www.nature.com/natureneuroscience

EDITORIAL OFFICE [email protected] 75 Varick Street, Fl 9, New York, NY 10013-1917 Tel: (212) 726 9319, Fax: (212) 696 0978 Editor: Kalyani Narasimhan Associate Editors: Hannah Bayer, Min Cho, Annette Markus, Charvy Narain Assistant Editor: Kathleen Dave Copy Editors: Anita Gould, David Lechtenberg Production Editors: Sabina Eberle, Jamel Wooten Senior Illustrator: Katie Vicari Illustrator: Kimberly Caesar Cover Design: Erin Boyle Editorial Assistant: Elizabeth Patrick

© 2009 Nature America, Inc. All rights reserved.

MANAGEMENT OFFICES NPG New York 75 Varick Street, Fl 9, New York, NY 10013-1917 Tel: (212) 726 9200, Fax: (212) 696 9006 Publisher: Stephanie Diment Executive Editor: Linda Miller Chief Technology Officer: Howard Ratner Head of Nature Research & Reviews Marketing: Sara Girard Marketing Manager: Amy Maurer Production Coordinator: Diane Temprano Head of Web Services: Anthony Barrera Web Production Manager: Susan Kline NPG London The Macmillan Building, 4 Crinan Street, London N1 9XW Tel: 44 207 833 4000, Fax: 44 207 843 4996 Managing Director: Steven Inchcoombe Publishing Director: Alison Mitchell Editor-in-Chief, Nature Publications: Philip Campbell Marketing Director: Della Sar Director of Web Publishing: Timo Hannay NPG Nature Asia-Pacific Chiyoda Building, 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843 Tel: 81 3 3267 8751, Fax: 81 3 3267 8746 Publishing Director — Asia-Pacific: David Swinbanks Associate Director: Antoine E. Bocquet Manager: Koichi Nakamura Operations Director: Hiroshi Minemura Marketing Manager: Masahiro Yamashita Asia-Pacific Sales Director: Kate Yoneyama Asia-Pacific Sales Manager: Ken Mikami DISPLAY ADVERTISING [email protected] (US/Canada) [email protected] (Europe) [email protected] (Asia) Global Head of Advertising and Sponsorship: Dean Sanderson, Tel: (212) 726 9350, Fax: (212) 696 9482 Global Head of Display Advertising: Andrew Douglas, Tel: 44 207 843 4975, Fax: 44 207 843 4996 Asia-Pacific Sales Manager: Ken Mikami, Tel: 81 3 3267 8765, Fax: 81 3 3267 8746 Display Account Managers: New England: Sheila Reardon, Tel: (617) 399 4098, Fax: (617) 426 3717 New York/Mid-Atlantic/Southeast: Jim Breault, Tel: (212) 726 9334, Fax: (212) 696 9481 Midwest: Mike Rossi, Tel: (212) 726 9255, Fax: (212) 696 9481 West Coast South: George Lui, Tel: (415) 781 3804, Fax: (415) 781 3805 West Coast North: Bruce Shaver, Tel: (415) 781 6422, Fax: (415) 781 3805 Germany/Switzerland/Austria: Sabine Hugi-Fürst, Tel: 41 52761 3386, Fax: 41 52761 3419 United Kingdom/Ireland: Jeremy Betts, Tel: 44 207 843 4968, Fax: 44 207 843 4749 Scandinavia/Iceland/Spain/Portugal: Evelina Rubio-Hakansson, Tel: 44 207 843 4079, Fax: 44 207 843 4749 France/Belgium/The Netherlands/Italy/Israel/Eastern Europe: Nicola Wright, Tel: 44 207 843 4959, Fax: 44 207 843 4749 Greater China/Singapore: Gloria To, Tel: 852 2811 7191, Fax: 852 2811 0743 NATUREJOBS [email protected] (US/Canada) [email protected] (Europe) [email protected] (Asia) US Sales Manager: Ken Finnegan, Tel: (212) 726 9248, Fax: (212) 696 9482 European Sales Manager: Dan Churchward, Tel: 44 207 843 4966, Fax: 44 207 843 4596 Asia-Pacific Sales Manager: Ayako Watanabe, Tel: 81 3 3267 8765, Fax: 81 3 3267 8746 SITE LICENSE BUSINESS UNIT Americas: Tel: (888) 331 6288 Asia/Pacific: Tel: 81 3 3267 8751 Australia/New Zealand: Tel: 61 3 9825 1160 India: Tel: 91 124 2881054/55 ROW: Tel: 44 207 843 4759

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

CUSTOMER SERVICE www.nature.com/help Senior Global Customer Service Manager: Gerald Coppin For all print and online assistance, please visit www.nature.com/help Purchase subscriptions: Americas: Nature Neuroscience, Subscription Dept., 342 Broadway, PMB 301, New York, NY 10013-3910. Tel: (866) 363 7860, Fax: (212) 689 9108 Europe/ROW: Nature Neuroscience, Subscription Dept., Macmillan Magazines Ltd., Brunel Road, Houndmills, Basingstoke RG21 6XS, United Kingdom. Tel: 44 1256 329 242, Fax: 44 1256 812 358 Japan: Nature Neuroscience, NPG Nature Asia-Pacific, Chiyoda Building, 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843. Tel: 81 3 3267 8751, Fax: 81 3 3267 8746 India: Nature Neuroscience, NPG India, 3A, 4th Floor, DLF Corporate Park, Gurgaon 122002, India. Tel: 91 124 2881054/55, Fax: 91 124 2881052 REPRINTS [email protected] Nature Neuroscience Reprint Department, Nature Publishing Group, 75 Varick Street, Fl 9, New York, NY 10013-1917, USA. For commercial reprint orders of 600 or more, please contact: UK Reprints: Tel: 44 1256 302 923, Fax: 44 1256 321 531 US Reprints: Tel: (212) 726 9278, Fax: (212) 679 0843

volume 12 number 4 APRIL 2009

E d i t o r ia l 363

Making the most of reviewer resources

© 2009 Nature America, Inc. All rights reserved.

co r r e s p on d e n c e Correlated network activity is important in the development of many neural circuits. Watt et al. characterize monsynaptic connections between Purkinje cells of the juvenile cerebellum and use these measurements to model the generation of traveling waves of activity between connected Purkinje cells. They validate their model with observations in juvenile cerebellar cortex. (p 463)

365

First report of action potentials in a C. elegans neuron is premature

book review 367

The Confabulating Mind: How the Brain Creates Reality by Armin Schnider Reviewed by Daniel L Schacter & Brendan Gaesser

news and views 369

Fine control: microRNA regulation of adult neurogenesis Qin Shen & Sally Temple  see also p 399

371

Proteoglycans specify Sonic Hedgehog effect Catherine Vaillant & Denis Monard  see also p 409

372

Neuronal communication: a detailed balancing act Emilio Salinas  see also p 483

374

It’s not you, it’s me. Really. Garrett B Stanley  see also p 492

c o m m e n ta r y 377

The quest for action potentials in C. elegans neurons hits a plateau Shawn R Lockery & Miriam B Goodman

review Manipulations of retinal activity modulate thalamic firing patterns (p 390)

379

Endoplasmic reticulum stress in disorders of myelinating cells Wensheng Lin & Brian Popko

Nature Neuroscience (ISSN 1097-6256) is published monthly by Nature Publishing Group, a trading name of Nature America Inc. located at 75 Varick Street, Fl 9, New York, NY 10013-1917. Periodicals postage paid at New York, NY and additional mailing post offices. Editorial Office: 75 Varick Street, Fl 9, New York, NY 10013-1917. Tel: (212) 726 9319, Fax: (212) 696 0978. Annual subscription rates: USA/Canada: US$225 (personal), US$3,060 (institution). Canada add 7% GST #104911595RT001; Euro-zone: €287 (personal), €2,430 (institution); Rest of world (excluding China, Japan, Korea): £185 (personal), £1,570 (institution); Japan: Contact NPG Nature Asia-Pacific, Chiyoda Building, 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843. Tel: 81 (03) 3267 8751, Fax: 81 (03) 3267 8746. POSTMASTER: Send address changes to Nature Neuroscience, Subscriptions Department, 342 Broadway, PMB 301, New York, NY 10013-3910. 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 © 2009 Nature Publishing Group. Printed in USA.

i

volume 12 number 4 APRIL 2009

b r i e f c o m m u n i c at i o n s

© 2009 Nature America, Inc. All rights reserved.

MicroRNA regulates adult neurogenesis (pp 369 and 399)

387

A dual leucine kinase–dependent axon self-destruction program promotes Wallerian degeneration B R Miller, C Press, R W Daniels, Y Sasaki, J Milbrandt & A DiAntonio

390

Thalamic activity that drives visual cortical plasticity M L Linden, A J Heynen, R H Haslinger & M F Bear

393

D2R striatopallidal neurons inhibit both locomotor and drug reward processes P F Durieux, B Bearzatto, S Guiducci, T Buch, A Waisman, M Zoli, S N Schiffmann & A de Kerchove d’Exaerde

396

Pharmacological REM sleep suppression paradoxically improves rather than impairs skill memory B Rasch, J Pommer, S Diekelmann & J Born

articles

Paracrine control of oligodendrocyte differentiation (p 418)

399

miR-124 regulates adult neurogenesis in the subventricular zone stem cell niche L-C Cheng, E Pastrana, M Tavazoie & F Doetsch  see also p 369

409

Proteoglycan interactions with Sonic Hedgehog specify mitogenic responses J A Chan, S Balasubramanian, R M Witt, K J Nazemi, Y Choi, M F Pazyra-Murphy, C O Walsh, M Thompson & R A Segal  see also p 371

418

Paracrine control of oligodendrocyte differentiation by SRF-directed neuronal gene expression C Stritt, S Stern, K Harting, T Manke, D Sinske, H Schwarz, M Vingron, A Nordheim & B Knöll

428

Trans-synaptic adhesion between NGL-3 and LAR regulates the formation of excitatory synapses J Woo, S-K Kwon, S Choi, S Kim, J-R Lee, A W Dunah, M Sheng & E Kim

438

Altered chloride homeostasis removes synaptic inhibitory constraint of the stress axis S A Hewitt, J I Wamsteeker, E U Kurz & J S Bains

444

Tuning of synapse number, structure and function in the cochlea A C Meyer, T Frank, D Khimich, G Hoch, D Riedel, N M Chapochnikov, Y M Yarin, B Harke, S W Hell, A Egner & T Moser

454

Phosphodiesterase 1C is dispensable for rapid response termination of olfactory sensory neurons K D Cygnar & H Zhao

463

Traveling waves in developing cerebellar cortex mediated by asymmetrical Purkinje cell connectivity A J Watt, H Cuntz, M Mori, Z Nusser, P J Sjöström & M Häusser

Tonotopic differences in synaptic properties (p 444)

nature neuroscience

iii

volume 12 number 4 APRIL 2009

© 2009 Nature America, Inc. All rights reserved.

Odor representations downstream of the olfactory bulb (p 474)

474

Transformation of odor representations in target areas of the olfactory bulb E Yaksi, F von Saint Paul, J Niessing, S T Bundschuh & R W Friedrich

483

Gating multiple signals through detailed balance of excitation and inhibition in spiking networks T P Vogels & L F Abbott  see also p 372

492

Phase-to-rate transformations encode touch in cortical neurons of a scanning sensorimotor system J C Curtis & D Kleinfeld  see also p 374

502

Interval time coding by neurons in the presupplementary and supplementary motor areas A Mita, H Mushiake, K Shima, Y Matsuzaka & J Tanji

508

A neural mechanism of first impressions D Schiller, J B Freeman, J P Mitchell, J S Uleman & E A Phelps

515

Hierarchical cognitive control deficits following damage to the human frontal lobe D Badre, J Hoffman, J W Cooney & M D’Esposito

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.

Somatosensory cortex neuron responses to active touch (pp 374 and 492)

nature neuroscience

v

e d i to r i a l

Making the most of reviewer resources

© 2009 Nature America, Inc. All rights reserved.

In 2008, Nature Neuroscience joined a community consortium aimed at making peer review more efficient by allowing reviews to be transferred between consortium journals. We look back at our experience with the Neuroscience Peer Review Consortium over the last year.

L

ast April, Nature Neuroscience announced that it would join the newly established Neuroscience Peer Review Consortium (NPRC) and that we would offer our authors whose papers were no ­longer under ­consideration an opportunity to transfer their reviews when ­submitting their paper to another consortium journal1. The editors of the ­consortium ­journals were recently polled to evaluate the success of NPRC over the last year and, ­surprisingly, most journals reported that only 1–2% of the ­manuscripts they received had been forwarded from another ­consortium journal or sent out from the journal to other ­participants. Our experience with the consortium has not been much different, with only a handful of papers being transferred from Nature Neuroscience to another consortium ­journal. We will continue our ­participation in the NPRC for this year and encourage more authors to take advantage of this process. The NPRC was officially launched in January 2008 with the aim of providing a system that would speed up the review process and reduce the workload for reviewers and editors. Over 33 journals now ­participate in the consortium (http://nprc.incf.org). Similar to the Nature journals’ transfer system, the NPRC system is completely ­voluntary for authors. Editors at one journal only know that a paper was reviewed ­elsewhere if the author chooses to inform them. Likewise, referees also have the option of opting out. Many journals in the consortium ask the ­referees to state at the time of reviewing a paper whether the editors may release their names along with the review in the event that a paper is transferred to another journal. At Nature Neuroscience, the editors contact the ­referees and ask for their permission to release their identities whenever authors ask for their papers to be transferred to another consortium journal. If a reviewer declines to participate, the reviews (comments to authors only) are transferred anonymously. Comments to editors are never transferred, even when the referee agrees to be identified to the receiving journal. Finally, the editors have full discretion in deciding how to use the transferred reviews. To date, we have had only a handful of transfers to other member journals (all to the Journal of Neuroscience), representing less than 1% of manuscripts that are eventually rejected after review. However, for the papers that were eventually published in the Journal of Neuroscience, the authors reported that the paper had been expedited. Even in the cases where new referees were solicited, authors felt that transferring the reviews from Nature Neuroscience had saved them both time and labor. Our experience with the NPRC may be less typical than that of other journals in the consortium (we have had no transfers to Nature Neuroscience, for example), but other journals also saw fewer than expected transfers. David Linden, the editor-in-chief of the Journal of Neurophysiology, felt that the effect of NPRC was fairly minor last year, representing less than 2% of total submissions. Moreover, from Linden’s experience, many authors were not using the ­system nature neuroscience volume 12 | number 4 | april 2009

e­ ffectively, with several authors choosing the transfer process when it clearly was not beneficial to them (for example, transferring their paper without making a serious attempt to address the referees’ concerns from the earlier round). As a result, the Journal of Neurophysiology found that, in 2008, the accept rate of transferred manuscripts was not much different from that of de novo submissions. Given the potential savings in time and labor, why are so few authors using the NPRC option? Authors may simply not be aware of NPRC or may not know what journals participate in it. Transfer rates may pick up as more authors learn of the consortium. At Nature Neuroscience, we have noticed an increase in the number of referees that state in ­comments to the editors whether they wish for their identities to be released to other ­consortium journals or not, suggesting a growing awareness of the NPRC. It could also be that there are not that many papers that lend ­themselves well to this process. Many of our authors who have had papers rejected may prefer to take their chances with new referees at another ­journal, rather than making substantial revisions in response to the concerns raised by our referees. Certainly, our authors appear to be more ­conservative when deciding to transfer their reviews, preferentially choosing to ­utilize the NPRC transfer option when the reviewers reject the paper on ­conceptual grounds and not for technical reasons. Another factor that influences the success of the transfer is whether the referees allow the release of their identities to receiving consortium ­journals. Previous reviews are clearly less useful to the receiving ­journal if the ­editors do not know who the reviewers were. Estimates at some ­consortium ­journals suggest that 25–30% of reviewers chose to not have their ­identities revealed to other consortium journals. Although we have had very few transfers to other NPRC journals, only a handful of our ­referees declined to have their names released. It could be that some ­referees feel confident releasing their identities when they know exactly which ­journal the paper is being transferred to, but prefer to opt-out when they are only given the option of ticking a box before submitting their review. The consortium estimates that only about 10% of rejected ­manuscripts would be good candidates for NPRC transfers in any case, but it is clear that the current transfer rate remains far lower than expected. It is ­therefore premature to gauge whether this system truly could save ­referees, authors and editors substantial time and effort. The members of the NPRC decided last November to extend the life of the consortium by at least another year. We are pleased to continue our participation in the NPRC for 2009 and invite authors who have not yet used the NPRC to try it. We shall evaluate the success of the NPRC and our participation in the consortium on an ongoing basis and we greatly encourage our authors, referees and readers to share their comments with us by emailing the editors directly, or by contacting us at [email protected]. L 1. Nat. Neurosci. 11, 375 (2008).

363

co r r e s p on d e n c e

© 2009 Nature America, Inc. All rights reserved.

First report of action potentials in a C. elegans neuron is premature To the editor: Since the publication of an essentially ­complete wiring diagram of the C. elegans nervous ­system more than 20 years ago1, there has been intense interest in the ­question of whether C. elegans neurons fire action ­potentials. Thus, the recent report by Mellem et al.2 of action potentials in a C. ­elegans ­neuron is likely to receive ­considerable attention. Having ­carefully reviewed the data, however, we find that the regenerative events described by Mellem et al.2 were incorrectly labeled as action potentials and are more accurately described as graded regenerative potentials. Action potentials are one instance of a broad class of regenerative events. Such events are caused by intrinsic positive ­feedback as a result of a voltage-activated depolarizing ­current; this current is usually carried by voltage-gated Na+ channels, voltage-gated Ca2+ ­channels or both. The other types of regenerative events are graded regenerative ­potentials, ­intrinsic oscillations and ­plateau ­potentials. An action potential per se has three ­distinguishing ­features (Fig. 1a–c). First, its amplitude is invariant with respect to the amplitude, ­duration and waveform of the stimulus that evoked it; once triggered it goes to ­completion. Second, it is ­intrinsically self-­terminating as a result of events set in motion by the action potential upstroke, such as ­activation of ­hyperpolarizing current and inactivation of depolarizing ­current. Third, it has a stereotyped waveform that, as with amplitude, is invariant with respect to the amplitude, duration and waveform of the stimulus. Thus, every action potential is a regenerative event, but not every ­regenerative event is an action potential. Recording in situ from the motor neuron class RMD, Mellem et al.2 observed voltage transients in response to current ­injection in the form of a rising and falling ramp (Fig. 1a of ref. 2) or a 50-ms depolarizing step (Figs. 2a,b and 3a,b,d–f of ref. 2). They reported that approximately half of these events were ­followed by a stable plateau potential (as in Fig. 2 of ref. 2). This is the first ­demonstration

a

b

c 20 mV 40 nA 2 ms

d

e

f 20 mV 40 nA 2 ms

Figure 1 Simulated action potentials and graded regenerative potentials. (a–c) Canonical Hodgkin-Huxley model of the action potential. Upper traces are voltage and lower traces are injected current. The amplitude and waveform of the action potentials are essentially invariant with respect to the amplitude (a), duration (b) and waveform (c) of the stimulus. In a–c, the onset of repolarization is independent of stimulus offset, indicating that the action potential is self-terminating. (d–e) Modified Hodgkin-Huxley model illustrating the main features of graded regenerative responses. The modified model uses the same equations as the canonical model, but with the maximum Na+ and K+ conductances reduced to produce graded responses instead of action potentials. In the modified model, amplitude and waveform are now stimulus dependent and the onset of repolarization always coincides with stimulus offset, indicating that the events are no longer self-terminating, as in Mellem et al.2.

of plateau potentials in a C. elegans neuron. Mellem et al.2 used the term “action ­potential” to refer to the ­initial ­voltage transient. The authors reported that plateau potentials occurred ­spontaneously (Figs. 1e and 3c of ref. 2), but they did not report the observation of ­spontaneous action ­potentials. The voltage transients described by Mellem et al.2 fail to meet the criteria for action ­potentials in three key respects. First, the ­amplitude of the transients was strongly dependent on the ­amplitude and duration of the stimulus (Figs. 2 and 3a of ref. 2). Second, there was no ­evidence that they were ­terminated ­intrinsically, as defined above. On the contrary, the onset of ­repolarization always coincided with the ­offset of the step (Figs. 2 and 3a,b,d–f of ref. 2) or the falling phase of the ramp (Fig. 1c of ref. 2). Third, the waveform of the transients was not invariant with respect to changes in the ­stimulus ­(compare Figs. 1a,c and 2 of ref. 2). Thus, the data in Mellem et al.2 do not ­constitute the demonstration of action ­potentials in RMD neurons.

nature neuroscience volume 12 | number 4 | APRIL 2009

The voltage transients in question instead appear to be graded regenerative potentials. Such potentials, which have been ­documented ­previously in nematodes3,4, superficially ­resemble action potentials, but unlike the ­latter, their amplitude and waveform are highly ­sensitive to the size, duration and waveform of the stimulus. Graded ­potentials occur in cells in which the regenerative ­current is too small to drive an initial ­depolarization to ­completion each time the current is ­activated (Fig. 1d–f). The question of when to use the term action potential transcends semantics. Action ­potentials support coding schemes that utilize firing rate or spike timing, whereas graded regenerative potentials do not. Thus, it would be wrong to conclude from the data of Mellem et al.2 that neuronal ­signaling in C. elegans ­necessarily involves ­coding schemes familiar to us from spiking nervous systems. Action ­potentials may yet be found under different ­conditions in RMD or in other classes of C. elegans ­neurons. But until then, the ­distinction between graded ­regenerative potentials and action ­potentials

365

co r r e s p on d e n c e should be kept in mind by ­theorists and ­experimentalists that are ­interested in both information processing and the molecular basis of neuronal signaling in C. elegans. Shawn R Lockery1, Miriam B Goodman2 & Serge Faumont1 1Institute of Neuroscience, University of Oregon, Eugene, Oregon, USA. 2Department of Molecular and Cellular Physiology, Stanford University, Stanford, California, USA. e-mail: [email protected]

© 2009 Nature America, Inc. All rights reserved.

1. White, J.G., Southgate, E., Thomson, J.N. & Brenner, S. Philos. Trans. R. Soc. Lond. B 314, 1–340 (1986). 2. Mellem, J.E., Brockie, P.J., Madsen, D.M. & Maricq, A.V. Nat. Neurosci. 11, 865–867 (2008). 3. Davis, R.E. & Stretton, A.O.W. J. Neurosci. 9, 415–425 (1989). 4. Goodman, M.B., Hall, D.H., Avery, L. & Lockery, S.R. Neuron 20, 763–772 (1998).

Mellem et al. reply: Recently, we reported the first evidence of action potentials in the ­nervous ­system of C. elegans1. Approximately 50% of the ­regenerative events that we observed in response to ­current stimulation were ­isolated action potentials. The ­remaining 50% were ­followed by ­progression into a ­plateau ­potential and were altered by genetic or ­pharmacological ­manipulations. We also observed spontaneous all-or-none changes between hyperpolarized and ­depolarized states. Lockery et al. argue that the term ‘action ­potential’ should be reserved for a ­specific type of regenerative event that is associated with rate or spike ­timing ­coding schemes. In their view, a ­regenerative event

366

should be called an action potential only if it meets three ­criteria: its ­amplitude is ­invariant, its ­waveform is ­invariant and it is ­intrinsically self-­terminating. However, the broader ­scientific ­community does not use this ­narrow definition. In fact, action ­potentials are used to describe a wide ­variety of ­regenerative events found in ­neurons, ­adrenal and ­pituitary cells2,3, and even in algae and plant cells4. Probably the best-known examples of long and ­variable duration action potentials are found in cardiac cells. Another example is the action potential described in retinal ­horizontal cells, which is variable in ­magnitude, ­waveform and response to ­depolarizing ­stimuli5. A final example is the variable ­duration action ­potential ­identified in C. elegans pharyngeal muscle6. Indeed, the diversity of action ­potentials is enormous and recent studies have begun to elucidate the molecular mechanisms that lead to this diversity6,7. The regenerative events that we have described in C. elegans are diverse: at times they have reproducible, stereotyped ­waveforms (data not shown), at others they are associated with long-lived plateau states (Fig. 3g of ref. 1), and at yet other times they show ­spontaneous fluctuations of invariant amplitude (Fig. 1e of ref. 1). These findings should be of ­considerable importance to current efforts to model ­network activity in C. elegans. Because an earlier study suggested that all neurons in C. elegans shared “…a ­common mechanism of ­sensitivity and dynamic range”8, ­modeling studies might make

the ­assumption that all responses are simply graded in C. ­elegans; ­however, our ­studies clearly indicate the ­presence of ­regenerative, highly nonlinear voltage changes in at least some neurons1. These are exciting times for C. elegans ­neurobiology. Our initial observations raise a multitude of fascinating questions about the cellular and synaptic signaling ­mechanisms that contribute to action ­potential ­generation and plateau potentials, as well as the ­importance of this digital-like signaling for ­circuit ­function. Undoubtedly, future ­studies that address these questions and others will help to reveal ­fundamental strategies for ­information processing by nervous systems. Jerry E Mellem, Penelope J Brockie, David M Madsen & Andres V Maricq Department of Biology, University of Utah, Salt Lake City, Utah, USA. e-mail: [email protected] 1. Mellem, J.E., Brockie, P.J., Madsen, D.M. & Maricq, A.V. Nat. Neurosci. 11, 865–867 (2008). 2. Tabares, L. & Lopez-Barneo, J. Pflugers Arch. 407, 163–165 (1986). 3. Korn, S.J., Bolden, A. & Horn, R. J. Physiol. (Lond.) 439, 423–437 (1991). 4. Fromm, J. & Lautner, S. Plant Cell Environ. 30, 249–257 (2007). 5. Tachibana, M. J. Physiol. (Lond.) 321, 141–161 (1981). 6. Davis, M.W., Fleischhauer, R., Dent, J.A., Joho, R.H. & Avery, L. Science 286, 2501–2504 (1999). 7. Schroeder, B.C., Cheng, T., Jan, Y.N. & Jan, L.Y. Cell 134, 1019–1029 (2008). 8. Goodman, M.B., Hall, D.H., Avery, L. & Lockery, S.R. Neuron 20, 763–772 (1998).

volume 12 | number 4 | APRIL 2009 nature neuroscience

book review

Memory and reality The Confabulating Mind: How the Brain Creates Reality By Armin Schnider

© 2009 Nature America, Inc. All rights reserved.

