turns. If the requirement was not reached in less than 30 s, the shock was terminated and the requirement reduced to a single quarter turn. This procedure was used to insure that the ES rats learned an operant response. Shock intensity was 1.0 mA for the first 30 trials, 1.3 mA for the sec-ond 30 trials and 1.6 mA for the last 40 trials. Nonshocked home cage control (HCC) rats remained undisturbed in the colony, except during the microdialysis experiments, where they remained undisturbed in the dialysis room.

c-Fos immunohistochemistry. Tissue preparation. Two hours after the last tailshock, rats were deeply anesthetized with sodium pentobarbital (Nembutal) and transcardially perfused with 100 ml of 0.9% saline followed by 250 ml of 4% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4). Brains were removed and postfixed in the same fixative overnight. After postfixation, brains were transferred to 30% sucrose and stored at 4 °C until sectioning. Brains were rapidly frozen in −40 °C isopentane and 35-µm sections were obtained in a −20 °C

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

cryostat. Free-floating sections were stored at 4 °C in cryoprotectant until stain-ing. Immunohistochemistry (IHC). Immunohistochemical staining for c-Fos and 5-HT was conducted sequentially as previously described6. Briefly, staining for c-Fos was conducted first using rabbit polyclonal Fos primary antibody (1:15,000; Santa Cruz) and biotinylated goat anti-rabbit secondary antibody (Jackson Laboratories) with 3,3′-diaminobenzidine (DAB) and nickel as chromogens. Staining for 5-HT used rabbit 5-HT primary antibody (1:10,000; ImmunoStar) and non-biotinylated goat anti–rabbit IgG (Jackson Laboratories) secondary antibody with peroxidase anti-peroxidase (PAP) and DAB as chromogens. Image analysis. Sections were assessed for the number of 5-HT–stained cells and the number of cells double-labeled for both 5-HT and c-Fos. c-Fos-stained nuclei were identified by dark brown or black ovoid particles. Larger reddish-tan par-ticles, with or without Fos-stained nuclei, were counted as 5-HT positive cells. Rostral DRN sections were comparable to an anterior-posterior coordinate of −1.36 mm from interaural zero. Caudal DRN sections were comparable to the −0.7 mm anterior-posterior coordinate. Histology. Verification of cannulae placement in the mPFC was conducted on frontal sections stained with cresyl violet and visualized under a light microscope.

In vivo microdialysis. The afternoon before the experiment, a CMA 12 microdialysis probe (0.5 mm in diameter, 1 mm membrane with a 20-kD molec-ular weight cut-off; CMA/Microdialysis), was introduced through the cannula guide so that the membranous tip of the probe was within the DRN. A portion of a 15-ml Eppendorf tube was screwed onto the skull-mounted screw cap, through which the dialysis tubing, protected within a metal spring, entered and attached to the probe. Each animal was placed individually in a Plexiglas bowl (Bioanalytical Systems) and infused with isotonic Ringer’s solution (Baxter) at a rate of 0.2 µl min–1 overnight. At about 0900 the next day, the flow rate was increased to1.5 µl min–1 and a 90-min stabilization period was allowed. The infusion flow remained constant throughout the experiment. Samples were col-lected every 20 min. After stabilization, three baseline samples were collected, then muscimol or vehicle was injected in the mPFC and three additional baseline samples were collected. Next the rats were moved into a room containing the shock equipment and placed in Plexiglas wheel-turn boxes that were designed to accommodate the dialysis tubing. There they received 100 ES and yoked IS tail shocks. Five samples were collected during the session. After this, the rats were transferred back to the Plexiglas bowls where three postshock samples were collected. Two HCC control groups received bilateral mPFC injections of either vehicle or muscimol.

5-H T analysis. 5-HT concentration was measured in dialysates by HPLC with electrochemical detection. The system consisted of an ESA 5600A Coularray detector with an ESA 5014B analytical cell and an ESA 5020 guard cell. The column was an ESA MD-150 (C-18, 3 µm, 150 × 3.2 mm) maintained at 26 °C, and the mobile phase was the ESA buffer MD-TM. The analytical cell potentials were kept at −75 mV and +250 mV and the guard cell at +300mV. Dialysate (23 µl) was injected with an ESA 542 autosampler that kept the dialysates at 6 °C. External standards (Sigma) were run each day to quantify 5-HT.

Cannula and dialysis probe verification. At the end of the experiment an overdose of pentobarbital was administered and brains were removed and frozen. A cryostat was used to take 40-µm sections, which were then stained with cresyl violet for cannula placement verification. The mPFC injections were considered successful if the injector tip was within the prelimbic or infralimbic regions of the mPFC, at about 2.2 mm rostral to bregma. Only rats with the dialysis probe at least 70% within the intermediate and caudal DRN (−7.8 to −8.5 from bregma) were included.

Fear conditioning and shuttlebox escape learning. Fear conditioning and escape learning occurred in shuttleboxes using procedures previously described11. Freezing was measured for the first 5 min after placement in the shuttleboxes. Each subject’s behavior was scored every 8 s as being either freez-ing or not freezing. Freezing was defined as the absence of all movement except that required for respiration. The observer was blind with regard to treatment condition, and inter-rater reliability has been calculated to be greater than 0.92. This observation period was followed by two scrambled footshocks (0.6 mA) that could be terminated by crossing to the other side of the shuttlebox (FR-1 trials). IS does not alter FR-1 shuttlebox escape latencies29, and therefore IS and

other rats are here exposed to shocks of equal duration. FR-1 latencies were measured in the present experiment, and, as is typical, there were no group differences. These two footshocks were followed by a 20-min observation period during which freezing was scored. This observation period was followed by three further FR-1 escape trials and then 25 FR-2 escape trails. On FR-2 trials, the rats were required to cross to the other side of the shuttlebox and back to terminate each shock. It was here that IS-induced escape deficits typically occurred. Each shock terminated after 30 s if an escape response had not occurred.

ACKNOWLEDGMENTSWe would like to thank J. Rudy for many helpful comments and discussions. This work was supported by US National Institutes of Health grants DA13159 and MH50479.


Received 11 November 2004; accepted 19 January 2005Published online at http://www.nature.com/natureneuroscience/

1. Weiss, J.M. & Simson, P.G. Depression in an animal model: focus on the locus ceru-leus. Ciba Found. Symp. 123, 191–215 (1986).

2. Basoglu, M. & Mineka, S. in Torture and its Consequences: Current Treatment Approaches (eds. Basoglu, M. & Mineka, S.) 182–225 (Cambridge Univ. Press, Cambridge, Massachusetts, USA,1992).

3. Seligman, M.E. & Maier, S.F. Failure to escape traumatic shock. J. Exp. Psychol. 74, 1–9 (1967).

4. Weiss, J.M. Effects of coping responses on stress. J. Comp. Physiol. Psychol. 65, 251–260 (1968).

5. Maier, S.F. & Watkins, L.R. Stressor controllability, anxiety, and serotonin. Cognit. Ther. Res. 22, 595–613 (1998).

6. Grahn, R.E. et al. Activation of serotonin-immunoreactive cells in the dorsal raphe nucleus in rats exposed to an uncontrollable stressor. Brain Res. 826, 35–43 (1999).

7. Maswood, S., Barter, J.E., Watkins, L.R. & Maier, S.F. Exposure to inescapable but not escapable shock increases extracellular levels of 5-HT in the dorsal raphe nucleus of the rat. Brain Res. 783, 115–120 (1998).

8. Amat, J., Matus-Amat, P., Watkins, L.R. & Maier, S.F. Escapable and inescapable stress differentially alter extracellular levels of 5-HT in the basolateral amygdala of the rat. Brain Res. 812, 113–120 (1998).

9. Bland, S.T. et al. Stressor controllability modulates stress-induced dopamine and serotonin efflux and morphine-induced serotonin efflux in the medial prefrontal cortex. Neuropsychopharmacology 28, 1589–1596 (2003).

10. Bland, S.T., Twining, C., Watkins, L.R. & Maier, S.F. Stressor controllability modulates stress-induced serotonin but not dopamine efflux in the nucleus accumbens shell. Synapse 49, 206–208 (2003).

11. Maier, S.F., Grahn, R.E. & Watkins, L.R. 8-OH-DPAT microinjected in the region of the dorsal raphe nucleus blocks and reverses the enhancement of fear conditioning and interference with escape produced by exposure to inescapable shock. Behav. Neurosci. 109, 404–412 (1995).

12. Weiss, J.M., Bailey, W.H., Pohorecky, L.A., Korzeniowski, D. & Grillione, G. Stress-induced depression of motor activity correlates with regional changes in brain norepi-nephrine but not in dopamine. Neurochem. Res. 5, 9–22 (1980).

13. Grahn, R.E. et al. Blockade of alpha1 adrenoreceptors in the dorsal raphe nucleus prevents enhanced conditioned fear and impaired escape performance following uncontrollable stressor exposure in rats. Behav. Brain Res. 134, 387–392 (2002).

14. Mesulam, M.M. in Principles of Frontal Lobe Function (eds. Stuss, D.T. & Knight, R.T.) 8–30 (Oxford Univ. Press, Oxford, 2002).

15. Davidson, R.J. Anxiety and affective style: role of prefrontal cortex and amygdala. Biol. Psychiatry 51, 68–80 (2002).

16. Peyron, C., Petit, J.M., Rampon, C., Jouvet, M. & Luppi, P.H. Forebrain afferents to the rat dorsal raphe nucleus demonstrated by retrograde and anterograde tracing methods. Neuroscience 82, 443–468 (1998).

17. Vertes, R.P. Differential projections of the infralimbic and prelimbic cortex in the rat. Synapse 51, 32–58 (2004).

18. Jankowski, M.P. & Sesack, S.R. Prefrontal cortical projections to the rat dorsal raphe nucleus: ultrastructural features and associations with serotonin and gamma-aminobu-tyric acid neurons. J. Comp. Neurol. 468, 518–529 (2004).

19. Hajos, M., Richards, C.D., Szekely, A.D. & Sharp, T. An electrophysiological and neu-roanatomical study of the medial prefrontal cortical projection to the midbrain raphe nuclei in the rat. Neuroscience 87, 95–108 (1998).

20. Robbins, T.W. From arousal to cognition: the integrative position of the prefrontal cortex. Prog. Brain Res. 126, 469–483 (2000).

21. Hoffman, G.E., Smith, M.S. & Verbalis, J.G. c-Fos and related immediate early gene products as markers of activity in neuroendocrine systems. Front. Neuroendocrinol. 14, 173–213 (1993).

22. Tao, R., Ma, Z. & Auerbach, S.B. Differential effect of local infusion of serotonin reuptake inhibitors in the raphe versus forebrain and the role of depolarization-induced release in increased extracellular serotonin. J. Pharmacol. Exp. Ther. 294, 571–579 (2000).

23. Maier, S.F. Learned helplessness and animal models of depression. Prog. Neuropsychopharmacol. Biol. Psychiatry 8, 435–446 (1984).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

24. Maier, S.F. Role of fear in mediating shuttle escape learning deficit produced by inescap-able shock. J. Exp. Psychol. Anim. Behav. Process. 16, 137–149 (1990).

25. Fanselow, M.S., Lester, L.S. & Helmstetter, F.J. Changes in feeding and foraging patterns as an antipredator defensive strategy: a laboratory simulation using aversive stimulation in a closed economy. J. Exp. Anal. Behav. 50, 361–374 (1988).

26. Amat, J. et al. The role of the habenular complex in the elevation of dorsal raphe nucleus serotonin and the changes in the behavioral responses produced by uncontrollable stress. Brain Res. 917, 118–126 (2001).

27. Puig, M.V., Artigas, F. & Celada, P. Modulation of the activity of pyramidal neurons in rat prefrontal cortex by raphe stimulation in vivo: involvement of serotonin and GABA. Cereb Cortex 15, 1–14 (2004).

28. McDonald, A.J., Mascagni, F. & Guo, L. Projections of the medial and lateral prefron-tal cortices to the amygdala: a Phaseolus vulgaris leucoagglutinin study in the rat. Neuroscience 71, 55–75 (1996).

29. Maier, S.F. et al. The role of the amygdala and dorsal raphe nucleus in mediating the behavioral consequences of inescapable shock. Behav. Neurosci. 107, 377–388 (1993).

30. Graeff, F.G., Guimaraes, F.S., De Andrade, T.G. & Deakin, J.F. Role of 5-HT in stress, anxiety, and depression. Pharmacol. Biochem. Behav. 54, 129–141 (1996).

31. Yin, H.H., Knowlton, B.J. & Balleine, B.W. Lesions of dorsolateral striatum preserve out-come expectancy but disrupt habit formation in instrumental learning. Eur. J. Neurosci. 19, 181–189 (2004).

32. Corbit, L.H. & Balleine, B.W. The role of prelimbic cortex in instrumental conditioning.

Behav. Brain Res. 146, 145–157 (2003).33. Balleine, B.W. & Dickinson, A. Goal-directed instrumental action: contingency and

incentive learning and their cortical substrates. Neuropharmacology 37, 407–419 (1998).

34. Quirk, G.J., Russo, G.K., Barron, J.L. & Lebron, K. The role of ventromedial prefrontal cortex in the recovery of extinguished fear. J. Neurosci. 20, 6225–6231 (2000).

35. Quirk, G.J. Memory for extinction of conditioned fear is long-lasting and persists follow-ing spontaneous recovery. Learn. Mem. 9, 402–407 (2002).

36. Morgan, M.A., Schulkin, J. & LeDoux, J.E. Ventral medial prefrontal cortex and emotional perseveration: the memory for prior extinction training. Behav. Brain Res. 146, 121–130 (2003).

37. Quirk, G.J. & Gehlert, D.R. Inhibition of the amygdala: key to pathological states? Ann. NY Acad. Sci. 985, 263–272 (2003).

38. Royer, S. & Pare, D. Bidirectional synaptic plasticity in intercalated amygdala neu-rons and the extinction of conditioned fear responses. Neuroscience 115, 455–462 (2002).

39. Bremner, J.D. et al. Neural correlates of exposure to traumatic pictures and sound in Vietnam combat veterans with and without posttraumatic stress disorder: a positron emission tomography study. Biol. Psychiatry 45, 806–816 (1999).

40. Miller, E.K. & Cohen, J.D. An integrative theory of prefrontal cortex function. Annu. Rev. Neurosci. 24, 167–202 (2001).

41. Paxinos, G. & Watson, C. The Rat Brain in Stereotaxic Coordinates (Academic, San Diego, 1998).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Receptive field structure varies with layer in the primary visual cortexLuis M Martinez1, Qingbo Wang2, R Clay Reid3, Cinthi Pillai2, José-Mañuel Alonso4, Friedrich T Sommer5 & Judith A Hirsch2

Here we ask whether visual response pattern varies with position in the cortical microcircuit by comparing the structure of receptive fields recorded from the different layers of the cat’s primary visual cortex. We used whole-cell recording in vivo to show the spatial distribution of visually evoked excitatory and inhibitory inputs and to stain individual neurons. We quantified the distribution of ‘On’ and ‘Off’ responses and the presence of spatially opponent excitation and inhibition within the receptive field. The thalamorecipient layers (4 and upper 6) were dominated by simple cells, as defined by two criteria: they had separated On and Off subregions, and they had push-pull responses (in a given subregion, stimuli of the opposite contrast evoked responses of the opposite sign). Other types of response profile correlated with laminar location as well. Thus, connections unique to each visual cortical layer are likely to serve distinct functions.

How does connectivity in striate cortex correlate with receptive field structure and, ultimately, with neural selectivity for elements of the visual scene? Anatomical studies show that each of the six cortical layers has a unique pattern of inputs and outputs1–4. Thus, it is possible to investi-gate the function of specific components of the cortical microcircuit by comparing neural response patterns at different laminar positions5–20. We took this approach to ask whether there are response properties exclusive to the first stage of cortical integration, where new response properties such as orientation sensitivity emerge12.

Early studies suggested that orientation selectivity depends on the structure of the simple receptive field, an arrangement of elongated On and Off subregions with an antagonistic effect on one another12,21–23. This idea came from observations of responses evoked by stimuli placed at different positions in visual space. For instance, a bright contour aligned lengthwise with an On subregion produced strong excitation that diminished when the stimulus was rotated towards the orthogonal angle or was moved sideways to cover larger portions of an adjacent Off subregion12. The geometry of the simple cell’s response was thought to result from an orderly pattern of convergence from On and Off thalamic relay cells12,23–26.

Later studies suggested that the two main physiological types of cell in the visual cortex, simple and complex, were generated at all levels of cor-tical processing and represented two ends of a continuous spectrum27–32 (M.S. Jacob et al., Soc. Neurosci. Abstr. 910.13, 2003). An argument made to advance this view is that values for some parameters used to dis-tinguish simple from complex cells are distributed unimodally rather than bimodally29,32. Yet, if the distribution of values for a given set of parameters is unimodal, but all cases that fall to one side of a cutoff are

restricted to a particular layer, it could nonetheless be possible to cor-relate type of visual response with location in the cortical circuit.

Thus, to measure quantitatively the receptive fields of neurons at estab-lished laminar positions, we combined intracellular staining, whole-cell recording and a spatial mapping protocol. Over time, we were able to obtain information about anatomically identified cells in each cortical layer. We used two main measures to describe receptive field structure. First, we used an overlap index to assess the spatial segregation of On and Off subregions33. Second, we used a push-pull index to determine the presence and relative weight of antagonistic responses to stimuli of the opposite contrast within individual subfields21–23,34–38. Our find-ing is that cells with simple receptive fields, as judged by scores for both indices, are found exclusively in thalamorecipient zones, where they are the majority. Complex cells are found throughout the cortical depth, though their response characteristics change with laminar location. All told, we show that the simple receptive field is a unique feature of regions that receive thalamic input. More generally, our results support the view that each stage of the cortical microcircuit is designed to analyze differ-ent aspects of the visual stimulus.

RESULTSTo explore how cortical receptive fields vary with position in the cortical microcircuit, we mapped the spatial distribution of excitation and inhibition in the receptive fields of neurons at identified anatomical sites. We also studied thalamic relay cells, which supply visual cortex. Our sample, 88 cells in 58 adult cats, included neurons in the thalamus (n = 25), layer 4 and its borders (n = 34), layers 2+3 (n = 12), layer 5 (n = 6) and layer 6 (n = 11).

1Department of Medicine, Campus de Oza, Universidad A Coruña, 15006, Spain. 2Department of Biological Sciences, University of Southern California, 3641 Watt Way, Los Angeles, California 90089-2520, USA. 3Department of Neurobiology, Harvard Medical School, 220 Longwood Ave., Boston, Massachusetts 02115, USA. 4Department of Biological Sciences, State University of New York College of Optometry, 33 West 42nd St., New York, New York 10036, USA. 5Redwood Neuroscience Institute, 1010 El Camino Real, Menlo Park, California 94025, USA. Correspondence should be addressed to J.A.H. ([email protected]).


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Synaptic structures of receptive fieldsAn overview of the receptive fields we recorded is provided by Figures 1 and 2. Each figure is organized according to station in the microcircuit, from the thalamus, to layer 4, to the superficial layers (layer 2+3) to the deep layers (layers 5 and 6). The stimulus was sparse noise (individually flashed bright and dark squares). The receptive fields are shown as grids in which each coordinate is represented by a pair of traces that show the averaged response to bright and dark stimuli. The dashed blue and red contours outline the general shape of the Off and On subregions, respectively.