Oxford University Press, 2008 284 pp, hardcover, $57.50 ISBN 0199206759

Reviewed by Daniel L Schacter & Brendan Gaesser Mrs. B was a 63-year-old psychiatrist who lived in Switzerland and was ­married to a state official. In an interview at a Geneva hospital, she ­complained about the rupture of a vessel in her left leg, recalled that her mother and brother had visited her earlier that day, and looked forward to a ­reception she would host at her home that evening. Despite her ­sincere belief that she was telling the truth, none of these events were real. Mrs. B was in fact a patient at the hospital, where she was ­recovering from a severe hemorrhage of a vessel in her brain. The ­striking disconnection between Mrs. B’s beliefs and her current reality provides the jumping off point for Armin Schnider’s fascinating new book, The Confabulating Mind. The basis for understanding the misguided convictions of patients such as Mrs. B goes all the way back to 1932, when the British ­psychologist Frederic C. Bartlett published a landmark ­volume, Remembering. In it, he argued that human memory is not a ­simple reproductive ­system, but instead involves complex constructive ­processes that are prone to error: when we remember, we piece together fragments of stored ­information under the influence of general ­knowledge and beliefs. Although the field took time to adopt this view, many cognitive ­psychologists endorsed Bartlett’s constructive approach by the 1970s and the psychological study of memory distortions has flourished ever since. The picture looks a bit different when we turn to neuroscience. For much of the twentieth century, neuroscientists paid little ­attention to the kinds of memory ­distortions that Bartlett and later cognitive psychologists believed would yield crucial insights into the nature of memory. Yet there was one ­distortion exhibited by brain-damaged patients that attracted the ­attention of investigators interested in the brain and ­memory: ­confabulation, the production of fabricated ­narratives and ­experiences. Schnider’s Mrs. B constitutes a classic example of a ­confabulating patient: she has difficulty ­remembering what ­actually happened in her past but strongly believes in the reality of her ­confabulations. As Schnider points out in a chapter that contains an informative history of confabulation, neurologists had begun to provide vivid case reports of brain-damaged patients who were similar to Mrs. B as early as the 1880s. One point noted by early clinical observers, and ­emphasized in this book, is that multiple forms of confabulation can be ­distinguished, The authors are at the Department of Psychology, Harvard University, Cambridge, Massachusetts, USA. e-mail: [email protected]

nature neuroscience volume 12 | number 4 | APRIL 2009

ranging from very brief or momentary ­confabulation to bizarre, ­fantastic confabulation that occurs in psychotic or demented states. The book focuses on what the author calls behaviorally ­spontaneous confabulation, in which patients such as Mrs. B ­spontaneously and frequently generate confabulated tales when attempting to deal with ongoing reality that seem plausible to an uninformed observer. One feature that is common to such patients, and that looms large in the author’s theoretical approach, is the fact that the confabulations are not invented out of whole cloth; they draw on bits of actual ­experiences that are incorrectly assembled and confused with ­current reality. Mrs. B, for instance, had hosted many receptions as the wife of a state official that probably contributed to her believing she would host another that evening; her brain injury and a prior surgery probably had something to do with her belief in a nonexistent leg injury. The book proposes that such confabulators have lost access to ­temporal information that normally allows a healthy ­individual to ­distinguish past from present, creating a ‘temporal context ­confusion’ that is a characteristic feature of behaviorally ­spontaneous ­confabulation. The author reviews evidence that the condition is ­consistently ­associated with damage to the posterior medial ­orbitofrontal cortex and to ­structures directly connected to it, such as the medial hypothalamus. The book also describes empirical investigations of the author’s ideas using a controlled experimental procedure that shows ­convincingly that temporal context confusion in confabulators is greater than in nonconfabulating amnesic patients or non-brain-damaged controls. Neuroimaging studies using a similar procedure with healthy ­volunteers indicate that the posterior medial orbitofrontal region is activated when memory for temporal context is required, suggesting to Schnider that this region allows us to filter memories according to their relevance to present reality, which is precisely what confabulators cannot do. Although this argument is compelling, one might question the strength of the link between the experimental results and ­everyday ­confabulation, as the former involves a temporal scrambling of events encoded after brain damage, whereas the latter typically comprises ­confusions among remote events that occurred well before brain ­damage. The author links his findings and ideas to those of others, ­providing a broad overview that will bring interested readers to the cutting edge of research on confabulation. He also covers studies of neurological patients who show related memory distortions, such as ­pathological false recognition, the experience of inappropriate familiarity in response to novel events. Recent developments in the analysis of what are termed ‘normal false memories’ is also nicely integrated in this book and this is one area where both behavioral and neuroimaging studies have greatly expanded our knowledge of the cognitive and neural mechanisms underlying memory distortion in the healthy brain. The Confabulating Mind succeeds in showing us that ­confabulation is not merely a neurological curiosity, but fits into an emerging ­theoretical picture in which neuroscience has an increasingly prominent role in ­illuminating the constructive nature of memory. We suspect that Bartlett would have been quite interested in what this book has to say and that ­investigators of memory and the brain, including researchers in the areas of ­memory and memory disorders and graduate students in neuroscience and ­psychology, will find much of value here.

367

news and views

Fine control: microRNA regulation of adult neurogenesis Qin Shen & Sally Temple

© 2009 Nature America, Inc. All rights reserved.

Neural stem cells transition through several progenitor stages before finally generating postmitotic neurons. New work shows that one of these steps, the generation of neuroblasts from transient amplifying precursors in the adult subventricular zone, requires downregulation of the transcription factor Sox9 by the microRNA miR-124. Over the past decade, we have made exciting progress in understanding the identity and location of ­neural stem cells in the adult brain. Now ­attention is turning to the mechanisms that control the ­progression from stem cells to ­differentiated progeny. Recently, ­microRNAs (miRNAs), a class of small, noncoding RNAs, have been identified as ­important ­regulators of many ­biological processes, ­including ­organogenesis and disease development. In an elegant study in this issue, Cheng et al.1 found that miR-124 ­stimulates adult neurogenesis by down-regulating the SRY-box transcription ­factor Sox9 (Fig. 1). Adult neurogenesis mainly occurs in two regions of the forebrain: the ­subventricular zone (SVZ) of the striatum and the ­hippocampal dentate gyrus. In the adult SVZ, the largest adult neurogenic niche, the slowly dividing type B stem cells express the astrocyte marker glial fibrillary acidic ­protein (GFAP). Type B stem cells produce more ­rapidly ­proliferating transit ­amplifying type C cells that expand the progenitor pool and in turn generate type A neuroblasts that express ­immature ­neuronal markers such as ­doublecortin and TuJ1. Type A­ neuroblasts divide as they migrate in the ­rostral ­migratory stream toward the olfactory bulb. There they differentiate into granule neurons that ­integrate into the ­granule layer or ­periglomerular ­neurons in the ­glomerular layer2. Transition between these stem cell ­lineage ­compartments is regulated by both exogenous and cell-intrinsic factors. Epigenetic factors such as DNA methylation, histone

The authors are at the New York Neural Stem Cell Institute, Rensselaer, New York, USA. e-mail: [email protected] or [email protected]

GFAP+ astrocytes

GFAP+ EGFR+ stem cell astrocytes

GFAP– EGFR+ transit amplifying cells

Dcx+ Ki67+ dividing neuroblasts

Dcx+ Ki67– neuroblasts

NeuN+ mature neurons

Sox9 protein Sox9 mRNA miR-124 REST? PTBP1? Figure 1 miR-124 regulates the lineage progression of adult neural stem cells in the SVZ via repression of Sox9 translation. Sox9 mRNA is expressed in GFAP+ type B stem cells and throughout most of the progenitor-cell lineage. miR-124 is turned on toward the end of the progenitor lineage, so that in neuroblasts it represses the translation of Sox9 mRNA into protein, allowing neurogenesis to proceed. Other mechanisms could involve targets such as REST/NRSF, which suppresses neurogenesis genes in glial cells and is both a target and a repressor of miR-124 (ref. 15), and PTBP1, a gene that is important for neuronal differentiation 4 and is repressed by miR-124.

­ odification and regulatory noncoding RNAs m affect neural stem cell fate ­specification3. ­miRNAs have the potential to specifically ­regulate a large set of target molecules, ­possibly affecting cell fate in a programmatic way, and the role of miRNAs among stem cell gene ­networks is being actively explored. miRNAs are 22-nucleotide-long small ­noncoding RNA sequences that alter gene ­expression by post-transcriptional inhibition or degradation of complementary mRNA sequences4. A miRNA recognizes its target mRNA through a ‘seed match’ between the seed, a 6 nucleotide stretch at the 5′ end of the miRNA, and a matching region at the 3′ ­untranslated region of the mRNA. The short seed match and imperfect base-pairing

nature neuroscience volume 12 | number 4 | april 2009

between miRNAs and their targets allows an miRNA to regulate many genes, possibly even hundreds, making them ideal ­candidates to act as master ­controllers. This feature is ­particularly ­advantageous for ­regulating the ­multifaceted and intertwined pathways involved in ­producing CNS cells, with their exquisite variety. Indeed, the brain expresses a more complex miRNA signature than other organs. miR-124 is one of these ­signature ­miRNAs that is enriched in the brain5. miR-124 is expressed at a low level in ­embryonic stem cells and neuronal precursors, but is markedly upregulated in differentiated neurons6. In tissue culture, miR-124 enhances neuronal differentiation and promotes ­neurite outgrowth (for a review, see ref. 6). The

369

© 2009 Nature America, Inc. All rights reserved.

news and views in vivo functions of miR-124 in the developing ­nervous system are less clear. In the embryonic chick spinal cord, miR-124 has been found to modestly enhance neuronal differentiation7 or to have no effect on neurogenesis8. Cheng et al.1 focused on the function of miR-124 in the adult SVZ. Using in situ ­hybridization, a challenging technique for detecting ­miRNAs that this group helped to pioneer, they found low levels of miR-124 in the SVZ and ­rostral migratory stream and high levels in the ­olfactory bulb. miR-124 was expressed in ­neuroblasts and mature ­neurons, but not in stem cells and type C cells. This ­expression ­pattern is ­consistent with the increasing ­expression of miR-124 that was observed as ­neurons ­differentiated ­during embryonic forebrain development. Using ­fluorescent-activated cell sorting (FACS), Cheng et al.1 separated SVZ cells derived from the hGFAP-gfp ­transgenic mouse into stem cell astrocytes (GFP+ EGFR+ mCD24–), other ­astrocytes (GFP+ EGFR– mCD24–), transit amplifying type C cells (GFP– EGFR+ mCD24–) and type A neuroblasts (GFP– EGFR– ­mCD24low) and confirmed that miR-124 is highly expressed in neuroblasts as, compared to other SVZ cells, using quantitative RT-PCR. The expression pattern of miR-124 in the SVZ suggests that it could regulate adult ­neurogenesis. The authors tested this ­hypothesis using a ­penetratin-conjugated 2′-O-methyl (2′OMe) antisense miR-124 (AS-124) to block ­endogenous miR-124 and a replication ­incompetent retrovirus (RV-124) to ­ectopically express miR-124 in ­dividing cells. miR-124 knockdown in cultures of FACS-sorted stem cell astrocytes reduced the ­number of ­postmitotic neurons and increased the ­number of both type C and dividing type A cells that developed. In ­contrast, ectopic expression of miR-124 ­promoted cell cycle exit, neuronal ­differentiation and loss of ­astrocyte ­markers. Delivery of AS-124 using a mini-osmotic pump or injection of RV-124 into the SVZ caused similar fate changes in the stem cell lineage in vivo. To further determine the role of miR-124 in SVZ cells at different stages, the authors used FACS-sorted cell ­populations and a ­neurosphere assay in which ­progenitor cells are grown in nonadherent ­conditions to assess their proliferative ability. In ­normal ­conditions, type B and type C cells, but not type A cells, can form neurospheres. Knockdown of miR-124 in type C cells resulted in substantially more ­neurospheres, but knockdown of miR-124 in type A ­neuroblasts did not endow them with the ­ability to form neurospheres. Thus, the authors concluded that miR-124 acts on ­transit amplifying cells to promote their ­transition to

370

neuroblasts, but the lineage progression ­cannot be reversed by reducing miR-124. How does miR-124 affect SVZ ­regeneration? Treatment with the antimitotic cytosine-b-darabinofuranoside for 1 week via minipump kills the rapidly dividing neuroblasts and transit amplifying cells, after which the stems cells are activated and regenerate the system with remarkable veracity9. Applying AS-124 via minipump ­during the ­regeneration period increased the number of dividing cells in the SVZ, and delayed, but did not prevent, their eventual differentiation. This ­indicates that miR-124 controls the timing of ­lineage ­progression of SVZ cells rather than ­determining their fate. What are the critical gene targets of miR-124 that mediate this effect? Cheng et al.1 ­identified and validated three miR-124 ­targets: Dlx2 and Jag1, two genes that are important for ­neurogenesis in the SVZ, and Sox9, whose function in the SVZ has not been studied. The authors demonstrated that Sox9 mRNA is a bona fide functional target of miR-124 in the adult SVZ and that ­downregulation of Sox9 by miR-124 is required for neuron ­production in cell ­culture. In the developing spinal cord, Sox9 is ­necessary for glial cell generation, and it ­suppresses neuron generation, thus helping to control the neuro-glial cell fate switch10. In the adult SVZ, Sox9 promotes the generation of GFAP-positive cells while suppressing ­neuronal production. Given that a variety of SVZassociated cells express GFAP—type B stem cells, other type B ­astrocytes with supportive functions in the niche and ­differentiated astrocytes that lie just a few microns below the ventricular surface—it will be ­interesting to determine whether Sox9 maintains the SVZ stem cell fate or drives cells toward ­differentiation of other astrocyte ­populations. Sox9 could have multiple roles, as it is critical for stem cell maintenance in the hair follicle11, but is required for ­differentiation of ­intestinal stem cells12. Effective SVZ ­neurogenesis ­probably requires repression of both stem cell and glial cell fates; perhaps by targeting Sox9, miR-124 can accomplish both. Overexpression of miR-124 represses more than 100 genes simultaneously in HeLa cells13; however, the extent of repression of ­different ­targets may differ14. Moreover, one or two of these targets might be more ­important ­targets. When the authors ­introduced a form of Sox9 that is not subject to miR-124 ­repression into SVZ cells, neuron ­differentiation was inhibited, as expected. Notably, ­concomitant ­overexpression of miR-124 in this ­context could not rescue the ­phenotype. This ­supports the notion that Sox9 is a key target of ­miR-124

in adult ­neurogenesis, but also raises ­questions about the importance of other targets, such as Dlx2 and Jagged1. This outcome could be explained by relative ­effectiveness: miR-124 appears to eliminate Sox9 protein ­expression, but only partially repress Dlx2. Or perhaps Sox9 is a more ­fundamental ­regulator of ­neurogenesis and these other targets are ­further down the ­hierarchy. Thus, it would be ­interesting to ­identify the downstream ­effectors of Sox9 in SVZ cells. In addition to Sox9, Dlx2 and Jagged1, other targets of miR-124 have been reported, ­including PTBP1 and SCP1, a component of the RE1-silencing transcription repressor/­neuron ­restrictive silencing factor (REST/NRSF)4. PTBP1 and REST/NRSF are two potent ­repressors of neuronal ­differentiation, acting through different pathways. PTBP1 represses neuron-specific alternative exon ­inclusion in the nervous system. REST/NRSF is a ­critical ­transcriptional ­regulator of ­neuronal ­differentiation, repressing ­neuron-specific genes in non-neuronal cells4. In theory, miR-124 could influence a broad array of fate-­determining genes via regulating these ­additional targets (Fig. 1), and it will be ­interesting to explore these ­possibilities ­further in the adult SVZ system. In conclusion, adult stem cell ­populations must be finely controlled, balancing the need for new cells with the need to keep tight reins on cell division. miRNAs are prime ­candidates for this application, as they enable delicate gene ­regulation. This pioneering study1 ­identifies miR-124 and its target Sox9 as being ­important elements of the ­neurogenesis ­regulatory ­network and paves the way for a fuller ­understanding of the role of miRNAs in this process, ­adding a microscale, but powerful, tool to ­combat aging and ­neurodegenerative disease. 1. Cheng, L., Pastrana, E., Tavazoie, M. & Doetsch, F. Nat. Neurosci. 12, 399–408 (2009). 2. Alvarez-Buylla, A. & Lim, D.A. Neuron 41, 683–686 (2004). 3. Namihira, M., Kohyama, J., Abematsu, M. & Nakashima, K. Epigenetic mechanisms regulating fate specification of neural stem cells. Philos. Trans. R Soc. Lond. B Biol. Sci. 363, 2099–2109 (2008). 4. Makeyev, E.V. & Maniatis, T. Science 319, 1789–1790 (2008). 5. Lagos-Quintana, M. et al. Curr. Biol. 12, 735–739 (2002). 6. Zeng, Y. Mol. Pharmacol. 75, 259–264 (2009). 7. Visvanathan, J., Lee, S., Lee, B., Lee, J.W. & Lee, S.K. Genes Dev. 21, 744–749 (2007). 8. Cao, X., Pfaff, S.L. & Gage, F.H. Genes Dev. 21, 531–536 (2007). 9. Doetsch, F., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. Proc. Natl. Acad. Sci. USA 96, 11619–11624 (1999). 10. Stolt, C.C. et al. Genes Dev. 17, 1677–1689 (2003). 11. Nowak, J.A., Polak, L., Pasolli, H.A. & Fuchs, E. Cell Stem Cell 3, 33–43 (2008). 12. Bastide, P. et al. J. Cell Biol. 178, 635–648 (2007). 13. Lim, L.P. et al. Nature 433, 769–773 (2005). 14. Bartel, D.P. & Chen, C.Z. Nat. Rev. Genet. 5, 396–400 (2004). 15. Conaco, C., Otto, S., Han, J.J. & Mandel, G. Proc. Natl. Acad. Sci. USA 103, 2422–2427 (2006).

volume 12 | number 4 | april 2009 nature neuroscience

news and views

Proteoglycans specify Sonic Hedgehog effect Catherine Vaillant & Denis Monard

© 2009 Nature America, Inc. All rights reserved.

How can the multifunctional factor Sonic Hedgehog (SHH) elicit specific responses from its target cells? A study now pinpoints proteoglycans as crucial anchors and modulators of SHH signaling, eliciting a proliferation response.

Catherine Vaillant is a member of the Developmental Genetics group, Department of Biomedicine, University of Basel, Basel, Switzerland. Denis Monard is at the Friedrich Miescher Institute for Biomedical Research, Basel, Switzerland. e-mail: [email protected] or [email protected]

SHHAla

SHH CardinWeintraub motif

Smo

GPI anchor

Heparan sulfate

Ptc1

Core protein Heparan sulfate PG

Proteoglycans have been recognized as being major functional components of the cell ­surface and the extracellular matrix. Classical ­biochemical studies, many ­dealing with blood homeostasis, have indicated that ­proteoglycans can promote or inhibit ­numerous protein-­p rotein ­i nteractions. However, the ­complexity of proteoglycan ­biochemistry, the diversity and complexity of the attached ­g lycoseaminoglycan chains, and their ­interactions with a ­multitude of cell ­surface receptors, growth factors and proteases or their inhibitors 1 make the design of effective experimental approaches ­challenging. Thus, we are still far from a clear and ­comprehensive understanding of ­proteoglycan function. Chan et al.2 show how proteoglycan ­interaction determines the function of SHH in the ­developing cerebellum. SHH has dual ­functions in embryonic and ­postnatal ­development, promoting both precursor ­proliferation and morphogenetic ­patterning3. The molecular basis of these distinct and opposing effects of SHH on its target cells has remained obscure. The same group had ­previously shown that ­proteoglycan binding of the SHH N-terminal Cardin-Weintraub motif is required for SHH’s mitogenic effect on ­postnatal cerebellar granule cell ­precursors4. Now, Chan et al.2 mutated the CardinWeintraub motif in mice, ­abolishing ­binding of proteoglycans in vivo. This approach is ­particularly elegant insofar as it targets not the proteoglycans but the way SHH ­interacts with them. Thus, ­neither proteoglycans’ actions on other GPI-linked factors or on other types of proteoglycan-­dependent modulation are impaired, nor is any proteoglycan­independent SHH ­signaling. In vitro and ex vivo, the proteoglycan ­binding–deficient SHH-Ala protein had decreased binding affinity for heparin, but its interactions with the SHH receptor Ptc were

Gli1

Gli1

Gli3

Gli2-A

Gli3 Gli2

Gli2 Gli2-R

Transcription of cell cycle genes

Repression of cell cycle genes

Proliferation

Patterning

Figure 1 SHH binding to heparan sulfate proteoglycans elicits a proliferative response. The membrane of cerebellar granule neuron precursors is rich in heparan sulfate proteoglycans, with the chains being bound to a core protein, which is linked to the membrane through a glycosylphosphatidylinositol (GPI) anchor. The Cardin-Weintraub motif present in the N terminus of the SHH ligand binds to heparan sulfate. This not only allows anchoring and locally increased concentration of SHH to the pericellular membrane, but also triggers a specific transduction cascade. In its activator form, the downstream transcriptional factor Gli2 (Gli2-A) regulates the transcriptional expression of some cell cycle genes, driving continuous proliferation. The mutated SHH-Ala protein lacks the Cardin-Weintraub motif. It is nevertheless still able to bind to its receptor, Ptc1, enhancing the levels of Gli2 in its repressor form (Gli2-R) and inhibiting the expression of a set of cell cycle genes, leading to final cell cycle exit. PG, proteoglycan.

intact. The Shh-Ala mutant mice showed ­ roliferation defects that affected the CNS, p ­particularly in the mitogenic niches located in the hippocampus, subventricular zone and cerebellum, which are known SHH ­targets5. Overall CNS patterning was not ­disturbed in the Shh-Ala mice, in contrast with the ­previously reported phenotype of Shh ­ablation6. Cerebellar granule neuron ­precursors undergo maximal proliferation in the ­postnatal outer external granule layer (EGL), followed by differentiation and migration into the internal granule layer7. Chan et al.2 found that wild-type, but not Shh-Ala, EGL promoted proliferation of both wild-type and mutant ­precursors. Furthermore, wild-type SHH ­protein ­accumulated in the outer EGL, but much less SHH-Ala accumulated there.

nature neuroscience volume 12 | number 4 | april 2009

This suggests that the proteoglycans function to anchor SHH near its target cells. SHH responses are mediated by the ­t ranscription factors Gli1, Gli2 and Gli3 (Fig. 1) 8. Gli transcription factors can ­function as either transcriptional activators or ­repressors9. Gli1 exclusively functions as an activator10. Gli2 can function as ­activator and repressor, and distinct isoforms have been identified that subserve these different functions11. The Gli2 activator/­repressor ratio is altered in vivo in Shh-Ala mice. SHH interacts with proteoglycans on ­cerebellar granule neuron precursors to alter the nature and timing of Gli2-dependent ­transcription. Analyzing a panel of genes ­downstream of SHH signaling, Chan et al.2 identified cyclin D1, cyclin D2 and Bmi-1

371

© 2009 Nature America, Inc. All rights reserved.

news and views as being the major effectors that were affected by SHH-proteoglycan interactions. Consequently, these genes may be critical for SHH-controlled neuronal proliferation. Their expression was consistently detected in in vitro and in vivo models, reinforcing the strength of the ­finding. These data led the authors to ­propose that an SHH-dependent expression profile ­signature is associated with neural progenitor ­proliferation that is distinct from that involved in ­t issue patterning. This documentation of SHH ­proteoglycan–dependent gene expression is a landmark for the neurosciences. The Chan et al. study2 represents a major step toward a full understanding of SHH ­signaling complexity by assessing its role in the ­establishment and maintenance of a ­favorable microenvironment for ­neural ­p recursor ­p roliferation. Moreover, the authors ­demonstrate that proteoglycan-SHH ­interactions are ­essential for the signaling of proliferation ­versus ­differentiation. The results presented by Chan et al.2 also ­suggest that proteoglycans might ­contribute to the developmental restriction of SHHdriven proliferation in mitogenic niches,

albeit through very different interactions. An ­increasing ­number of endogenous ­inhibitory ligands have been identified that attenuate SHH ­mitogenicity during ­cerebellar ­postnatal ­development. Some of these also contain Cardin-Weintraub motifs and bind to ­heparan sulfate. For example, ­vitronectin12, protease nexin-1 (ref. 13) and FGF-2/FGFR1 (ref. 14) signaling can stop the expansion of granular precursors by ­specifically ­antagonizing the SHH pathway. It is tempting to speculate that these ligands might ­compete with SHH for ­proteoglycan binding sites. This ­occupancy would chase away SHH, thus ­depleting the proliferative niche and ­possibly triggering downstream pro-­differentiating ­signaling. This ­hypothesis ­obviously requires further ­experimental ­evidence. Furthermore, both proteoglycans and SHH have been implicated in cancer. Tumor growth, angiogenesis and ­metastasis ­originating from tumorigenic cell lines are attenuated in mice lacking the heparan ­sulfate ­proteoglycan ­g lypican-1 (ref. 15). Thus, in addition to advancing our understanding of Shh ­signaling, Chan et al.’s2 approach of ­interfering with ­proteoglycan-modulated

signaling via the Cardin-Weintraub motif of proteoglycan-­binding proteins may well inspire ­developmental and cell biologists and have an effect on studies investigating the role of SHH in cancer. 1. Kreuger, J., Spillmann, D., Li, J.P. & Lindahl, U. J. Cell Biol. 174, 323–327 (2006). 2. Chan et al. Nat. Neurosci. 12, 409–417 (2009). 3. Varjosalo, M. & Taipale, J. Genes Dev. 22, 2454–2472 (2008). 4. Rubin, J.B., Choi, Y. & Segal, R.A. Development 129, 2223–2232 (2002). 5. Ahn, S. & Joyner, A.L. Nature 437, 894–897 (2005). 6. Lewis, P.M., Gritli-Linde, A., Smeyne, R., Kottmann, A. & McMahon, A.P. Dev. Biol. 270, 393–410 (2004). 7. Vaillant, C. & Monard, D. Cerebellum published online, doi:10.1007/s12311-009-0094-8 (18 February 2009). 8. Ulloa, F. & Briscoe, J. Cell Cycle 6, 2640–2649 (2007). 9. Sasaki, H., Nishizaki, Y., Hui, C., Nakafuku, M. & Kondoh, H. Development 126, 3915–3924 (1999). 10. Aza-Blanc, P., Lin, H.Y., Ruiz i Altaba, A. & Kornberg, T.B. Development 127, 4293–4301 (2000). 11. Pan, Y., Bai, C.B., Joyner, A.L. & Wang, B. Mol. Cell. Biol. 26, 3365–3377 (2006). 12. Pons, S., Trejo, J.L., Martinez-Morales, J.R. & Marti, E. Development 128, 1481–1492 (2001). 13. Vaillant, C. et al. Development 134, 1745–1754 (2007). 14. Fogarty, M.P., Emmenegger, B.A., Grasfeder, L.L., Oliver, T.G. & Wechsler-Reya, R.J. Proc. Natl. Acad. Sci. USA 104, 2973–2978 (2007). 15. Aikawa, T. et al. J. Clin. Invest. 118, 89–99 (2008).

Neuronal communication: a detailed balancing act Emilio Salinas What controls the functional connections between sending and receiving neurons? A new model suggests that each receiver circuit has a local switch that is controlled by the balance between excitation and inhibition. Learning and experience can modify the ­physical connections between neurons in the ­mammalian brain over time scales of ­minutes to days. However, the flexibility of everyday ­behavior ­provides abundant evidence that the ­functional connections between ­sensory and motor ­systems can change virtually ­instantaneously. If, for ­example, you are ­typing a document, I can ask you to press the H key with your right index ­finger or with any other finger, and vice versa, I can ask you to use the right index finger to press any key. Thus, the map between letters and ­finger presses can change at any moment and any ­possible ­combination can be executed on demand. At the neural level, this means that the ­sensory afferents ­conveying visual ­information about

Department of Neurobiology and Anatomy, Wake Forest University School of Medicine, Winston-Salem, North Carolina, USA. e-mail: [email protected]

372

the ­keyboard must be able to ­communicate with all the ­efferent motor circuits that ­control the fingers. Therefore, a mechanism must exist to gate the ­underlying ­connections according to the ­ongoing behavioral task so that only the ­appropriate ­communication channel is switched on at any moment. Understanding how this ­switchboard works is a ­fundamental, but ­difficult, problem in systems and ­computational ­neuroscience. In this issue, Vogels and Abbott1 describe a ­biophysically plausible account of how this switchboard may work. According to their new model, each ­excitatory input to a ­downstream group of receiver neurons is ­carefully ­cancelled by local inhibition, except when those receiver neurons need to ­replicate one of the inputs; in that case, that one ­signal is ‘gated on’ by ­upsetting the detailed balance between ­excitation and ­inhibition. Thus, their model provides a ­specific ­mechanism for how neural signals may route through different ­circuits.

This study1 is the result of advances in understanding two phenomena in ­cortical ­neurophysiology: the balance between ­excitation and inhibition in local circuits and the ­transmission of information from one circuit to another. The first issue has been thoroughly ­studied by both ­theoreticians and ­experimentalists. Modeling studies have shown that the simultaneous action of ­excitation and inhibition generates large ­fluctuations in ­membrane potential, which are critical for explaining the high ­variability that ­characterizes cortical ­neuron spike trains2. Modeling has also revealed that the ­dynamics of both single neurons3,4 and whole networks5 are extremely ­different in the ­balanced ­(fluctuation driven) and unbalanced (excitation driven) regimes. Correspondingly, experiments have shown6,7 that feedback mechanisms in local ­microcircuits ensure that excitation and ­inhibition stay ­notably proportional, ­maintaining a delicate balance throughout a large range of activation levels.

volume 12 | number 4 | april 2009 nature neuroscience

© 2009 Nature America, Inc. All rights reserved.

news and views a

b

Gates off

100

Sender 1

Sender 2

Receiver

Firing rate (spikes per s)

The second issue, relating to ­information transmission, is a computational ­problem that, on the surface, seems to be rather ­innocuous. Suppose that the neurons in layer L1 of a ­network respond to a ­stimulus and that this signal goes from L1 to L2, from L2 to L3, and so on. Under what conditions will the ­signal be faithfully transmitted? Or, how many ­layers can the original ­signal reach ­w ithout it being corrupted? This problem dates back to a study that proposed8 that faithful ­t ransmission may be achieved if the neurons that are active in a given layer fire volleys of synchronous spikes that excite the next layer, and so on. This model was ­developed further9 and an ­alternative proposal10 later emerged ­suggesting that ­s ynchronicity is not critical and that ­asynchronous changes in firing rate can also be transmitted in a stable manner. However, in either case, there are two severe complications. First, real neuronal networks are not purely feedforward (L1 → L2 → ... → LN), but instead have ­extensive ­feedback ­connections; in fact, cortical areas that ­communicate with each other are ­t ypically interconnected ­bidirectionally11. Second, ­neurons do not ­simply relay ­information through waves of ­excitation; instead, when a neuron is ­activated, it ­normally receives large quantities of both excitation and ­inhibition. Therefore, the ­stable propagation of activity needs to ­coexist with the ­irregular ­f luctuations that ­characterize ­b alanced ­networks. Reliable ­signal ­t ransmission is tricky, because the ­signal may either fade away and die amidst the sea of ­fluctuations or else grow uncontrollably into a sort of shock wave. Nevertheless, recent ­simulation studies 12,13 have found conditions for ­reliable signal ­propagation under ­realistic ­constraints, both for ­synchronous ­volleys of activity and for ­v ariations in ­firing rate. Vogels and Abbott developed one of those models13, which is the starting point for their current work. The network constructed by Vogels and Abbott1 contains about 20,000 neurons that produce spikes and interact through model synaptic conductances. It ­represents a sparse description of a large swath of cortex ­spanning many areas. Without any input, the recurrent connectivity of the net produces irregular, asynchronous activity with low ­firing rates around 8 spikes per s. Embedded in this structure, there are smaller ­subnetworks ­representing distinct cortical areas, ­designated as ‘senders’ and ‘receivers’ (Fig. 1a). The ­crucial part of the model’s design is that a sender group excites both excitatory and inhibitory neurons in a

Gate 1 on

Gate 2 on

0 100

0 100

0 100

0

0

Time (ms)

900

Figure 1 Gating neuronal inputs through detailed balance of excitation and inhibition. (a) Schematic of the model by Vogels and Abbott1. Receiver neurons get excitatory input from two groups of sender neurons (red and purple). Senders contact both excitatory (orange) and inhibitory receiver neurons (dark and light blue). (b) Responses as functions of time. The excitatory receiver neurons (orange trace) may either not respond (gates off), follow the signal from sender 1 (gate 1 on) or follow the signal from sender 2 (gate 2 on), depending on the gain of the local inhibitory neurons (blue traces).

receiver group such that, by default, when the senders are highly active, the receiving ­excitatory neurons remain rather quiet, ­firing at low rates. This is because the ­receiving inhibitory neurons act locally to ­cancel the senders’ excitation. The authors refer to this local equilibrium as ‘detailed ­balance’. Now, a particular receiver subnetwork can be connected to many sender subnetworks, but as long as each sender’s signal is balanced by a number of local inhibitory neurons, ­nothing much happens to the excitatory neurons at the receiving end. However, if the detailed balance of a particular ­incoming signal is disrupted (for example, if the gain of the corresponding local interneurons is decreased) then that signal is gated on and drives the receiver excitatory neurons, which reproduce it quite faithfully (Fig. 1b). What is interesting about this circuit organization is that by selectively altering the local balance of the receiver subnetwork any of the sender signals can be gated on. Thus, the excitatory receiver neurons may respond to any one of a large number of sender inputs without scrambling their signals. There are three aspects of this work that are notable. First, the sender and receiver ­populations are part of a larger network with realistic dynamics (irregular, ­asynchronous and low firing) that does not require an ­external source of noise and the activation of these subnetworks does not disrupt the ­activity of the rest of the model neurons. Thus, the overall dynamical regime is quite accurate. Second, the idea that gating functions can be implemented by tweaking the ­balance between excitation and inhibition is an old one, but the authors, without ­invoking any

nature neuroscience volume 12 | number 4 | april 2009

exotic ­assumptions, have achieved a ­specific and ­reasonable account of how this may ­happen. In particular, they performed a ­beautiful ­analysis of the number of input ­signals that can be ­cancelled and gated by a local ­population of inhibitory neurons. They demonstrate that if a receiver signal is ­composed of N sender signals (firing rates), there is a ­fundamental limitation in the ­accuracy with which a single ­component ­signal can be extracted that depends on N and is ­independent of the extraction method. This in itself is a remarkable result about ­neural ­coding. What it means for the detailed-balance gating mechanism is that as long as no more than ~5 sender inputs are active ­simultaneously, 200 local ­inhibitory neurons are capable of gating on, with high accuracy, any one of 600 ­possible sender inputs (see Fig. 7 in Vogels and Abbott1). Third, the model makes a specific and testable prediction: at least some inhibitory neurons should be strongly active when their nearby excitatory targets are quiet (gating off). The converse pattern should also be seen, albeit not always; when those excitatory targets are strongly active (gating on), the inhibitory responses should decrease ­sometimes, but not always, because although the excitatory ­neurons should be able to react to any sender, each inhibitory neuron should correlate only with some of them. Relatively few studies14,15 have ­investigated mechanisms for switching ­information from one cortical circuit to another. Notably, ­however, one of them14 works in the ­opposite way, modulating the gain of the sender ­neurons. In contrast, in Vogels and Abbott’s model1, routes already exist from the ­senders

373

© 2009 Nature America, Inc. All rights reserved.

news and views to the receiver, but they are shut down by default; the relevant ­connection is activated locally, at the receiver end, ­without the need to alter the sender’s activity. It is possible that real ­circuits exploit both ­strategies, ­depending on the ­current ­behavior, the ­circuit’s ­function or its ­connectivity ­constraints. These ­mechanisms could be tested by recording from suspected sender and receiver neurons in tasks in which the relationship between multiple sensory ­signals and motor ­effectors can change, as in our ­keyboard example. In addition, the ­mechanism put forth by Vogels and Abbott1 should have a distinct ­microanatomical ­substrate reflecting the ­convergence of ­excitation and inhibition driven by the same input. The model by Vogels and Abbott1 may solve part of the neuronal switchboard ­puzzle, but many questions still remain. How is the detailed balanced between excitation and ­inhibition maintained? How are the ­connection pathways established and kept

separate? Perhaps the most pressing issue is the nature of the control ­signal that disrupts the local balance of excitation and inhibition to open each gate. The authors showed that it may work by acting on very few interneurons (<50) per receiver net. Therefore, this ­control needs to be both highly selective, to alter each interneuron by the right proportion, and fairly strong, to be able to nearly silence some interneurons. It is not clear how this can be implemented and the solution will ­probably require complementary new modeling and new experimental approaches. This may be the next dot that needs to be connected in the study of signal transmission. These new results suggest that the neural circuits that give rise to flexible, ­nonreflexive behavior may be both highly complex and exquisitely specific. More generally, they should spur further investigations into the neural basis of such flexibility, which remains a fundamental neuroscientific challenge.

1. Vogels, T.P. & Abott, L.F. Nat. Neurosci. 12, 483–491 (2009). 2. Shadlen, M.N. & Newsome, W.T. Curr. Opin. Neurobiol. 4, 569–579 (1994). 3. Troyer, T.W. & Miller, K.D. Neural Comput. 9, 971–983 (1997). 4. Salinas, E. & Sejnowski, T.J. J. Neurosci. 20, 6193–6209 (2000). 5. van Vreeswijk, C. & Sompolinsky, H. Science 274, 1724–1726 (1996). 6. Shu, Y., Hasenstaub, A. & McCormick, D.A. Nature 423, 288–293 (2003). 7. Haider, B., Duque, A., Hasenstaub, A.R. & McCormick, D.A. J. Neurosci. 26, 4535–4545 (2006). 8. Abeles, M. Corticonics: Neural Circuits of the Cerebral Cortex (Cambridge University Press, New York, 1991). 9. Diesmann, M., Gewaltig, M.-O. & Aertsen, A. Nature 402, 529–533 (1999). 10. van Rossum, M.C.W., Turrigiano, G.G. & Nelson, S.B. J. Neurosci. 22, 1956–1966 (2002). 11. Felleman, D.J. & Van Essen, D.C. Cereb. Cortex 1, 1–47 (1991). 12. Kumar, A., Rotter, S. & Aertsen, A. J. Neurosci. 28, 5268–5280 (2008). 13. Vogels, T.P. & Abbott, L.F. J. Neurosci. 25, 10786–10795 (2005). 14. Olshausen, B.A., Anderson, C.H. & Van Essen, D.C. J. Neurosci. 13, 4700–4719 (1993). 15. Salinas, E. BMC Neurosci. 5, 47 (2004).

It’s not you, it’s me. Really. Garrett B Stanley A subset of neurons in rat barrel cortex integrate information about the object a whisker contacts with the motion of the whisker at the time of contact, setting the stage for a highly specialized object localization system. The information flow entering the ­sensory ­pathways of the brain is shaped by our ­interactions with the environment 1,2. For example, when we look around the room, we move our eyes, head and body to control the images that fall on our retinas. When we want to feel a texture, we actively move our ­fingertips across the surface to gather ­information. When we want to locate an object in the dark, we reach out and move our arms until our hand comes into contact with the object. The ­signals that the brain has access to are ­therefore ­determined by both the ­features of the outside world and how we use our muscles to move our ­sensors ­(retina, ­fingertips, etc.) to meet the outside world. How do we interpret the outside world when the information that we receive is tangled up with the manner in which we gather it? Feeling around in the dark with one’s hand to determine the location of an object requires

The author is in the Coulter Department of Biomedical Engineering, Georgia Institute of Technology and Emory University, Atlanta, Georgia, USA. e-mail: [email protected]

374

π

a

b

π

Whisk cycle

Whisk cycle

0 0

2π

2π

Figure 1 Object localization in a head-centered coordinate system. (a) A single vibrissa on the right side of the face and the different positions of the vibrissa during the whisk cycle are shown. Contact with an object (black dot) occurs at a particular phase of the whisk cycle, given as a fraction of the entire cycle from 0 to 2π. For the example shown, assuming contact in the forward sweep, the phase is approximately 34 π at the point of contact. The phase of the contact is the same, regardless of the whisking frequency. (b) Whisking over a different amplitude leads to a different phase at the point of contact for the same object shown in a. Now the phase is closer to π, resulting in an ambiguity in object localization relative to the face.

some implicit knowledge of the motion and position of our arms. The ­physical contact (hard object meets soft skin) may be the same ­regardless of position, but when placed in the context of the position and trajectory of the arm, we ­unambiguously identify the ­location of the object. In this issue, Curtis and Kleinfeld3 investigate ­sensory signals in the primary ­sensory ­cortex that reflect

the ­combination of elements originating strictly from objects in the ­environment and elements ­associated with active, motordriven inputs to the ­sensory pathway. They utilized the rat vibrissa system in which the rats actively bring their facial ­whiskers ­(vibrissae) in contact with objects during exploratory behavior, a model system that has ­previously been shown to be capable of

volume 12 | number 4 | april 2009 nature neuroscience

© 2009 Nature America, Inc. All rights reserved.

news and views v­ arious forms of ­distance ­discrimination4–7. They report that the ­neural activity ­measured in the ­corresponding ­primary cortical ­representation of each ­whisker (the ­barrel ­cortex) depends on vibrissa contact with an object and on the moment at which the ­whisker contacts the object during ­rhythmic ­movement of the vibrissae ­(whisking). This activity ­represents a combination of ­afferent (contact) and ­reafferent (self-motion) ­signals that could be used for perception of the ­correct position of the object relative to the face of the rat. In an elegant set of experiments, Curtis and Kleinfeld3 implanted arrays of micro­electrodes in the barrel cortex of awake rats that were trained to palpate a sensor that reports ­whisker contacts. The ­spiking ­activity of single neurons in the ­barrel ­cortex was ­studied in relation to these ­contacts, as well as in ­relation to the ­whisking cycle (assessed by ­electromyogram recording of the facial ­musculature responsible for the active ­movement of the whiskers). The goal of the study was to ­understand how afferent touch ­signals are integrated with ­reafferent signals that are ­associated with the rat’s own ­generated motion of the ­follicles ­during active sensing. The ­assertion is that the ­mixture of afferent and self-­generated ­reafferent signals allows the rat to ­unambiguously identify object position in head-centered ­coordinate space. Given that the active whisking is ­approximately ­sinusoidal in the ­rostral-caudal (nose-to-tail) plane, with ­frequencies between 6 and 12 Hz ­(sweeping front to back 6 to 12 times per s), the ­contact of the whisker with an object occurs at a ­particular phase of the cyclic back-and-forth motion of the whisker (Fig. 1a). The phase is reported as a number in radians, ranging from 0 to 2π, to ­indicate whether the ­contact occurs at the ­beginning or at the end of the cycle, ­respectively. The main finding is that cortical neurons are tuned such that the touch response is greatest at a ­particular phase of the whisk cycle, ­suggesting that the ­reafferent ­signal ­associated with the active movement of the whiskers gates the ­cortical activity. The ­implication is that this could lead to a representation of the object in headcentered coordinates. The influence of motor signals in a primary sensory area ­supports the growing feeling in the sensory and motor communities that the artificial lines between historically designated sensory and motor areas of the brain are starting to blur. Although these data suggest a ­potential mechanism for integrating afferent ­signals

with internal representations of our own motion, we maintain the ability to ­recognize an object ­independent of its ­location in space. This ­presents a ­paradox in the ­influence of our own actions in ­acquiring sensory ­information that might be best summed up as we can’t live with it and we can’t live without it. On the one hand, we have the perception of an invariant world whose stability seems unthreatened no ­matter how erratic our behavior. When ­asking the ­question ‘What is that object?’, an apple is still an apple, ­regardless of how much our eyes, head and body move while we take in the visual ­information. The sandpaper still has the same feeling of roughness, as long as we are ­moving our fingertip across it in some ­reasonable way. The elements of our own part of the ­gathering of ­information are thus removed from our ­sensory ­experience. On the other hand, when we are locating an object relative to ourselves, the active ­element of the sensing becomes ­i mportant for answering the question ‘Where is the object?’. The position of my center of gaze or hand, or how far I have moved my eyes or hand, ­suddenly become ­relevant for ­answering the question, and thus we no ­longer wish to remove this from our ­experience, but instead to somehow ­incorporate this ­information for context, as has been ­pursued in the study of the preparation and guidance of movement8. The vibrissa ­pathway has been implicated in a rich continuum of ­behaviors, from inferring ­distance during locomotion9 to generating ­representations for texture discrimination10,11. The various behaviors are often framed ­simplistically in terms of questions of ‘what’ and ‘where’, although the ­convenience of such an artificial ­dichotomy probably breaks down outside of the ­laboratory ­setting, as things often do. For the vibrissa system, how might this all work when the rubber hits the road? If the ­cortical neurons are tuned to respond ­preferentially to contact at a particular phase of the whisk cycle, independent of the ­frequency of a particular whisking bout, as is shown for a subset of the neurons in the Curtis and Kleinfeld study3, then the relative ­activity of a population of cortical neurons tuned for ­different phases would ­unambiguously ­identify the position of the object in face-­centered ­coordinates. The ­contact with the object would occur at the same phase of the whisk cycle, just at ­different ­absolute times. However, the study3 also ­suggests that the representation is ­invariant to the ­amplitude of the whisk cycle,

nature neuroscience volume 12 | number 4 | april 2009

an ­observation that makes the framework for the neural ­coding more ­complicated. For two ­situations with the same whisking frequency, but of different ­amplitudes (Fig. 1a,b), the ­contact would occur at ­different phases of the whisk cycle, thus ­recruiting a ­different ­subpopulation of ­cortical cells. In this case, the position can still be ­unambiguously identified, but not from the observation of these cortical units alone. They would need to be judged in the context of an ­internal ­representation of the actual ­whisking motion (amplitude and frequency). As previous studies have reported, however, and Curtis and Kleinfeld3 confirm, cortical activity is modulated in a way to reflect the nature of the whisking even in the absence of ­contact, thus potentially providing just such a ­contextual signal. As with any ambitious study, Curtis and Kleinfeld’s3 findings inspire a whole range of further questions. From a ­circuitry ­perspective, given the ­observation that the phase­dependent cortical units ­predominately reside in cortical layer 4 (the input layer to ­primary cortices), but less so in other layers that form ­projections back to subcortical areas, what is the role of ­cortical feedback in ­establishing this neural code? How do we connect this finding, which relates to active whisking, to other aspects of active ­sensing, such as head and body ­movement? How might this coding operate during more ­continuous ­contact with objects during navigation? The vibrissa system is an enormously rich model system in which to pursue these types of ­questions, which are central to the way in which we, as organisms, interact with the external world. Continued exploration in this area may help us begin to identify the mechanisms by which we create ­representations of the environment in the ­context of our own movements within it.