The receptive field of an Off-center thalamic relay cell is shown in Figure 1a. Within each subregion, center and surround, stimuli of the reverse contrast evoked responses of the opposite sign: a push-pull pat-tern21–23,35–37. Dark squares at the center coordinates evoked an initial depolarization followed by a hyperpolariza-tion that corresponded to withdrawal of the stimulus. Bright squares flashed at the same positions evoked the opposite response: a hyperpolarization succeeded by a depolariza-tion. The responses from the surround, though weak (small spots are suboptimal stimuli for the surround), showed a push-pull pattern as well.

The majority of receptive fields (26 of 38) in thalamorecipient zones, layer 4 and its borders, and upper layer 6, were built of adjacent On and Off subregions; each sub-region had a push-pull pattern, as in the thalamus. In cortex, however, the subregions lay side by side. This qualitative arrange-ment resembles simple receptive fields as

first described12 (Figs. 1b–d). A receptive field of a cell in layer 4 (Fig. 1b) had a strong Off subregion flanked by a smaller On subregion. Throughout the Off subregion, dark squares evoked a strong initial depolarization whereas bright squares flashed in the same positions produced a hyperpolarization. A complementary pattern was seen in the On subregion. Push-pull was present for cells with different numbers of subregions or anatomical profiles. For example, push-pull was seen in all three subregions of the receptive field of a spiny stellate cell (Fig. 1d) and throughout the receptive field of a basket cell (Fig. 1c; see refs. 35,36).

Most remaining cells (n = 37) lacked adjoining On and Off subregions, a spatial profile often termed complex9,12,34. Such cells responded in one of three main ways to the sparse-noise stimulus (Fig. 2). One pattern, typical of thalamorecipient zones, is shown for a spiny stellate neuron

DarkBright

a

c

b

d

10 mV200 ms

Figure 1 Receptive fields with a push-pull arrangement of synaptic inputs. (a–d) Receptive fields of a thalamocortical neuron in the lateral geniculate nucleus (a), two spiny cells (b,d) and a smooth cell (c), all in layer 4. The receptive fields are shown as arrays of trace pairs in which each position in the stimulus grid is represented by averages of the corresponding responses to dark (black traces) and bright (gray traces) squares. The boundaries of On (red) and Off (blue) subregions are approximated by dashed circles or ovals. In all panels, stimuli of the reverse contrast evoked responses of the opposite sign (push-pull) in each subregion. The small vertical dashes indicate the onset of the stimulus, which was flashed for 31 or 47 ms; stimulus size was 0.85° or 1.7° and grid spacing was 0.85° (that is, each square in the array represents 0.85° of visual angle).

b

d

cDark

Bright10 mV

200 ms

a

Figure 2 Receptive fields with push-push or push-null configurations. (a–c) Receptive fields of a spiny stellate cell (a) and a smooth cell (c) in layer 4, a pyramidal cell in layers 2+3 (b) and a pyramidal cell in lower layer 6 (d); conventions as for Figure 1. Excitation to bright and dark stimuli was spatially overlapping (push-push) in the receptive fields from layer 4 (a,c). Outside layer 4, cells rarely responded to both polarities of the stimulus, so receptive fields often had just one subregion (push-null) (b,d) or could not be mapped with the sparse noise (not shown). Stimulus size was 0.85° or 1.7° and grid spacing was 0.85°.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

(Fig. 2a) and for a smooth cell (Fig. 2c). For both neurons, bright and dark squares produced excitatory responses throughout the field: a push-push rather than a push-pull profile. As for all cells at the first stage of cortical processing, the time course of the response followed the temporal envelope of thalamic activity11. A second group of cells, in layers 2+3, 5 or 6, responded only to one polarity of the stimulus: a push-null profile (Fig. 2b,d). The responses of these cells are brief and irregular, as is typical of cells that do not receive contact from the thalamus11. Last, many cells failed to respond to the sparse noise, though they responded vigorously to moving stimuli11. Such cells occupied later cortical stages: the upper tier of layers 2+3, layer 5 or the bottom half of layer 6. Thus, the response profile of complex cells at the thalamocorti-cal level distinguishes them from complex cells in regions that do not receive direct afferent input.

The synaptic structure of the receptive field predicted the suprathreshold pattern of response. For simple cells (Fig. 3a), contour plots of the receptive field are shown in a single plot. For complex cells (Fig. 3b), maps of bright and dark responses are shown separately, in adjacent plots, so that On and Off responses can be compared. The receptive fields made with spikes were often smaller than those made with synaptic potentials, but the overall shape was similar. In rare instances, however, weak subregions remained subthreshold.

Spatial distribution of On and Off responsesWe used an overlap index33 to measure the spatial segregation of subregions within the receptive field (Figs. 3c,d). Values 0 indicate separated subregions and those ≈1 denote symmetrically overlapped subregions (Fig. 3c, legend). The index did not resolve potential overlap between the outermost regions of each subfield (see Methods), and only cells that responded to dark and bright spots could be included. The

cells qualitatively described as simple cells had values between –0.40 and 0.09 (–0.09 ± 0.12, mean ± s.d.; n = 26) and cells with scores from 0.32 to 0.84 (0.67 ± 0.16; n = 15) corresponded to a subset of those described qualitatively as complex.

We next compared the overlap indices from synaptic responses to those measured from spikes (Fig. 3d). (We were able to use only a subset of the population; for some cells, action potentials were blocked with QX-314, and in rare cases, subregions remained below spike thresh-old.) The distributions of the values for subthreshold (Fig. 3d, top) and suprathreshold (Fig. 3d, right) responses were similar (correlation coefficient r = 0.90; P < 0.0001). Still, most points in the scatter plot (Fig. 3d, center) that compares the two values for each cell fell below the line of unit slope; it is likely that the reduced width of the spike subfields emphasized even small disparities between the peaks of largely cospatial On and Off subfields and widened the distance between segregated On and Off subregions.

We further analyzed receptive field structure to include inhibition, using a push-pull index (measurements were restricted to the center of each subfield; see Methods). If stimuli of the opposite contrast evoked comparable amounts of push and pull, the index value was ≈0; a value ≈1 indicated push-null (numbers are absolute values) and a score ≈2 denoted push-push (Fig. 4a, bottom). Cells with separated On and Off subregions (Fig. 3a) had index values <<1 (histogram, Fig. 4a; range, 0.00–0.57; mean, 0.22 ± 0.17; n = 26). Conversely, almost all cells that lacked segregated On and Off subregions had values ≈1 (range, 0.91–1.10; mean, 0.97 ± 0.07; n = 8) or values approaching 2 (range 1.27–1.93; mean, 1.62 ± 0.20; n = 15); the two outliers had strong push-pull but only a single prominent response area.

We then compared the pattern of push and pull for simple cells to that for thalamic relay cells (Fig. 4b,c), which are widely held to have

No.

of c

ells

–0.5 0 10

7c d

On-off fieldOn subregion Off subregion

–1 0 1Overlap index (graphical explanation)

No.

of c

ells

Ove

rlap

inde

x (s

pike

s)

Overlap index (PSPs) No. of cells

0

6

0 5

1

–1

1–1

a

b

Synaptic potentials

Spikes

Synapticpotentials

Spikes

Overlap index

Figure 3 The spatial arrangement of On and Off subregions in cortical receptive fields. (a,b) Contour plots of the receptive fields of five simple (a) and four complex cells (b) compare maps constructed from synaptic potentials to those made from spikes. For simple cells, On (red) and Off (blue) responses are shown in the same plot, and for complex cells, maps of On and Off responses are shown in separate panels. Each contour was smoothed and represents a 10% decrement relative to the peak (brightest) value; the maps were thresholded by 10%. The responses constructed from spikes were normalized separately from those made from synaptic potentials. Stimulus size was 0.85° or 1.7° and grid spacing was 0.85° (that is, the space between each line on the overlay is 0.85°). Overlap index values for simple cells from left to right were (synaptic potentials and spikes, respectively) –0.20, –0.34; 0.07, –0.08; –0.16, –0.29; –0.22, –0.82; –0.18, –0.04. For complex cells, they were 0.80, 0.53; 0.79, 0.76; 0.82, 0.66; 1 polarity, 1 polarity. (c) Histogram showing the distribution of values of overlap index (bin size = 0.1) for the entire population with a graphical explanation of the index below. Only cells that responded to both polarities of the stimulus were included. The distribution of values was not unimodal (probability of rejection 0.99, Hartigan’s dip test). (d) The histograms at top and right show index values for synaptic excitation (as in a) and spikes (as in b), respectively. The central scatter plot compares the two sets of values for each cell; the red line indicates unit slope.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

segregated On and Off subregions (Fig. 1a). The values for the push-pull index, measured from the center subregion of the thalamic field, were 0.00–0.59 (0.10 ± 0.14; n = 25; Fig. 4b). To estimate the relative spatial distribution of push and pull for simple and relay cells, we expanded the use of the overlap index (Fig. 4c). For each subregion, we compared the area of push response (excitation evoked by stimuli of the preferred contrast, fitted with an elliptical Gaussian) with the corresponding pull response (inhibition evoked by stimuli of the opposite contrast, fitted with an elliptical Gaussian). The results show that the push and pull largely overlap; values ranged from 0.30–0.91 (0.59 ± 0.17; n = 26) for simple cells and 0.17–0.96 (0.72 ± 0.19; n = 25) for relay cells. Note that these values may underestimate the actual overlap because weak inhibi-tion was sometimes difficult to visualize. Overall, the scores for simple and relay cells were similar but not identical; the slight disparity might

reflect mild asymmetries in the arrangement of excitation and inhibition in the two types of receptive fields.

Laminar distribution of receptive fieldsHow do these different receptive field profiles correlate with position in the cortical microcircuit? We plotted the distribution of the overlap index (Fig. 5a), number of segregated On and Off subregions (Fig. 5b) and push-pull index (Fig. 5c) according to laminar location. The profiles show that values for each parameter vary with depth in the cortical column. Cells whose receptive fields had scores that indicated simpleness (small values of overlap and push-pull indices and multiple subregions) were located only in layer 4 and the upper half of layer 6, where the thalamic afferents terminate (see Fig. 7). By contrast, neurons whose scores indicated complexness were found in all layers, though response pattern varied with laminar location. For example, com-plex cells in thalamorecipient zones always responded to both bright and dark sparse-noise stimuli, so the push-pull and overlap indices were near the maximum values. Conversely, cells in positions farther removed from the thalamus (layers 2+3, layer 5 and lower layer 6) seldom responded to the sparse noise; when they did respond, the push-pull index values were ≈1 because responses were limited to stimuli of one polarity: either bright or dark.

The relationship between scores for the overlap and push-pull indices for cells in the different cortical layers is depicted in a scatter plot (Fig. 6; color-coded for laminar location). The resulting distribution forms two clouds that represent statistically significant groups. If simple recep-tive fields are defined as having separated On and Off subregions with

0 1 2Null-null

–1–2Pull-pull Null-pull Push-pull Push-null Push-push

(NR)

Push-pull index (graphical explanation)

0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.00

11

No.

of c

ells

Push-pull indexNR

a

*

0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.00

17

No.

of c

ells

Push-pull indexNR

Thalamic receptive fieldsSimple receptive fields

b

c

No.

of cells

–0.5 0 10

7

Overlap index(push-pull)

Figure 4 Excitation and inhibition within single subregions of the receptive field. (a) Histogram of absolute values of the push-pull index (bin size = 0.1) with a graphical explanation of the index below. Filled bars, cells with segregated On and Off subregions (overlap index ≤0.09); open bars, cells with overlapping On and Off subfields (overlap index >0.3) or with just one subregion; NR indicates that there was no response to the flash stimulus. The asterisk marks a pyramid in layer 2+3 whose dendrites extended into layer 4 and whose receptive field had push-pull in only one of two subregions. The distribution of values was not unimodal, probability of rejection 0.99 (Hartigan's dip test). (b) Comparison of values of push-pull index for thalamic receptive fields (gray) and simple cortical receptive fields with segregated On and Off subregions (black); bin size = 0.1. (c) Overlap index values of excitatory and inhibitory responses to stimuli of the opposite contrast in thalamic receptive field centers (gray) and in the individual subfields of cortical cells with separated subregions (black); bin size = 0.1.

2+3

4 (+ borders)

5

6Lam

inar

loca

tion

0

1

0 NR23

Subfield segregation

Simpleness Complexness

Overlap index

2+3

4 (+ borders)

5

6

Lam

inar

loca

tion

–0.5 0 1.0 1P NR0

1Simpleness Complexness

0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0Push-pull index

NR0

12+3

4 (+ borders)

5

6Lam

inar

loca

tion Simpleness Complexness

*

*

a

b

c

Figure 5 Correlation between receptive field structure and cortical layer. (a–c) Histograms show the distribution of values for the overlap index (a), number of subregions (b) and push-pull index (c) in the different cortical layers. 1P indicates cells that responded to only one polarity of the sparse noise, and NR denotes cells that did not respond to the static stimulus at all; bin size was 0.1 in a and c. In each histogram, for each layer, the bin with the greatest number of cells is shaded black, and the gray level in the remaining bins is normalized to that maximum. For b, the bins are labeled by the number (2 or 3) of separated On and Off subregions; 0 includes cells with overlapped On and Off subregions and those that responded to only one stimulus polarity; asterisk same as for Figure 4.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

push-pull, the plot shows that simple cells are confined to thalamor-ecipient zones. The remaining heterogeneous group of cells, which we call complex, is distributed through the cortical depth. Notably, layer 4 and bordering regions contain the cells with the greatest degree of separation between On and Off subregions (5 of 9 smooth cells and 17 of 25 spiny neurons, mostly spiny stellate cells from layer 4) as well as those with the highest degree of overlap7,11 (4 of 9 smooth cells and 5 of 25 spiny neurons, 3 pyramids at the borders of layer 4 and 2 spiny stellate cells). The remaining neurons had receptive fields composed of a single region, 2 (in layer 4) with a push-pull profile and 1 (at the 4–5 border) with a push-null profile.

Last, we used Pearson’s correlation coefficient to compare the raw responses to bright and dark stimuli point by point, as others29,32 have done to measure the segregation of On and Off subregions. The resulting distribution (not shown) was similar to that for the overlap index (r = 0.94783, P < 0.0001) as well as to that for the push-pull index (r = 0.93202, P < 0.0001).

A plot of the laminar position of cells with simple receptive fields (Fig. 7) gives the strong impression that simple cells in lower layer 4 have more compact subregions than those in the middle or upper parts of the layer, although our sample was not large enough to allow us to establish definitive sublaminar patterns. Also, multiple subregions were more common in the middle to upper half of layer 4. Last, receptive fields built of very long subregions were found in layers 4 and 6, as indicated in an earlier physiological study39.

Morphology and receptive field structureFinally, we found no systematic associations between receptive field structure and general anatomical class, except for stereotyped lami-nar variations in morphology1,3,4,10 (see summary of reconstructions, Fig. 8). We found simple (Fig. 8a) and complex (Fig. 8b) spiny stel-late cells and smooth cells in layer 4, simple pyramids at the borders of layer 4 or the upper half of layer 6, and complex pyramids and smooth cells throughout the cortical depth. On a more subtle level, our past work40 has shown that simple pyramidal cells in layer 6 have different dendritic branching patterns and axonal termination zones from complex pyramidal cells in layer 6. Perhaps future studies will show similar trends for neurons in other layers.

DISCUSSIONEach layer of cortex is characterized by a unique profile of connections1,3,4,10,40. The goal of our study was to understand the struc-ture of the receptive fields that these different circuits build. The approach we used, whole-cell recording with dye-filled electrodes, pro-vided two key advantages over traditional

extracellular recordings. First, it was possible to label the cells from which we recorded to ascertain their laminar location and, hence, their position in the microcircuit. Second, the method revealed the synaptic structure of responses by showing subthreshold excitation as well as inhibition. Two main indices, an overlap index33 and a push-pull index, allowed us to quantify the spatial relationship between On and Off responses and the presence of excitation (push) and/or inhibition (pull) within each receptive field. Cells with simple receptive fields (adjacent On and Off subregions with push-pull) were restricted to the first stage of cortical processing. The receptive fields of the remaining neurons, namely the complex cells, were heterogeneous, although stereotyped patterns correlated with separate positions in the microcircuit. Thus, different neural circuits play distinct roles in cortical processing.

Receptive field structure at the first cortical stageBy combining morphological identification with quantitative mapping of the receptive field, our experiments show that simple cells are confined to regions that receive direct thalamic input: layer 4, its borders and upper layer 6. Simple receptive fields have scores for the overlap index (<<0.1) indicating segregated On and Off subre-gions. Additionally, the scores for the push-pull index (<<1) show that stimuli of the reverse contrast evoke responses of the opposite

–0.5

01 2

1

1P

NR

NR

Push-pull index

Ove

rlap

inde

x

Layer 6Layer 5Layer 4 (+ borders)Layers 2+3

*

Figure 6 Comparison of subregion overlap and push-pull. A scatter plot of subregion overlap versus push-pull; the results are color coded for layer. The points above the plot show push-pull index values for cells that responded to only one polarity of the stimulus (1P); the label NR indicates cells (upper right) that did not respond to the sparse noise. The intersection of the crosses in each cluster of points corresponds to the mean, and the length of each line to the 95% confidence intervals (calculated with a bootstrap method). Concentric symbols are used when multiple cells shared the same coordinate; asterisk as for Figure 4.

1

4

2+3

5

6

** **

*

5o

Figure 7 Laminar distribution of receptive fields with push-pull. Receptive fields with a push-pull organization were found in layer 4, its borders or in upper layer 6, with one exception: a pyramid in layer 2+3 with dendrites extending into layer 4. The receptive fields are ordered from left to right according to depth of the soma. All but three of the receptive fields with push-pull had 2 subregions; On and Off subregions are red and blue, respectively, and asterisks indicate cells with only one obvious subregion. The scale bar (5°) indicates the size of the receptive fields.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

sign. Thus, a clear view of the simple receptive field emerges when the spatial distribution of excitation and inhibition are taken into account. Furthermore, we have found that the values of the push-pull index measured from simple cells are similar to those measured from the centers of thalamic relay cells, indicating that this arrangement is car-ried forth from geniculate to cortex. Our current study, which places simple cells in thalamorecipient zones, is consistent with the idea that simple receptive fields are built by the convergence of thalamic inputs12,23–26,41. Thus, our results can be understood in the context of feed-forward models of orientation selectivity26,37,42.

Layer 4 and its borders also contain a second, smaller population of cells (24% of the spiny neurons and 44% of the smooth cells) whose receptive fields have largely superimposed On and Off subfields (overlap index >0.35; push-pull index >1.34). Indeed, these cells have scores for the overlap and push-pull indices at the upper bounds of both distributions. Such ‘first-order’ complex cells may well correspond to types of nonsimple cells (complex and/or non-oriented concentric cells) that receive synaptic contacts from the lateral geniculate nucleus7,43. Although the layout of the receptive fields of these first-order complex cells is different from that of neighboring simple cells, both groups share a common synaptic physiology; as well the responses of all cells in layer 4 follow the time-course of thalamic drive11.

A natural question is whether the first-order complex cells are orientation selective. Recent studies have shown that inhibitory complex cells in layer 4 were not tuned for stimulus orientation and have suggested that they might serve various roles in the global regulation of excitability36,37,43. As yet, it is unclear whether spiny com-plex cells in layer 4 are orientation selective and what their functional role might be.

Finally, we wonder whether the sublaminar differences in receptive field structure that we have observed relate to the anatomical organization of the primate’s visual cortex. Specifically, we have found that simple cells with compact subfields are more common in the deeper aspect of layer 4, whereas those with narrower subregions are more frequent in the upper half of the layer. In the monkey, receptive fields in lower layer 4 are rounder than in the higher tiers, a difference

that covaries with the distribution of parvocel-lular versus magnocellular inputs13,44.

Receptive field structure at later cortical stagesThe robust On and Off responses that sparse-noise stimuli routinely evoke in layer 4 are rare in layers 2+3, 5 and lower 6. At these later stages of cortical processing, most cells respond primarily to flashed stimuli of only one polarity (push-null) or do not respond to the static stimulus at all. Evoked responses are briefer and less reliable

than in layer 4. Although cells in all layers respond vigorously to moving bars, these stimuli never evoked a push-pull pattern of response in layers 2+3, 5 or lower 6 (ref. 16). In general, responses at later stages of processing seem heterogeneous, unlike the situation in layer 4, where receptive fields divide into one of two clusters (Fig. 6). Similar differences in complex cell profile have been reported earlier, notably between the classes C2 and C1 (ref. 22). Here we extend this observation by demonstrating a correlation between response type and cortical location.

Laminar distribution of simple and complex cellsOur results support some earlier studies in the cat that have placed simple cells at the first stage of cortical processing and have found a broader distribution of various classes of complex cells7,9,12. Other studies have reported simple and complex cells throughout the cortical depth15,32 (M.S. Jacob et al., Soc. Neurosci. Abstr. 910.13, 2003). We believe this dis-crepancy reflects variations in the nomenclature or criteria used for clas-sification. First, there are different definitions of simple receptive fields, some of which include cells that are excited by only one polarity of the stimulus (S1). Thus, a cell classified as S1 by some investigators could be classified as complex by others, including ourselves. Indeed there are many terms for cells that respond to only one stimulus polarity, including complex9,12, A45, C1 or S1 (refs. 7,15,22,45), discrete complex38, Eon or Eoff46 and monocontrast13. It seems likely, therefore, that the push-null responses that we have recorded from layers 2+3, 5 or 6 might have been called simple by others.

A second method of distinguishing simple from complex cells is based on response linearity rather than spatial structure of the receptive field. Earlier results have suggested that drifting sinusoidal gratings evoke linear responses from cells with separate On and Off subregions but nonlinear responses from cells with cospatial On and Off subfields38,47. Recent work in primates, however, has challenged the assumption that the spatial structure of the receptive field and linearity of response neces-sarily correlate13. Also, we think that the cells sensitive to only one stimu-lus contrast (such as push-null) would probably respond to gratings in a linear fashion. Hence, the observation that cells with linear responses are found throughout the cortical depth is not in apparent conflict with

Figure 8 Morphology and receptive field structure. (a,b) The figure shows a sample of our three-dimensional reconstructions taken from the simple cell (a) and complex cell (b) populations. The figure shows coronal views (from left to right, top) of a pyramid in upper layer 6, a pyramid at the 4–5 border, a spiny stellate cell in layer 4, a smooth cell in layer 4 and a pyramid at the 3–4 border; and (from left to right, bottom) of a pyramid in mid layer 6, a pyramid in layer 5, two pyramids in the superficial layers; a basket cell in layer 4 and a spiny stellate cell in the same layer. Cell bodies and dendritic arbors are gray, and axons are black.

2+3

4

5

6

m1

b1

m1

m1

b1

b1

b1

b1

m1

m1

m1

b1

b1

m1b1

b1

m1

b2

b2

b2

m2

m2

m2

b2

m2

m2

m2

m2

m2

m1

m1

b1

m1

m1

m1

m1

m1

m1

b1t2

m1

b1

m1

m1

b1

b1

b1

b1

m1

m1

m1

b1

b1

m1b1

b1

m1

b2

b2

b2

m2

m2

m2

b2

m2

m2

m2

m2

m2

m1

m1

b1

m1

m1

m1

m1

m1

m1

b1t2

Simple cells

Complex cells1

1

2+3

6

5

4n

n

n

n

a

b

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

our results. Rather, it seems likely that there may be multiple ways to generate linear responses in the different cortical layers.

Are simple cells a distinct population?There is current debate about whether simple and complex cells divide into two general classes or represent two ends of a continu-ous distribution29,32,48. Although the parameters we have mea-sured are not unimodally distributed, we understand the potential for nonlinear output effects to change the shape of the distribu-tions and current limitations of recording and spatial mapping techniques (see Methods). In any case, the strength of our result is that cells with low scores for both the overlap and the push-pull indices are found only in layer 4. Had we obtained unimodal dis-tributions of both indices but found that all cells that fell below a certain cutoff were in layer 4, whereas all others were distributed in other cortical layers, our conclusions would be the same: that simple receptive fields represent an exclusive feature of the first stages of cortical processing. Thus, model circuits for simple cells in the cat should be constrained to components of the layers where thalamic afferents terminate.

Comparable studies of diverse species and sensory systemsAnatomical evidence emphasizes the importance of laminar specialization: the projection patterns of axons and dendrites grow more spatially precise with progression from rodent to carnivore to primate1–3,15. Our results contribute to the idea that connections in different layers are specialized for different tasks. Although we do not anticipate that receptive field structure per se will always vary with cortical layer, there is substantial evidence that different response properties arise at successive stages of cortical processing. The tree shrew’s visual cortex provides a dramatic example. There, orienta-tion tuning (absent in layer 4) emerges in the superficial layers, a development thought to arise from specific patterns of inter- and intralaminar convergence17,49. Likewise, in the monkey, orientation tuning and dynamics change as a function of laminar location20. Furthermore, we have previously shown that the relative orientation tuning of excitatory and inhibitory inputs varies substantially from the superficial to the deep layers in the cat16.

Systematic changes in response properties are observed in other sensory modalities as well. For example, in rodent somatosensory cortex, receptive field structure, orientation selectivity and plastic-ity change from layer 4 to the superficial and deep layers5,6,50. In the auditory cortex, response properties such as inhibitory-side band structure (analogous to the antagonistic subregions of the simple receptive field) and bandwidth seem to vary with laminar loca-tion14. Our hope is that a better understanding of the structure and function of the visual cortical microcircuit will expose fundamental principles of neocortical processing.

METHODSPhysiological preparation. Anesthetized adult cats (1.5–3.5 kg) were pre-pared as described earlier35. All procedures were in accordance with the guidelines of the US National Institute of Health and the Institutional Animal Care and Use Committees of the Rockefeller University and the University of Southern California.

Recording, data acquisition and membrane properties. The meth-ods for recordings were identical to those used in earlier stud-ies11,16,35,36,40. Voltage-current relationships were measured before and after each stimulus cycle to monitor changes in access and input resistances, spike threshold and membrane time constant (6–28 ms, smooth cells; 11–32 ms, spiny cells). Recordings lasted from 0.3 to 5 h. It was

often impractical to assign absolute resting potential, as the ratio of access to seal resistance led to a voltage division in the neural signal.

Receptive field mapping. Receptive fields were hand plotted to position the stimulus monitor and then mapped quantitatively with modified35 sparse noise21; individual light and dark squares were flashed briefly (31–47 ms) in pseudorandom order, 16 times each on a 16 × 16 square grid; stimulus size was 0.85 or 1.7°, and contrast was 50 or 70%. Plots of receptive fields were made in two ways: as contour plots, where each concentric line represents a 10% reduction in response strength, or as arrays of trace pairs. For cells with spatially segregated On and Off subregions, contour plots were made by subtracting dark from bright responses. For overlapping On and Off subfields, plots for bright and dark stimuli, centered on the same spatial coordinates, are shown separately. For the arrays of trace pairs, each position on the stimulus grid is represented by two stacked traces showing the (spike-subtracted) average to all bright and dark stimuli flashed there.

Measuring synaptic excitation and inhibition. For responses evoked from each coordinate within the receptive field, excitation (net depolarization) was defined as the area of the synaptic response that was above rest (the averaged membrane potential in the prestimulus condition) within a fixed time interval centered near the peak response (∼20–80 ms after stimulus presentation)11,35,36. Inhibition (net hyperpolarization) was defined as the area between rest and more negative voltages in the same time window. We sometimes made recordings at different membrane levels for the same cell. In those instances, the visually evoked hyperpolarization grew smaller at more negative membrane potentials, consistent with the idea that it was produced by inhibitory inputs rather than withdrawal of excitation16,35.

Measuring receptive field structure. We combined different measures to analyze the spatial relationship between excitation and inhibition in the receptive fields. First, we gauged the spatial relationship of On and Off excitatory responses by means of an overlap index33:

(1)

0.5Wp + 0.5Wn – d

0.5Wp + 0.5Wn + dOverlap index =

where Wp and Wn are the widths of the On and Off subregions, respectively, and d is the distance between the peak positions of each subregion (measured from the synaptic response areas, as above). The value of the index is ≤0 for separated subre-gions and approaches 1 for subregions that overlap symmetrically.

The parameters for the overlap index, Wp, Wn and d, were determined by separately fitting each On and Off excitatory response region with an elliptical Gaussian:

(2)f(x,y) =

A

2 ab

x'2 y'2

2a2 2b2exp(– – )

for which A determines the maximum amplitude, a and b are the half axes of the ellipse, and x′ and y′ are transformations of the stimulus coordinates x and y, taking into account the angle θ and the offset (xc and yc) of the ellipse. Thus, there were six free parameters in the fitting procedure: A, a, b, θ, xc and yc .

Any measure of subfield overlap is subject to nonlinearities that could bias results towards greater segregation (simpleness) or overlap (complexness). For instance, the method we used is based on Gaussian fits of the subregions; thus, it can overesti-mate the degree of overlap if On and Off subfields have very different amplitudes. Additionally, it cannot account for the weak borders of the subregions, which fall below the cutoff of the fitted Gaussian. Nonetheless, we prefer the overlap index to alternative measures such as the correlation coefficient32, which can exaggerate actual overlap because stimuli that straddle the borders between subregions will evoke On and Off responses simultaneously35. This problem is exacerbated by the voltage depen-dence of the amplitudes of synaptic potentials. Still, to relate our measures to others (e.g., ref. 32), we evaluated our results with Pearson’s correlation coefficient:

(Roni – Ron)(Roffi

– Roff) Σn

i = 1

(Roni – Ron)2 (Roffi

– Roff)2 Σn

i = 1 i = 1Σ

nr = (3)

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

where Ron is the response to all bright squares and Roff the response to all dark squares that fell within the receptive field.

Our second measure, the push-pull index, gauged the presence and relative magnitude of antagonistic responses to stimuli of the reverse contrast within each subfield:

Push–pull index = | P + N | (4)

where P and N represent synaptic responses to bright and the dark stimuli, respectively. The value of the index was ≈0 when stimuli of the opposite con-trast evoked excitatory and inhibitory responses of comparable magnitude (Fig. 4a, bottom). Values ≈1 indicated that stimuli of only one contrast gen-erated significant responses11 and scores ≈2 indicated that bright and dark stimuli evoked responses of the same sign and similar strength. We confined our measurement to the center of each subregion because our stimuli were often large and sometimes seemed to overlap subregions35. We normalized the mea-sures for push and pull because the relative strength of each depends strongly on membrane potential35. Although we recorded at membrane levels set to show both excitation and inhibition, it was impossible to achieve equivalent recording conditions in every cell. Notably, however, we calculated the index without normalizing and found little change in the shape of the distribution of values (not shown).

Last, we expanded the use of the overlap index to estimate the relative spatial distribution of push and pull within individual subfields (that is, to compare excitatory and inhibitory responses evoked by bright and dark spots flashed in the same subregion). If the push and pull were largely cospatial, the values of the index approached 1.

Histology. After histological processing35, labeled neurons were drawn using a computerized three-dimensional reconstruction system (Microbrightfield).

ACKNOWLEDGMENTSWe thank T.N. Wiesel for support over the years and C.G. Marshall, K.D. Naik and J.M. Provost for assistance with the anatomical reconstructions. Supported by US National Institutes of Health grant EY09593 to J.A.H.



1. Callaway, E.M. Local circuits in primary visual cortex of the macaque monkey. Annu. Rev. Neurosci. 21, 47–74 (1998).

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. Lund, J.S., Henry, G.H., MacQueen, C.L. & Harvey, A.R. Anatomical organization of the primary visual cortex (area 17) of the cat. A comparison with area 17 of the macaque monkey. J. Comp. Neurol. 184, 599–618 (1979).

4. Binzegger, T., Douglas, R.J. & Martin, K.A. A quantitative map of the circuit of cat primary visual cortex. J. Neurosci. 24, 8441–8453 (2004).

5. Brecht, M., Roth, A. & Sakmann, B. Dynamic receptive fields of reconstructed pyramidal cells in layers 3 and 2 of rat somatosensory barrel cortex. J. Physiol. (Lond.) 553, 243–265 (2003).

6. Brumberg, J.C., Pinto, D.J. & Simons, D.J. Cortical columnar processing in the rat whisker-to-barrel system. J. Neurophysiol. 82, 1808–1817 (1999).

7. Bullier, J. & Henry, G.H. Ordinal position of neurons in cat striate cortex. J. Neurophysiol. 42, 1251–1263 (1979).

8. Contreras, D. & Palmer, L.A. Response to contrast of electrophysiologically defined cell classes in primary visual cortex. J. Neurosci. 23, 6936–6945 (2003).

9. Gilbert, C.D. Laminar differences in receptive field properties of cells in cat primary visual cortex. J. Physiol. (Lond.) 268, 391–421 (1977).

10. Gilbert, C.D. & Wiesel, T.N. Morphology and intracortical projections of functionally characterised neurones in the cat visual cortex. Nature 280, 120–125 (1979).

11. Hirsch, J.A. et al. Synaptic physiology of the flow of information in the cat’s visual cortex in vivo. J. Physiol. (Lond.) 540, 335–350 (2002).

12. Hubel, D.H. & Wiesel, T.N. Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex. J. Physiol. (Lond.) 160, 106–154 (1962).

13. Kagan, I., Gur, M. & Snodderly, D.M. Spatial organization of receptive fields of V1 neurons of alert monkeys: comparison with responses to gratings. J. Neurophysiol. 88, 2557–2574 (2002).

14. Linden, J.F. & Schreiner, C.E. Columnar transformations in auditory cortex? A com-parison to visual and somatosensory cortices. Cereb. Cortex 13, 83–89 (2003).

15. Martin, K.A. & Whitteridge, D. Form, function and intracortical projections of spiny neurons in the striate visual cortex of the cat. J. Physiol. (Lond.) 353, 463–504

(1984).16. Martinez, L.M., Alonso, J.M., Reid, R.C. & Hirsch, J.A. Laminar processing of stimulus

orientation in cat visual cortex. J. Physiol. 540, 321–333 (2002).17. Mooser, F., Bosking, W.H. & Fitzpatrick, D. A morphological basis for orientation tuning

in primary visual cortex. Nat. Neurosci. 8, 872–879 (2004).18. Nowak, L.G., Azouz, R., Sanchez-Vives, M.V., Gray, C.M. & McCormick, D.A.

Electrophysiological classes of cat primary visual cortical neurons in vivo as revealed by quantitative analyses. J. Neurophysiol. 89, 1541–1566 (2003).

19. Swadlow, H.A. & Hicks, T.P. Somatosensory cortical efferent neurons of the awake rabbit: latencies to activation via supra- and subthreshold receptive fields. J. Neurophysiol. 75, 1753–1759 (1996).

20. Ringach, D.L., Shapley, R. & Hawken, M.J. Orientation selectivity in macaque V1: diversity and laminar dependence. J. Neurosci. 22, 5639–5651 (2002).

21. Jones, J.P. & Palmer, L.A. The two-dimensional spatial structure of simple receptive fields in cat striate cortex. J. Neurophysiol. 58, 1187–1211 (1987).

22. Palmer, L.A. & Davis, T.L. Receptive-field structure in cat striate cortex. J. Neurophysiol. 46, 260–276 (1981).

23. Ferster, D. Spatially opponent excitation and inhibition in simple cells of the cat visual cortex. J. Neurosci. 8, 1172–1180 (1988).

24. Chapman, B., Zahs, K.R. & Stryker, M.P. Relation of cortical cell orientation selectivity to alignment of receptive fields of the geniculocortical afferents that arborize within a single orientation column in ferret visual cortex. J. Neurosci. 11, 1347–1358 (1991).

25. Reid, R.C. & Alonso, J.M. Specificity of monosynaptic connections from thalamus to visual cortex. Nature 378, 281–284 (1995).

26. Ferster, D. & Miller, K.D. Neural mechanisms of orientation selectivity in the visual cortex. Annu. Rev. Neurosci. 23, 441–471 (2000).

27. Chance, F.S., Nelson, S.B. & Abbott, L.F. Complex cells as cortically amplified simple cells. Nat. Neurosci. 2, 277–282 (1999).

28. Borg-Graham, L.J., Monier, C. & Fregnac, Y. Visual input evokes transient and strong shunting inhibition in visual cortical neurons. Nature 393, 369–373 (1998).

29. Mata, M.L. & Ringach, D.L. Spatial overlap of On and Off subregions and its relation to response modulation ratio in macaque primary visual cortex. J. Neurophysiol. (in the press).

30. Rivadulla, C., Sharma, J. & Sur, M. Specific roles of NMDA and AMPA receptors in direction-selective and spatial phase-selective responses in visual cortex. J. Neurosci. 21, 1710–1719 (2001).

31. Tao, L., Shelley, M., McLaughlin, D. & Shapley, R. An egalitarian network model for the emergence of simple and complex cells in visual cortex. Proc. Natl. Acad. Sci. USA 101, 366–371 (2004).

32. Priebe, N.J., Mechler, F., Carandini, M. & Ferster, D. The contribution of spike threshold to the dichotomy of cortical simple and complex cells. Nat. Neurosci. 7, 1113–1122 (2004).

33. Schiller, P.H., Finlay, B.L. & Volman, S.F. Quantitative studies of single-cell prop-erties in monkey striate cortex. I. Spatiotemporal organization of receptive fields. J. Neurophysiol. 39, 1288–1319 (1976).

34. De Angelis, G.C., Ohzawa, I. & Freeman, R.D. Receptive field dynamics in central visual pathways. Trends Neurosci. 18, 451–458 (1995).

35. Hirsch, J.A., Alonso, J.M., Reid, R.C. & Martinez, L.M. Synaptic integration in striate cortical simple cells. J. Neurosci. 18, 9517–9528 (1998).

36. Hirsch, J.A. et al. Functionally distinct inhibitory neurons at the first stage of visual cortical processing. Nat. Neurosci. 6, 1300–1308 (2003).

37. Lauritzen, T.Z. & Miller, K.D. Different roles for simple- and complex-cell inhibition in V1. J. Neurosci. 23, 10201–10213 (2003).

38. Dean, A.F. & Tolhurst, D.J. On the distinctness of simple and complex cells in the visual cortex of the cat. J. Physiol. (Lond.) 344, 305–325 (1983).

39. Grieve, K.L. & Sillito, A.M. A re-appraisal of the role of layer VI of the visual cortex in the generation of cortical end inhibition. Exp. Brain Res. 87, 521–529 (1991).

40. Hirsch, J.A., Gallagher, C.A., Alonso, J.M. & Martinez, L.M. Ascending projections of simple and complex cells in layer 6 of the cat striate cortex. J. Neurosci. 18, 8086–8094 (1998).

41. Movshon, J.A., Thompson, I.D. & Tolhurst, D.J. Spatial summation in the receptive fields of simple cells in the cat’s striate cortex. J. Physiol. (Lond.) 283, 53–77 (1978a).