1. Gibson, J.J. Psychol. Rev. 61, 304–314 (1954). 2. Gibson, J.J. Psychol. Rev. 69, 477–491 (1962). 3. Curtis, J. & Kleinfeld, D. Nat. Neurosci. 12, 492–501 (2009). 4. Krupa, D.J., Matell, M.S., Brisben, A.J., Oliveira, L.M. & Nicolelis, M.A. J. Neurosci. 21, 5752–5763 (2001). 5. Knutsen, P.M., Pietr, M. & Ahissar, E. J. Neurosci. 26, 8451–8464 (2006). 6. Hentschke, H., Haiss, F. & Schwarz, C. Cereb. Cortex 16, 1142–1156 (2006). 7. Mehta, S.B., Whitmer, D., Figueroa, R., Williams, B.A. & Kleinfeld, D. PLoS Biol. 5, e15 (2007). 8. Graziano, M.S.A., Yap, G.S. & Gross, C.G. Science 266, 1054–1057 (1994). 9. Vincent, S.B. Behav. Monographs 1, 7–85 (1912). 10. Guic-Robles, E., Valdivieso, C. & Guajardo, G. Behav. Brain Res. 31, 285–289 (1989). 11. Carvell, G.E. & Simons, D.J. J. Neurosci. 10, 2638–2648 (1990).

375

c o m m e n ta ry

The quest for action potentials in C. elegans neurons hits a plateau © 2009 Nature America, Inc. All rights reserved.

Shawn R Lockery & Miriam B Goodman The small size and high resistance of C. ­elegans neurons makes them ­sensitive to the ­random opening of single ion ­channels, ­probably ­rendering codes that are based on ­classical, all-or-none action ­potentials ­unworkable. The recent discovery in C. ­elegans of a special class of regenerative events known as ­plateau ­potentials introduces the possibility of ­digital neural codes. Such codes would solve the ­problem of ­representing information in ­nervous systems in which action potentials are unreliable. The ­publication in 1986 of an essentially ­complete wiring diagram of the C. elegans nervous ­system1 raised the ­prospect of the first ­comprehensive account of the behavior of an entire ­organism. It quickly became ­apparent, however, that a ­wiring diagram by itself is ­insufficient to explain behavior, even for an organism as simple as a nematode. Clearly, one also needs to know the intrinsic electrical ­properties of the neurons. Do they act as ­passive or active nodes? If active, do they fire all-or-none action potentials, graded ­regenerative responses or something else? With the ­introduction of specialized ­electrophysiological methods for patch-­clamping C. elegans ­neurons2–5, ­several laboratories are now addressing these ­questions directly. A recent study6 reported the ­surprising ­finding that the answer, at least in some C. ­elegans ­neurons, is something else. Recording in situ from the motor ­neuron class RMD, these researchers noted a ­distinctive regenerative response to current injection. A brief pulse of positive current in an RMD ­neuron (or a puff of glutamate) ­elicited a ­depolarization of more than 30 mV that was stable for at least 1 min (Fig. 1a) but could be terminated by a brief pulse of ­negative ­current. Thus, RMD neurons have two stable resting potentials, one near –70 mV and one

Shawn R Lockery is at the Institute of Neuroscience, University of Oregon, Eugene, Oregon, USA. Miriam B Goodman is in the Department of Molecular and Cellular Physiology, Stanford University, Stanford, California, USA. e-mail: [email protected]

a

b

C. elegans RMD Neuron

Action potential

Graded potential

Vm #2 ≈ -35 mV Vm Im

Vm #1 ≈ -70 mV

Vm Im

Figure 1 Schematic representations of Plateau potential Intrinsic oscillation regenerative events. (a) A quasi-stable plateau potential in C. elegans RMD neurons (as shown in ref. 6). A positive current pulse (lower trace) elicits a regenerative Vm depolarizing response followed by a Im sustained plateau (upper trace). Thus, RMD has two resting membrane potentials (Vm). The plateau can be terminated by a negative current pulse (after Mellem et al. 2008). (b) Four common types of regenerative neuronal activity. Intrinsic oscillations and plateau potentials are frequently accompanied by trains or bursts of spikes in neurons capable of firing action potentials (vertical lines). Dashed lines indicate amplitudes of all-or-none events.

near –35 mV. Notably, shifts between the two resting potentials were also observed to occur ­spontaneously, indicating that these events probably happen naturally in RMD neurons. Electrophysiologists have described four main types of regenerative events: action ­potentials, graded potentials, ­intrinsic ­oscillations and ­plateau ­potentials (Fig. 1b). The defining ­features of action ­potentials are well known; they are allor-none ­depolarizations with a ­stereotypical ­waveform that is ­independent of the ­amplitude and ­waveform of the ­triggering ­stimulus. The defining ­features of other types of ­regenerative events may be less well known. Graded ­potentials resemble action ­potentials, but their amplitude and waveform are ­sensitive to the amplitude and waveform of the ­stimulus. Intrinsic ­oscillations,

nature neuroscience volume 12 | number 4 | APRIL 2009

such as those that ­underlie endogenous ­bursting, are slow, cyclic ­alterations in ­membrane ­potential that are caused by ­antagonistic voltage­dependent currents. Plateau potentials are ­prolonged, all-or-none ­depolarizations that can be ­triggered and ­terminated by brief positiveand ­negative-current pulses, respectively7,8. Regenerative activity has been seen before in nematode neurons. In Ascaris, motor ­neurons fire graded potentials in response to the offset of hyperpolarizing current pulses9. Inhibitory motor neurons show slow ­oscillations that are probably intrinsic10, as do certain ventral cord interneurons11,12. There is also indirect ­evidence of Ca2+-dependent action ­potentials in the ­ventral nerve cord13. In C. elegans, ­depolarizing current pulses in chemosensory

377

© 2009 Nature America, Inc. All rights reserved.

c o m m e n ta r y neurons elicits active responses that outlast the ­stimulus pulse2. The events recorded by the above-mentioned study6 fit the criteria of plateau potentials ­perfectly in that they are long-lasting, all-or-none events that can be terminated by a n ­ egative-current pulse. In addition, the genetic ­manipulations and ionsubstitution experiments performed in that study6 suggest that RMD plateau potentials depend on a current that is carried mainly by Na+ and Ca2+ ions, which is consistent with the biophysical mechanisms of plateau potentials in other organisms14–19. The discovery of plateau potentials in C. ­elegans neurons is an important ­development, for it greatly expands the ­computational ­repertoire of the C. elegans nervous ­system. Plateau ­potentials are the ­biological ­equivalents of Schmitt ­triggers, which have many ­interesting ­applications, including ­oscillators, ­timers and flip-flop ­elements. Thus, it is not ­surprising that plateau potentials have been reported in a wide range of vertebrate and ­invertebrate ­organisms20–26, where they have been ­implicated in such ­functions as ­pattern ­generation and short-term memory. At ­present, almost ­nothing is known about either function in C. elegans, despite long-standing behavioral ­evidence of multiple pattern ­generators27–29 and ­several forms of ­associative and ­nonassociative ­learning30. The findings of the new study6 raise many new questions that the field will now be keen to answer. Do most C. elegans ­neurons exhibit plateau potentials? If not, which ­functional subsets of neurons exhibit them and what purposes do they serve? Is plateau ­behavior an intrinsic property of C. elegans neurons or does it arise, at least in part, through synaptic ­interactions? The latter is a possibility because at least some C. elegans ­neurons appear to release neurotransmitter at the resting potential31,32. What ion channels are required for plateau potentials and how might they be regulated by learning and other forms of experience? The quest for action potentials in C. ­elegans neurons is really just beginning. To date, ­current-clamp recordings have been ­published for only 4 of the 118 anatomical neuron classes in C. elegans hermaphrodites2,6,33,34; thus, action potentials may yet be found in some classes of neurons. But if action potentials turn out to be the main currency of information ­transfer in C. elegans, it may come as ­something of a ­surprise, at least to theoreticians who have ­considered the ­signaling properties of ­neurons near what might be termed the small-cell limit. At this spatial scale, there may be only a few tens or hundreds of sodium and ­potassium ­channels in an entire neuron and input ­resistance is so high that the opening of a single sodium ­channel can trigger an action ­potential, or what is left of

378

one. Many C. ­elegans neurons are likely to be nearly isopotential2 and ­reducing ­isopotential model neurons to the small-cell limit has ­profound effects on signaling. For example, when a Hodgkin-Huxley model is reduced, an otherwise silent neuron fires tonically at average rates of tens of spikes per second35,36. Moreover, firing is so irregular that the s.d. of interspike intervals approaches the size of the intervals themselves. Together, these two effects render coding schemes that are based on firing rate and spike timing ­essentially unworkable. Additional problems arise when one considers spike ­propagation along thin axons (diameter < 500 nm), ­including ­spurious or deleted spikes and the redistribution of spike arrival times37. Evidence that C. elegans neurons operate at the small-cell limit is already quite strong. Measured input resistances in medium-sized C. elegans neurons are on the order of Gohms2 and the true input resistance is probably much higher when one takes into account the Gohm order leak of the seal resistance in a whole-cell patch-clamp experiment. In this range, the opening of a single ion channel can cause ­voltage fluctuations on the order of millivolts to tens of millivolts. Process diameters of most C. elegans neurons are in the range of 100–200 nm38 and it is ­estimated that chemosensory ­neurons, whose size and morphology are ­typical of most neurons in the worm’s head, have only about 50 voltage-­dependent calcium ­channels2, the probable carrier of inward current in C. ­elegans39. Coding schemes that are based on the ­amplitude or ­duration of graded ­potentials are also ­probably unworkable, as these ­parameters would also be strongly ­sensitive to the ­random behavior of small numbers of highly ­effective channels. Thus, however C. elegans neurons may code information, it is probably ­different from the way it is done in the nervous systems with which we are currently more ­familiar and that have much larger neurons. Perhaps the most important aspect of the ­finding of plateau potentials in C. ­elegans ­neurons is the prospect of an ­elegant ­solution to the small-cell coding ­problem: how is ­information ­represented when both action potentials and graded ­potentials are ­unreliable? In C. elegans, the small cell ­problem is ­particularly acute because there are at most six neurons in any functional class, a fact that ­eliminates ­averaging across ­multiple neurons as a remedy. Action ­potentials and graded ­potentials support ­analog coding schemes, ­firing rate and spike timing as being analog ­quantities, whereas plateau ­potentials might support a digital coding scheme in which ­neuronal state (depolarized or ­hyperpolarized) stores the sign of the most recent synaptic input. Digital codes are, of course, famous for their ­immunity to noise, spurious ­voltage offsets

and other ­probable afflictions of small-neuron networks. There are already ­tantalizing hints that the ­direction of locomotion in C. elegans (forward versus reverse) is coded ­digitally40. If it turns out that plateau ­potentials are the ­dominant mode of electrical signaling in C. elegans, we might soon be getting our first look at a digital ­nervous system. 1. White, J.G., Southgate, E., Thomson, J.N. & Brenner, S. Phil. Trans. R. Soc. Lond. B 314, 1–340 (1986). 2. Goodman, M.B., Hall, D.H., Avery, L. & Lockery, S.R. Neuron 20, 763–772 (1998). 3. Brockie, P.J., Mellem, J.E., Hills, T., Madsen, D.M. & Maricq, A.V. Neuron 31, 617–630 (2001). 4. Christensen, M. et al. Neuron 33, 503–514 (2002). 5. Nickell, W.T., Pun, R.Y., Bargmann, C.I. & Kleene, S.J. J. Membr. Biol. 189, 55–66 (2002). 6. Mellem, J.E., Brockie, P.J., Madsen, D.M. & Maricq, A.V. Nat. Neurosci. 11, 865–867 (2008). 7. Marder, E. Curr. Biol. 1, 326–327 (1991). 8. Russell, D.F. & Hartline, D.K. J. Neurophysiol. 48, 914–937 (1982). 9. Davis, R.E. & Stretton, A.O.W. J. Neurosci. 9, 415–425 (1989). 10. Angstadt, J.D. & Stretton, A.O.W. J. Comp. Physiol. [A] 166, 165–177 (1989). 11. Angstadt, J.D., Donmoyer, J.E. & Stretton, A.O. J. Comp. Neurol. 284, 374–388 (1989). 12. Holden-Dye, L. & Walker, R.J. Parasitology 108, 81–87 (1994). 13. Davis, R.E. & Stretton, A.O.W. J. Comp. Physiol. [A] 171, 17–28 (1992). 14. Lee, C.R. & Tepper, J.M. J. Neurosci. 27, 6531–6541 (2007). 15. Lo, F.S., Ziburkus, J. & Guido, W. J. Neurophysiol. 87, 1175–1185 (2002). 16. Otsuka, T., Abe, T., Tsukagawa, T. & Song, W.J. J. Neurophysiol. 92, 255–264 (2004). 17. Simon, M., Perrier, J.F. & Hounsgaard, J. Eur. J. Neurosci. 18, 258–266 (2003). 18. Amat, C., Lapied, B., French, A.S. & Hue, B. J. Neurophysiol. 80, 2718–2726 (1998). 19. Zhang, B. & Harris-Warrick, R.M. J. Neurophysiol. 74, 1929–1937 (1995). 20. Mercer, A.R., Kloppenburg, P. & Hildebrand, J.G. J. Neurophysiol. 93, 1949–1958 (2005). 21. Derjean, D., Bertrand, S., Nagy, F. & Shefchyk, S.J. J. Physiol. (Lond.) 563, 583–596 (2005). 22. Angstadt, J.D. & Choo, J.J. J. Neurophysiol. 76, 1491–1502 (1996). 23. Di Prisco, G.V., Pearlstein, E., Robitaille, R. & Dubuc, R. Science 278, 1122–1125 (1997). 24. Susswein, A.J., Hurwitz, I., Thorne, R., Byrne, J.H. & Baxter, D.A. J. Neurophysiol. 87, 2307–2323 (2002). 25. Sierra, F., Comas, V., Buno, W. & Macadar, O. J. Comp. Physiol. A Neuroethol. Sens. Neural. Behav. Physiol. 191, 1–11 (2004). 26. Scroggs, R.S. & Anderson, E.G. Brain Res. 485, 391–395 (1989). 27. Niebur, E. & Erdos, P. Biophys. J. 60, 1132–1146 (1991). 28. Thomas, J.H. Genetics 124, 855–872 (1990). 29. Hart, A.C., Sims, S. & Kaplan, J.M. Nature 378, 82–85 (1995). 30. Rankin, C.H. Curr. Biol. 14, R617–R618 (2004). 31. Chalasani, S.H. et al. Nature 450, 63–70 (2007). 32. Suzuki, H. et al. Nature 454, 114–117 (2008). 33. O’Hagan, R., Chalfie, M. & Goodman, M.B. Nat. Neurosci. 8, 43–50 (2005). 34. Ramot, D., Macinnis, B.L. & Goodman, M.B. Nat Neurosci. 11, 908–915 (2008). 35. Strassberg, A.F. & DeFelice, L.J. Neural Comput. 5, 843–855 (1993). 36. Faisal, A.A., White, J.A. & Laughlin, S.B. Curr. Biol. 15, 1143–1149 (2005). 37. Faisal, A.A. & Laughlin, S.B. PLoS Comput. Biol. 3, e79 (2007). 38. Hall, D.H. & Altun, Z. C. elegans Atlas (Cold Spring Harbor Press, Woodbury, New York, 2008). 39. Bargmann, C.I. Science 282, 2028–2033 (1998). 40. Chronis, N., Zimmer, M. & Bargmann, C.I. Nat. Methods 4, 727–731 (2007).

volume 12 | number 4 | APRIL 2009 nature neuroscience

REVIEW

Endoplasmic reticulum stress in disorders of myelinating cells

© 2009 Nature America, Inc. All rights reserved.

Wensheng Lin1 & Brian Popko2 Myelinating cells, oligodendrocytes in the CNS and Schwann cells in the peripheral nervous system produce an enormous amount of plasma membrane during the myelination process, making them particularly susceptible to disruptions of the secretory pathway. Endoplasmic reticulum stress, initiated by the accumulation of unfolded or misfolded proteins, activates the unfolded protein response, which adapts cells to the stress. If this adaptive response is insufficient, the unfolded protein response activates an apoptotic program to eliminate the affected cells. Recent observations suggest that endoplasmic reticulum stress in myelinating cells is important in the pathogenesis of various disorders of myelin, including Charcot-Marie-Tooth disease, Pelizaeus-Merzbacher disease and Vanishing White Matter Disease, as well as in the most common myelin disorder, multiple sclerosis. A better understanding of endoplasmic reticulum stress in myelinating cells has laid the groundwork for the design of new therapeutic strategies for promoting myelinating cell survival in these disorders.

Myelinating cells, oligodendrocytes in the CNS and Schwann cells in the peripheral nervous system (PNS) produce a vast amount of myelin as an extension of their plasma membrane (Fig. 1)1,2. Myelin, which is a unique, lipid-rich multilamellar sheath that wraps around axons in the CNS and PNS, is not only essential for the fast conduction of the action potential along axons, but also for the maintenance of axonal integrity3. According to estimates from morphometric analyses, the mean surface area of myelin per mature myelinating cell is thousands of times greater than the surface area of a typical mammalian cell1. Thus, during the active phase of myelination, each myelinating cell must synthesize an enormous amount of myelin membrane proteins, cholesterol and membrane lipids through the secretory pathway2. Thus, it is not surprising that myelinating cells are highly sensitive to the disruption of the secretory pathway, particularly the homeostasis of the endoplasmic reticulum. Recent studies indicate that this increased susceptibility contributes to the pathogenesis of a number of myelin disorders4–11. The endoplasmic reticulum has three essential functions12,13. Secretory proteins and proteins destined for the cell surface and other intracellular organelles are synthesized by ribosomes on the cytosolic surface of the endoplasmic reticulum and translocated into the endoplasmic reticulum lumen through a pore in the endoplasmic reticulum membrane. Inside the lumen, they are properly modified and folded. In addition, the biosynthesis of steroids, cholesterol and other lipids takes place on the cytoplasmic side of the endoplasmic reticulum membrane. Moreover, cellular calcium is 1Department

of Cell Biology and Neuroscience, University of South Alabama, Mobile, Alabama, USA. 2The Jack Miller Center for Peripheral Neuropathy, Department of Neurology, The University of Chicago, Chicago, Illinois, USA. Correspondence should be addressed to W.L. ([email protected]) or B.P. ([email protected]). Published online 15 March 2009; doi:10.1038/nn.2273

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

mainly stored in the endoplasmic reticulum lumen. The concentration of calcium in the cytoplasm is primarily regulated by Ca2+ transport into and out of the endoplasmic reticulum. A number of cell stress conditions, such as a perturbed calcium homeostasis or redox status, elevated secretory protein synthesis rates, altered glycosylation levels and abnormally high cholesterol levels, can interfere with oxidative protein folding and can subsequently lead to the accumulation of unfolded or misfolded proteins in the endoplasmic reticulum lumen. This phenomenon has been referred to as endoplasmic reticulum stress14–16. As a consequence, the eukaryotic cell has evolved an adaptive coordinated response, the unfolded protein response (UPR; Fig. 2 and Table 1), to limit further accumulation of unfolded proteins in the endoplasmic reticulum. Three endoplasmic reticulum–resident transmembrane proteins involved in the UPR have been identified: pancreatic endoplasmic reticulum kinase (PERK), the kinase encoded by the inositol requiring 1 (IRE1, also known as ERN1) gene and activating transcription factor 6 (ATF6). At the onset of endoplasmic reticulum stress, the most rapidly activated pathway is the PERK pathway, which couples protein folding in the endoplasmic reticulum with protein synthesis by phosphorylating the a subunit of eukaryotic translation initiation factor 2 (eIF2a)17,18. The activation of the PERK-eIF2a pathway attenuates the initiation of translation in response to endoplasmic reticulum stress and enhances the stress-induced expression of numerous cytoprotective genes19–21. Activation of IRE1 initiates the splicing of X box– binding protein (XBP-1) mRNA, producing an active transcription factor, spliced XBP-1 (sXBP-1)22,23. In addition, ATF6 becomes an active transcription factor by transiting to the Golgi complex, where it is cleaved by the proteases S1P and S2P before translocating to the nucleus24. The activation of IRE1 signaling and the cleavage of ATF6 promote endoplasmic reticulum expansion and the expression of endoplasmic reticulum–localized chaperones, which facilitate protein folding in the endoplasmic reticulum18,25. In multicellular organisms, if these adaptive responses are not sufficient to resolve the folding

379

REVIEW

Smooth ER Myelin Golgi Nucleus Axon

Mitochondria

© 2009 Nature America, Inc. All rights reserved.

Rough ER

Figure 1 Oligodendrocyte and myelin. Oligodendrocytes produce as an extension of their plasma membrane vast amounts of myelin, a unique, lipid-rich, multilamellar sheath that wraps axons in the CNS. The secretory pathways for proteins and lipids, including the rough endoplasmic reticulum (ER), smooth endoplasmic reticulum and Golgi apparatus, are well developed in oligodendrocytes. Evidence is accumulating that oligodendrocytes, as well as their PNS counterparts, the Schwann cells, rank among the cells that are most sensitive to the disruption of the secretory pathway.

problems in the endoplasmic reticulum, the UPR will trigger an apoptotic program to eliminate the cells26,27. It has become increasingly clear that endoplasmic reticulum stress is an important feature of a number of human diseases, particularly those involving cells dedicated to the synthesis of secreted proteins, such as diabetes mellitus, liver diseases and inflammatory disorders28–30. In this review, we discuss the current understanding of the contribution of endoplasmic reticulum stress in myelinating cells to the pathogenesis of inherited myelin disorders, such as Charcot-Marie-Tooth disease, Pelizaeus-Merzbacher disease (PMD), Vanishing White Matter Disease, and multiple sclerosis, an autoimmune demyelination disorder of the CNS. We also examine the unique features of endoplasmic reticulum stress in myelinating cells and how the increased sensitivity of these cells to endoplasmic reticulum stress might provide selective therapeutic opportunities. The secretory pathway in Charcot-Marie-Tooth disease Charcot-Marie-Tooth disease is one of the most common inherited neurological disorders in humans, affecting approximately 1 in 2,500 individuals. Charcot-Marie-Tooth disease is actually a group of disorders of the PNS that are caused by mutations in a number of distinct genes and can be categorized as primarily originating from either the myelinating cell (Charcot-Marie-Tooth 1) or the axons (CharcotMarie-Tooth 2)31. Mutations in genes encoding peripheral myelin proteins account for the majority of Charcot-Marie-Tooth 1 cases. Peripheral myelin protein zero (MPZ), a transmembrane glycoprotein, comprises greater than 50% of the total protein of peripheral myelin and a number of distinct point mutations in MPZ have been shown to be responsible for a dominant form of Charcot-Marie-Tooth 1B32. There is strong evidence that mutations in MPZ can lead to endoplasmic reticulum stress in myelinating Schwann cells4,5. A previous study generated an authentic mouse model of Charcot-Marie-Tooth 1B that expresses a human disease-causing form of MPZ with a deletion of S63 (P0S63del)4. Studies with this mutant demonstrated that this mutation causes a perturbed alignment of hydrophobic residues in b strand C of MPZ that results in the retention of the mutant protein in the endoplasmic reticulum of Schwann cells and leads to increased levels

380

of phosphorylated eIF2a (p-eIF2a), ATF4, ATF3 and CAATT enhancer binding protein homologous protein (CHOP) in the nerve, suggesting activation of the PERK pathway5. The spliced form of XBP-1 is also increased in the nerves of these animals, suggesting that the IRE1 pathway is activated. Moreover, cleavage of ATF6 was detected in the P0S63del nerve. Notably, the activation of the PERK, IRE1 and ATF6 pathways in P0S63del nerves is dose dependent, which strongly suggests that the retention of P0S63del in the endoplasmic reticulum results in the activation of the UPR in Schwann cells. It has been shown that CHOP, a transcription factor that is downstream of the PERK-eIF2a pathway, is important in the apoptosis that results from endoplasmic reticulum stress and CHOP deletion protects various cell types from endoplasmic reticulum stress–induced apoptosis in cell culture and animal models25,33,34. A previous study showed that CHOP is undetectable in Schwann cells of normal nerves, but is present in the nuclei of P0S63del Schwann cells5. To determine the role of the CHOP pathway in the pathogenesis of the myelin abnormalities that occurs in the P0S63del nerve, P0S63del mice were crossed with Chop / (also known as Ddit3) mice5. In the absence of CHOP activity, the number of demyelinated fibers in P0S63del nerves is reduced and the behavioral and electrophysiological abnormalities are attenuated. Nevertheless, CHOP inactivation did not affect the number of apoptotic Schwann cells or the degree of hypomyelination in the P0S63del nerve5. The mechanism accounting for the detrimental effects of CHOP in Charcot-Marie-Tooth 1B remains to be determined. Mutations in genes encoding peripheral myelin protein 22 (PMP22) are the most common cause for Charcot-Marie-Tooth 1 cases32. PMP22 is a highly hydrophobic transmembrane protein with four putative transmembrane domains. In humans, the molecular lesion that is responsible for the vast majority of Charcot-Marie-Tooth 1 cases is a duplication of the region of chromosome 17 that contains the PMP22 gene, resulting in increased expression of this membrane protein35. Similarly, enforced expression of wild-type PMP22 in Schwann cells of transgenic mice and rats results in a Charcot-Marie-Tooth–like disorder36. Cumulative evidence indicates that this is a gain-of-function disorder in which the increased expression of PMP22 has a deleterious effect on myelinating Schwann cells37. In addition, spontaneous dominant mouse mutations have been identified in the Pmp22 gene, Trembler and Trembler-J (Tr-J), which introduce nonconservative amino acid changes into hydrophobic domains of this protein. It has been demonstrated that misfolded Pmp22Trembler and Pmp22Tr-J associate with calnexin, an endoplasmic reticulum chaperone, and accumulate in the endoplasmic reticulum of Schwann cells38,39. Moreover, a previous study has shown that treatment with the chemical compound curcumin, which is capable of promoting the transport of misfolded proteins from the endoplasmic reticulum to the plasma membrane, substantially attenuates Schwann cell apoptosis and improves the neuropathologic phenotype in Pmp22Tr-J mice40. Taken together, these data indicate that the accumulation of mutant PMP22 in the endoplasmic reticulum causes the apoptosis of Schwann cells, resulting in myelin abnormalities. Nevertheless, this accumulation does not lead to increased expression of the endoplasmic reticulum stress markers binding immunoglobulin protein (BIP) or CHOP39,40. These data suggest that the role of the UPR in the pathogenesis of Charcot-Marie-Tooth 1A mutations needs to be examined directly and in greater detail. Severe endoplasmic reticulum stress in PMD PMD is an X chromosome–linked dysmyelinating disease of the CNS with a broad range of clinical severity that is caused by more than 60 known mutations, including missense and nonsense mutations, in the proteolipid protein 1 (PLP1) gene41–43. In addition, PLP1 gene

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

REVIEW Unfolded/misfolded proteins

ER Caspase-12 ATF6

PERK Cleaved caspase-12

Apoptosis pathway Apoptosis

© 2009 Nature America, Inc. All rights reserved.

CHOP

eIF2 P

eIF2 GADD34-PP1

Preferential translation of ATF4

IREI

Proteolysis

Inhibition of translation

Transcription of Reduced cytoprotective genes ER load

XBP-1

sXBP-1

Transcription of chaperones

ER expansion

Adaptive pathway

Figure 2 The UPR pathway in eukaryotic cells. Endoplasmic reticulum (ER) stress triggers PERK dimerization and autophosphorylation. Activated PERK attenuates protein biosynthesis by phosphorylating eIF2a. Phosphorylated eIF2a also activates ATF4, which enhances the stress-induced expression of numerous cytoprotective genes. In addition, ATF4 increases the expression of CHOP, which induces GADD34 expression. GADD34 binds to PP1 and forms the GADD34-PP1 complex that specifically dephosphorylates eIF2a. Endoplasmic reticulum stress also triggers IRE1dimerization and autophosphorylation. Activation of IRE1 initiates the splicing of XBP-1 mRNA, producing an active transcription factor, sXBP-1. In addition, ATF6 becomes an active transcription factor by transiting to the Golgi complex, where it is cleaved by the proteases S1P and S2P. The activation of IRE1 signaling and the cleavage of ATF6 promote endoplasmic reticulum expansion and the expression of endoplasmic reticulum–localized chaperones. In multicellular organisms, if these adaptive responses are not sufficient to resolve the folding problems in the endoplasmic reticulum, the UPR will trigger an apoptotic program, such as activation of caspase 12 or CHOP, to eliminate the cells.

duplications are responsible for approximately 50% of PMD cases. PLP and its alternatively spliced isoform, DM20, are believed to contain four membrane-spanning domains and together comprise approximately 50% of total CNS myelin protein. It is believed that mutant PLP has a dominant gain of toxic function, as PLP1 deletion does not affect oligodendrocyte development and myelination and Plp1 / mice have normal neurological function until late adulthood41. Moreover, the enforced expression of a wild-type Plp1 gene is not sufficient to rescue the Plp1 mutant phenotype41–43. A previous study demonstrated that PLP mutations cause oligodendrocyte death during the myelination process, at the developmental time point when oligodendrocytes are extending extensive processes, making contact with nearby axons and initiating the synthesis of the myelin proteins44. Several studies have shown that the mutant PLP proteins are not properly folded and accumulate in the endoplasmic reticulum42,45,46. COS-7 cells that are transfected with a number of distinct PLP mutants show accumulation of mutant proteins in the endoplasmic reticulum7,42. In addition, immunohistochemical studies show perinuclear accumulation of mutant PLP proteins in the oligodendrocytes of PLP mutant mice44,47. This was further supported by electron microscopy analyses, which provided direct evidence that mutant PLP proteins accumulate in the endoplasmic reticulum of oligodendrocytes, where they disturb the endoplasmic reticulum ultrastructure42,48. Strong evidence has also recently indicated that the accumulation of mutant PLP in the endoplasmic reticulum activates the UPR and leads to myelinating oligodendrocyte death7,49. Mutant PLP that is expressed by transfected fibroblasts accumulates in the endoplasmic reticulum and substantially increases the expression of endoplasmic reticulum stress markers such as CHOP, BIP and ATF3, which are also upregulated in the oligodendrocytes of PLP mutant mice7. Notably, CHOP is localized to the nucleus in mutant oligodendrocytes, which is strongly indicative of the endoplasmic reticulum stress response7. Similarly,

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

increased CHOP expression is seen in the oligodendrocyte nuclei of PMD patients7. Lastly, a previous study showed that oligodendrocytes from rats transgenically overexpressing PLP are characterized by swollen endoplasmic reticulum and increased expression of BIP, as well as disulfide isomerase, an endoplasmic reticulum enzyme involved in protein folding6. To determine the involvement of endoplasmic reticulum stress in oligodendrocyte death in Plp1 mutant mice, one study used a genetic approach, crossing Chop / mice with Plp1 mutant mice7. The study found that the absence of CHOP notably exacerbated the clinical phenotype and increased oligodendrocyte apoptosis in Plp1 mutant mice. This protective effect of CHOP on oligodendrocytes during endoplasmic reticulum stress is in marked contrast with its proapoptotic effect on other cell types. The mechanisms accounting for the protective effect of CHOP on oligodendrocytes remain a mystery. The above-mentioned study showed that CHOP induction in Plp1 mutant mice does not enhance expression of proteins that normally act downstream of CHOP, such as DOC (downstream of CHOP) 1, 4, and 6 (ref. 7). It is possible that CHOP has a distinct set of target genes depending on cell type, which is reflected in its cell-specific function. The UPR and Vanishing White Matter Disease Vanishing White Matter Disease is an inherited autosomal-recessive hypomyelinating disorder. The discovery of missense mutations in the ubiquitously expressed genes encoding the five subunits of eIF2B in individuals with Vanishing White Matter Disease led to the identification of the genetic causes of Vanishing White Matter Disease50,51. This is the first human disease that has been shown to be caused by a mutation in a gene encoding a protein synthesis factor. eIF2B functions in translation as a guanine nucleotide exchange factor, promoting the release of GDP from eIF2 and the formation of an active eIF2-GTP complex52. eIF2-GTP then binds aminoacylated initiator methionyl tRNA, forming a ternary complex that is essential for each translation initiation event. p-eIF2a reduces the activity of eIF2B, with the two proteins forming a nonproductive p-eIF2–eIF2B complex. eIF2B mutations may impair its response to p-eIF2a, particularly during cell stress53. Although oligodendrocytes have been described as being abnormal and ‘foamy’ in appearance, no evidence has been presented that demonstrates a reduction of oligodendrocyte numbers in Vanishing White Matter Disease54,55. Notably, a recent study presented evidence for the activation of the UPR in oligodendrocytes in individuals with Vanishing White Matter Disease, including increased levels of phosphorylated PERK, p-eIF2a, ATF4 and CHOP8,9. Moreover, it has been shown that primary cultured fibroblasts from individuals with Vanishing White Matter Disease respond normally to endoplasmic reticulum stress by reduced global translation rates. Nevertheless, the induction of ATF4 is substantially enhanced in these cells compared with normal controls, despite there being equal levels of cell stress and p-eIF2a56. A recent study has shown that the enforced expression of a Vanishing White Matter Disease–causing eIF2B mutation in an oligodendroglial cell line leads to elevated basal levels of the endoplasmic reticulum stress markers ATF4, GADD34 and BIP and results in the hyperinduction of these markers in response to a pharmacological stress agent, indicating that the UPR adapts oligodendrocytes to Vanishing White Matter Disease–causing eIF2B mutations57. The UPR and immune-mediated demyelinating disorders Multiple sclerosis is the most common neurological disorder affecting young adults, with an estimated incidence of approximately 1 in 500 individuals. It is generally believed that multiple sclerosis and its animal

381

REVIEW Table 1 The functions of major effecter components of the UPR UPR effecter molecule PERK

Function At the onset of endoplasmic reticulum stress, the most rapidly activated endoplasmic reticulum stress sensor is PERK, which couples protein folding in the endoplasmic reticulum with protein synthesis by phosphorylating eIF2a19,20.

eIF2a

eIF2a phosphorylation promotes a stress-resistant state by global attenuation of protein biosynthesis and induction of numerous stress-induced cytoprotective genes21,88. Four different kinases are known to phosphorylate eIF2a. They are PERK, general control non-derepressible 2 kinase (GCN2), which is activated by amino acid starvation, double-stranded RNA-dependent protein kinase (PKR), which is activated by viral infection, and heme-regulated inhibitor (HRI), which is activated by iron deficiency88.

ATF4

ATF4 mRNA contains a unique 5¢ untranslated region that binds to p-eIF2a, leading to ATF4 protein expression. The gene expression

© 2009 Nature America, Inc. All rights reserved.

program activated by ATF4 is involved in regulating amino acid CHOP

metabolism and resistance to oxidative stress89. CHOP is a transcription factor downstream of ATF4, and was thought to promote apoptotic cell death, as its deletion diminishes cell death during endoplasmic reticulum stress. Recent work, however, demonstrated that the deleterious effects of CHOP are associated with the induction of the protein growth arrest and DNA damage 34 (GADD34), rather than the activation of an apoptotic program25,33,34.

GADD34

A regulatory subunit of protein phosphatase 1 (PP1) that specifically dephosphorylates phosphorylated eIF2a. GADD34 blockage increases the levels of phosphorylated eIF2a in stressed cells and protects cells from the stresses34,90.

ATF6

Activation of ATF6 promotes the expression of endoplasmic reticulum–localized chaperones18,25.

IRE1

Activation of IRE1 initiates the splicing of X-box–binding protein (XBP-1) mRNA, producing an active transcription factor, spliced

XBP-1

XBP-1 (refs. 22,23). Spliced XBP-1 promotes endoplasmic reticulum expansion and the

Caspase-12

expression of endoplasmic reticulum–localized chaperones18,25. Activation of caspase-12, an endoplasmic reticulum–localized caspase, has been shown to be associated with cell apoptosis that is induced by endoplasmic reticulum stress79,80.

model, experimental autoimmune encephalomyelitis (EAE), are Th1 T cell–mediated autoimmune diseases58,59. The pathological hallmarks of multiple sclerosis and EAE include inflammation, demyelination, oligodendrocyte loss and axonal degeneration. Evidence is emerging that the UPR is involved in the disease pathogenesis of multiple sclerosis and EAE10,11,60. Gene chip analyses have shown upregulation of the endoplasmic reticulum stress–responsive genes ATF4 and HSP70 (also known as HSPA4) in multiple sclerosis demyelinated lesions61,62. Another recent report has shown elevated expression of CHOP, BIP and XBP-1 in multiple cell types in multiple sclerosis demyelinated lesions, including oligodendrocytes, astrocytes, T cells and microglia10. Moreover, elevated levels of p-eIF2a have been observed in oligodendrocytes and infiltrating T cells in the CNS during the course of EAE63,64. Recent evidence has suggested that activation of the PERK-eIF2a pathway of the UPR in oligodendrocytes protects against EAE-induced oligodendrocyte death and demyelination65. Interferon-g (IFN-g), a T cell–derived pleiotropic cytokine, is believed to be important in the pathogenesis of multiple sclerosis and EAE11,66. Delivery of IFN-g to the CNS before EAE onset ameliorated disease severity and

382

prevented EAE-induced oligodendrocyte loss, demyelination and axonal damage65. Notably, the protective effects of this cytokine are accompanied with the activation of the PERK-eIF2a pathway in oligodendrocytes and are abrogated in PERK-deficient animals65. Although evidence has shown that IFN-g is capable of increasing eIF2a phosphorylation through activation of double-stranded RNAdependent protein kinase (PKR)67 and the level of phosphorylated PKR is elevated in oligodendrocytes in the course of EAE63, we have previously demonstrated that IFN-g–induced eIF2a phosphorylation in oligodendrocytes is mediated by PERK and that the protective effect of IFN-g in EAE is dependent on PERK65,68. Thus, these data provide strong evidence that the UPR that is induced by IFN-g is involved in the pathogenesis of immune-mediated demyelination diseases. Remyelination is sufficiently robust to repair myelin damage and restore neurological function in some animal models of CNS demyelination. The remyelination process also occurs in the CNS of individuals with multiple sclerosis. Nevertheless, this remyelination is generally considered to be insufficient69,70. Several reports have shown that IFN-g is an important cytokine that suppresses oligodendrocyte regeneration and inhibits remyelination in multiple sclerosis and EAE demyelinated lesions71–73. The detrimental effects of IFN-g on myelin repair are mediated by endoplasmic reticulum stress in the remyelinating oligodendrocytes73. Transgenic mice that ectopically express IFN-g in the CNS during development have a tremoring phenotype, oligodendrocyte loss and hypomyelination68,74,75. Activation of the PERK-eIF2a pathway is detected in the myelinating oligodendrocytes of these transgenic mice68. In addition, decreased activity of the PERK-eIF2a pathway via PERK inactivation exacerbates IFN-g–induced oligodendrocyte apoptosis and hypomyelination68, whereas increased activity of the PERK-eIF2a pathway via GADD34 inactivation has the opposite effect76. Moreover, the detrimental effects of IFN-g on remyelination in demyelinated lesions are associated with endoplasmic reticulum stress in remyelinating oligodendrocytes, as evidenced by the upregulation of BIP and CHOP and an elevated level of p-eIF2a73. Notably, PERK deficiency substantially exacerbates remyelinating oligodendrocyte apoptosis and remyelination failure induced by IFN-g in demyelinated lesions73. It appears that endoplasmic reticulum stress has both beneficial and detrimental effects on oligodendrocyte survival in immune-mediated demyelination diseases. Endoplasmic reticulum stress induction in fully myelinated mature oligodendroytes promotes cell survival, but in actively myelinating/remyelinating oligodendrocytes leads to cell death. Although it is largely unknown how the UPR selectively initiates the apoptotic and adaptive pathways77, we speculate that the outcomes of endoplasmic reticulum stress in oligodendrocytes are probably determined by the developmental status of the cells11,60,65. During active myelination, the endoplasmic reticulum in oligodendrocytes is largely occupied with the production of enormous amounts of myelin proteins and lipids1,2. Therefore, its adaptive capacity may be limited at this stage. It is likely that the marked upregulation of protein synthesis initiated by the inflammatory response overloads the endoplasmic reticulum of myelinating cells, causing severe endoplasmic reticulum stress and cell death. In contrast, mature oligodendrocytes in adult mice produce just enough myelin proteins and lipids to maintain myelin structure homeostasis78. The endoplasmic reticulum in mature oligodendrocytes may have sufficient adaptive capacity to handle the increased protein load. In this situation, the resulting endoplasmic reticulum stress would be moderate and would not trigger the apoptotic program. Instead, the adaptive response activated by endoplasmic reticulum stress could act protectively against subsequent insults.

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

© 2009 Nature America, Inc. All rights reserved.

REVIEW The unique features of the UPR in myelinating cells As described in the discussion of myelin disorders above, there is evidence to suggest that myelinating cells respond to endoplasmic reticulum stress in a manner that is somewhat distinct from that observed in other cell types. For example, CHOP, which is a downstream effecter molecule of the PERK-eIF2a pathway, participates in the induction of the apoptotic response in most cell types, but does not appear to contribute to the UPR-induced apoptotic demise of myelinating cells5,7,33. In fact, activation of CHOP promotes oligodendrocyte survival during the endoplasmic reticulum stress that is induced by PLP mutations7. Moreover, it has been demonstrated that CHOP is important in demyelination in the P0 mutant nerve, but is not involved in Schwann cell apoptosis5. Caspase-12, an endoplasmic reticulum–localized caspase, has been shown to be important in apoptosis that is elicited by endoplasmic reticulum stress in neurodegenerative diseases79,80. Although activation of caspase-12 is detected in the oligodendrocytes of PLP mutant mice, caspase-12 deletion does not substantially affect the clinical phenotype, oligodendrocyte apoptosis or myelin abnormalities in these mice81. In addition, the upregulation of ATF3 has been shown to occur in the oligodendrocytes of PLP mutant mice, but the deletion of ATF3 also fails to have an effect on oligodendrocyte apoptosis in these animals82. Taken together, these data indicate that the apoptotic process that is elicited by endoplasmic reticulum stress in myelinating cells is unique. A more detailed understanding of these unique features could be critical when considering potential therapeutic approaches for myelin disorders. Perhaps, the considerable demand on the secretory pathway of actively myelinating cells has resulted in an alteration of the typical response to endoplasmic reticulum stress that is seen in other cell types. A more thorough assessment of the UPR in myelinating cells is essential for a better understanding of the response of these cells to endoplasmic reticulum stress. One difficulty in the study of the UPR in myelinating cells is that they appear to be most susceptible to secretory pathway perturbations while they are actively myelinating axons, diminishing the utility of examining their response to stresses in heterologous systems or in isolation. Therefore, the most informative studies have been carried out in vivo using mouse models. Nevertheless, recent advances in the development of in vitro myelination systems, in combination with siRNA technology, should allow for a more rapid characterization of the UPR in myelinating cells83. Therapeutic potential and future directions Considerable progress has now been made toward a detailed understanding of the signaling pathways of the endoplasmic reticulum stress response. Recent studies indicate that manipulating these signaling pathways has the potential to be therapeutic. The chemical chaperones 4-phenylbutyric acid and taurine-conjugated ursodeoxycholic acid, which reduce the phosphorylation of PERK and IRE1 during endoplasmic reticulum stress, both substantially improve glucose tolerance and insulin sensitivity in insulin-resistant obese mice84. Another chemical chaperone, the resveratrol tetramer vaticanol B, which suppresses the expression of CHOP and BIP during endoplasmic reticulum stress, prevents endoplasmic reticulum stress–induced apoptosis85. In addition, treatment with salburinal, a small-molecule inhibitor of eIF2a dephosphorylation, results in sustained eIF2a phosphorylation and protects cells from endoplasmic reticulum stress and viral infection86. Currently, there is no effective therapy for patients with disorders of myelinating cells. Therapeutic strategies that facilitate the transport of misfolded proteins out of the endoplasmic reticulum could be

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

beneficial for these patients. It has been shown that treatment with curcumin promotes mutant PMP22 transport from the endoplasmic reticulum to the plasma membrane and attenuates Schwann cell apoptosis in Pmp22Tr-J mice40. Moreover, our recent data demonstrate that treatment with salburinal promotes the survival of myelinating oligodendrocytes that have been exposed to IFN-g, implying that therapeutic strategies that enhance the PERK-eIF2a pathway could promote oligodendrocyte survival in immune-mediated demyelination diseases76. The manipulation of endoplasmic reticulum stress for therapeutic purposes, without causing severe side effects, is a formidable challenge. Nevertheless, just as actively myelinating cells show increased sensitivity to perturbations to the secretory pathway, these cells may also show selective sensitivity and benefit from modulators of the UPR. Future studies will need to address the many open questions concerning the physiological importance of the various endoplasmic reticulum stress signaling pathways during the myelination process and the pathological importance of these pathways in myelin disorders. For example, considering the relative prevalence of the disorder, it is critical to know whether endoplasmic reticulum stress is an important participant in the pathogenesis of Charcot-Marie-Tooth 1 in patients that overexpress the membrane protein PMP22. Moreover, the unique aspects of the UPR in myelinating cells need to be better understood. The knowledge gained from such studies will provide a strong foundation for the design of therapeutic strategies for these diseases. In addition, insights gained from these efforts might prove to be applicable to neurodegenerative disorders that show the selective vulnerability of specific neuronal subtypes. Neuronal populations with excessive secretory requirements might have increased sensitivity to factors, genetic and environmental, that disrupt endoplasmic reticulum function87. Future investigations into this possibility might be enlightening. ACKNOWLEDGMENTS We thank D. Douglas for critical comments on the manuscript. W.L. is supported by a National Multiple Sclerosis Society Career Transition Fellowship grant (TA 3026-A-1). B.P. is supported by grants from the US National Institutes of Health (NS34939 and 027336) and the Myelin Repair Foundation. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Pfeiffer, S.E., Warrington, A.E. & Bansal, R. The oligodendrocyte and its many cellular processes. Trends Cell Biol. 3, 191–197 (1993). 2. Anitei, M. & Pfeiffer, S.E. Myelin biogenesis: sorting out protein trafficking. Curr. Biol. 16, R418–R421 (2006). 3. Simons, M. & Trotter, J. Wrapping it up: the cell biology of myelination. Curr. Opin. Neurobiol. 17, 533–540 (2007). 4. Wrabetz, L. et al. Different intracellular pathomechanisms produce diverse Myelin Protein Zero neuropathies in transgenic mice. J. Neurosci. 26, 2358–2368 (2006). 5. Pennuto, M. et al. Ablation of the UPR-mediator CHOP restores motor function and reduces demyelination in Charcot-Marie-Tooth 1B mice. Neuron 57, 393–405 (2008). This paper shows that the UPR is involved in the demyelination process in CharcotMarie-Tooth 1B mice. 6. Bauer, J. et al. Endoplasmic reticulum stress in PLP-overexpressing transgenic rats: gray matter oligodendrocytes are more vulnerable than white matter oligodendrocytes. J. Neuropathol. Exp. Neurol. 61, 12–22 (2002). 7. Southwood, C.M., Garbern, J., Jiang, W. & Gow, A. The unfolded protein response modulates disease severity in Pelizaeus-Merzbacher disease. Neuron 36, 585–596 (2002). This paper shows that oligodendrocyte apoptosis in PMD is mediated by the UPR. 8. van der Voorn, J.P. et al. The unfolded protein response in vanishing white matter disease. J. Neuropathol. Exp. Neurol. 64, 770–775 (2005). 9. van Kollenburg, B. et al. Glia-specific activation of all pathways of the unfolded protein response in vanishing white matter disease. J. Neuropathol. Exp. Neurol. 65, 707–715 (2006). 10. Mha´ille, A.N. et al. Increased expression of endoplasmic reticulum stress-related signaling pathway molecules in multiple sclerosis lesions. J. Neuropathol. Exp. Neurol. 67, 200–211 (2008).

383

© 2009 Nature America, Inc. All rights reserved.