42. Troyer, T.W., Krukowski, A.E., Priebe, N.J. & Miller, K.D. Contrast-invariant orienta-tion tuning in cat visual cortex: thalamocortical input tuning and correlation-based intracortical connectivity. J. Neurosci. 18, 5908–5927 (1998).

43. Usrey, W.M., Sceniak, M.P. & Chapman, B. Receptive fields and response proper-ties of neurons in layer 4 of ferret visual cortex. J. Neurophysiol. 89, 1003–1015 (2003).

44. Bullier, J. & Henry, G.H. Ordinal position and afferent input of neurons in monkey striate cortex. J. Comp. Neurol. 193, 913–935 (1980).

45. Henry, G.H. Receptive field classes of cells in the striate cortex of the cat. Brain Res. 133, 1–28 (1977).

46. Toyama, K., Kimura, M. & Tanaka, K. Organization of cat visual cortex as investigated by cross-correlation analysis. J. Neurophysiol. 46, 202–214 (1981).

47. Skottun, B.C. et al. Classifying simple and complex cells on the basis of response modulation. Vision Res. 31, 1079–1086 (1991).

48. Mechler, F. & Ringach, D.L. On the classification of simple and complex cells. Vision Res. 42, 1017–1033 (2002).

49. Chisum, H.J., Mooser, F. & Fitzpatrick, D. Emergent properties of layer 2/3 neurons reflect the collinear arrangement of horizontal connections in tree shrew visual cortex. J. Neurosci. 23, 2947–2960 (2003).

50. Diamond, M.E., Huang, W. & Ebner, F.F. Laminar comparison of somatosensory cortical plasticity. Science 265, 1885–1888 (1994).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Multiple periods of functional ocular dominance plasticity in mouse visual cortexYoshiaki Tagawa1,2,3, Patrick O Kanold1,3, Marta Majdan1 & Carla J Shatz1

The precise period when experience shapes neural circuits in the mouse visual system is unknown. We used Arc induction to monitor the functional pattern of ipsilateral eye representation in cortex during normal development and after visual deprivation. After monocular deprivation during the critical period, Arc induction reflects ocular dominance (OD) shifts within the binocular zone. Arc induction also reports faithfully expected OD shifts in cat. Shifts towards the open eye and weakening of the deprived eye were seen in layer 4 after the critical period ends and also before it begins. These shifts include an unexpected spatial expansion of Arc induction into the monocular zone. However, this plasticity is not present in adult layer 6. Thus, functionally assessed OD can be altered in cortex by ocular imbalances substantially earlier and far later than expected.

Sensory experience can modify structural and functional connectiv-ity in cortex1,2. Many previous studies of highly binocular animals have led to the current consensus that visual experience is required for maintenance of precise connections in the developing visual cortex and that competition-based mechanisms underlie ocular dominance (OD) plasticity during a critical period3–6. To elucidate cellular and molecular mechanisms underlying the critical period, recent experiments have used mouse visual system as a model6–9. In mouse, retinal projections to the LGN are almost completely crossed. Consequently, visual cortex contains a large monocular zone that receives inputs only from the contralateral eye, and a small binocular zone (BZ) that receives inputs from both eyes10,11 (Fig. 1a). Within this BZ, the effects of monocular deprivation (MD) can be detected physiologically after eye closure. Microelectrode recordings have shown that 3–5 d MD during a defined critical period between P25 and P35 shifts OD towards the open eye12,13, consistent with previous observations in higher mammals. However, clear anatomical rearrangement of thalamocortical projections cannot be detected in mice without long periods of MD (40 d)14. This con-trasts with the rapid rearrangements in OD columns that occur within 4–7 d after MD in cats and monkeys15. Although there are similarities between mice and higher mammals in visual plasticity, there are also substantial differences14.

Our understanding of visual plasticity in mouse has been based primarily on single-unit extracellular recordings from cortical neu-rons, in which the relative balance of inputs representing each eye is assessed within the BZ12,13. This method requires visually driven inputs to be strong enough for postsynaptic spiking. However, visually evoked potential (VEP) recordings, thought to reflect primarily excit-atory synaptic inputs, have recently revealed surprising OD plasticity in the BZ of adult mice9. Moreover, the less-represented ipsilateral eye inputs, which could be more susceptible to visual deprivation, have not

been studied extensively. Here, a functional technique based on in situ hybridization for the immediate early gene Arc16 is used to investigate pathways representing the ipsilateral eye in developing and adult mouse visual cortex and after visual deprivation. We find multiple periods of susceptibility to visual deprivation in mouse visual cortex.

RESULTSArc maps functional connections driven by each eye in VCTo determine the spatial pattern of functional inputs representing each eye across visual cortex (VC), we used immediate early gene activation, which leaves a lasting trace in tissue sections17–19. Arc was selected because it is highly and rapidly upregulated after visual stimulation. The BZ was clearly evident after 30 min of monocular visual stimula-tion as a small patch of Arc expression at the border between V1 and V2 ipsilateral to the stimulated eye; Arc expression was present through-out VC contralateral to the stimulated eye (Fig. 1b; briefer periods did not induce robust levels of Arc mRNA by the ipsilateral eye; data not shown). The expected broader representation of the BZ at more caudal visual cortical locations14,20,21,22 was also present (Fig. 1c). Induction of Arc depended on visual stimulation, as there was no induction in binocularly enucleated mice (Fig. 1d).

Arc mRNA hybridization signal after induction was heavy in two bands of cortex corresponding to layers 2–4 (ref. 23) and layer 6, and it was lowest in layer 5 (ref. 24; Fig. 1e). Double in situ hybridization showed that induction was restricted to CaMKIIα-expressing neurons, not GAD67-expressing neurons (Fig. 1f), indicating that Arc mRNA could be induced in a large cohort of excitatory neurons25.

Arc induction was compared to the anatomical pattern of geniculo-cortical projections (Fig. 1g). As expected14,26,27, transneuronal label was widely distributed throughout layer 4 of VC contralateral to the injected eye and was restricted to a distinct patch ipsilateral to the

1Department of Neurobiology, Harvard Medical School, 220 Longwood Avenue, Boston, Massachusetts 02115, USA. 2Current address: Department of Biophysics, Kyoto University Graduate School of Science, Kitashirakawa-Oiwake-cho, Sakyo-ku, Kyoto 606-8502, Japan. 3These authors contributed equally to this work. Correspondence should be addressed to C.J.S. ([email protected]).


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

injected eye (the BZ). This pattern was similar to Arc induction pat-terns (Fig. 1b,1g), though slightly more restricted horizontally, since Arc shows ipsilateral eye activation of cortical neurons, whereas the transneuronal transport method exclusively reflects the pattern of presynaptic geniculocortical axons in layer 4.

The spatial pattern of Arc induction also reflects accurately known OD distributions in the cat, where stimulation of one eye activates vertical columns throughout VC3. Monocular TTX injection rap-idly downregulates BDNF mRNA within OD columns linked to the inactive eye28, with high expression maintained within active eye columns (Fig. 1h). An almost identical columnar pattern was pres-ent after Arc induction (Fig. 1h). To confirm that vertical columns of Arc mRNA belonged to the stimulated eye, transneuronal tracer was injected into one eye26,29, and the other eye was stimulated visually for 30 min, resulting in interdigitating columns of label representing inputs from either eye (Fig. 1i). The low Arc signal in OD columns belonging to the unstimulated eye also suggested that the visual stimulation protocol did not saturate or elevate background nonspecifically. Collectively, Arc induction provides an excellent signal-to-noise ratio, and its spatial patterns resemble known functional and anatomical distributions of inputs representing the two eyes to mouse and cat VC9,20–22.

Developmental changes in the ipsilateral representationIn the formation of OD columns in higher mammals, geniculocortical connections representing the two eyes are initially more widespread and

then are remodelled26,30,31. Transneuronal transport of WGA-HRP or [3H]proline has been used in mouse to monitor developmental changes in the ipsilateral eye representation and formation of the BZ32. However, these anatomical methods are unreliable at early ages. Therefore, we used Arc induction as a different, functional (rather than anatomical) means for assessing ipsilateral eye input at early ages.

At P14, Arc induction in response to monocular stimulation was no higher than background expression (data not shown). By P17, Arc mRNA was induced across the entire mediolateral VC contralateral to the stimulated eye (Fig. 2a), as expected given the extensive representa-tion of the contralateral eye. Notably, a similar, widespread pattern of Arc induction was also present ipsilateral to the stimulated eye. This pattern at P17 did not reflect high basal levels of Arc mRNA in immature cortex, as no Arc signal could be detected in the absence of visual stimulation (Fig. 2b). Thus, Arc signal at P17 reflects the functional representation of the stimulated eye, suggesting that the ipsilateral eye’s influence on cortical neurons is not restricted to the BZ at this early age but instead may be broadly distributed across VC. An adult-like restriction of the ipsilateral eye’s representation emerged during the next week (Fig. 2a), so that by P25, the Arc induction pattern narrowed into the BZ; this restricted pattern persisted into adulthood. In contrast, the contralateral eye representation was distributed across the entire VC at all ages.

Anatomically, at P17, retinal inputs have already segregated into LGN layers for more than 1 week33–35. It is possible, however, that ipsilateral retinal inputs are still able functionally to activate neurons throughout

Figure 1 The pattern of Arc induction reliably reflects the pathways activated by visual stimulation. (a) Illustration of mouse visual system. Binocular zone (arrowhead) is located at the border between primary (V1) and secondary (V2) visual cortex. (b) In situ hybridization showing pattern of Arc mRNA induction in P34 mouse brain after monocular visual stimulation. Arrowhead: restricted distribution of Arc mRNA in binocular zone of visual cortex ipsilateral to stimulated eye. Cx, cortex. Hc, hippocampus. Scale bars (b,c,d), 500 µm. (c) Representation of ipsilateral eye in rostral, middle and caudal regions of visual cortex as assessed by Arc induction. Arrowheads indicate BZ. (d) Lack of Arc mRNA induction after binocular enucleation. (e) Laminar analysis of Arc induction (left). Cresyl violet staining (CV) and in situ hybridization with layer-specific probes (RORβ, layer 4 specific23; ER81, layer 5 specific24) were performed on adjacent sections. Scale bar, 200 µm. (f) Visual stimulation upregulates Arc expression specifically in excitatory neurons: double in situ hybridization with 35S-labeled Arc probe and DIG-labeled cell type–specific markers (CaMKIIα, excitatory; GAD67, inhibitory). Arc expression coincides with most (87 of 91) CaMKIIα-expressing neurons (arrowheads), and never (0 of 82) with GAD67 expression (small arrows, Arc positive and GAD67 negative; large arrows, Arc negative and GAD67 positive). In this bright-field autoradiograph, silver grains appear black. Scale bar, 25 µm. (g) Pattern of thalamocortical projection visualized by transneuronal transport of [3H]proline injected into one eye (3H-AA). Note band of strong signal in cortex contralateral to injected eye (contra), and small patch (arrowhead) in cortex ipsilateral to injected eye (ipsi). Scale bar, 500 µm. (h) Coexpression of Arc and BDNF mRNAs in OD columns in cat VC. Activity in one eye was blocked for 48 h by a single intraocular injection of TTX at P43. Adjacent sections were hybridized with either Arc or BDNF riboprobe; in situ hybridization shown in dark-field optics. Merged image: green, Arc mRNA; red, BDNF mRNA. (i) Complementary pattern of OD column labeling and Arc induction in adult cat visual cortex. Geniculocortical axons in layer 4 (arrowheads) were labeled by transneuronal transport of [3H]proline (see Methods) and adjacent sections hybridized for Arc mRNA (arrows). Merged image: green, Arc mRNA; red, transneuronally transported label. Scale bar (h,i), 500 µm.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

the LGN. Thus, we carried out in situ hybridization for the immediate early gene c-fos (also known as Fos) in adjacent LGN sections (Arc is not expressed in the thalamus at any age). Within the dorsal LGN ipsilateral to the stimulated eye, c-fos induction was confined to a small zone; con-tralateral to the stimulated eye, it was induced in a large region of the LGN with a gap corresponding to the projection from the ipsilateral eye (Fig. 2c). This pattern mirrors the adult patterns of segregated retinal axons from the two eyes. Thus, the widespread pattern of Arc induction in VC is not a consequence of unrestricted activation of LGN neurons.

These observations were validated quantitatively using two methods. First, the ratio of Arc signal intensities in monocular versus binocular zones was compared at each age (Fig. 2d; Methods). The ratio did not change with age in cortex contralateral to the stimulated eye, but ipsilat-erally, the ratio decreased between P17 and P25, validating the observed decline in Arc mRNA in situ hybridization signal outside the BZ at progressively older ages. Second, to eliminate any bias introduced by having to decide the location of monocular or binocular zones at early ages, densitometric scans were made (Fig. 2e; Methods) of autoradio-graphs of Arc mRNA in situ hybridizations along layer 4 across all of VC (including V1, V2 and the intervening BZ; Fig. 1a). Arc signal induced

by the ipsilateral eye increases from a minimum in V2 to a maximum and then decreases again, a pattern evident in averaged scans at all ages examined (Fig. 2f). The width of the zone activated by the ipsilateral eye declines from P17 to P25 and remains stable thereafter (Fig. 2f,g). The intensity of the hybridization signal (Fig. 2f,g) also showed a pro-gressive increase in the strength of Arc induction between P17 and P34. Collectively, these observations suggest that between P17 and P34 there is a period of functional sculpting and strengthening of the ipsilateral eye representation to form the BZ within VC.

To further support this suggestion, retrograde tracer was injected into the VC to assess thalamocortical projections at P19 anatomically. In adult, projections from LGN neurons representing the ipsilateral eye are restricted to the BZ. If the widespread Arc induction at P19 reflects direct projections from LGN neurons, the extent of retrograde labeling in the LGN after tracer injection into the BZ should be greater than that seen at older ages. Indeed, red retrobeads injected into the lateral part of VC (Fig. 1a: BZ) labeled about two times the area of LGN at P19 than at P37 (Supplementary Fig. 1 and Supplementary Methods online). This implies that the widespread pattern of Arc induction seen at P19 reflects, at least in part, thalamocortical projections that are also

Figure 2 Developmental restriction of ipsilateral eye representation in visual cortex between P17 and P25. (a) Arc mRNA in situ hybridization after 30-min monocular visual stimulation at P17 (top), P22 (middle) or P25 (bottom) in cortex contralateral (left) or ipsilateral (right) to the stimulated eye. Scale bar, 500 µm. (b) No Arc induction was seen in the absence of visual stimulation at P17. (c) Pattern of c-fos induction in LGN after visual stimulation of right eye at P17. Top: diagram of retinogeniculate projection. Retinal axons from contralateral eye occupy most of the LGN, with a small zone receiving input from the ipsilateral eye. Bottom: in situ hybridization showing restricted pattern of c-fos mRNA induction in LGN ipsilateral to the stimulated (right) eye. Arrows indicate ipsilateral eye projection zones. dLGN is outlined. Scale bar, 300 µm. (d) Ratio of Arc mRNA signal intensity in monocular zone (MZ) relative to binocular zone (BZ). Error bars, s.d. 8–12 sections each were scanned from three mice at each age. Ipsilateral eye ratio (filled squares) decreases until P25, whereas contralateral eye ratio (open squares) does not. (e) Densitometric line scans of Arc mRNA signal in layer 4 across the entire mediolateral extent of V1 and V2 from a representative in situ hybridization section after stimulation of ipsilateral eye at P17, P25 or P96. The region of ipsilateral eye representation, determined by an automated program, is shown in each image. Raw trace (thin line) and low-pass filtered trace (solid line) are indicated. Horizontal line marks threshold; vertical lines mark borders of Arc induction above threshold in V1 and V2; scan peak is located in center of BZ (see Methods). (f) Averaged line scans at multiple ages (P17, P22, P25, P56) aligned to the background value in V2. The average scans show a similar pattern of induction as individual scans above. The signal increases from minima in V1 and V2 to a maximum. Vertical lines indicate the borders of Arc signal induction above background in V1 and V2. Horizontal lines indicate width of Arc induction at the different ages and show that width of the BZ decreases with age. (g) Top: width of zone representing ipsilateral eye in layer 4 declines until P25, as assessed in densitometric scans of autoradiographs of Arc mRNA in situ hybridizations. Bottom: Arc induction by ipsilateral eye, as assessed by densitometric scans of mRNA signal intensity in layer 4, increases between P17 and P25. At each age, 11–18 sections from three mice were scanned.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

more broadly distributed than at later ages. However, it is possible that horizontal connections within cortex also contribute to the widespread Arc induction at the youngest ages.

Arc induction detects OD shifts during a critical periodIn rodents, as in higher mammals, monocular deprivation (MD) dur-ing a critical period causes functional shifts in connectivity in favor of the nondeprived eye, as shown by microelectrode and VEP record-ings9,12,13,36. To assess whether Arc induction can detect such functional plasticity, we first examined mice after long periods (2 weeks) of depri-vation. One eye was enucleated (ME) or sutured closed (MD) at P18, and mice were reared in a normal environment until P33; mice were then placed in the dark at P33, and Arc induction was examined 24 h

later by visually stimulating either the nonde-prived eye (NDE), or the deprived eye (DE) in MD mice, for 30 min. We began deprivation at P18 based on observations (discussed above) of an early period during which the ipsilat-eral eye input is widespread. Deprivation was ended at P34 because the critical period for the effects of brief MD concludes at about this time9,12,13. The intensity of Arc induction by the ipsilateral NDE was increased after either ME or MD in these mice as compared with unmanipulated mice at P34 (Fig. 3a, top and middle, versus Fig. 3b, normal), consistent with the strengthening of ipsilateral NDE inputs shown previously by VEP or intrinsic signal optical recordings9,14.

Arc induction also extends our understand-ing of the functional organization of the DE and NDE after deprivation (Figs. 3 and 4). First, after MD, the ipsilateral, DE represen-tation in the BZ was weaker than normal, as assessed by the intensity of Arc mRNA signal, and was restricted to a very narrow region at the V1-V2 border (Fig. 3a, bottom). Many pre-vious studies have shown a weakening of the DE input to contralateral cortex during the critical period9,12–14. However, weakening of the deprived ipsilateral eye, as observed here with Arc induction, has been reported only with single-unit13 or optical imaging meth-ods14. Second, we found, unexpectedly, that the spatial extent of Arc induction by the ipsi-lateral NDE was much broader than normal, resembling the distribution seen at earlier developmental ages (compare Fig. 3a,b with Fig. 2 at P17). In contrast, signal intensity and spatial distribution of Arc induction contralat-eral to the NDE did not seem to change.

To validate these impressions, cortical sec-tions from ME or MD mice were mounted on the same glass slide with those from nor-mally reared mice and processed simultane-ously for direct, pairwise comparison (Figs. 3b,c). After ME or MD from P18 to P34, Arc induction by stimulation of the ipsilateral NDE was 200% of normal (Fig. 3c). In con-trast, Arc induction by the DE decreased to about 50% of normal (Fig. 3c).

The changes in strength and spatial extent of the ipsilateral eye rep-resentation in cortex are unlikely to be due to deprivation-induced changes in the LGN, as c-fos induction was indistinguishable in the LGN from deprived and normally reared mice, whether the stim-ulated ipsilateral eye was deprived or nondeprived (Fig. 3d). The expanded spatial distribution of Arc induction by the ipsilateral NDE might also reflect a widespread distribution of LGN projections to layer 4. Yet the geniculocortical projection pattern from the NDE to both hemispheres seems indistinguishable from normal (Figs. 1g and 3e), consistent with previous studies of mice that underwent MD for a similar duration14. Thus, the changes in Arc induction intensity and spatial extent are likely to reflect changes at or beyond thalamocortical connections.