REVIEW This is the first study to show elevated levels of endoplasmic reticulum stress markers in multiple sclerosis demyelinated lesions. 11. Lees, J.R. & Cross, A.H. A little stress is good: IFN-gamma, demyelination, and multiple sclerosis. J. Clin. Invest. 117, 297–299 (2007). 12. Kaufman, R.J. Stress signaling from the lumen of the endoplasmic reticulum: coordination of gene transcriptional and translational controls. Genes Dev. 13, 1211–1233 (1999). 13. Schro¨der, M. Endoplasmic reticulum stress responses. Cell. Mol. Life Sci. 65, 862–894 (2008). 14. Ma, Y. & Hendershot, L.M. The unfolding tale of the unfolded protein response. Cell 107, 827–830 (2001). 15. Rutkowski, D.T. & Kaufman, R.J. A trip to the ER: coping with stress. Trends Cell Biol. 14, 20–28 (2004). 16. Ron, D. & Walter, P. Signal integration in the endoplasmic reticulum unfolded protein response. Nat. Rev. Mol. Cell Biol. 8, 519–529 (2007). 17. Ron, D. & Harding, H. PERK and translational control by stress in the endoplasmic reticulum. in Translational Control (eds. Hershey, J., Mathews M., & Sonenberg, N.) 547–560 (Cold Spring Harbor Laboratory Press, Cold Spring Harbor, NY, 2000). 18. Schro¨der, M. & Kaufman, R.J. Divergent roles of IRE1alpha and PERK in the unfolded protein response. Curr. Mol. Med. 6, 5–36 (2006). 19. Harding, H.P., Zhang, Y. & Ron, D. Protein translation and folding are coupled by an endoplasmic-reticulum-resident kinase. Nature 397, 271–274 (1999). 20. Harding, H.P., Zhang, Y., Bertolotti, A., Zeng, H. & Ron, D. Perk is essential for translational regulation and cell survival during the unfolded protein response. Mol. Cell 5, 897–904 (2000). 21. Harding, H.P. et al. An integrated stress response regulates amino acid metabolism and resistance to oxidative stress. Mol. Cell 11, 619–633 (2003). 22. Yoshida, H., Matsui, T., Yamamoto, A., Okada, T. & Mori, K. XBP1 mRNA is induced by ATF6 and spliced by IRE1 in response to ER stress to produce a highly active transcription factor. Cell 107, 881–891 (2001). 23. Calfon, M. et al. IRE1 couples endoplasmic reticulum load to secretory capacity by processing the XBP-1 mRNA. Nature 415, 92–96 (2002). 24. Ye, J. et al. ER stress induces cleavage of membrane-bound ATF6 by the same proteases that process SREBPs. Mol. Cell 6, 1355–1364 (2000). 25. Marciniak, S.J. & Ron, D. Endoplasmic reticulum stress signaling in disease. Physiol. Rev. 86, 1133–1149 (2006). 26. Szegezdi, E., Logue, S.E., Gorman, A.M. & Samali, A. Mediators of endoplasmic reticulum stress–induced apoptosis. EMBO Rep. 7, 880–885 (2006). 27. Faitova, J., Krekac, D., Hrstka, R. & Vojtesek, B. Endoplasmic reticulum stress and apoptosis. Cell. Mol. Biol. Lett. 11, 488–505 (2006). 28. Scheuner, D. & Kaufman, R.J. The unfolded protein response: a pathway that links insulin demand with beta-cell failure and diabetes. Endocr. Rev. 29, 317–333 (2008). 29. Ji, C. Dissection of endoplasmic reticulum stress signaling in alcoholic and nonalcoholic liver injury. J. Gastroenterol. Hepatol. 23, S16–S24 (2008). 30. Todd, D.J., Lee, A.H. & Glimcher, L.H. The endoplasmic reticulum stress response in immunity and autoimmunity. Nat. Rev. Immunol. 8, 663–674 (2008). 31. Skre, H. Genetic and clinical aspects of Charcot-Marie-Tooth’s disease. Clin. Genet. 6, 98–118 (1974). 32. Berger, P., Niemann, A. & Suter, U. Schwann cells and the pathogenesis of inherited motor and sensory neuropathies (Charcot-Marie-Tooth disease). Glia 54, 243–257 (2006). 33. Oyadomari, S. & Mori, M. Roles of CHOP/GADD153 in endoplasmic reticulum stress. Cell Death Differ. 11, 381–389 (2004). 34. Marciniak, S.J. et al. CHOP induces death by promoting protein synthesis and oxidation in the stressed endoplasmic reticulum. Genes Dev. 18, 3066–3077 (2004). 35. Lupski, J.R. & Chance, P.F. Hereditary motor and sensory neuropathies involving altered dosage or mutation of PMP22: the CMT1A duplication and HNPP deletion. in Peripheral Neuropathy (eds. Dyck, P.J. & Thomas, P.K.) 1659–1680 (Elsevier Saunders, Philadelphia, 2005). 36. Sereda, M.W. & Nave, K.A. Animal models of Charcot-Marie-Tooth disease type 1A. Neuromolecular Med. 8, 205–216 (2006). 37. Snipes, G.J., Orfali, W., Fraser, A., Dickson, K. & Colby, J. The anatomy and cell biology of peripheral myelin protein-22. Ann. NY Acad. Sci. 883, 143–151 (1999). 38. Colby, J. et al. PMP22 carrying the trembler or trembler-J mutation is intracellularly retained in myelinating Schwann cells. Neurobiol. Dis. 7, 561–573 (2000). 39. Dickson, K.M. et al. Association of calnexin with mutant peripheral myelin protein-22 ex vivo: a basis for ‘‘gain-of-function’’ ER diseases. Proc. Natl. Acad. Sci. USA 99, 9852–9857 (2002). 40. Khajavi, M. et al. Oral curcumin mitigates the clinical and neuropathologic phenotype of the Trembler-J mouse: a potential therapy for inherited neuropathy. Am. J. Hum. Genet. 81, 438–453 (2007). This report demonstrates the therapeutic benefit of facilitated endoplasmic reticulum export in a CMT animal model. 41. Garbern, J.Y. Pelizaeus-Merzbacher disease: Genetic and cellular pathogenesis. Cell. Mol. Life Sci. 64, 50–65 (2007). 42. Southwood, C. & Gow, A. Molecular pathways of oligodendrocyte apoptosis revealed by mutations in the proteolipid protein gene. Microsc. Res. Tech. 52, 700–708 (2001). 43. Gow, A. & Lazzarini, R.A. A cellular mechanism governing the severity of PelizaeusMerzbacher disease. Nat. Genet. 13, 422–428 (1996). 44. Gow, A., Southwood, C.M. & Lazzarini, R.A. Disrupted proteolipid protein trafficking results in oligodendrocyte apoptosis in an animal model of Pelizaeus-Merzbacher disease. J. Cell Biol. 140, 925–934 (1998).

384

45. Swanton, E., Holland, A., High, S. & Woodman, P. Disease-associated mutations cause premature oligomerization of myelin proteolipid protein in the endoplasmic reticulum. Proc. Natl. Acad. Sci. USA 102, 4342–4347 (2005). 46. Dhaunchak, A.S. & Nave, K.A. A common mechanism of PLP/DM20 misfolding causes cysteine-mediated endoplasmic reticulum retention in oligodendrocytes and PelizaeusMerzbacher disease. Proc. Natl. Acad. Sci. USA 104, 17813–17818 (2007). 47. Koeppen, A.H., Barron, K.D., Csiza, C.K. & Greenfield, E.A. Comparative immunocytochemistry of Pelizaeus-Merzbacher disease, the jimpy mouse and the myelin-deficient rat. J. Neurol. Sci. 84, 315–327 (1988). 48. Roussel, G., Neskovic, N.M., Trifilieff, E., Artault, J.C. & Nussbaum, J.L. Arrest of proteolipid transport through the Golgi apparatus in Jimpy brain. J. Neurocytol. 16, 195–204 (1987). 49. McLaughlin, M. et al. Genetic background influences UPR, but not PLP, processing in the rumpshaker model of PMD/SPG2. Neurochem. Res. 32, 167–176 (2007). 50. Schiffmann, R. & Elroy-Stein, O. Childhood ataxia with CNS hypomyelination/vanishing white matter disease—a common leukodystrophy caused by abnormal control of protein synthesis. Mol. Genet. Metab. 88, 7–15 (2006). 51. Leegwater, P.A. et al. Subunits of the translation initiation factor eIF2B are mutant in leukoencephalopathy with vanishing white matter. Nat. Genet. 29, 383–388 (2001). 52. Mohammad-Qureshi, S.S., Jennings, M.D. & Pavitt, G.D. Clues to the mechanism of action of eIF2B, the guanine nucleotide exchange factor for translation initiation. Biochem. Soc. Trans. 36, 658–664 (2008). 53. van der Knaap, M.S., Pronk, J.C. & Scheper, G.C. Vanishing white matter disease. Lancet Neurol. 5, 413–423 (2006). 54. Wong, K. et al. Foamy cells with oligodendroglial phenotype in childhood ataxia with diffuse central nervous system hypomyelination syndrome. Acta Neuropathol. 100, 635–646 (2000). 55. Van Haren, K. et al. The life and death of oligodendrocytes in vanishing white matter disease. J. Neuropathol. Exp. Neurol. 63, 618–630 (2004). 56. Kantor, L. et al. Heightened stress response in primary fibroblasts expressing mutant eIF2B genes from CACH/VWM leukodystrophy patients. Hum. Genet. 118, 99–106 (2005). 57. Kantor, L. et al. A point mutation in translation initiation factor 2B leads to a continuous hyper stress state in oligodendroglial-derived cells. PLoS ONE 3, e3783 (2008). This paper provides evidence that the UPR adapts oligodendrocytes to eIF2B mutations in Vanishing White Matter Disease. 58. Frohman, E.M., Racke, M.K. & Raine, C.S. Multiple sclerosis—the plaque and its pathogenesis. N. Engl. J. Med. 354, 942–955 (2006). 59. Trapp, B.D. & Nave, K.A. Multiple sclerosis: an immune or neurodegenerative disorder? Annu. Rev. Neurosci. 31, 247–269 (2008). 60. Zhang, K. & Kaufman, R.J. From endoplasmic reticulum stress to the inflammatory response. Nature 454, 455–462 (2008). 61. Mycko, M.P., Papoian, R., Boschert, U., Raine, C.S. & Selmaj, K.W. Microarray gene expression profiling of chronic active and inactive lesions in multiple sclerosis. Clin. Neurol. Neurosurg. 106, 223–229 (2004). 62. Cwiklinska, H. et al. Heat shock protein 70 associations with myelin basic protein and proteolipid protein in multiple sclerosis brains. Int. Immunol. 15, 241–249 (2003). 63. Chakrabarty, A., Danley, M.M. & LeVine, S.M. Immunohistochemical localization of phosphorylated protein kinase R and phosphorylated eukaryotic initiation factor-2 alpha in the central nervous system of SJL mice with experimental allergic encephalomyelitis. J. Neurosci. Res. 76, 822–833 (2004). 64. Chakrabarty, A., Fleming, K.K., Marquis, J.G. & LeVine, S.M. Quantifying immunohistochemical staining of phospho-eIF2alpha, heme oxygenase-2 and NADPH cytochrome P450 reductase in oligodendrocytes during experimental autoimmune encephalomyelitis. J. Neurosci. Methods 144, 227–234 (2005). 65. Lin, W. et al. The integrated stress response prevents demyelination by protecting oligodendrocytes against immune-mediated damage. J. Clin. Invest. 117, 448–456 (2007). This paper shows the beneficial effects of the UPR in an immune-mediated demyelinating disease. 66. Imitola, J., Chitnis, T. & Khoury, S.J. Cytokines in multiple sclerosis: from bench to bedside. Pharmacol. Ther. 106, 163–177 (2005). 67. Su, Q. et al. Interferons induce tyrosine phosphorylation of the eIF2alpha kinase PKR through activation of Jak1 and Tyk2. EMBO Rep. 8, 265–270 (2007). 68. Lin, W., Harding, H.P., Ron, D. & Popko, B. Endoplasmic reticulum stress modulates the response of myelinating oligodendrocytes to the immune cytokine interferon-gamma. J. Cell Biol. 169, 603–612 (2005). 69. Bruck, W., Kuhlmann, T. & Stadelmann, C. Remyelination in multiple sclerosis. J. Neurol. Sci. 206, 181–185 (2003). 70. Franklin, R.J. Why does remyelination fail in multiple sclerosis? Nat. Rev. Neurosci. 3, 705–714 (2002). 71. Panitch, H.S., Hirsch, R.L., Schindler, J. & Johnson, K.P. Treatment of multiple sclerosis with gamma interferon: exacerbations associated with activation of the immune system. Neurology 37, 1097–1102 (1987). 72. Renno, T. et al. Interferon-gamma in progression to chronic demyelination and neurological deficit following acute EAE. Mol. Cell. Neurosci. 12, 376–389 (1998). 73. Lin, W. et al. Interferon-gamma inhibits central nervous system remyelination through a process modulated by endoplasmic reticulum stress. Brain 129, 1306–1318 (2006).

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

REVIEW 82. Sharma, R., Jiang, H., Zhong, L., Tseng, J. & Gow, A. Minimal role for activating transcription factor 3 in the oligodendrocyte unfolded protein response in vivo. J. Neurochem. 102, 1703–1712 (2007). 83. Watkins, T.A., Emery, B., Mulinyawe, S. & Barres, B.A. Distinct stages of myelination regulated by gamma-secretase and astrocytes in a rapidly myelinating CNS coculture system. Neuron 60, 555–569 (2008). 84. Ozcan, U. et al. Chemical chaperones reduce ER stress and restore glucose homeostasis in a mouse model of type 2 diabetes. Science 313, 1137–1140 (2006). 85. Takano, K. et al. A dibenzoylmethane derivative protects dopaminergic neurons against both oxidative stress and endoplasmic reticulum stress. Am. J. Physiol. Cell Physiol. 293, C1884–C1894 (2007). 86. Boyce, M. et al. A selective inhibitor of eIF2alpha dephosphorylation protects cells from ER stress. Science 307, 935–939 (2005). This paper demonstrates the potential of small molecules to protect cells from endoplasmic reticulum stress. 87. Lindholm, D., Wootz, H. & Korhonen, L. ER stress and neurodegenerative diseases. Cell Death Differ. 13, 385–392 (2006). 88. Proud, C.G. eIF2 and the control of cell physiology. Semin. Cell Dev. Biol. 16, 3–12 (2005). 89. Rutkowski, D.T. & Kaufman, R.J. All roads lead to ATF4. Dev. Cell 4, 442–444 (2003). 90. Novoa, I., Zeng, H., Harding, H.P. & Ron, D. Feedback inhibition of the unfolded protein response by GADD34-mediated dephosphorylation of eIF2alpha. J. Cell Biol. 153, 1011–1022 (2001).

© 2009 Nature America, Inc. All rights reserved.

This paper shows that endoplasmic reticulum stress induction in actively remyelinating oligodendrocytes leads to cell death and remyelination failure in an immune-mediated demyelinating diseases. 74. Corbin, J.G. et al. Targeted CNS expression of interferon-gamma in transgenic mice leads to hypomyelination, reactive gliosis and abnormal cerebellar development. Mol. Cell. Neurosci. 7, 354–370 (1996). 75. LaFerla, F.M., Sugarman, M.C., Lane, T.E. & Leissring, M.A. Regional hypomyelination and dysplasia in transgenic mice with astrocyte-directed expression of interferongamma. J. Mol. Neurosci. 15, 45–59 (2000). 76. Lin, W. et al. Enhanced integrated stress response promotes myelinating oligodendrocyte survival in response to interferon-gamma. Am. J. Pathol. 173, 1508–1517 (2008). 77. Boyce, M. & Yuan, J. Cellular response to endoplasmic reticulum stress: a matter of life or death. Cell Death Differ. 13, 363–373 (2006). 78. Morell, P. & Quarles, R.H. Myelin formation, structure and biochemistry. in Basic Neurochemistry: Molecular, Cellular and Medical Aspects (eds. Siegel, G.J., Agranoff, B.W., Albers, R.W., Fisher, S.K. & Uhler, M.D.) 69–93 (Lippincott-Raven Publishers, Philadelphia, 1999). 79. Szegezdi, E., Fitzgerald, U. & Samali, A. Caspase-12 and endoplasmic reticulum stress–mediated apoptosis: the story so far. Ann. NY Acad. Sci. 1010, 186–194 (2003). 80. Nakagawa, T. et al. Caspase-12 mediates endoplasmic reticulum–specific apoptosis and cytotoxicity by amyloid-beta. Nature 403, 98–103 (2000). 81. Sharma, R. & Gow, A. Minimal role for caspase 12 in the unfolded protein response in oligodendrocytes in vivo. J. Neurochem. 101, 889–897 (2007).

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

385

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

Axon degeneration underlies many common neurological disorders, but the signaling pathways that orchestrate axon degeneration are unknown. We found that dual leucine kinase promoted degeneration of severed axons in Drosophila and mice, and that its target, c-Jun N-terminal kinase, promoted degeneration locally in axons as they committed to degenerate. This pathway also promoted degeneration after chemotherapy exposure and may be a component of a general axon selfdestruction program. Axons degenerate in a range of neurological disorders, such as mechanical injury, chemotherapy-induced neuropathy, hereditary neuropathies, glaucoma, diabetes and the neurodegenerative diseases Alzheimer’s disease and Parkinson’s disease1. Degenerating axons follow a stereotyped pathological progression that was first described in the 1850s for the breakdown of the distal segments of severed axons and was termed Wallerian degeneration2. This degeneration probably results from an active self-destruction program rather than from passive deterioration2, as it is delayed by overexpression of the chimeric Wallerian degeneration slow3 (Wlds) protein and its component nicotinamide mononucleotide adenylyltransferase4,5, by preventing Figure 1 Neuronal Wnd promotes axon degeneration in Drosophila. (a) GFP was expressed in a subpopulation of ORNs using Or47b-Gal4 to visualize ORN axons. Before severing, the commissure shown in the boxed region was present in all genotypes. (b,c) Higher-power images of the commissural region taken 24 h after axotomy showed that no axons remained in the commissure of most wild-type flies (14 of 46 flies had a commissure, b) and that axons were significantly preserved in the commissure of most wnd mutant flies (33 of 49 flies had a commissure, c). This reflects a delay of degeneration, as remaining commissures were thinner than non-axotomized commissures and wnd mutant flies reached wild-type levels of degeneration after 48 h (data not shown). (d) When Wnd was exclusively expressed in the GFP-expressing subpopulation of ORNs of otherwise wnd mutant flies, no axons were detected in the commissure of most flies (3 out of 20 flies had a commissure). Thus, Wnd promotes degeneration cell autonomously. (e) Fraction of flies of each genotype with and without an ORN commissure. Axons were significantly preserved in genotype C compared with both genotype B and genotype D (P o 0.005) and axons in genotype D were not preserved compared with genotype B (P 4 0.25; w2, two degrees of freedom). Scale bar represents 25 mm (a) and 10 mm (b–d).

a

b c d

e Fraction of flies with commissure

© 2009 Nature America, Inc. All rights reserved.

Bradley R Miller1, Craig Press2, Richard W Daniels1, Yo Sasaki2, Jeffrey Milbrandt2 & Aaron DiAntonio1

Ca2+ influx6, by blocking protein degradation7, and by disrupting the phagocytic clearance of axon fragments8. A variety of insults trigger axon degeneration2,9 and identifying the components of the hypothesized axon self-destruction program could have broad clinical value. To date, however, no loss-of-function mutations have been identified that disrupt the internal axon breakdown machinery and the internal signaling pathways that orchestrate axon breakdown in injury and disease remain unknown. Candidate components of axon breakdown pathways should be present in axons and should be activated by diverse cellular insults. One such candidate is dual leucine kinase (DLK), a mitogen-activated protein kinase kinase kinase (MAP3K)10. One of DLK’s downstream targets, the mitogen-activated protein kinase (MAPK) c-Jun Nterminal kinase (JNK), is activated following axonal injury11. We tested the hypothesis that DLK promotes axon degeneration using a wellestablished Drosophila olfactory receptor neuron (ORN) axotomy model8,12 (Fig. 1, complete methods are provided in the Supplementary Methods online). We expressed green fluorescent protein (GFP) in ORNs to visualize their axons, which extend from cell bodies in the antennae into the antennal lobes of the brain and across a midline commissure (Fig. 1a). To sever ORN axons and induce degeneration, we removed the antennae from wild-type flies and mutants lacking the Drosophila ortholog of DLK, wallenda (wnd)13. Most wild-type axons degenerated in 24 h (Fig. 1b), whereas wnd mutant axons were significantly preserved (P o 0.001; Fig. 1c). Wnd is therefore required for normal axon degeneration in Drosophila. Wnd could act in neurons to promote breakdown after injury or in surrounding cells to promote axon clearance. To distinguish between these possibilities, we expressed Wnd in the GFP-expressing subpopulation of ORNs in wnd mutant flies. Such Wnd expression was not

Fraction without commissure

A dual leucine kinase–dependent axon self-destruction program promotes Wallerian degeneration

0.8 C 0.6 0.4 0.2

B D

0.2 0.4 0.6 0.8

1Departments of Developmental Biology and 2Pathology, Hope Center for Neurological Disorders, Washington University School of Medicine, St. Louis, Missouri, USA. Correspondence should be addressed to A.D. ([email protected]).

Received 3 December 2008; accepted 6 February 2009; published online 15 March 2009; doi:10.1038/nn.2290

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

387

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

Vincristine

c

d

e

f

DLK mutant

Wild type

b

Wild type axotomized

DLK axotomized

g

h

i

j

k

l

Toluidine blue Electron microscopy

© 2009 Nature America, Inc. All rights reserved.

Wild type non-axotomized

sufficient to induce degeneration in the absence of injury. However, we found that the Wnd-expressing axons of these otherwise wnd mutant flies were not preserved 24 h after axotomy (Fig. 1d). Thus, Wnd functions in an internal neuronal pathway that promotes injury-induced axon degeneration. Wnd may selectively promote injury-induced axon degeneration, as we found no defects in the developmental pruning of mushroom body gamma-lobe axons (data not shown). To determine whether DLK promotes Wallerian degeneration in mammals, we used dorsal root ganglion (DRG) cultures from littermate wild-type (Fig. 2a–c) and DLK-deficient (Fig. 2d–f) embryos14 (Supplementary Fig. 1 online). We severed DRG axons to induce degeneration and evaluated the degeneration of the distal axon segment. Wild-type axons distal to the transection deteriorated into axon fragments 24 h after severing, whereas DLK-deficient axons retained their continuity (Fig. 2b,e). To quantify the extent of axon fragmentation, we measured the fraction of total axonal area that was occupied by axon fragments (degeneration index) and found that DLK-deficient axons were significantly preserved (P o 0.001; Fig. 2b,e). This delay in fragmentation persisted for 48 h

Figure 3 JNK promotes Wallerian degeneration and is critical in the first 3 h after axotomy. (a) Shaded areas designate the drug treatment period. kP38 refers to the P38 inhibitor SB203580 (20 mM) and kJNK refers to the JNK inhibitor SP600125 (15 mM). (b) Degeneration index of the conditions shown in c–h. (c–h) Phase-contrast images of axotomized wild-type DRG axons with the indicated drug treatment. Shown are wild-type DRG axons distal to the axotomy 24 h post-axotomy. Conditions e and g showed significantly less degeneration than conditions c, d, f and h (P o 0.001, ANOVA and post hoc Tukey test). Error bars represent s.e.m. Scale bar represents 20 mm.

388

Figure 2 Normal axon degeneration in mice requires DLK in vitro and in vivo. (a–f) Phase contrast images of DRG axons from littermate wild-type (a–c) and DLK mutant embryos (d–f). DLK-deficient axons had a 65 ± 3.2% (s.e.m.) degeneration decrease relative to controls (P o 0.001, Student’s t test) 24 h after axotomy (b,e). Vincristine induced 59 ± 6.3% (s.e.m.) less degeneration in DLK mutants than in littermate controls after 48 h (P o 0.002, Student’s t test; c,f). (g–l) In vivo sciatic nerve transections in adult mice. Sciatic nerve cross-sections distal to the transection site were stained with Toluidine blue (g–i) or imaged by electron microscopy (j–l). Axons were significantly preserved in DLK-deficient nerves 52 h post-axotomy (208 ± 22% axon profiles per mm2 compared with wild-type nerves, P o 0.007, Student’s t test, n ¼ 4 wild-type animals and 5 DLK mutant animals; h,i). Electron microscopy showed that these profiles contained axons with mitochondria and a cytoskeletal network (j–l; see Supplementary Fig. 3). Scale bars represent 20 mm (a–i) and 2 mm (j–l). All mouse experiments were approved by the Washington University School of Medicine Animal Studies Committee.

(Supplementary Fig. 2 online). Because non-neuronal cells are eliminated in this DRG culture system, DLK must operate in mammalian neurons to promote axon breakdown. Neuronal DLK therefore promotes axon fragmentation after injury in flies and mice. To determine whether DLK promotes degeneration in response to multiple insults, we assessed the response of DLK-deficient DRG axons to vincristine, a chemotherapeutic drug that induces axon degeneration in vitro and whose dose-limiting side effects in patients include neuropathy15. We found that DLK-deficient axons were significantly protected from vincristine-induced fragmentation (P o 0.002; Fig. 2c,f), suggesting that DLK operates in a general axon breakdown program. To determine whether disrupting DLK protects injured axons in vivo, we transected the sciatic nerves of littermate wild-type and DLKdeficient adult mice (Fig. 2g–l). Wild-type axons degenerated 52 h post-transection, whereas DLK-deficient axons were significantly preserved (P o 0.007; Fig. 2h,i). Electron microscopy revealed that preserved axon profiles contained mitochondria and a cytoskeleton (Fig. 2l and Supplementary Fig. 3 online). Thus, normal Wallerian degeneration in vivo in adult mice requires DLK.

a

Axotomy

c Vehicle d P38 e JNK f JNK g h (h) –24

Image

JNK JNK

0

3

24

d

e

f

g

h

[

NUMBER 4

[

1.0 0.8 0.6 0.4 0.2 c d e f g h

c

VOLUME 12

b Degeneration index

Nontreated

a

APRIL 2009 NATURE NEUROSCIENCE

© 2009 Nature America, Inc. All rights reserved.

B R I E F C O M M U N I C AT I O N S DLK is a MAP3K that can activate JNK and p38 via intermediary MAP2Ks10. To determine whether either downstream kinase promotes axon degeneration, we inhibited JNK and p38 in the DRG axotomy model using wild-type cultures. Inhibition of JNK, but not p38, significantly protected transected axons from fragmentation (P o 0.001; Fig. 3a–e) and a significant delay in fragmentation persisted for over 48 h (P o 0.001; Supplementary Fig. 2). Thus JNK, similar to DLK, acts in neurons to promote axon degeneration. Axon degeneration is hypothesized to comprise at least three distinct phases: competence to degenerate, much of which is determined transcriptionally before axotomy, commitment to degenerate, which occurs in the substantial delay period between injury and axon fragmentation, and the execution phase, when axons fragment1. If JNK’s primary role were to promote competence to degenerate, then JNK activity should be required before axotomy. This is not the case, as applying the JNK inhibitor 24 h before axotomy and then removing it just before axotomy was not protective (Fig. 3f). In contrast, JNK inhibition started concurrently with axotomy was significantly protective (P o 0.001; Fig. 3g). Thus, JNK promotes axon fragmentation after the competence period and acts in the severed distal axon segment. JNK could commit axons to degenerate during the delay between injury and breakdown or it could operate during the subsequent execution phase of axon breakdown. To test whether JNK activity is required during the execution phase, we added the JNK inhibitor 3 h after axotomy, which is approximately 9 h before the onset of fragmentation. This treatment schedule spans the transition from the proposed commitment phase to the execution phase and the entire execution phase itself. Continuous JNK inhibition beginning 3 h postaxotomy did not delay axon fragmentation (Fig. 3h). Therefore, JNK activity is not required during the execution phase of axon fragmentation. Instead, JNK activity is required during the early response to injury that commits the axon to breakdown hours later. Converging lines of evidence suggest that there is a general internal axon self-destruction program, but its molecular components are unknown. We found that the MAP3K DLK and its downstream MAPK JNK are important elements of such a program. Disrupting

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

this pathway delayed axon fragmentation in response to both axotomy and the neurotoxic chemotherapeutic agent vincristine. Thus, a common self-destruction program may promote axon breakdown in response to diverse insults and may be targetable in multiple clinical settings. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank the members of our laboratories, V. Cavalli and E.M. Johnson. This work was supported by US National Institutes of Health grants P30 NS057105 to Washington University, NS040745 (J.M.), AG13730 (J.M.), DA 020812 (A.D.), Washington University Alzheimer’s Disease Research Center National Institute on Aging grant NIA P50 AG05681-25 (A.D.), the Hope Center for Neurological Disorders and the Keck Foundation (A.D.). AUTHOR CONTRIBUTIONS B.R.M. designed and conducted all of the experiments and co-wrote the manuscript. C.P. helped to design and to conduct the in vitro DRG experiments. R.W.D. contributed to the electron microscopy analysis. Y.S. developed the degeneration index algorithm. J.M. helped to supervise the project. A.D. supervised the project and co-wrote the manuscript. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Saxena, S. & Caroni, P. Prog. Neurobiol. 83, 174–191 (2007). 2. Coleman, M. Nat. Rev. Neurosci. 6, 889–898 (2005). 3. Mack, T.G. et al. Nat. Neurosci. 4, 1199–1206 (2001). 4. Araki, T., Sasaki, Y. & Milbrandt, J. Science 305, 1010–1013 (2004). 5. Wang, J. et al. J. Cell Biol. 170, 349–355 (2005). 6. George, E.B., Glass, J.D. & Griffin, J.W. J. Neurosci. 15, 6445–6452 (1995). 7. Zhai, Q. et al. Neuron 39, 217–225 (2003). 8. MacDonald, J.M. et al. Neuron 50, 869–881 (2006). 9. Luo, L. & O’Leary, D.D. Annu. Rev. Neurosci. 28, 127–156 (2005). 10. Gallo, K.A. & Johnson, G.L. Nat. Rev. Mol. Cell Biol. 3, 663–672 (2002). 11. Cavalli, V., Kujala, P., Klumperman, J. & Goldstein, L.S. J. Cell Biol. 168, 775–787 (2005). 12. Hoopfer, E.D. et al. Neuron 50, 883–895 (2006). 13. Collins, C.A., Wairkar, Y.P., Johnson, S.L. & DiAntonio, A. Neuron 51, 57–69 (2006). 14. Bloom, A.J., Miller, B.R., Sanes, J.R. & DiAntonio, A. Genes Dev. 21, 2593–2606 (2007). 15. Wang, M.S., Wu, Y., Culver, D.G. & Glass, J.D. J. Neuropathol. Exp. Neurol. 59, 599–606 (2000).

389

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

Thalamic activity that drives visual cortical plasticity Manipulations of activity in one retina can profoundly affect binocular connections in the visual cortex. Retinal activity is relayed to the cortex by the dorsal lateral geniculate nucleus (dLGN). We compared the qualities and amount of activity in the dLGN following monocular eyelid closure and monocular retinal inactivation in awake mice. Our findings substantially alter the interpretation of previous studies and define the afferent activity patterns that trigger cortical plasticity. The quality of sensory experience during early postnatal life has a crucial role in the development of cortical circuitry and function. In the visual system, this role has been investigated by comparing the consequences of temporary monocular eyelid closure and pharmacological inactivation of one retina with those of normal visual experience

Figure 1 Firing rate and ISI distributions before Open Open and after visual manipulation. (a) Peristimulus O O O O time histograms and raster plots from 12 representative neurons for each experimental 0.8 group. Stimuli were presented at 0 or 901, 1-Hz 8 1.2 phase reversing. Arrowheads in this and 0.4 4 subsequent figures indicate time of stimulus phase reversal. Spike waveforms are recording 0 0 0 2 104 1 10 session averages. Scale bars represent 100 mV Open Closed and 500 ms. Left, data were obtained during O C O C baseline. Right, data were obtained after eye 12 manipulation. Top, control group; middle, 0.8 monocular eyelid closure group; bottom, retinal 8 2 inactivation group. (b) Firing rates (recording 0.4 4 session average) for each neuron in each group. Connected circles represent the same 0 0 0 1 102 104 neuron recorded before and after eye Open Inactivated manipulation (O, open; C, closed; I, O I O I inactivated). Black lines indicate median 12 values (control: n ¼ 22 neurons (9 mice), 0.8 P 4 0.2 Wilcoxon sign-rank; eyelid closure: 8 1.6 n ¼ 24 neurons (12 mice), P 4 0.3; retinal 0.4 4 inactivation: n ¼ 19 neurons (8 mice), P 4 0.3). See Supplementary Figure 6 for 0 0 0 0 0.5 1.0 0 0.5 1.0 1 102 104 similar results with natural visual stimuli. Time (s) Time (s) Time (ms) (c) ISI distributions during baseline (thick black line) and after eye manipulation (thin black line). Note that retinal inactivation increased the probability of observing short ISIs (P o 10
5). Inset, probability density function (y axis, 0–0.14; x axis, 0–20 ms); the curves differ significantly from 2–4 ms (P o 0.01, Wilcoxon sign-rank).

b

c

ISI cum. prob.

ISI cum. prob. ISI cum. prob.

Firing rate (Hz)

Spikes

Firing rate (Hz)

Spikes

Firing rate (Hz)

a

Spikes

© 2009 Nature America, Inc. All rights reserved.

Monica L Linden1,2, Arnold J Heynen1, Robert H Haslinger2,3 & Mark F Bear1,2

(NVE). Previous studies have shown that eyelid closure and retinal inactivation have very different effects on visual cortex1–4. A brief period of lid closure causes long-term synaptic depression (LTD) of deprived-eye responsiveness, whereas a comparable period of retinal inactivation has no effect on deprived-eye responsiveness and instead causes an increase in the responses to stimulation of the nondeprived eye (see Supplementary Fig. 1 online). Understanding how lid closure and retinal inactivation differ from one another and from NVE is of great interest because it may reveal how deprivation triggers LTD and causes visual disability. We examined this in awake mice at the age of maximal sensitivity to visual deprivation. Bundles of electrodes were inserted in the dLGN (see Supplementary Fig. 2 and Supplementary Methods online) at the age of maximal sensitivity to monocular deprivation (Bpostnatal day 28, P28)5. Baseline recordings were made with the contralateral eye viewing and the ipsilateral eye occluded. Activity approximating that during NVE was recorded in response to phase-reversing sinusoidal gratings and natural scene stimuli. Unless otherwise indicated, results using grating stimuli are illustrated, as results from natural scenes did not differ qualitatively. Following the baseline recording session, we briefly anesthetized the mice and carried

1Howard Hughes Medical Institute, The Picower Institute for Learning and Memory, 2Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA. 3Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts, USA. Correspondence should be addressed to M.F.B. ([email protected]).

Received 10 October 2008; accepted 27 January 2009; published online 1 March 2009; doi:10.1038/nn.2284

390

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

B R I E F C O M M U N I C AT I O N S Figure 2 Analysis of bursting activity before and Open Closed Inactivated Inactivated (48 h) after visual manipulation. (a) Raster plots of 80 stimulus trials from representative neurons (those nearest the median) in each experimental group. Black squares represent spikes in bursts and gray squares represent non-burst spikes. (b) Bursting of a representative neuron 48 h after intraocular TTX injection (see also Supplementary Fig. 3). (c) Burst percentage for each neuron in each * group. Connected circles represent the same TTX injection neuron recorded before and after eye 40 manipulation. Black lines indicate the median 40 values (control: n ¼ 22 neurons (9 mice), 20 P 4 0.7; eyelid closure: n ¼ 24 neurons (12 mice), P 4 0.2); retinal inactivation: n ¼ 19 0 neurons (8 mice), P o 10
3 Wilcoxon sign-rank). 0 –2 2 24 48 120 O O O C O I (d) Burst percentage as a function of the duration Time (h) of retinal inactivation. Circles represent individual neurons. Black lines indicate the median values. At 2, 24 and 48 h, the percentage of spikes in 8 * * * * bursts was significantly different from both the 4 baseline and recovery (120 h) time points (P o 20 40 0.05 for all comparisons, Mann-Whitney U test). 4 The baseline and recovery time points were not significantly different (P 4 0.7, Mann-Whitney 0 0 0 0 U test), nor were the time points during retinal O I O I O I O I inactivation (P 4 0.4 for all comparisons, MannWhitney U test). (e,f) Inactivation of the contralateral eye increased firing rate and bursting of neurons in the dLGN ipsilateral core (n ¼ 9 neurons (4 mice), P o 0.02 and 0.01, respectively, Wilcoxon sign-rank). Data are presented as in c (see also Supplementary Fig. 4). (g,h) Neuronal firing rate and the percentage of spike in bursts decreased significantly after retinal inactivation in Nembutal-anesthetized mice (n ¼ 9 neurons (3 mice), P o 0.05 and 0.03, respectively, Wilcoxon sign-rank).

b

c

d

Burst percent

a

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

Burst percent

Firing rate (Hz)

Burst percent

was a shift to the left in the ISI distribution after retinal inactivation (P o 0.00001). The shift corresponded to an increased probability of observing ISIs from 2–4 ms, suggesting that dLGN neurons tend to fire in bursts when the retina is inactivated. Thalamic bursts have been defined in previous studies as an initial period of quiescence (4100 ms) followed by two or more spikes with an ISI o 4 ms6. We analyzed burst firing using this definition and found a significant increase following retinal inactivation (P o 0.001; Fig. 2). In a subset of mice, we confirmed that increased burst firing persisted for the entire duration of retinal inactivation (Z 48 h) and returned to baseline values following TTX wash out (Fig. 2b,d and Supplementary Fig. 3 online). Occasionally, we recorded from the ipsilateral core of the dLGN instead of our intended target of the contralateral shell. Notably, even though the retinal input to these neurons had not been changed, we

0.7 Inactivated ρ

a

Closed ρ

Figure 3 Eyelid closure and retinal inactivation have opposite effects on correlative dLGN firing. (a) Scatter plots of the area under the crosscorrelogram from pairs of simultaneously recorded neurons before and after visual manipulation for each pair of simultaneously recorded neurons. Gray line represents unity. Note that nearly all points fall below the unity line following eyelid closure (center panel), indicating a decrease in correlative firing (r indicates correlation coefficient). (b) Eyelid closure and retinal inactivation had opposite effects on spike correlation (P o 10
4 KruskalWallis test). Bars represent the median change in area under the peak of the cross-correlogram (±10 ms) following visual manipulation. Error bars show the interquartile range. Eyelid closure and retinal inactivation induced significant changes in correlation (control: n ¼ 22 neuron pairs (6 mice), P 4 0.2 Wilcoxon sign-rank; eyelid closure: n ¼ 20 neuron pairs (6 mice), P o 10
3; retinal inactivation: n ¼ 18 neuron pairs (6 mice), P o 0.01; see also Supplementary Fig. 6). (c) Bursts contributed to increased correlation following retinal inactivation. Data are represented as in b. The bursts group considered only the first spikes in each burst (n ¼ 18 neuron pairs (6 mice), P o 0.03); the non-bursts group considered all spikes not contained in bursts (P 4 0.9).

h

0

0

0.7

0.7

0

0

0.7

Open ρ

b

* 0.1

c

*

0.1

∆(ρ)

out eyelid closure, intraocular tetrodotoxin (TTX) injection or no manipulation (control). After 430 min of recovery from anesthesia, stimuli were presented for a second recording session. Although eyelid closure and retinal inactivation abolished the visually evoked responses, these manipulations had no effect on spontaneous activity, and therefore had no effect on recording session averages of firing rate (Fig. 1a,b). These findings invalidate the assumptions that visual deprivation simply reduces activity that is afferent to the cortex and that silencing the retina silences the input to cortex. To determine whether the temporal patterning of spikes was affected by deprivation, we analyzed the distribution of interspike intervals (ISIs) before and after deprivation (Fig. 1c). Again, we were surprised to find that there was no significant effect of eyelid closure on the distribution of ISIs (P 4 0.08). Even more unexpected, however,

g

f

Open ρ

Firing rate (Hz)

© 2009 Nature America, Inc. All rights reserved.

e

* –0.1

0 O

*

C

I

All Bursts Nonspikes bursts

391

© 2009 Nature America, Inc. All rights reserved.

B R I E F C O M M U N I C AT I O N S found clear evidence for robust changes in the firing properties of these neurons after inactivation of the contralateral eye (Fig. 2e,f and Supplementary Fig. 4 online). Bursting following retinal inactivation or eye enucleation has been described previously in the ferret dLGN, but these observations were made in very young animals, before natural eye opening and the developmental onset of sensitivity to monocular deprivation7. It has long been assumed that by adolescence, monocular TTX treatment simply reduces activity in the central visual system1–4,8–13. The experimental basis for this assumption can be traced to studies in anesthetized cats1,8. We therefore examined the effects of anesthesia and found that retinal inactivation significantly decreased dLGN activity in the anesthetized mouse (P o 0.05; Fig. 2g,h and Supplementary Fig. 5 online). These findings illustrate the importance of using an awake preparation when determining the patterns of activity that drive cortical plasticity14. None of the properties of individual spike trains (firing rate and burst percentage) differentiated NVE and eyelid closure. Therefore, we also investigated the correlative firing between simultaneously recorded neurons (Fig. 3 and Supplementary Fig. 6 online). This analysis revealed a significant decrease in simultaneously active dLGN neurons in the eyelid closed condition relative to both NVE and retinal inactivation (P o 0.001). Thus, the manipulation of vision that triggered robust LTD in visual cortex (lid closure; see Supplementary Fig. 1) also decorrelated the input to cortex. Monocular retinal inactivation, which does not trigger response depression (Supplementary Fig. 1), actually significantly increased simultaneous firing of neuron pairs (P o 0.01; Fig. 3b), largely as a consequence of synchronous bursting activity (P o 0.03; Fig. 3c). Our results show that the markedly different consequences on visual cortex of deprivation by eyelid closure and retinal inactivation1–3 are accounted for by equally marked differences in dLGN activity.

392

Although neither eyelid closure nor retinal inactivation caused a decrease in the amount of firing, lid closure resulted in a decrease of correlative firing between pairs of simultaneously recorded neurons and retinal inactivation caused an increase in thalamic bursting. These results overturn previous assumptions, provide new insight into the mechanisms that drive ocular dominance plasticity and suggest fruitful avenues for future research (see Supplementary Discussion online). Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank M. Shuler, J. Coleman, M. Lamprecht, B. Blais, H. Shouval, E. Sklar, K. Oram and S. Meagher. This work was partly supported by grants from the National Eye Institute and a National Research Service Award fellowship from the US National Institute of Neurological Disorders and Stroke (M.L.L.). Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/

1. Rittenhouse, C.D., Shouval, H.Z., Paradiso, M.A. & Bear, M.F. Nature 397, 347–350 (1999). 2. Heynen, A.J. et al. Nat. Neurosci. 6, 854–862 (2003). 3. Frenkel, M.Y. & Bear, M.F. Neuron 44, 917–923 (2004). 4. Maffei, A. & Turrigiano, G.G. J. Neurosci. 28, 4377–4384 (2008). 5. Gordon, J.A. & Stryker, M.P. J. Neurosci. 16, 3274–3286 (1996). 6. Lu, S.M., Guido, W. & Sherman, S.M. J. Neurophysiol. 68, 2185–2198 (1992). 7. Weliky, M. & Katz, L.C. Science 285, 599–604 (1999). 8. Stryker, M.P. & Harris, W.A. J. Neurosci. 6, 2117–2133 (1986). 9. Greuel, J.M., Luhmann, H.J. & Singer, W. Brain Res. 431, 141–149 (1987). 10. Catalano, S.M., Chang, C.K. & Shatz, C.J. J. Neurosci. 17, 8376–8390 (1997). 11. Caleo, M., Lodovichi, C. & Maffei, L. Eur. J. Neurosci. 11, 2979–2984 (1999). 12. Desai, N.S., Cudmore, R.H., Nelson, S.B. & Turrigiano, G.G. Nat. Neurosci. 5, 783–789 (2002). 13. Young, J.M. et al. Nat. Neurosci. 10, 887–895 (2007). 14. Greenberg, D.S., Houweling, A.R. & Kerr, J.N. Nat. Neurosci. 11, 749–751 (2008).

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

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

D2R striatopallidal neurons inhibit both locomotor and drug reward processes

© 2009 Nature America, Inc. All rights reserved.

Pierre F Durieux1, Bertrand Bearzatto1, Stefania Guiducci2, Thorsten Buch3, Ari Waisman4, Michele Zoli2, Serge N Schiffmann1,5 & Alban de Kerchove d’Exaerde1,5 The specific functions of dopamine D2 receptor–positive (D2R) striatopallidal neurons remain poorly understood. Using a genetic mouse model, we found that ablation of D2R neurons in the entire striatum induced hyperlocomotion, whereas ablation in the ventral striatum increased amphetamine conditioned place preference. Thus D2R striatopallidal neurons limit both locomotion and, unexpectedly, drug reinforcement. The striatum is critically involved in motor and motivational functions1,2. The dorsal striatum, caudate-putamen, is primarily implicated in motor control and the learning of habits and skills, whereas the ventral striatum, the nucleus accumbens (NAc), is essential for motivation and drug reinforcement1,3. Striatal dysfunction has been demonstrated in movement disorders, including Parkinson’s and Huntington’s disease, and in psychiatric disorders, such as schizophrenia and drug addiction4. The GABA medium-sized spiny neurons (MSNs, about 95% of striatal neurons), which are targets of the cerebral cortex and the midbrain dopaminergic neurons, form two pathways5. The dopamine D1 receptor–positive (D1R) striatonigral MSNs project to the medial globus pallidus and substantia nigra pars reticulata (direct pathway) and coexpress dopamine D1 receptors and substance P, whereas D2R striatopallidal MSNs project to the lateral globus pallidus (indirect pathway) and coexpress dopamine D2 receptor, adenosine A2A receptor (A2AR) and enkephalin (Enk). The specific role of the two efferent pathways in motor and motivational control remains poorly understood. D1R striatonigral and D2R striatopallidal neurons, which are intermingled and morphologically indistinguishable, cannot be functionally dissociated with techniques such as chemical lesions or surgery and the currently available tools for selective targeting of these populations are unsatisfactory. The Drd1a- and Drd2-egfp transgenic mice obtained by BAC transgenesis6 have recently shed some light on the role of MSN subpopulations or genes in striatal pathophysiology7–10. In regards to their role in motivation and drug addiction, current studies are focused mostly on the D1R striatonigral neurons2. To assess the role of D2R striatopallidal neurons, we selectively ablated these cells in adult mice by Cre-mediated expression of a

diphtheria toxin receptor (DTR) and diphtheria toxin injection11 (Supplementary Methods online). All animal procedures were approved by the Universite´ Libre de Bruxelles School of Medicine Ethical Committee. We generated mice expressing Cre recombinase under the control of the Adora2a (A2AR) promoter (Adora2a-cre mice, Supplementary Fig. 1 online) by BAC transgenesis. A2AR was chosen because it is expressed more in D2R neurons than in any other brain area12 and, in contrast to D2R, A2AR is supposed to not be expressed in striatal cholinergic interneurons and mesostriatal dopaminergic cells. In Adora2a-cre mice mated with a Rosa26-LacZ reporter strain, b-galactosidase staining was only found in striatal Enk-positive cells (Supplementary Fig. 1), indicating that Cre was selectively expressed in D2R striatopallidal neurons. Adora2a-cre mice were crossed with mice in which the expression of a simian DTR (Hbegf ) gene from a ubiquitously active promoter is prevented by a loxP-flanked stop cassette (inducible DTR mice, iDTR)11, leading to double transgenic Adora2a-cre/+; iDTR/+ mice (DTR-positive mice) that selectively expressed DTR in D2R striatopallidal neurons (Supplementary Fig. 2 online). Toxin was stereotaxically injected into the striatum to produce unilateral or bilateral ablation of D2R neurons in the entire striatum (full ablation) or in the NAc (NAc ablation). We analyzed the kinetics of D2R striatopallidal ablation in DTR-positive mice with unilateral striatum diphtheria toxin injections by quantifying A2AR binding sites from 3–28 d after diphtheria toxin injections (Fig. 1a–c). The injected side showed a 65% and 45% reduction in A2AR binding in the dorsal and ventral striatum, respectively, at 7 d after injections, and an almost complete loss of striatal A2AR binding (97% in dorsal and 87% in ventral striatum) from 14 d after diphtheria toxin injections, with no reduction in striatonigral neuron–specific D1R binding (Fig. 1a–c). We found no changes in binding sites in diphtheria toxin–injected control littermates lacking DTR expression (Adora2a-cre–/–; iDTR/+; DTR negative mice) or in saline-injected DTR-positive mice. The disappearance of D2R striatopallidal neuron mRNAs (D2R, A2AR and Enk) and the preservation of D1R striatonigral neuron mRNAs (D1R and substance P) on the injected side at 14 d after injections (Fig. 1d,e) confirmed the specificity of the lesion. The same pattern of ablation was detected from the anterior to the posterior striatum (Supplementary Fig. 3 online). The four subpopulations of striatal interneurons remained intact (Supplementary Fig. 4 online). In light of the connections between midbrain dopamine and striatal neurons, we investigated the consequences of D2R striatopallidal neuron loss on the mesostriatal dopamine system. Bilateral full ablation of D2R striatopallidal neurons did not modify cell body or terminal dopaminergic markers (Supplementary Fig. 5 online). To assess dopaminergic function in vivo, we carried out intrastriatal microdialysis in bilateral diphtheria toxin–injected DTR-positive and

1Laboratory

of Neurophysiology, Universite´ Libre de Bruxelles, Brussels, Belgium. 2Department of Biomedical Sciences, Section of Physiology, University of Modena and Reggio Emilia, Modena, Italy. 3Department of Pathology, Institute of Experimental Immunology, University of Zurich, Zurich, Switzerland. 4First Medical Department, Johannes Gutenberg University of Mainz, Mainz, Germany. 5These authors contributed equally to this work. Correspondence should be addressed to A.d.K.d.E. ([email protected]). Received 19 December 2008; accepted 28 January 2009; published online 8 March 2009; doi:10.1038/nn.2286

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

393

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

+

DTR saline D1R +

c

80 60

0

–

DTR DT D1R –

DTR DT A2AR

+

DTR saline D1R +

DTR saline A2AR

*** Dorsal striatum

***

** ***

3 5 7 9 11 13 15 17 28 Days after unilateral DT or saline injections

100 80

D1R

100

20

+

120 Realtive binding levels (%)

Realtive binding levels (%)

120

40

+

DTR DT D1R DTR DT A2AR

DTR saline A2AR

*** 60 40 20

Ventral striatum

***

** ***

0 3 5 7 9 11 13 15 17 28 Days after unilateral DT or saline injections

e

–

DTR + DTR

Relative A2AR mRNA levels (%)

A2AR –

DTR DT A2AR

Enk

–

DTR DT D1R

+

Substance P

+

DTR DT D1R DTR DT A2AR

© 2009 Nature America, Inc. All rights reserved.