Figure 3 Arc induction shows OD plasticity induced by ME or MD during the known ‘critical period’. (a) ME (top) or MD (middle and bottom; diagonal bar indicates deprived eye) was performed at P18, and Arc induction in VC was assessed by stimulating either the nondeprived eye (top, middle) or deprived eye (bottom) at P34 (stimulated eye is shown in white). Compared with that in normally reared mice (Fig. 1b), the spatial extent of Arc induction in VC ipsilateral to the nondeprived eye was broader (top and middle); induction in VC ipsilateral to the deprived eye was more restricted (bottom). Scale bars in a and b, 500 µm. (b) Pairwise comparison of Arc induction. Sections from ME or MD mice were paired with those from normal mice on the same glass slide and processed in parallel. Each pair shows Arc induction in the binocular zone after 30 min visual stimulation of the ipsilateral eye at P34. Top pair: nondeprived (ND) eye was stimulated after ME at P18. Second pair: MD was performed at P18 and ND eye was stimulated at P34. Third pair: MD at P18 followed by stimulation of the deprived (D) eye at P34. Bottom pair: sections from different normal mice show reproducibility of Arc induction. (c) Relative Arc induction calculated from paired images such as those in b: ME stimulus to ND, 2.30 ± 0.25, P = 0.017; MD stimulus to ND, 1.85 ± 0.34, P = 0.017; MD stimulus to D, 0.66 ± 0.10, P = 0.014; n = 7–8 pairs of sections in four mice for each treatment. Error bars, s.d. *P < 0.05, Mann-Whitney test. (d) c-fos induction in the LGN (outlined) ipsilateral to the stimulated eye after various treatments. Scale bar, 300 µm. (e) A representative autoradiograph of VC from a mouse that received ME at P18, [3H]proline injection into the ND eye at P35 and was examined at P40. The pattern of thalamocortical projection does not seem altered by ME at P18 (compare with Fig. 1g). Scale bar, 500 µm. Arrowheads in a,b,e indicate location of binocular zone.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Previous single-unit recording studies have reported a weakening of the DE inputs to contralateral cortex within the mouse BZ dur-ing but not after the critical period12,13, but in situ hybridizations for Arc mRNA did not show a similar decrease in Arc induction by the contralateral DE (compare Fig. 3a, bottom left, with Fig. 1b). To clarify this discrepancy between methods, we quantified Arc mRNA in cortex after MD from P18–34 by real-time PCR (Methods). In ipsilateral cortex after stimulation of the DE, real-time PCR detected the expected decrease in Arc mRNA (P34: 67% of unmanipulated; n = 6 pairs of mice; P < 0.05). However, in cortex contralateral to the DE, there was no significant decrease in Arc mRNA relative to unma-nipulated controls (P > 0.2). Thus, two independent methods (PCR and in situ hybridization) show that there is no detectable change in Arc mRNA contralateral to the DE. The difference between Arc in situ hybridization and single-unit recordings may be due to the fact that the single-unit recordings give a local ratio of the balance between the DE and NDE at one location (such as in the BZ), whereas the Arc method is averaged across both MZ and BZ. In addition, visual stimulation used in this study was optimized to assess the pattern of Arc induction by the ipsilateral eye; shorter visual exposures might show a weakening of Arc induction by the contralateral DE. These further considerations imply that Arc induction ipsilateral to the stimulated eye, whether the eye is deprived or nondeprived, provides a reliable assessment of the functional state of ocular inputs to mouse visual cortex.

OD formation and maintenance require visual experienceAs shown above, the representation of the ipsilateral eye in cortex is widespread at P17 and then restricted by P25 during normal devel-opment. Visual deprivation could alter normal development from P17–P25, change the representation of the ipsilateral eye at later times (P25–34) or affect both periods. To examine these possibilities, the deprivation period was subdivided into two separate 1-week periods. After MD from P17–25, Arc induction by the ipsilateral NDE extended well beyond the normal BZ (compare Fig. 4a and 2a). Normally, Arc induction at P17 is widespread, so this result suggests that MD at these early ages prevents the normal developmental restriction of ipsilateral inputs representing the NDE into the BZ. Furthermore, Arc induction ipsilateral to the DE was lower than normal, suggesting that inputs from the DE have weakened.

Because the normal adult-like restriction of the ipsilateral eye to the BZ occurs by P25 (Fig. 2), we examined the effect of later deprivation (Fig. 4b). ME or MD from P25–P34 causes widespread induction of Arc across VC after stimulation of the ipsilateral NDE; this is a genu-ine expansion of the normally restricted pattern at P25 (compare Fig. 2). Arc induction by the ipsilateral DE after MD from P25–34 was fainter than normal at P25 (Fig. 4b, bottom), suggesting a loss of connectivity.

Quantitative analyses of Arc induction in layer 4 supported these observations (Figs. 4c,d). Densitometric scans of Arc in situ hybridiza-tion signals showed that the ipsilateral NDE occupied a broader-than-

Figure 4 Visual experience alters initial formation and subsequent maintenance of ipsilateral eye representation within visual cortex. (a) ME or MD was performed at P17, and Arc induction after 30-min stimulation of ipsilateral eye in VC was assessed at P25. (b) After developmental sculpting was completed around P25 (see Fig. 2), 1 week of ME or MD could still modify the representation of the ipsilateral eye as assessed by Arc induction. Scale bar, 500 µm. (c) Representative examples of densitometric scans of Arc mRNA signal in layer 4 across the entire mediolateral extent of V1 and V2 after stimulation of ipsilateral eye under different rearing conditions. Top: ME at P25 and ND eye stimulated for 30 min at P34. Middle: MD at P25 and D eye stimulated. Bottom: normally reared until Arc induction experiment at P34. The area of ipsilateral eye representation, determined by an automated program, is shown in each image. Arrows indicate periodic fluctuation seen in MD animal (see text). Raw trace (thin line) and low-pass filtered trace (solid line) are shown (see Methods for more details). (d) Width of the area of ipsilateral eye representation in layer 4 measured from densitometric scans of Arc in situ signal. Solid line, normal development (from Fig. 2g); open squares, ND eye; filled squares, D eye. At each age, deprivation was performed 7–11 d before Arc induction experiment. For each time point, 11–18 sections from three animals were scanned. The ipsilateral, nondeprived eye was consistently greater than normal at every eye including at P96 (P < 0.001 at all ages). (e) Pattern of in situ hybridization for Arc mRNA (green) and transneuronal labeling of LGN axons in layer 4 (red) of cat visual cortex at P42. Left: columnar organization of Arc induction at P42 after visual stimulation of one eye in an unmanipulated animal; the other eye was injected with [3H]proline for transneuronal labeling of LGN projections (red). Middle: MD from P32–42 resulted in expanded Arc induction by the NDE and narrower LGN projections representing the DE in layer 4. Right: narrow columns of Arc induction by the DE; LGN projections representing the NDE were expanded. Layer 4 borders are indicated to the right of panels. Scale bar, 1 mm.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

normal territory across VC (Fig. 4c, top and bottom). ME from P17–25 also results in a wider ipsilateral NDE representation than normal in P17 (Fig. 4d). In contrast, we did not detect a change in the overall spatial extent of the ipsilateral DE representation after MD. However, there was decreased Arc signal intensity and a marked periodic fluctua-tion of the in situ hybridization signal for the ipsilateral DE (Fig. 4c, middle). Consequently, overall Arc induction as measured by normalized pixel intensity was lower: 56.8% of unmanipulated control (P = 0.009). Similar results were seen with visual deprivation from P25–34 (53.4% of unmanipulated control, P < 0.001). These observations suggest that after 1 week of MD during the critical period, the ipsilateral represen-tation of the DE within cortex weakens considerably. Thus, in mouse, deprivation-induced OD shifts can be detected in the BZ by Arc induc-tion during the known critical period (P25–P34), consistent with results from single unit recordings13. However, we now find that not only the relative physiological balance of inputs representing the DE and NDE, but also the spatial extent of the input from the ipsilateral eye, depends on visual experience during this period.

In mouse, MD during the critical period expands and strengthens Arc induction ipsilateral to the NDE, and it weakens but does not decrease the spatial extent of Arc induction ipsilateral to the DE. However, in cat or monkey after MD, OD columns representing the deprived eye shrink anatomically and physiologically3,29,30. To examine whether Arc induction could detect this species difference, cats received MD from P32 to P42 (the height of the critical period). Then Arc induction by NDE (n = 2 animals) or the DE (n = 2 animals) was assessed. Arc in situ hybridization showed that OD columns driven by stimulation of the

NDE expanded in all layers of cortex, whereas columns driven by the DE were narrower than normal (Fig. 4e). Thus, Arc induction patterns in the cat after MD reported expected changes in OD column width for both the deprived and the nondeprived eyes. Although the way

in which the functional representation of the ipsilateral DE changes after MD may differ somewhat between cat and mouse, in both cases Arc induction indicated a decrease.

OD plasticity in adult VC and before the critical periodIn mouse, single-unit microelectrode recordings have defined P25–P35 as a ‘critical period’ for OD plasticity after 3–5 d of MD; OD shifts have not been observed by this method after similar MD at older ages12,13. However, VEP recordings detected OD shifts towards the NDE after MD in adults9. To clarify these observations, we examined whether longer periods of MD or ME resulted in detectable OD shifts assessed by Arc induction. Eleven days of MD begun at P44 (data not shown) or in adults at 13 weeks of age (Fig. 5) resulted in shifts in the representation of the ipsilateral eye within VC (Fig. 5a). Arc induction after stimulation of the ipsilateral NDE appeared more intense and extended beyond the limits of the normal BZ (Fig. 5b). In contrast, Arc induction after stimulation of the ipsilateral DE was fainter than normal (Fig. 5c) (P56: 63.5% of normal, P = 0.005; 13 weeks: 72.3% of normal, P = 0.007). Densitometric scanning of layer 4 confirmed these observations quantitatively (Fig. 4d). In contrast, similar scans confined to layer 6 did not show significant changes in the width of Arc induction by the ipsilateral NDE at either P56 or 13 weeks, whereas changes in layer 6 were detected at P34 (155% of normal, P < 0.005). Together, these results suggest that a robust form of OD plasticity exists in the superficial (but not deep) cortical layers at older ages in mouse VC, at least with longer periods of monocular deprivation.

In view of the plasticity revealed by Arc induction at older ages (Fig. 4d), we sought to validate the effects of brief 3–5 d periods of

Figure 5 Arc induction shows OD plasticity earlier and beyond the known ‘critical period’. (a) Normal pattern of Arc induction at 13 weeks. (b,c) Effect of long-term (11-d) ME or MD on Arc induction. After ME, extent of Arc induction after 30-min stimulation of the ipsilateral NDE is increased and expanded (b), whereas Arc induction is decreased after stimulation of the ipsilateral, DE after MD (c). Scale bar, 500 µm. (d,e) Short-term (4-d) manipulations cause similar effects as long-term. (f) Normal pattern of Arc induction at P22. Area of Arc induction is not yet restricted to a binocular zone, but there is a clear peak of Arc signal (arrowhead). (g,h) Changes in extent of Arc induction similar to those at older ages are evident after 4-d ME or MD begun at P17. (i) Comparison of the effects of 4-d MD on Arc induction before, during or long after the ‘critical period’. MD done at P17, P28, or 13 weeks, and Arc induction assessed 4 d later. Coronal sections from MD animals were mounted on same glass slide with those from age-matched controls and processed together for in situ hybridization. In each pair, the ratio of the intensity of Arc induction by DE to normal was calculated (see Methods). P17, 0.72 ± 0.05, P = 0.014; P28, 0.63 ± 0.16, P = 0.011; 13 weeks, 0.72 ± 0.12, P = 0.041; n = 8–10 pairs of sections in four animals for each age. Error bars: s.d. *P < 0.05, Mann-Whitney test.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

MD or ME on OD during the known critical period (Fig. 5). As with microelectrode recordings, Arc induction also showed that 4 d of MD between P28 and P32 weakens ipsilateral inputs driven by the DE within the binocular zone of cortex (Fig. 5i). As with the longer-duration deprivation experiments (Fig. 4), 4 d of MD also caused the pattern of Arc induction within the BZ to become patchy, consistent with a report13 that brief MD during the critical period in mouse VC weakens ipsilateral DE within the BZ. In addition, Arc induction showed shrinkage of the spatial representation of the DE within VC after very brief periods of MD during the critical period (Fig. 5i). Both weakening and shrinkage could account for the shift in OD towards the NDE during the critical period seen with microelectrode recordings.

Notably, in mice at 13 weeks, even a 4-d deprivation resulted in clear OD plasticity: Arc hybridization signal after stimulation of the ipsilateral DE was spatially more restricted (Fig. 5e) and weaker (Fig. 5i). Real-time PCR also detected this decrease in Arc induction by the DE in ipsilateral cortex (30% of unmanipulated; P < 0.05, n = 5 pairs of mice) after 4 d of MD in adults. No decrease was seen in Arc induction by the DE in contralateral cortex (P > 0.2), consistent with the in situ hybridization observations. In addition, Arc induction increased in intensity and expanded spatially after stimulation of the ipsilateral NDE (Figs. 5a,d). These changes are clearest in the superficial cortical layers (2–4) and not obvious in layer 6. OD plasticity has been reported recently in the adult mouse VC using VEP recordings, but was thought to be confined to the BZ and to represent strengthening of the ipsilateral NDE exclusively9. Expanded representation of the ipsilateral NDE into the MZ, normally driven by the contralateral eye, after as little as 4 d of MD is an unex-pected outcome of using Arc induction to monitor OD shifts.

We also examined whether 4 d of MD at earlier ages, before the critical period, result in OD shifts. Arc induction by the ipsilateral NDE after MD from P17 to P21 was more intense and widespread in most of VC (Figs. 5f,g); Arc induction by the ipsilateral DE was weaker (Fig. 5h,i). This result implies that certain forms of OD plas-ticity have gone undetected in mouse VC using microelectrode or VEP recordings.

DISCUSSIONHere we have used Arc induction to monitor the functional representa-tion of eye input in the mouse VC during normal development and also after manipulations that create imbalances in inputs from the two eyes, such as MD or ME. Our observations using this method not only confirm several prior key observations on OD plasticity but also include unex-pected findings suggesting that mouse VC is capable of plastic changes earlier and much later than previously described. In addition, because Arc induction permits a direct functional assessment of ipsilateral eye input, we discovered a previously unknown malleability in the spatial extent of the ipsilateral eye representation within VC. MD or ME, both during and after the known critical period (P25–34; ref. 12), cause an expansion of the ipsilateral NDE representation far beyond the normal confines of the BZ, which has not been reported in any species to date. This implies that mouse VC, while presenting important opportunities for genetic manipulation, may not be identical to the VC of highly binocular mam-mals such as cat or monkey with regard to critical period mechanisms.

Arc induction reports OD reliablyThe validity of our conclusions rests on how faithfully Arc induction mirrors the functional state of inputs from the two eyes within VC. We believe this is the case. First, the original experiments in hippocampus demonstrated a faithful readout of physiological activity in Arc mRNA induction patterns16,19. Second, Arc induction patterns in adult mouse VC after monocular visual stimulation are accurate reflections of the

well-known representations of ipsilateral and contralateral eyes based on microelectrode13,22, VEP9,37 or optical14,20,21 recordings, or anatomi-cal transneuronal tracing methods14,27. Furthermore, using the tightly organized system of OD columns in cat VC, we show that stimulation of one eye results in a columnar pattern of Arc induction that coincides with known OD column patterns, both with normal rearing and also after MD during the critical period26,28. Third, Arc induction during the known critical period in mouse VC (P25–P34) reports shifts in OD expected from electrophysiological recording studies9,12,13. Fourth, cortical Arc induction patterns are unlikely to be an indirect reflection of changes in the LGN because c-fos induction in the LGN in the same experiments does not increase the amount of LGN territory activated by the open (stimulated) eye. We stress that this method augments rather than replaces other techniques, and in particular, it adds a way to assess functionally the spatial extent of ipsilateral eye representation in VC.

We report that monocular eye enucleation or brief eye closure (4 d) in the adult (13 weeks) causes a clear OD shift (Fig. 5i) as assessed by examining Arc induction in layers 2–4. This is unexpected because it indicates that adult mouse VC is capable of more plasticity than pre-viously imagined, but it supports a recent observation based on VEP recordings from the BZ also showing a shift in OD towards the open eye9. These two experiments argue for caution in interpreting OD shifts in adult mouse VC38.

The Arc induction method also provides new information about the spatial extent and laminar pattern of plasticity in each experimental condition. We find OD plasticity in layers 2–4 of adult mouse VC, but Arc induction within layer 6 after MD in adult mouse was not substan-tially different. Because previous electrophysiological recordings9 have been made only from the upper cortical layers, this essential laminar difference was overlooked. Our results suggest that in adulthood, the deeper cortical layers have more limited plasticity than the upper lay-ers. In this regard, layer 6 of adult mouse VC resembles the cortex of higher mammals39.

Ipsilateral eye representation is labileNotably, the cortical representation of the ipsilateral eye is labile, not only during development but also in adult layers 2–4. Past electrophysiological and optical imaging experiments have focused on the representation of the contralateral eye and have shown that after MD during the critical period, the deprived eye’s ability to drive cortical neurons weakens12–14. We report a similar weakening of the ipsilateral eye after MD not only during the critical period but also in adulthood. This observation does not agree with a VEP-based assessment of OD plasticity in adult mouse VC9, which reported strengthening of the NDE, but not weakening of the DE, at this late age. Despite methodological differences in the two techniques, together, the results are consistent with the persistence of OD plasticity in the superficial layers of adult mouse VC.

A new finding is that the representation of the ipsilateral NDE is capable of expanding spatially well beyond the normal limits of the BZ after MD during the critical period; this expansion can occur even in adult mice. Strengthening of NDE inputs has been reported using electrophysiological methods9, but these studies have focused on cortex within the BZ at the border between V1 and V2, missing the expansion seen in our study. In addition, Arc induction permits assessment of the state of OD in unanesthetized animals; this could help explain differ-ences with VEP and microelectrode recordings40. The mechanism for the expansion—whether it is driven by changes at thalamocortical syn-apses in layer 4, via horizontal connections within and beyond layer 4, or both—is not clear, but this expansion of the ipsilateral eye repre-sentation detected with Arc induction will be useful in characterizing molecular mechanisms of OD plasticity in mutant mice.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Re-evaluation of the critical period for MDThe Arc induction method also permits assessment of the representa-tion of the ipsilateral eye in cortex at times in neonatal development that cannot be studied using VEP, microelectrode recording, optical imaging or even anatomical methods. We have found that at early ages shortly after eye opening (P17), the ipsilateral eye functionally can acti-vate widespread areas of VC. This early widespread, and later restricted, pattern of Arc induction by the ipsilateral eye in mouse VC is reminis-cent of the remodeling of thalamocortical axons known to occur in the formation of OD columns in the VC of higher mammals1,4,31.

During this same time period (P14–P21), synaptic scaling in response to monocular deprivation occurs in layer 4 of rodent VC41,42, and 4-d MD can alter the pattern and intensity of Arc induction in VC by the ipsilateral eye, resulting in a spatially expanded representation of the NDE and a weakening in intensity and shrinkage of Arc induction by the DE (Figs. 4 and 5). Our and others’ results41,42 imply that the criti-cal period for the effects of monocular visual deprivation on cortical physiology begins at or shortly after the time of eye opening: at P14–17 rather than later. Thus, the mouse VC is capable of multiple periods of OD plasticity both during development and, in a more restricted fashion, in adulthood. Although Arc induction can show the existence of these periods, the underlying cellular and molecular mechanisms will require additional experiments that examine not only age but also laminar and spatial location within cortex.

METHODSAll experiments were performed according to the Harvard Medical School Institutional Animal Care and Use Committee Protocol.

Mouse surgery. C57Bl/6 mice were used. For monocular enucleation experi-ments, mice were anesthetized with isofluorane, one eye was removed and pieces of gelfoam inserted in the cavity. Eyelids were trimmed and sutured with 6-0 ster-ile surgical silk. For monocular deprivation experiments, eyelids were trimmed and sutured. Ophthalmic ointment (Pharmaderm) was used to prevent infection. A drop of Vetbond (3M) was put on sutured eyelids to prevent reopening.

Arc induction experiments in mouse. One eye was enucleated 24 h before visual stimulation, and mice were put in total darkness. Mice were returned to a lighted environment for 30 min in the alert condition. After light exposure, mice received an overdose of sodium pentobarbital (1 mg g–1 body weight); brains were removed, flash-frozen in M-1 mounting medium (ThermoShandon) and sectioned (coronal plane) at 16 µm for in situ hybridization.

Transneuronal transport of [3H]proline. To visualize the pattern of geniculocor-tical projections to layer 4 of cat or mouse VC, 2 mCi (cat) or 150–200 µCi (mouse) of L-[2,3,4,5-3H]proline (Amersham) was injected intraocularly26,36. At 10–14 d later (cat) or 6–8 d later (mouse), animals were given an overdose of sodium pentobarbital (200 mg kg–1 cat, 1 mg g–1 mouse); VC was cut and frozen in M-1 mounting medium (ThermoShandon). Cryostat sections (16–25 µm) were fixed in 4% paraformaldehyde/0.1 M sodium phosphate-buffered saline (PBS), pH 7.0, washed twice in PBS and dehydrated through an ethanol series. Sections were coated with NTB-2 emulsion (Kodak) and developed after 2–3 months.

Ocular dominance column labeling experiments in cat. The normal pattern of Arc induction was compared to that of the thalamocortical projection in one adult cat and one P42 cat. Arc induction was done 9 d to 2 weeks after mon-ocular [3H]proline injection (see above). Under anesthesia with inhaled isoflu-orane, 3 mM TTX solution (1 µl per 100 g body weight; Sigma) was injected into the eye that had received the [3H]proline injection previously. The cat was put in total darkness for 24 h and then returned to a lighted environment for 30 min in the alert condition. After light exposure, an intraperitoneal overdose of euthasol (200 mg kg–1 to effect) was given; the brain was removed and frozen in M-1 medium.

In a second experiment, the pattern of Arc expression in VC was compared to that of BDNF28. One eye was injected with TTX (3 mM; 1 µl per 100 g body

weight; Sigma) at P43 to block retinal activity. The cat was reared until P45 and euthanized; its brain was removed, frozen in M-1 medium and prepared for in situ hybridization. Four additional cats received MD from P32–42. Under general anesthesia with isofluorane (see above), one eye was sutured and an intraocular injection of [3H]proline was made into one eye (n = 2 animals, deprived eye; n = 2 animals, nondeprived eye). Under isofluorane anesthesia 9 d later, TTX solution was injected into the eye that had received the [3H]proline injection previously. The cat was placed in total darkness for 24 h and then returned to a lighted environment for 30 min in the alert condition. In two animals the DE was exposed to light; in the other two animals the nondeprived eye was exposed to light. Animals were euthanized and brain tissue removed as described above.

In situ hybridization. The following were used as probes: full-length mouse Arc cDNA (gift from P. Worley, Johns Hopkins University), full-length mouse Rorb (RORβ) cDNA (IMAGE Consortium Clone ID5358124; Open Biosystems), mouse c-fos (Fos) cDNA (Ambion), mouse Camk2a and GAD67 (Gad1) cDNAs (gifts from S. Nakanishi, Kyoto University), and cat BDNF cDNA28. A mouse ER81 (Etv1) cDNA was obtained by RT-PCR. In situ hybridization was per-formed as described previously28. Images were taken using dark-field optics with a cooled CCD camera (SPOT, Diagnostic instruments), then analyzed using Photoshop, NIH Image and MATLAB software. Double in situ hybridization was performed as described previously43. Sense probes yielded only background levels of signal (data not shown).

Quantification of Arc induction across multiple cortical layers. The ratio of signal intensity of Arc in situ hybridization in monocular to binocular zone (Fig. 2d) was calculated as follows: images of Arc in situ hybridization were transferred to NIH image software for box scans. A box was drawn to enclose hybridization signal in layers 2–4 of the BZ in each image, and average signal intensity of Arc in situ hybridization in the box was calculated. Adjacent areas of cortex were scanned in similar fashion through the entire primary VC, including the MZ. Ratio of signal intensity in MZ to BZ was calculated by selecting the box with minimum value and that with maximum value. Three animals (8–12 sections) were scanned at each age.

For pairwise analyses (Figs. 3c and 5i), a section from an experimental (ME or MD) animal was mounted with that from a normally reared animal on the same glass slide and processed for in situ hybridization simultaneously. In each pair of sections, mean signal intensity of Arc signal was measured in VC across all layers in V1 and V2; then relative Arc induction (experimental to normal) was calculated pairwise. At least 7–8 pairs of sections in 4 animals were scanned.

Densitometric scans of Arc induction in specific cortical layers. Quantitative analysis of Arc expression was performed in MATLAB (Mathworks) by line scans in layers 2/3, 4 or 6 (Figs. 2e and 4c). At each age, 11–18 sections from three animals were scanned. Analyses were performed blind to age and manipulation; slides from different animals and manipulations were interleaved and only reas-sembled once decoded. For each section, a line along the center of the chosen layer was generated by selecting 20–100 points and performing a cubic spline interpolation between these points. At every pixel along this line, a perpendicular line through the layer (30–60 pixels long; 1 pixel = 1.75 µm) was computed, and the average signal intensity of pixels along this line was measured. The resulting intensity line scan was low-pass filtered (7 pt triangular), generating a curve of Arc signal intensity versus distance along the layer of interest. Arc signal rose from a minimum in both V1 and V2 to a maximum within the binocular zone (Fig. 2e). BZ width was measured as the region around the intensity maximum in which signal intensity was greater than 2 s.d. of the Arc background signal intensity (determined as average intensity of 30 pixels in the region of minimum Arc induction outside the BZ; this method would, if anything, underestimate BZ width). The area of V1 and V2 in which Arc induction is at a minimum is defined as the monocular zone. To quantify the strength of Arc induction, average signal intensity above background within the BZ was computed; normalized intensity = total signal/width of BZ, where normalized intensity was interpreted as a measure of strength of eye input. In addition, scans from each age were aligned and aver-aged to generate a curve of average ipsilateral eye representation (Fig. 2f).

Real-time quantitative PCR analyses of Arc mRNA. Measurements of Arc mRNA were made after 30-min light exposure in two separate experiments:

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

(i) in visual cortex of P34 mice monocularly deprived beginning at P18 or (ii) in visual cortex of adult mice (P96) that had received 4-d prior MD. Each experi-mental condition was paired with measurements from VC receiving no manipu-lation at the corresponding age (P34 or P96) (n = 5–6 mice in each condition). Mice were euthanized and brains removed. V1 and V2 were microdissected from coronal 2-mm-thick brain slices cut on an acrylic matrix (Ted Pella) and frozen immediately on dry ice. Total RNA was isolated using Trizol (Gibco-BRL). cDNA was synthesized using the iScript cDNA Synthesis Kit (Bio-Rad). Real-time PCR reactions were carried out on a SmartCycler system (Cepheid). A reaction mix contained 1× iQ SYBR Green Supermix (BioRad), 100 nM of each oligonucleotide primer and 10 ng of cDNA in 25 µl total volume. The relative amount of Arc was normalized to the level of internal control message, for hypoxanthine phosphori-bosyltransferase (HPRT). Primers used were Arc forward, 5′-gaaggagtttctgcaata-cagtgag-3′; Arc reverse, 5′-acatactgaatgatctcctcctcct-3′; HPRT (Hprt1) forward, 5′-tgctcgagatgtcatgaagg-3′; HPRT reverse, 5′-tatgtcccccgttgactgat-3′. Real-time PCR was performed according to the comparative threshold cycle (CT) method (SmartCycler manufacturer’s instructions). Differences in threshold crossing cycle between Arc and HPRT (equal to dArc) were calculated for each condi-tion; then the levels of Arc expression were computed as 2–dArc, and ratios of Arc mRNA levels in deprived to unmanipulated cortex were computed.


ACKNOWLEDGMENTSWe thank M. Marcotrigiano, B. Printseva and Y. Kim for technical assistance, P. Worley and S. Nakanishi for providing plasmids (Arc, CaMKIIα, GAD67) and Shatz lab members for helpful discussions. This work was supported by US National Institute of Health grants to C.J.S. (NEI R01, EY02858) and P.O.K. (F32 EY1352), an Uehara Memorial Foundation fellowship to Y.T. and a Canadian Institute for Health Research fellowship to M.M.



1. Katz, L.C. & Crowley, J.C. Development of cortical circuits: lessons from ocular domi-nance columns. Nat. Rev. Neurosci. 3, 34–42 (2002).

2. Feller, E. & Feldman, D.E. Synaptic basis for developmental plasticity in somatosen-sory cortex. Curr. Opin. Neurobiol. 14, 89–95 (2004).

3. Hubel, D.H., Wiesel, T.N. & LeVay, S. Plasticity of ocular dominance columns in mon-key striate cortex. Philos. Trans. R. Soc. Lond. B Biol. Sci. 278, 377–409 (1977).

4. Katz, L.C. & Shatz, C.J. Synaptic activity and the construction of cortical circuits. Science 274, 1133–1138 (1996).

5. Crair, M.C. Neuronal activity during development: permissive or instructive? Curr. Opin. Neurobiol. 9, 88–93 (1999).

6. Hensch, T.K. Controlling the critical period. Neurosci. Res. 47, 17–22 (2003).7. Gordon, J.A., Cioffi, D., Silva, A.J. & Stryker, M.P. Deficient plasticity in the primary

visual cortex of alpha-calcium/calmodulin-dependent protein kinase II mutant mice. Neuron 17, 491–499 (1996).

8. Hensch, T.K. et al. Local GABA circuit control of experience-dependent plasticity in developing visual cortex. Science 282, 1504–1508 (1998).

9. Sawtell, N.B.et al. NMDA receptor-dependent ocular dominance plasticity in adult visual cortex. Neuron. 38, 977–985 (2003).

10. Drager, U.C. Receptive fields of single cells and topography in mouse visual cortex. J. Comp. Neurol. 160, 269–290 (1975).

11. Caviness, V.S., Jr. Architectonic map of neocortex of the normal mouse. J. Comp. Neurol. 164, 247–263 (1975).

12. Fagiolini, M., Pizzorusso, T., Berardi, N., Domenici, L. & Maffei, L. Functional post-natal development of the rat primary visual cortex and the role of visual experience: dark rearing and monocular deprivation. Vision Res. 34, 709–720 (1994).

13. Gordon, J.A. & Stryker, M.P. Experience-dependent plasticity of binocular responses in the primary visual cortex of the mouse. J. Neurosci. 16, 3274–3286 (1996).

14. Antonini, A., Fagiolini, M. & Stryker, M.P. Anatomical correlates of functional plasticity in mouse visual cortex. J. Neurosci. 19, 4388–4406 (1999).

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

16. Lyford, G.L. et al. Arc, a growth factor and activity-regulated gene, encodes a novel cytoskeleton-associated protein that is enriched in neuronal dendrites. Neuron 14, 433–445 (1995).

17. Morgan, J.I. & Curran, T. Stimulus-transcription coupling in the nervous system: involvement of the inducible proto-oncogenes fos and jun. Annu. Rev. Neurosci. 14, 421–451 (1991).

18. Kaczmarek, L. & Chaudhuri, A. Sensory regulation of immediate-early gene expression in mammalian visual cortex: implications for functional mapping and neural plasticity. Brain Res. Brain Res. Rev. 23, 237–256 (1997).

19. Guzowski, J.F., McNaughton, B.L., Barnes, C.A. & Worley, P.F. Environment-specific expression of the immediate-early gene Arc in hippocampal neuronal ensembles. Nat. Neurosci. 2, 1120–1124 (1999).

20. Schuett, S., Bonhoeffer, T. & Hubener, M. Mapping retinotopic structure in mouse visual cortex with optical imaging. J. Neurosci. 22, 6549–6559 (2002).

21. Kalatsky, V.A. & Stryker, M.P. New paradigm for optical imaging: temporally encoded maps of intrinsic signal. Neuron 38, 529–545 (2003).

22. Hubener, M. Mouse visual cortex. Curr. Opin. Neurobiol. 13, 413–420 (2003).23. Schaeren-Wiemers, N., Andre, E., Kapfhammer, J.P. & Becker-Andre, M. The expres-

sion pattern of the orphan nuclear receptor RORβ in the developing and adult rat ner-vous system suggests a role in the processing of sensory information and in circadian rhythm. Eur. J. Neurosci. 9, 2687–2701 (1997).

24. Weimann, J.M. et al. Cortical neurons require Otx1 for the refinement of exuberant axonal projections to subcortical targets. Neuron 24, 819–831 (1999).

25. Liu, X.B. & Jones, E.G. Localization of alpha type II calcium calmodulin-dependent protein kinase at glutamatergic but not gamma-aminobutyric acid (GABAergic) synapses in thalamus and cerebral cortex. Proc. Natl. Acad. Sci. USA 93, 7332–7336 (1996).

26. LeVay, S., Stryker, M.P. & Shatz, C.J. Ocular dominance columns and their develop-ment in layer IV of the cat’s visual cortex: a quantitative study. J. Comp. Neurol. 179, 223–244 (1978).

27. Drager, U.C. Autoradiography of tritiated proline and fucose transported transneuro-nally from the eye to the visual cortex in pigmented and albino mice. Brain Res. 82, 284–292 (1974).

28. Lein, E.S. & Shatz, C.J. Rapid regulation of brain-derived neurotrophic factor mRNA within eye-specific circuits during ocular dominance column formation. J. Neurosci. 20, 1470–1483 (2000).

29. Shatz, C.J. & Stryker, M.P. Ocular dominance in layer IV of the cat’s visual cortex and the effects of monocular deprivation. J. Physiol. (Lond.) 281, 267–283 (1978).

30. LeVay, S., Wiesel, T.N. & Hubel, D.H. The development of ocular dominance columns in normal and visually deprived monkeys. J. Comp. Neurol. 191, 1–51 (1980).

31. Crair, M.C., Horton, J.C., Antonini, A. & Stryker, M.P. Emergence of ocular dominance columns in cat visual cortex by 2 weeks of age. J. Comp. Neurol. 430, 235–249 (2001).

32. Kageyama, G.H. & Robertson, R.T. Development of geniculocortical projections to visual cortex in rat: evidence early ingrowth and synaptogenesis. J. Comp. Neurol. 335, 123–148 (1993).

33. Godement, P., Salaun, J. & Imbert, M. Prenatal and postnatal development of reti-nogeniculate and retinocollicular projections in the mouse. J. Comp. Neurol. 230, 552–575 (1984).

34. Muir-Robinson, G., Hwang, B.J. & Feller, M.B. Retinogeniculate axons undergo eye-specific segregation in the absence of eye-specific layers. J. Neurosci. 22, 5259–5264 (2002).

35. Pham, T.A., Rubenstein, J.L., Silva, A.J., Storm, D.R. & Stryker, M.P. The CRE/CREB pathway is transiently expressed in thalamic circuit development and contributes to refinement of retinogeniculate axons. Neuron 31, 409–420 (2001).

36. Drager, U.C. Observations on monocular deprivation in mice. J. Neurophysiol. 41, 28–42 (1978).

37. Porciatti, V., Pizzorusso, T. & Maffei, L. The visual physiology of the wild type mouse determined with pattern VEPs. Vision Res. 39, 3071–3081 (1999).

38. Pizzorusso, T. et al. Reactivation of ocular dominance plasticity in the adult visual cortex. Science 298, 1248–1251 (2002).

39. Hubel, D.H. & Wiesel, T.N. The period of susceptibility to the physiological effects of unilateral eye closure in kittens. J. Physiol. (Lond.) 206, 419–436 (1970).

40. Pham, T.A. et al. A semi-persistent adult ocular dominance plasticity in visual cortex is stabilized by activated CREB. Learn. Mem. 11, 1–10 (2004).

41. Desai, N.S., Cudmore, R.H., Nelson, S.B. & Turrigiano, G.G. Critical periods for experience-dependent synaptic scaling in visual cortex. Nat. Neurosci. 5, 783–789 (2002).

42. Turrigiano, G. G & Nelson, S. B. Homeostatic plasticity in the developing nervous system. Nat. Rev. Neurosci. 5, 97–107 (2004).

43. Kaneko, S. et al. Synaptic integration mediated by striatial cholinergic interneurons in basal ganglia function. Science 289, 633–637 (2000).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Hierarchical and asymmetric temporal sensitivity in human auditory corticesAnthony Boemio1, Stephen Fromm2, Allen Braun2 & David Poeppel3

Lateralization of function in auditory cortex has remained a persistent puzzle. Previous studies using signals with differing spectrotemporal characteristics support a model in which the left hemisphere is more sensitive to temporal and the right more sensitive to spectral stimulus attributes. Here we use single-trial sparse-acquisition fMRI and a stimulus with parametrically varying segmental structure affecting primarily temporal properties. We show that both left and right auditory cortices are remarkably sensitive to temporal structure. Crucially, beyond bilateral sensitivity to timing information, we uncover two functionally significant interactions. First, local spectrotemporal signal structure is differentially processed in the superior temporal gyrus. Second, lateralized responses emerge in the higher-order superior temporal sulcus, where more slowly modulated signals preferentially drive the right hemisphere. The data support a model in which sounds are analyzed on two distinct timescales, 25–50 ms and 200–300 ms.

Structure, function and lateralization in human auditory cortex are the focus of much recent work1. One central issue concerns the origin and nature of lateralization. For example, there are subtle anatomic and physiological asymmetries in the afferent pathway, but compelling functional asymmetries attributable to cortical processing2,3. Where do such asymmetries originate? One hypothesis proposes that functional lateralization arises from differences in the early spectrotemporal computations performed in auditory cortices that transform sensory representations of signals into more abstract perceptual codes. A prevailing model is that temporal features are processed predominantly in the left hemisphere and spectral features in the right4. A second and different source of lateralization derives from the nature of the stored representations that the transformed sensory information must interface with for further processing—for example, lexical information in the left and affective prosodic information in the right hemisphere.

Both explanations have led to the notion that speech—whether resulting from lateralization of stored lexical representations or from early auditory cortical specialization for processing temporal signal attributes—is preferentially processed within the left hemisphere5–7, whereas processing of dynamic pitch and prosody—whether resulting from lateralized representation of higher-order phrase-level intonation or specialized analysis of spectral information—is carried out in the right hemisphere8–10.

We argue that both hemispheres together—including left and right non-primary auditory areas—participate in one critical intermediate com-putation, the analysis of the auditory signal on multiple timescales3,11,12,

with the relevant scales being 25–50 ms and 200–300 ms (ref. 3). In addi-tion, we propose that functional lateralization emerges from differential connectivity patterns linking temporal cortices along the afferent path-way such that information processed on the longer timescales is routed predominantly to higher-order right hemisphere cortices, whereas information resulting from processing on the shorter timescale primar-ily projects to the left. To evaluate these hypotheses, we varied a single stimulus para-meter, the temporal structure, and looked for differential activation along the afferent pathway and between the two hemispheres. The present design controls for potential spectral confounds in a fashion that was not possible in previous studies in which stimuli varied both temporally and spectrally4,13.

Fifteen participants listened passively to non-speech stimuli (Fig. 1) while we recorded the hemodynamic responses from the entire brain using a single-trial sparse acquisition fMRI design14,15. We cre-ated 9-s auditory signals by concatenating short-duration narrowband noise segments, spanning a range of segment transition rates from 3 to 83 segments per second, encompassing the syllabic to segmen-tal transition rates of speech16 (Fig. 1a,b; examples can be heard in Supplementary Audios 1–5 online). Each segment had a bandwidth of 125 Hz and a segment center frequency spanning a half-octave range from 1,000 Hz to 1,500 Hz. The bandwidth was chosen to be within the critical band at that frequency17 and interpretable in the context of the rate and bandwidth of speech formants16. Local spectrotem-poral variations were introduced by constructing two types of seg-ments. In one, the frequency remained constant throughout the signal

1Laboratory of Brain and Cognition, National Institute of Mental Health, National Institutes of Health, Bethesda, Maryland 20892, USA. 2National Institute of Deafness and Other Communication Disorders, National Institutes of Health, Bethesda, Maryland 20892, USA. 3Department of Linguistics and Department of Biology, University of Maryland, College Park, Maryland 20742, USA. Correspondence should be addressed to D.P. ([email protected]).


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

(TN; Fig. 1b); in the other, frequency was swept linearly upward or downward randomly (FM; Fig. 1a). A control stimulus (CN; Fig. 1c) was constructed from a single 9-s TN segment with center frequency in the middle of the half-octave range.

We show that early and higher-order auditory cortical areas are exqui-sitely sensitive to temporal structure bilaterally. In addition, local spectro-temporal structure is differentially processed within the superior temporal gyrus. Finally, in higher-order superior temporal sulcus, slowly modulated signals preferentially drive the right hemisphere. To account for these observations, we present a model involving cortical processing of audi-tory signals on short and long timescales, and a hypothesized differential connectivity pattern from lower- to higher-order auditory areas.

RESULTSThe stimuli were effective at activating auditory cortex selectively and robustly (Fig. 2). Two independent analyses of the fMRI data were car-ried out, one at the cohort level, the second a region-of-interest (ROI) analysis at the level of individual subjects. Both analyses were based on categorical contrasts between each of the six FM and TN stimuli and the one CN stimulus. The cohort analysis produced a single set of activation maps across all subjects, whereas the single-subject ROI analysis produced one set of maps for each subject. These contrasts were designed to identify cortical regions sensitive to segmental struc-ture. Given the range of controls designed into the stimuli, segmental structure should be the primary feature driving the response.

Cohort analysisContrasts between the active FM and TN stimuli and the CN control yielded bilateral activations (SPM 99, P < 0.05, corrected) in the transverse temporal gyrus (TTG), supe-rior temporal gyrus (STG) and superior temporal sulcus (STS). All areas showed a strong effect of segment duration, with longer durations yielding greater activa-tion (Fig. 2). Table 1 shows the Montreal Neurologic Institute (MNI) coordinates and number of suprathreshold voxels for all conditions producing suprathreshold activation. Several effects are visible: (i) the increasing spatial extent of activation from 12 to 300 ms for all conditions (ii) the greater activation observed for TN as compared to FM type for short segment duration, and (iii) the hemispheric asymmetry, in which

Figure 1 Concatenated narrow-band noise stimuli. (a,b) Spectrograms of (a) 6 FM type stimuli and (b) 6 TN type stimuli. Frequency (Hz) is plotted on the ordinate, time (ms) on the abscissa. Mean segment duration (ms) is shown at the top of the FM plots. Dotted lines connecting the segments in the 160 ms plots denote the common segmental structure of FM and TN type stimuli. (c) Spectrogram of CN stimulus. Although all 13 conditions were 9 s long in the study, only 1 s is shown here for clarity. (d) Temporal profiles of the six FM stimuli, obtained by summing over frequency, showing the common amplitude envelope across segment duration. (e) Spectral profiles of the 300-ms FM and TN segment types and CN stimulus, obtained by collapsing over time, showing the similar long-term spectrum across segment type. Amplitude (Amp) values are in arbitrary units (a.u.). (f) Number of segments contained in each of the six FM and TN conditions as a function of segment duration. Distributions are Gaussian with clipped tails to avoid overlap between conditions. Note the log scale on the abscissa.

Figure 2 Surface-mapped activation from the cohort analysis shown on the inflated N27 brain. Segment duration (ms) is specified on the left, segment type (FM versus TN) and hemisphere at the top. Threshold corresponds to P < 0.05, corrected.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

longer segment duration stimuli elicited greater activation in the right hemisphere.

Inspection of the activation maps show that the majority of the acti-vation was limited to STG and STS, with the remainder being generated by TTG, the putative core human auditory cortex18. The small activa-tion of TTG was not due to its smaller size relative to STG or STS, but to a paucity of suprathreshold voxels compared to the total number of voxels comprising this area. The modest activation is likely to be due to the nature of the contrast used, which was designed to identify areas that explicitly code temporal structure, because all stimuli had similar spectral envelope (Fig. 1e), temporal envelope (Fig. 1d) and RMS (root mean square) power. Thus, it seems that TTG codes signal properties common to all the stimuli, whereas areas STG and STS represent signal properties unique to the TN and FM types.

To obtain the perceptual correlate of the hemodynamic responses, subjects rated the perceptibility of the individual segments in an offline, single-interval, two-alternative forced-choice task. Figure 3 shows the results displayed as the proportion of responses for which individual segments were perceived. These data are plotted with the normalized hemodynamic response, expressed as the total number of suprathresh-old voxels from the cohort analysis. The two curves—which can be interpreted as transfer functions between the acoustic properties of the stimulus and the perceptual and physiological (hemodynamic) responses they elicit—are very similar, indicating a tight correlation between perception and underlying physiology as assessed here.

Single-subject ROI analysisResponses in TTG, STG and STS were subsequently analyzed on a subject-wise basis in the three ROIs identified in the cohort analysis. ROIs were defined by Talairach Daemon classification of the union of suprathresh-old voxels across all conditions from the cohort analysis. The mean con-trast value (mean voxel strength) in all three ROIs was computed for each condition and subject and then submitted to a repeated-measures ANOVA with the factors hemisphere, segment type and segment dura-tion. The results showed several robust effects. First, all three regions showed a compelling main effect of segment duration (TTG, P < 0.001; STG, P < 0.001; STS, P < 0.001; Fig. 4a), consistent with the cohort analy-sis. Second, a strong main effect of segment type was observed in STG (P < 0.001; Fig. 4b) but not in TTG (P < 0.12) or STS (P < 0.13), as was an interaction between segment type and segment duration (P < 0.05; Fig. 4c). The effect is greatest at 45 ms SOA (stimulus onset asynchrony) and falls off on both sides (Fig. 4, inset). Post-hoc testing showed that neighboring conditions rose to significance (P < 0.05) (asterisks) when compared to the 300 ms SOA condition. Third, and crucially, for STS alone, we observed a significant main effect of hemisphere (P < 0.001, Fig. 4d) as well as a significant interaction between segment duration and hemisphere (P < 0.001, Figs. 4e and 5), with slowly varying signals preferentially driving the right hemisphere. In contrast, in TTG and STG, neither the main effect of hemisphere (TTG, P < 0.86; STG, P < 0.33) nor the duration × hemisphere interaction reached significance (TTG, P < 0.14; STG, P < 0.07).

In summary, the ROI analysis showed a progressive elaboration of the representation of the acoustic signal along the afferent pathway from TTG to STG to STS (Table 2), in which TTG was weakly activated, STG showed a sensitivity to segment type (or stimulus identity, based on local spectrotemporal cues), and STS showed a hemispheric asymmetry to segment duration (or stimulus rate, based on durational cues).

DISCUSSIONWe focus on three important findings in turn: (i) the bilateral response to temporal structure (Figs. 2 and 4a) (ii) the differential sensitivity in

STG to FM and TN signals (Figs. 4b,c) and (iii) the hemispheric asym-metry in STS as a function of segment duration (Figs. 4e and 5).

Bilateral sensitivity to temporal structureA robust main effect of segment duration (Fig. 4a) was obtained in both hemispheres of all three ROIs, similar to the overall response pattern shown in the cohort analysis (Fig. 2). The observation that varying a narrowband signal along a single (temporal) dimension induces differ-ential processing in non-primary auditory cortices substantially extends the range of cortex to which we must attribute temporal sensitivity. The finding is consistent with prior data11,19 but new in demonstrating the extensive contribution of non-primary areas. Typically, core auditory fields18 are studied to investigate the representation of signals19–22. We show that temporal lobe structures within and beyond belt and parabelt projection areas also reflect the temporal properties of the stimulus19. Furthermore, both left and right non-primary areas showed a high sensitivity to this temporal structure, suggesting that the rightward lateralization of spectral (for example, pitch change) sensitivity2,4,8,9

Table 1 Cohort analysis summary

Left Right

Seg dur (ms) FM TN FM TN

12 – – – –

25 – –56 –28 4

6.88 (64)

–

6.36 (10)

52 –16 0

45 –40 –36 8

4.91 (7)

–56 –24 4

8.16 (97)

48 –20 0

5.30 (10)

52 –12 0

8.95 (61)

85 –60 –28 8

7.14 (58)

–60 –28 4

8.36 (136)

52 –8 0

8.27 (44)

56 –12 0

8.8 (163)

160 –56 –24 4

10.94 (242)

–56 –28 4

8.61 (226)

52 –12 0

9.58 (256)

52 –12 0

9.77 (328)

300 –60 –28 8

10.60 (252)

–56 –28

10.38 (248)

452 –12 0

9.52 (314)

52 –12 0

8.56 (229)

Seg dur, segment duration. Activation is organized by condition and hemisphere. Values in the cells in the left-most column denote segment duration in ms. Remaining cells contain (i) the MNI coordinates of the most activated voxel for the specified condition (ii) the t-score of the most activated voxel, and (iii) the total number of suprathreshold voxels at P < 0.05 corrected across all areas (parentheses). No suprathreshold activation was observed at 12 ms SOA for any condition.

Figure 3 Comparison of physiological (cohort analysis) and behavioral responses. Normalized aggregate hemodynamic response (left, solid line) and the proportion of individuated segments judgments (right, dashed line) are plotted together to show the tight correlation between perception and neural representation of stimulus segmental structure. The hemodynamic response was computed by normalizing the number of suprathreshold voxels from the cohort analysis across all 12 contrasts. The proportion of individuated segments was obtained by summing the number of times that segments were classified as perceptible for each condition and for each subject, normalizing across conditions, and averaging across all subjects.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

does not come at the cost of right auditory areas’ sensitivity to tim-ing information—a finding that may have important implications for theories of speech perception.

STG sensitivity to local spectrotemporal featuresSTG was strongly activated in both the cohort (Table 1) and ROI analy-ses, where main effects of segment duration (Fig. 4a) and segment type (Fig. 4b) were observed, in addition to an interaction between segment type and segment duration (Fig. 4c). Collectively, these results show that left and right STG are sensitive to both the local temporal structure (seg-ment duration) and the spectral structure (segment type) of the stimuli.

We argued that auditory signals are analyzed simultaneously on at least two independent timescales, namely 25–50 ms and 200–300 ms; here we provide the physiological basis for these claims. The notion of processing on a particular timescale here refers to the integration of auditory information over time and across frequency in cortex.

Evidence for processing on the 25–50 ms timescale derives prima-rily from the difference observed in the hemodynamic response as a function of both segment duration and segment type (Fig. 4b,c). This difference is greatest at 45 ms SOA (Fig. 4c, inset). The concept of a temporal window 25–50 ms in width—independently supported by psychophysical3 and neurophysiological21 considerations—provides an explanatory basis for this effect when considered in the context of the segmental structure of the stimuli. For long segment duration, an FM segment sweeps across the half-octave (1,000–1,500 Hz) frequency range in the longest amount of time, making its spectral slope (Hz/s) the least steep, and thus most like the TN segment, which has a slope of zero. At this segment duration, the smallest difference between the hemody-namic responses of the two segment types is observed (Fig. 4c). As the segment duration decreases, the difference between the two segment

types within the integration window increases: the FM segment sweeps across the same fre-quency span in less time. At 45 ms SOA, the duration of our hypothetical window, exactly one FM segment ‘fits’ within the window (spanning the entire half-octave frequency range), making it maximally different from the TN segment, and this corresponds to the largest observed difference in hemodynamic response between the two segment types. For further decreases in segment duration, the win-dow will contain more than one segment, and these begin to fuse into a single homogenous

stream, reducing the difference between the neural representation of the two segment types and thus the differential hemodynamic response.

Although analysis on the 25–50 ms timescale is manifested as a differ-ence between the hemodynamic responses of the FM and TN segments—suggesting processing dedicated to enhancing differences in transient signals like speech sounds—it might also reflect extraction of segment boundaries. Detection and explicit representation of the abrupt changes in spectrotemporal structure are important because these are potent cues in the estimation of the onset and offset of sound sources23.

Evidence supporting processing on the timescale of 200–300 ms is derived from the asymptote in the hemodynamic response for SOA >160 ms, evident in both the cohort (Fig. 3 and Table 1) and ROI (Fig. 4c) analyses, each of which were based on different measures of the hemodynamic response. Asymptotic behavior was observed in both the spatial extent (number of suprathreshold voxels) and magnitude (mean voxel strength) of activation. Here again the concept of a temporal analy-sis window—in which a restricted portion of the auditory representa-tion is processed—provides the basis for interpreting the physiological response. However, the specific features computed on this timescale are likely to differ from those computed on the 25–50 ms scale. For example, analysis within a 200–300 ms window may be optimized to represent

Figure 4 ROI analysis: STG sensitivity to segment type and STS hemispheric asymmetry. Normalized hemodynamic response (ordinate) for all plots represents the mean voxel strength of the contrast within the specified area. Values were averaged across subjects within each ROI and then normalized across conditions. When segment duration is plotted on the abscissa, it is in log scale. (a) Main effect of segment duration in all areas; longer segment duration produced greater activation. (b) Main effect of segment type in STG; overall, TN conditions produced greater activation than FM conditions. (c) Interaction between segment duration and segment type in STG ( left-TN, right-TN, left-FM, ∆ right-FM). Inset shows the difference between TN and FM conditions as a function of segment duration. The greatest difference occurs at 45 ms SOA. Asterisks denote significant differences relative to 300 ms SOA (P < 0.05). (d) Main effect of hemisphere in STS; overall, greater activation was observed in the right hemisphere. (e) Interaction between hemisphere and segment duration in STS ( left-TN, right-TN, left-FM, ∆ right-FM). The greatest disparity occurs at the longest segment durations where activation in right STS exceeds that in the left (Fig. 5).

Table 2 Summary of single-subject ROI analysis

Left Right

Seg dur Seg type Hemi asym Timescales Seg dur Seg type Hemi asym Timescales

TTG Yes No No <25 ms? Yes No No <25 ms?

STG Yes Yes No 25–50 ms, 200–300 ms

Yes Yes No 25–50 ms, 200–300 ms

STS Yes No Yes ≥200–300 ms Yes No Yes ≥200–300 ms

Seg dur, segment duration; seg type, segment type; hemi asym, hemispheric asymmetry. All three areas (TTG, STG and STS) showed a robust effect of segment duration. STG alone showed a main effect of segment type and an interaction between seg-ment type and segment duration. In STS, a main effect of hemisphere and an interaction of hemisphere and segment duration were observed. Time-scale columns denote the hypothesized period over which auditory information is combined across time and across frequency in cortex. Longer processing times are associated with greater displacement along the afferent auditory pathway (see Discussion for details).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

perceptual objects such as syllables and to mediate the processing of dynamic pitch and sentential prosody. In contrast, processing on the 25–50 ms timescale may be involved in extracting segment bounda-ries subserving (sub)segmental processing. Thus, the existence of two independent processing timescales facilitates the extraction of different features in a manner not possible with a single timescale. This greatly increases the efficacy of the transformation from an early sensory code—isomorphic with the acoustic signal—to a more stable percep-tual code—isomorphic with the percept—comprising features extracted during earlier processing.

The longest segment duration used in this study was 300 ms, and thus it is not possible to determine with certainty what happens for segment durations >300 ms. Will the asymptote in hemodynamic response and number of suprathreshold voxels observed between 160–300 ms persist, or will these values ultimately decrease to baseline? To answer this ques-tion, recall that the contrasts between the 12 active FM and TN condi-tions and the 1 CN control condition compared the hemodynamic response elicited by stimuli with many segments to a stimulus with just one (9-s) segment. Thus, it is certain that, if the segment duration of the FM and TN conditions were increased to values approaching 9 s, the contrast between these and the CN condition eventually would fall to zero because the hypothetical contrast would be comparing increasingly similar entities. So the question becomes, “Is the observed asymptote in response a relative maximum or an absolute maximum?” Based on previous electrophysiological24–26 and psychophysical27 evidence, we predict that the asymptote between 160–300 ms is an absolute maxi-mum and that the hemodynamic response will decrease to baseline as the segment duration is increased beyond 300 ms.

Hemispheric asymmetry in STSSTS—recently described as constituting the fourth level of processing in the primate auditory system28—showed a strong main effect of segment duration, like STG, demonstrating sensitivity to temporal structure over a range of more than an order of magnitude. In con-trast to STG, however, STS was found not to be sensitive to segment type, and it showed a smaller asymptote in hemodynamic response for segment duration >160 ms. Yet a marked hemispheric asym-metry was observed (Figs. 4d and 5), as was an interaction between hemisphere and segment duration (Figs. 4e and 5) such that stimuli with segment duration >85 ms SOA produced robust activation in the dorsal bank of right STS, in both the cohort and single-subject analyses (Supplementary Fig. 1 online). This asymmetry cannot be accounted for by appealing to signal complexity—because the stimuli used varied only along the temporal axis—or to speech specificity—given that the stimuli were neither speech signals nor perceived as

such. In a recent study addressing a similar issue4, a lateralization was observed consistent with that observed here. In that study, however, both temporal and spectral signal attributes were varied; we dem-onstrate that varying just a single dimension, temporal structure, can drive lateralization.

Hemispheric asymmetries of the auditory cortex have been docu-mented in anatomical29,30 and physiological studies, and it has been argued that they underlie lateralized auditory functions including the analysis of speech sounds31–34 and pitch8,9,35. The right-hemisphere advantage for slow modulation rate (long segment duration) observed in STS in the present study—although similar to findings from research on phrase-level prosody36 and the analysis of dynamic pitch—suggests a more general explanation.

The absence of sensitivity to segment type, the strong dependence on segment duration, and the slower progression to asymptote in the hemodynamic response between 160–300 ms relative to STG sug-gest that the integration of information in STS in both hemispheres occurs on a timescale equal to or greater than that in STG, that is ≥200–300 ms. To explain the observed hemispheric asymmetry, we hypothesize that left and right STS receive input differentially from STG through intra- and inter-hemispheric (transcallosal) fibers. Specifically, left STS receives contributions from left and right STG weighted toward processing on the 25–50 ms timescale, and right STS receives input weighted toward processing on the 200–300 ms timescale (Supplementary Fig. 2 online). Evidence supporting this claim is reflected in the markedly different slopes of the mean voxel strength in left and right STS as a function of segment duration (Fig. 4e). Two factors are likely to contribute to this disparity: the dif-ference in the magnitude of the input received from the two popula-tions in STG, and the differential effect of the two temporal windows. For long segment duration, the greater activation in right STS is a con-sequence of the larger magnitude of the predominantly 200–300 ms input received from STG. This results from integration over a lon-ger period relative to that received by the less activated left STS, which is smaller owing to integration in STG over only 25–50 ms. However, as the segment duration becomes shorter, the magnitude of the output from the 200–300 ms population in STG falls off faster than that from the 25–50 ms population, because individual short segments begin to fuse sooner in the longer window.

The appeal of a model based on differential cortical connectivity between areas STG and STS in explaining the asymmetry observed is fourfold. First, it is automatic in that the spectrotemporal structure of the stimulus drives the hemispheric asymmetry by selectively engaging two intrinsic processing timescales. Second, the confluence in STS of the independent representations of a signal, each extracted on different

Figure 5 STS activation from cohort analysis. Axial and coronal slices for all six TN stimuli shown in neurological convention. Slices were chosen by finding the maximum activation in right STS for the 300 ms condition (MNI coordinates 61, –10, –16). Comparison between the left and right hemisphere in the axial slices shows that two distinct clusters of activation exist only in the right hemisphere; the more anterior in STG, the more posterior in STS (arrows). Coronal slices show that STS activation is confined to the dorsal bank (arrows) and occurs only for the three longest segment durations. Similar patterns are observed for the FM stimuli (data not shown). Activation maps for individual subjects derived from the ROI analysis (Supplementary Fig. 1 online) show that the hemispheric asymmetry observed in STS in the cohort analysis is also observed at the individual subject level.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

timescales, to a single timescale in which the features from each repre-sentation are in temporal register facilitates the construction of more abstract representations of the signal. Third, our model explains why no obvious anatomic or physiologic asymmetries have been found in higher-order cortices—none are implicated. The model stipulates only the pattern of intra- and inter-hemispheric connectivity. No con-straints are placed either on the spatial distribution of circuits within STG that mediate processing on each timescale or on the fine structure of the afferents that route the resulting information to STS. Fourth, although an architecture based on differential cortical connectivity dif-fers mechanistically from previous proposals, it is consistent with them both conceptually and empirically. For example, the central hypothesis of the double filtering by frequency model37 is that the left hemisphere performs an operation akin to high-pass filtering of the sensory repre-sentation of a stimulus whereas the right hemisphere performs low-pass filtering. In the present model, the differential routing of information processed on the short 25–50 ms timescale is equivalent to high-pass fil-tering, whereas processing on the longer 200–300 ms timescale produces the effect of low-pass filtering. This parallels findings from electrophysi-ological recordings in marmosets showing two neuronal populations with distinct temporal processing profiles in STG38.

The present model is also consistent with the proposal suggesting that left auditory cortex specializes in processing stimuli requiring enhanced temporal resolution, whereas right auditory cortex specializes in pro-cessing stimuli requiring higher frequency resolution4. The output of the analysis on the 25–50 ms timescale will contain more transient features than that resulting from the 200–300 ms analysis, thus mani-festing enhanced temporal resolution. The parallel between analysis on the 200–300 ms timescale and enhanced frequency resolution is more subtle and derives from consideration of the minimum duration of a signal necessary to produce a reliable estimate of dynamic pitch and sentential prosody, which is on the order of 200–300 ms. The asymmet-ric sampling in time (AST) model3, although based on a different speci-fication of the spatial distribution of short- and long-duration analysis windows, agrees conceptually with the present model and makes similar predictions for right-hemisphere function. For example, AST predicts a rightward lateralization of function for long-duration segments, owing to temporal integration with a time constant of 200–300 ms in the right hemisphere, and a leftward lateralization for short segments owing to a time constant of 25–50 ms in the left hemisphere. Although a rightward lateralization for-long duration stimuli was observed in the present study, the corresponding leftward lateralization was not observed. That shortcoming of AST—the localized processing posited in each hemisphere—is circumvented here by attributing lateralization to differential routing of information.

SummaryWe propose a model that accounts for the (i) bilateral temporal sen-sitivity of auditory cortices (ii) differential sensitivity to local spec-trotemporal structure in STG, and (iii) hemispheric lateralization in STS. It is characterized by distributed and hierarchical processing on multiple timescales3,12,39, because different timescales carry distinct and functionally relevant information about a signal. We hypothesize that there exist (at least) two timescales in STG pertinent to cortical auditory processing, 25–50 ms and 200–300 ms, and that neuronal populations corresponding to each timescale differentially target STS, with the right hemisphere receiving afferents carrying information processed on the long timescale and the left hemisphere those resulting from processing on the short timescale.

The observed symmetries and asymmetries follow from the inter-action of the physiological properties of the neuronal ensembles that

mediate the analysis of auditory signals with the spectrotemporal pro-perties of these signals. The model that emerges is one of a progressive elaboration of the representation of an acoustic signal by a distri-buted and hierarchical cortical architecture in which sensory features extracted by earlier stages of processing facilitate the subsequent con-struction of stable perceptual codes. Such a model provides a unifying and neurophysiologically grounded account of early auditory cortical processing.

METHODSSubjects. Fifteen (9 female, 6 male) right-handed subjects aged 18–40 partici-pated in the study. All subjects were free of neurological or medical illnesses, had normal structural MRI and audiometric examinations, gave written informed consent and were paid for their participation.

Stimuli. A total of 13 stimulus conditions (6 tonal TN, 6 frequency- modulated FM and 1 control CN) were created by concatenating narrow-band noise segments consisting of a sum of 50 sinusoids with randomized amplitude phase and frequency. Segment bandwidth spanned a half-octave frequency range of 1,000–1,500 Hz and had a bandwidth of 125 Hz, corresponding to a typical second formant16 and remaining within the critical band at that frequency17. Two types of segments were used, one in which the frequency remained constant throughout the segment (Fig. 1b; TN) and one in which frequency was swept lin-early upward or downward randomly (Fig. 1a; FM). For the TN stimuli, the cen-ter frequency was drawn from a uniform distribution spanning the half-octave range. For the FM stimuli, all of the individual segments swept up or down over the same frequency range. For each of the two segment types, six mean segment durations of 12, 25, 45, 85, 160 and 300 ms were drawn from a Gaussian distribu-tion with the tails ‘clipped’ so as not to overlap each other, and with a standard deviation equal to 20% of the mean (Fig. 1f). A single control stimulus (CN) was constructed from a single 9-s TN segment with center frequency in the middle of the half-octave range. The amplitudes of all 13 stimuli were adjusted so that all had equal RMS power (Fig. 1d). When collapsed over the entire 9-s stimulus duration, all conditions had a nearly equal spectral profile (Fig. 1e).

Procedure. Presentation software (version 0.43, Neurobehavioral Systems) run-ning on a PC controlled stimulus delivery by synchronizing each trial to a series of TTL pulses produced during image acquisition. Sound was delivered binaurally to a Commander XG audio system (Resonance Technology). The electrostatic head-phones provide ∼30 dB of sound attenuation and reduce the ambient MRI scanner noise. The sound level was set to 80 dB SPL.

During behavioral testing, participants were presented a set of stimuli via Presentation identical to those used during imaging except with a segment duration range increased to 10–600 ms (comprising 11 conditions). Subjects were instructed to classify the perceptibility of the individual segments in a single-interval, two-alternative forced-choice task as either perceptible or not perceptible by pressing the appropriate button on a computer keyboard.

Image acquisition. All images were obtained from a 1.5-T GE Signa scanner (GE Medical Systems) equipped with a standard quadrature head-coil. Functional images were collected using a single-shot echoplanar pulse sequence (TE, 40 ms; TR, 11.4 s; flip angle, 90°). Subjects listened passively to a total of 24 trials of each of the 13 stimuli in 8 blocks. Blocks consisted of 39 pseudo-randomized trials during which each condition occurred three times, producing a total of 312 images per subject. Two images were added at the beginning of each block to allow the hemodynamic response to equilibrate, and subsequently discarded from further analysis. The 11.4 s ‘clustered’ volume acquisition consisted of 9 s of stimulus presentation followed by 2.4 s of slice acquisition to minimize con-tamination from artifacts induced by scanner14,15. Thus, subjects never heard (gradient coil) scanner noise while listening to the stimuli. A total of 24 slices were acquired to provide whole-brain coverage. Subject head movement was minimized by wrapping the head with a vacuum pillow.

Image analysis. Image preprocessing and statistical analysis were performed using SPM99b software. Anatomical segmentation was performed in Matlab version 6 (Mathworks) based on labeled voxels provided by the Talairach

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


A R T I C L E S

Daemon, version 1.1 (Research Imaging Center, University of Texas Health Science Center). Image preprocessing consisted of realigning the image time series to the first image to correct for subject movement, smoothing via con-volution with a Gaussian kernel (8 mm full-width half-maximum), high-pass filtering to remove slow drifts in signal intensity, and re-sampling from a grid size of 3.75 × 3.75 × 5 mm (24 slices) to 4 × 4 × 4 mm (34 slices). In the fixed-effects cohort analysis, categorical contrasts were constructed between each of the six TN and FM conditions and the CN stimulus, resulting in a t-statistic for each voxel and thresholded at P < 0.05 (corrected via Gaussian random fields for multiple comparisons). The thresholded t-score maps (SPMs) were then spatially normalized to the MNI T1 template. The coordinates of all suprathreshold voxels in MNI space were converted to Talairach coordinates and uploaded to the Talairach Daemon for anatomical classification. Although the Talairach Daemon conflates areas STS and MTG at some voxels, visual inspection of the co-registered EPI (echo planar imaging) and anatomical data showed that the activation occurred primarily in STS. Thus, we refer to these voxels simply as STS. In the subsequent single-subject ROI analysis, the effect of segment type and hemisphere were assessed. Mean voxel strength was computed by extracting all voxels from the categorical contrasts within a region defined by the union of suprathreshold voxels from the cohort analysis. Voxel strengths were then submitted to a three-way, full-factorial, repeated-measures ANOVA with hemisphere, segment type (TN or FM) and segment duration as factors.

Surface maps in Figure 2 were created by converting the cohort activation from MNI space to Talairach space in AFNI and overlaying it on the N27 Brain, also warped to Talairach space, using SUMA.

URLs. AFNI, http://afni.nimh.nih.gov/afni; N27 Brain, http://www.bic.mni.mcgill.ca and UCLA http://www.loni.ucla.edu; SUMA, http://afni.nimh.nih.gov/afni/suma.


ACKNOWLEDGMENTSWe thank P. Bandettini, J. Fritz, A.-L. Giraud, A. Martin and J. Rauschecker for insightful critical comments; F. Husain for help with experimental setup; and K.M. Boemio for her continued and continuous encouragement. A.B. and D.P. were supported by US National Institutes of Health R01 DC05660 to D.P. During the preparation of the manuscript, D.P. was a fellow at the Wissenschaftskolleg zu Berlin and the American Academy Berlin.


Received 20 December 2004; accepted 25 January 2005.Published online at http://www.nature.com/natureneuroscience/

1. Hall, D.A., Hart, H.C. & Johnsrude, I.S. Relationships between human auditory cortical structure and function. Audiol. Neurootol. 8, 1–18 (2003).

2. Zatorre, R., Belin, P. & Penhune, V. Structure and function of auditory cortex: music and speech. Trends Cogn. Sci. 6, 37–46 (2002).

3. Poeppel, D. The analysis of speech in different temporal integration windows: cere-bral lateralization as ‘asymmetric sampling in time’. Speech Commun. 41, 245–255 (2003).

4. Zatorre, R. & Belin, P. Spectral and temporal processing in human auditory cortex. Cereb. Cortex 11, 946–953 (2001).

5. Binder, J.R. et al. Human temporal lobe activation by speech and nonspeech sounds. Cereb. Cortex 10, 512–528 (2000).

6. Scott, S.K. & Johnsrude, I.S. The neuroanatomical and functional organization of speech perception. Trends Neurosci. 26, 100–107 (2003).

7. Hickok, G. & Poeppel, D. Towards a functional neuroanatomy of speech perception. Trends Cogn. Sci. 4, 131–138 (2000).

8. Johnsrude, I.S., Penhune, V.B. & Zatorre, R.J. Functional specificity in the right human auditory cortex for perceiving pitch direction. Brain 123, 155–163 (2000).

9. Scott, S.K., Blank, C.C., Rosen, S. & Wise, R.J. Identification of a pathway for intel-ligible speech in the left temporal lobe. Brain 123, 2400–2406 (2000).

10. Gandour, J. et al. A cross-linguistic FMRI study of spectral and temporal cues underly-ing phonological processing. J. Cogn. Neurosci. 14, 1076–1087 (2002).

11. Giraud, A-L. et al. Representation of the temporal envelope of sounds in the human brain. J. Neurophysiol. 84, 1588–1598 (2000).

12. Shamma, S. On the role of space and time in auditory processing. Trends Cogn. Sci. 5, 340–348 (2001).

13. Hall, D.A. et al. Spectral and temporal processing in human auditory cortex. Cereb. Cortex 12, 140–149 (2002).

14. Hall, D.A. et al. ‘Sparse’ temporal sampling in auditory fMRI. Hum. Brain Mapp. 7, 213–223 (1999).

15. Edmister, W.B., Talavage, T.M., Ledden, P.J. & Weisskoff, R.M. Improved auditory cortex imaging using clustered volume acquisitions. Hum. Brain Mapp. 7, 89–97 (1999).

16. Stevens, K.N. Acoustic Phonetics (MIT Press, Cambridge, Massachusetts, USA, 1998).

17. Moore, B.C.J. in Human Psychophysics (eds. Yost, W.A., Popper, A.N. & Fay, R.R.) (Springer, New York, 1993).

18. Hackett, T.A., Preuss, T.M. & Kaas, J.H. Architectonic identification of the core region in auditory cortex of macaques, chimpanzees, and humans. J. Comp. Neurol. 441, 197–222 (2001).

19. Griffiths, T.D., Buchel, C., Frackowiak, R.S. & Patterson, R.D. Analysis of temporal structure in sound by the human brain. Nat. Neurosci. 1, 422–427 (1998).

20. Harms, M.P. & Melcher, J.R. Sound repetition rate in the human auditory pathway: representations in the waveshape and amplitude of fMRI activation. J. Neurophysiol. 88, 1433–1450 (2002).

21. Wang, X., Lu, T. & Liang, L. Cortical processing of temporal modulations. Speech Commun. 41, 107–121 (2003).

22. Wessinger, C.M. et al. Hierarchical organization of the human auditory cortex revealed by functional magnetic resonance imaging. J. Cogn. Neurosci. 13, 1–7 (2001).

23. Yost, W.A. Auditory image perception and analysis: the basis for hearing. Hear. Res. 56, 8–18 (1991).

24. Yabe, H. et al. Organizing sound sequences in the human brain: the interplay of auditory streaming and temporal integration. Brain Res. 897, 222–227 (2001).

25. Winkler, I., Reinikainen, K. & Naatanen, R. Event-related brain potentials reflect traces of echoic memory in humans. Percept. Psychophys. 53, 443–449 (1993).

26. Sussman, E., Winkler, I., Ritter, W., Alho, K. & Naatanen, R. Temporal integration of auditory stimulus deviance as reflected by the mismatch negativity. Neurosci. Lett. 264, 161–164 (1999).

27. Zwislocki, J. Theory of temporal auditory summation. J. Acoust. Soc. Am. 32, 1046–1060 (1960).

28. Kaas, J.H. & Hackett, T.A. Subdivisions of auditory cortex and processing streams in primates. Proc. Natl Acad. Sci. USA 97, 11793–11799 (2000).

29. Geschwind, N. & Levitsky, W. Human brain: left-right asymmetries in temporal speech region. Science 161, 186–187 (1968).

30. Galuske, R.A., Schlote, W., Bratzke, H. & Singer, W. Interhemispheric asymmetries of the modular structure in human temporal cortex. Science 289, 1946–1949 (2000).

31. Jäncke, L., Wustenberg, T., Scheich, H. & Heinze, H.J. Phonetic perception and the temporal cortex. Neuroimage 15, 733–746 (2002).

32. Näätänen, R. et al. Language-specific phoneme representations revealed by electric and magnetic brain responses. Nature 385, 432–434 (1997).

33. Palva, S. et al. Distinct gamma-band evoked responses to speech and non-speech sounds in humans. J. Neurosci. 22, RC211 (2002).

34. Schwartz, J. & Tallal, P. Rate of acoustic change may underlie hemispheric specializa-tion for speech perception. Science 207, 1380–1381 (1980).

35. Divenyi, P.L. & Robinson, A.J. Nonlinguistic auditory capabilities in aphasia. Brain Lang. 37, 290–326 (1989).

36. Meyer, M., Alter, K., Friederici, A.D., Lohmann, G. & von Cramon, D.Y. FMRI reveals brain regions mediating slow prosodic modulations in spoken sentences. Hum. Brain Mapp. 17, 73–88 (2002).

37. Ivry, R.B. & Robertson, L.C. The Two Sides of Perception (Bradford Books, MIT Press, Cambridge, Massachusetts, 1997).

38. Lu, T., Liang, L. & Wang, X. Temporal and rate representations of time-varying signals in the auditory cortex of awake primates. Nat. Neurosci. 4, 1131–1138 (2001).

39. Smith, Z.M., Delgutte, B. & Oxenham, A.J. Chimaeric sounds reveal dichotomies in auditory perception. Nature 416, 87–90 (2002).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


CO R R I G E N DA

Corrigendum: High frequency, synchronized bursting drives eye-specific segregation of retinogeniculate projectionsChristine L Torborg, Kristi A Hansen & Marla B FellerNat. Neurosci. 8, 72-78 (2005)

A sentence in the discussion of this paper contained some inaccurate references.

The second sentence (first column) on page 76 should read “First, local infusion of general nAChR antagonists directly into the dLGN during the first postnatal week does not prevent eye-specific segregation2”.

The authors regret the error.

Corrigendum: A representation of the hazard rate of elapsed time in macaque area LIPPeter Janssen & Michael N ShadlenNat. Neurosci. 8, 234-241 (2005)

In the Methods section, equation 4 contained an error.

The corrected version should read as follows:


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


CO R R I G E N DA

Corrigendum: High frequency, synchronized bursting drives eye-specific segregation of retinogeniculate projectionsChristine L Torborg, Kristi A Hansen & Marla B FellerNat. Neurosci. 8, 72-78 (2005)

A sentence in the discussion of this paper contained some inaccurate references.

The second sentence (first column) on page 76 should read “First, local infusion of general nAChR antagonists directly into the dLGN during the first postnatal week does not prevent eye-specific segregation2”.


Corrigendum: A representation of the hazard rate of elapsed time in macaque area LIPPeter Janssen & Michael N ShadlenNat. Neurosci. 8, 234-241 (2005)

In the Methods section, equation 4 contained an error.

The corrected version should read as follows:


©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

Date post:	13-Apr-2016
Category:	Documents
Upload:	horacio-martinez
View:	27 times
Download:	0 times

Nature Neuroscience March 2005

Documents