DTR+

D 2R

D1R

b

DTR–

100 80 60 40 20 0

Relative D2R mRNA levels (%)

d

Day 28

100 80 60 40 20 0

Relative Enk mRNA levels (%)

Day 14

100 80 60 40 20 0

Relative D1R mRNA levels (%)

Day 7

A2AR

Day 3

100 80 60 40 20 0

Relative substance P mRNA levels (%)

a

100 80 60 40 20 0

***

***

Dorsal striatum

Ventral striatum

*** Dorsal striatum

***

*** Ventral striatum

***

Dorsal Ventral striatum striatum

Dorsal striatum

Ventral striatum

Dorsal striatum

Ventral striatum

Figure 1 Characterization of D2R striatopallidal neuron ablation after full striatum unilateral diphtheria toxin injections in DTR-positive mice (coronal sections, level +1.2 mm relative to bregma). (a) Autoradiograms of A2ARs and D1Rs, markers of D2R striatopallidal and D1R striatonigral neurons, respectively, from 3–28 d after unilateral full striatum diphtheria toxin injections in DTR-positive (DTR+) mice, showing kinetics and specificity of the D2R striatopallidal neuron ablation. (b,c) A2AR and D1R binding levels in the dorsal (b) and ventral (c) striatum of diphtheria toxin (DT)-injected DTR-positive mice and controls (diphtheria toxin–injected DTR-negative (DTR–) or saline-injected DTR-positive mice) (n ¼ 3
9 in each group). (d) In situ hybridization autoradiograms of D2R striatopallidal (A2AR, D2R and Enk) and D1R striatonigral (D1R and substance P) neuron mRNAs at 14 d after unilateral full striatum diphtheria toxin injections. (e) Striatopallidal (A2AR, D2R and Enk) and striatonigral (D1R and substance P) neuron mRNA levels at 14 d after diphtheria toxin injections (n ¼ 7 in each group). Data are expressed as optical density values of the injected side in percent of the uninjected side. Arrows indicate the injected side. Scale bars represent 1 mm. Data are reported as mean ± s.e.m. *** P o 0.001 (as compared with diphtheria toxin–injected DTR-injected mice); ** P o 0.001 (as compared with saline-injected DTR-positive mice).

DTR-negative mice. No difference in basal dopamine extracellular concentration or in amphetamine (AMPH)-induced dopamine overflow was found between the two genotypes (Supplementary Fig. 5). These results indicate that D2R striatopallidal neuron ablation does not induce major modifications in striatal dopaminergic function. As D2R neurons send GABAergic projections to GABA neurons of the globus pallidus, we assessed the effect of bilateral D2R neuron loss on glutamic acid decarboxylase isoform 67 mRNA levels in the globus pallidus as an indirect index of GABA neuron activity13 (Supplementary Fig. 6 online). We found an increase in glutamic acid decarboxylase isoform 67 mRNA in the globus pallidus of DTR-positive mice as compared with DTR-negative mice, confirming that the D2R striatopallidal neurons exert inhibitory control on globus pallidus GABA neuron activity. Locomotor activity in open field boxes was recorded daily for 30 min (Fig. 2) in mice that underwent bilateral full striatum diphtheria toxin injections. Starting at 6 d after diphtheria toxin injections, DTRpositive mice became threefold to fourfold more active than controls (Supplementary Video 1 online). This hyperactivity was stable through day 16 (Fig. 2c) and DTR-positive mice were still hyperactive (207 ± 19% of control level) 33 d after diphtheria toxin injection (data not shown). These results demonstrate the inhibitory function of the D2R striatopallidal neurons on locomotor activity. As the ventral striatum is the key neuronal substrate for drug reinforcement3, we carried out a NAc D2R striatopallidal neuron ablation in DTR-positive mice (Fig. 2d,e). Enk mRNA levels in the ventral striatum of diphtheria toxin–injected DTR-positive mice decreased by 80% compared with control diphtheria toxin–injected

394

DTR-negative mice. We found a 30% reduction in Enk mRNA in the dorsal striatum, selectively in its rostral part. Because dorsal striatum shows progressive rostro-caudal enlargement, this limited rostral loss modestly influenced the overall number of D2R striatopallidal neurons in the dorsal striatum. NAc ablation did not show spontaneous hyperlocomotion (Supplementary Fig. 7 online), demonstrating that the NAc ablation of D2R striatopallidal neurons is functionally different from the full ablation. NAc DTR-positive mice were examined in an AMPH (1, 3 or 5 mg per kg) conditioned place preference (CPP) procedure followed by an examination of CPP extinction over 1 week (Fig. 2f). The DTR-positive mice showed a higher preference for the AMPH-paired compartment as compared with controls on the first test day (2 d after the last AMPH injection) and maintained greater CPP on the following test days (4 and 9 d after last AMPH injection) (see Supplementary Methods for statistical analyses). Note that there was a loss of CPP for DTR-negative mice on day 9 (AMPH 1 and 3 mg per kg), whereas DTR-positive mice still showed a preference for the AMPH-paired compartment. In summary, our results provide direct experimental evidence that D2R striatopallidal neurons are critical for both the control of motor behavior and drug reinforcement. The Adora2a-cre/+; iDTR/+ mouse, allowing specific D2R striatopallidal neuron ablation, confirmed that D1R–substance P–expressing and D2R-A2AR-Enk–expressing neurons in the striatum are largely segregated. This model shows the advantage of A2AR- over D2R-targeted transgenic mice for targeting D2R striatopallidal neurons, as A2AR does not target striatal cholinergic interneurons and dopaminergic afferents14, thus avoiding cholinergic alterations and distinguishing between post- and presynaptic dopaminergic

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

b

DTR +

Relative Enk mRNA levels (%) DTR–

Relative Enk mRNA levels (%)

Enk caudal striatum

© 2009 Nature America, Inc. All rights reserved.

c

** ***

** ***

** ***

Dorsal Ventral Rostral striatum

Caudal striatum

DTR– DT DTR+ DT

e

DTR+

Enk rostral striatum

d

110 100 90 80 70 60 50 40 30 20 10 0

– DTR DT + DTR saline + DTR DT

110 100 90 80 70 60 50 40 30 20 10 0

f

***

*** Dorsal Ventral Caudal Rostral striatum striatum

20,000 18,000 16,000 14,000 12,000 10,000 8,000 6,000 4,000 2,000 0

DTR– DT DTR+ saline + DTR DT

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

Injections

–3 –2 –1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Days before and after bilateral DT or saline injections

Score (s)

DTR –

Enk Enk caudal striatum rostral striatum

a

Horizontal distance in 30 min (cm)

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

– DTR DT * DTR+ DT 300 250 200 150 100 50 0 0 1 3 5 0 1 3 5 –50 AMPH –100 (mg per kg) –150 Day 2 after last

AMPH injection

*

*

0 1 3 5 0 1 3 5

0 1 3 5 0 1 3 5

Day 4 after last AMPH injection

Day 9 after last AMPH injection

Figure 2 Behavioral consequences of D2R striatopallidal neuron removal and quantification of the D2R striatopallidal neuron ablation. (a) In situ hybridization autoradiograms of Enk mRNA in rostral (level +1.2 mm relative to bregma) and caudal (level
0.1 mm relative to bregma) coronal brain sections of full striatum diphtheria toxin–injected DTR-positive and DTR-negative mice. (b) Quantification of Enk mRNA levels in rostral and caudal striatum. Data are expressed as optical density values of the injected striatum in DTR-positive in percent of DTR-negative mice (n ¼ 6
11 per group). (c) Locomotor activity of DTR-positive and control mice 3 d before to 16 d after full striatum bilateral diphtheria toxin injections (n ¼ 17 in diphtheria toxin–injected DTR-positive group and n ¼ 6 per group in control DTR-negative diphtheria toxin–injected and saline–injected DTR-positive groups). (d,e) In situ hybridization autoradiograms (d) of Enk mRNA in rostral (level +1.2 mm relative to bregma) and caudal (level
0.1 mm relative to bregma) coronal brain sections of NAc diphtheria toxin–injected DTRpositive and DTR-negative mice and quantification of Enk mRNA levels (e). Data are expressed as optical density values of the injected striatum as a percentage of DTR-negative mice (n ¼ 34–39 per group). (f) CPP for AMPH of diphtheria toxin–injected DTR-positive and DTR-negative mice. Preference score was measured at days 2, 4 and 9 after the last AMPH injection (n ¼ 8–14 in each group). Arrows indicate the injected side. Scale bars represent 1 mm. Data are reported as mean ± s.e.m. Statistical comparisons were made between diphtheria toxin–injected DTR-positive mice and respective control mice. * P o 0.05, *** P o 0.001 (as compared with diphtheria toxin–injected DTR-positive mice); ** P o 0.001 (as compared with saline-injected DTR-positive mice).

mechanisms. The complete bilateral ablation of these neurons induces persistent spontaneous hyperlocomotion, demonstrating a functional effect of D2R striatopallidal neuron loss and validating the hypothesis that A2AR-D2R–expressing neurons normally inhibit motor activity1. The increase in AMPH CPP following ablation of D2R striatopallidal neurons mainly in the NAc was unanticipated and is, to the best of our knowledge, the first experimental demonstration that the pathway in which these neurons take part normally inhibits drug reinforcement. These data suggest that, similar to what has been observed for motor control, reciprocal antagonism between D2R striatopallidal and D1R striatonigral neurons is crucial for motivational processes and reinforcement. The involvement of D2Rs in drug reward processes is still puzzling15, as D2Rs are expressed at many sites in the striatal network. Our results suggest that the activation of postsynaptic D2Rs on D2R striatopallidal neurons in the NAc facilitates drug reinforcement by inhibiting these neurons4,15. Together, these data show that Adora2a-cre/+; iDTR/+ mice are a useful movement disorder model and underscore the need for characterization of the specific cellular and molecular modifications that are induced in D2R striatopallidal neurons by drugs of abuse. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank M. Picciotto for helpful and critical comments on the manuscript and D. Houtteman, S. Laghmiri and L. Cuvelier for expert technical assistance. P.F.D. is Research Fellow of the Fonds de la Recherche Scientifique (bourse de doctorat Fonds de la Recherche Scientifique) and A.d.K.d.E. is a Research Associate of the

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

Fonds de la Recherche Scientifique (Belgium). This study was supported by Fondation Me´dicale Reine Elisabeth (Belgium), Fonds de la Recherche Scientifique (Belgium), Fonds d’Encouragement a` la Recherche from the Universite´ Libre de Bruxelles, Action de Recherche Concerte´e from the Communaute´ Franc¸aise Wallonie Bruxelles and Ministero Italiano dell’Universita` e della Ricerca (grant number PRIN20072BTSR2) to M.Z. AUTHOR CONTRIBUTIONS P.F.D., S.N.S. and A.d.K.d.E. conceived and designed the experiments. P.F.D., A.d.K.d.E., B.B. and S.G. carried out the experiments. T.B. and A.W. contributed materials. P.F.D., M.Z., S.N.S. and A.d.K.d.E. analyzed the data and wrote the paper. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

1. Graybiel, A.M. Curr. Biol. 10, 509–511 (2000). 2. Hyman, S.E., Malenka, R.C. & Nestler, E.J. Annu. Rev. Neurosci. 29, 565–598 (2006). 3. Everitt, B.J. & Robbins, T.W. Nat. Neurosci. 8, 1481–1489 (2005). 4. Albin, R.L., Young, A.B. & Penney, J.B. Trends Neurosci. 12, 366–375 (1989). 5. Gerfen, C.R. et al. Science 250, 1429–1432 (1990). 6. Gong, S. et al. Nature 425, 917–925 (2003). 7. Lobo, M.K., Karsten, S.L., Gray, M., Geschwind, D.H. & Yang, X.W. Nat. Neurosci. 9, 443–452 (2006). 8. Shen, W. et al. Nat. Neurosci. 10, 1458–1466 (2007). 9. Kreitzer, A.C. & Malenka, R.C. Nature 445, 643–647 (2007). 10. Bertran–Gonzalez, J. J. Neurosci. 28, 5671–5685 (2008). 11. Buch, T. et al. Nat. Methods 2, 419–426 (2005). 12. Schiffmann, S.N. & Vanderhaeghen, J.J. J. Neurosci. 13, 1080–1087 (1993). 13. Chesselet, M.F. et al. Prog. Brain Res. 99, 143–154 (1993). 14. Sano, H. et al. J. Neurosci. 23, 9078–9088 (2003). 15. Self, D.W. Neuropharmacology 47, 242–255 (2004).

395

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

Pharmacological REM sleep suppression paradoxically improves rather than impairs skill memory

© 2009 Nature America, Inc. All rights reserved.

Bjo¨rn Rasch1,2, Julian Pommer1, Susanne Diekelmann1 & Jan Born1 Rapid eye movement (REM) sleep has been considered important for consolidation of memories, particularly of skills. Contrary to expectations, we found that REM sleep suppression by administration of selective serotonin or norepinephrine re-uptake inhibitors after training did not impair consolidation of skills or word-pairs in healthy men but rather enhanced gains in finger tapping accuracy together with sleep spindles. Our results indicate that REM sleep as a unitary phenomenon is not required for skill-memory consolidation. Sleep is important for consolidating memories1–3 and it has long been hypothesized that REM sleep subserves particular aspects in this process. In animals, learning increases subsequent REM sleep and selective REM sleep deprivation after learning impairs the retention of memories2,4. In humans, REM sleep has been related specifically to consolidation of procedural memories (that is, perceptual and motor skills5,6), although non-REM sleep and the electrophysiological features of non-REM (specifically sleep spindles) have also been consistently implicated in this process7–9 (for reviews, see refs. 1,10). REM sleep is hallmarked by a unique pattern of neurotransmitter activity comprising, in particular, minimum serotoninergic and norepinephrinergic activity11. Selective re-uptake inhibitors of serotonin (SSRI) and norepinephrine (SNRI), which are widely used as antidepressant drugs, enhance the availability of these neurotransmitters in the synaptic cleft and are well known for suppressing REM sleep11,12. Paradoxically, clinical studies in depressed patients have not revealed any impairment in skill performance accompanying prolonged treatment with such antidepressants, despite marked REM sleep suppression13, and, in the ongoing debate about contributions of sleep to memory, this has been repeatedly put forward as major argument against any involvement of REM sleep in memory consolidation14,15. As pointed out by others1, however, out of all of the available studies on this topic, ‘‘none have investigated sleep-dependent tasks, none tested memory after a post-training night of sleep, and none confirmed the degree of REM suppression.’’ Here, we investigated the effects of two antidepressant drugs, the SSRI fluvoxamine and the SNRI reboxetine, on procedural memory consolidation during sleep in healthy

volunteers. We used sleep-dependent tasks (mirror tracing and finger sequence tapping), tasks in which performance gains at delayed retesting are known to depend on post-training sleep, and obtained polysomnographic recordings during post-learning sleep (Fig. 1a, for details see Supplementary Methods online). As expected, the two substances strongly reduced the amount of time spent in REM sleep from 16.8 ± 1.0% during placebo nights to 12.5 ± 1.0% after fluvoxamine (P ¼ 0.002), and reboxetine led to a more complete reduction after administration (2.6 ± 1.1%, P o 0.0001; Fig. 1b). In the nights after reboxetine administration, the amount of time spent in stage 2 sleep, as well as the number and density (spindle count/number of epochs) of fast spindles (413 Hz), was significantly increased (all P o 0.03). SSRI and SNRI administration both increased the amount of time spent awake (P o 0.002) and stage 1 sleep (P ¼ 0.06), whereas no effects were observed for slow-wave sleep (SWS, all P 4 0.1; see Supplementary Results and Supplementary Table 1 online). In spite of the pronounced REM sleep suppression, neither SSRI nor SNRI administration impaired overnight memory consolidation of procedural motor skills (assessed after an additional night of recovery sleep). On the mirror tracing task, significant gains in mean draw time (speed) developed from training to retrieval testing after placebo (13.0 ± 2.3 s, P ¼ 0.001), SSRI (18.1 ± 4.7 s, P ¼ 0.003) and SNRI administration (15.6 ± 5.1 s, P ¼ 0.006). In parallel, the number of deviations from the trace (errors) was reduced at retrieval testing compared with the learning phase in all three conditions (–8.7 ± 1.4, –7.8 ± 1.9 and –10.0 ± 2.1 for placebo, SSRI and SNRI conditions, respectively; all P o 0.001). Analysis of variance revealed no significant differences in overnight changes in speed or accuracy between any of the treatment conditions (all P 4 0.4). On the finger sequence tapping task, overnight (training-to-retrieval) changes in performance profited from pharmacological REM sleep suppression. Gains in the number of correctly tapped sequences (speed) were highly significant after SSRI (3.0 ± 0.5 sequences, P o 0.0001) and SNRI administration (2.3 ± 0.6 sequences, P ¼ 0.001), and were greater on average than those observed across placebo nights (1.6 ± 0.6, P ¼ 0.02), although the effect failed to reach statistical significance (P ¼ 0.14; Fig. 1c). The enhancing effect of the re-uptake inhibitors on overnight gains in finger tapping skill was even more pronounced for performance accuracy. Error rates significantly decreased after SSRI (–1.6 ± 0.8%, P ¼ 0.02) and SNRI administration (–3.2 ± 1.2%, P ¼ 0.01), whereas decreases in the error rate after placebo remained non-significant (–0.2 ± 1.3%, P ¼ 0.7). The improvements in accuracy after SSRI and SNRI were significant in the overall analysis when compared with the effects of placebo (P ¼ 0.01; Fig. 1d) and were most pronounced after SNRI (post hoc comparison between SNRI and placebo, P ¼ 0.03). Notably, the difference in accuracy gains between active treatment and placebo conditions was positively correlated with the change in the number

1University

of Lu¨beck, Department of Neuroendocrinology, Haus 23a, Ratzeburger Allee 160, 23538 Lu¨beck, Germany. 2University of Basel, Department of Molecular Psychology, Missionsstrasse 60/62, 4055 Basel, Switzerland. Correspondence should be addressed to B.R. ([email protected]) or J.B. ([email protected]).

Received 24 July; accepted 9 September; published online 5 October 2008; doi:10.1038/nn.2206

396

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

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

Wake

Sleep Sleep

Re Re

Wake

Sleep

Re

Wake

Sleep

Re

15

***

10 5 0

© 2009 Nature America, Inc. All rights reserved.

2200

0700

2200

d 4 3 2 1 0

0700 h

0.06

0.04

0.02

0.00

e * **

Change in spindle density

SNRI Le Sleep Placebo Le Sleep

Wake

c

*** **

Gain of accuracy (errors per sequence)

SSRI Le Sleep Placebo Le Sleep

20

Gain of speed (sequences per 30 s)

b REM sleep (%)

a

1.0

r = .51, P = 0.002

Placebo SSRI SNRI

0.5 0.0 –0.5 –1.0 0.5 1.0 –0.5 0.0 Change in accuracy gains

Figure 1 Experimental procedure and main results. (a) We measured the effects of fluvoxamine (versus placebo, n ¼ 15) and reboxetine (versus placebo, n ¼ 13) administered orally after learning (Le) on memory consolidation during sleep in two separate experiments, each conducted according to a double-blind within-subject crossover design. Retrieval (Re) was tested 32 h after substance administration. Memory was tested using a finger sequence tapping task (with explicit feedback of performance provided after each block of tapping), mirror tracing and a declarative paired-associate learning task. Written informed consent was obtained from all subjects. (b) REM sleep (in percent of total sleep time) was significantly reduced after SSRI (**P ¼ 0.002) and almost completely suppressed after SNRI administration compared with placebo (***P o 0.0001). (c) Finger tapping speed remained unaffected (P 4 0.14). (d) Overnight gains in tapping accuracy were significantly enhanced in the active treatment conditions (**P ¼ 0.01, versus placebo), particularly after SNRI administration (*P ¼ 0.03, versus placebo). Absolute increases in the number of correctly tapped sequences (per 30-s trial) and absolute reductions in the number of errors (per tapped sequences) at retrieval testing compared with training are indicated as positive gains in speed and accuracy, respectively. Means ± s.e.m. are shown. (e) Changes (with reference to placebo) in gains of finger tapping accuracy after SSRI (gray dots) and SNRI (black dots) administration were correlated with changes in fast spindle density during stage 2 sleep and SWS.

of fast spindles (r ¼ 0.57, P ¼ 0.007) and fast spindle density (r ¼ 0.51, P ¼ 0.002; Fig. 1e) between active treatment versus placebo nights, whereas no such correlation was observed for other sleep parameters (all P 4 0.1). This result is consistent with numerous reports of positive associations between spindle activity and overnight improvements in fine motor skills8,9. It contrasts with our claim, based on previous studies (not examining spindles), of REM sleep facilitating of this form of memory6 but strongly supports the notion of an involvement of spindle activity in motor-skill consolidation during sleep. To test for the specificity of REM sleep suppression on procedural memory consolidation, we taught participants a declarative pairedassociate learning task that is known to depend on SWS rather than REM sleep3. As we expected, retention of word pairs was not affected by SSRI or SNRI administration (–0.6 ± 0.7, –2.0 ± 0.9 and +0.6 ± 0.9 word pairs for placebo, SSRI and SNRI conditions, respectively; all P 4 0.2). For all memory tests, performance during learning was comparable between active treatment and placebo conditions (all P 4 0.1; Supplementary Table 2 online). The lack of an impairing effect of REM sleep suppression on motorskill consolidation cannot be attributed to any confounding influence of the substances during retrieval because retrieval was tested a long time (32 h) after substance administration, including a recovery night, to ensure that the substances had cleared the system by the time of retrieval testing. At learning and retrieval testing, control measures of mood and vigilance (reaction time), as well as blood levels of cortisol or norepinephrine, did not differ between active agent and placebo (all P 4 0.1; Supplementary Table 3 online). However, it could be argued that impairments of skill consolidation as a result of pharmacological REM sleep suppression in the first post-training night were compensated by REM sleep that occurred in the second (recovery) night before retrieval. We therefore tested the effects of reboxetine (versus placebo) in an additional experiment using a shorter 24-h retention interval, omitting the night of recovery sleep. Consistent with our main study, this experiment did not indicate any impairing effect of pharmacological REM sleep blockade on consolidation of both skills or word pairs (all P 4 0.2) but enhanced overnight gains in finger sequence tapping accuracy (P ¼ 0.07) and speed (P ¼ 0.05) after SNRI administration (Supplementary Tables 4–6 online). Our data add to the ongoing debate about the contribution of REM sleep to memory consolidation. Consistent with clinical observations of

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

largely preserved procedural memory function in depressed patients even after long-term treatment with antidepressants and accompanying severe REM sleep suppression13, our findings clearly favor the view that phenotypic REM sleep as measured by polysomnography is not critical for the consolidation of procedural motor-skill memories14,15. Nevertheless, our data cannot be taken to completely exclude any contribution of REM sleep to memory. Unidentified processes that are critical to memory consolidation and normally associated with REM sleep (for example, high cholinergic activity and the expression of plasticity-related early genes) may persist during REM sleep suppression after SSRI or SNRI administration and such processes may be differentially affected by the procedure employed for depriving REM sleep. Thus, our results will help to disentangle the components of natural REM sleep that could mediate putatively beneficial effects on memory consolidation. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank E. Wiege, I. von Lu¨tzau, H. Ruf and A. Otterbein for technical assistance and L. Marshall for helpful discussions. This work was supported by a grant from the Deutsche Forschungsgemeinschaft SFB 654 ‘Plasticity and Sleep’. AUTHOR CONTRIBUTIONS B.R., J.P. and S.D. conducted the experiments and analyzed the data. B.R. and J.B. designed the experiments and wrote the manuscript. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

1. Walker, M.P. & Stickgold, R. Neuron 44, 121–133 (2004). 2. Peigneux, P. et al. Neuroreport 12, A111–A124 (2001). 3. Marshall, L. & Born, J. Trends Cogn. Sci. 11, 442–450 (2007). 4. Smith, C. Neurosci. Biobehav. Rev. 9, 157–168 (1985). 5. Karni, A. et al. Science 265, 679–682 (1994). 6. Born, J. et al. Neuroscientist 12, 410–424 (2006). 7. Robertson, E.M. et al. Curr. Biol. 14, 208–212 (2004). 8. Nishida, M. & Walker, M.P. PLoS ONE 2, e341 (2007). 9. Tamaki, M. et al. Sleep 31, 204–211 (2008). 10. Smith, C. et al. Psychol. Belg. 44, 81–104 (2004). 11. McCarley, R.W. Sleep Med. 8, 302–330 (2007). 12. Mayers, A.G. & Baldwin, D.S. Hum. Psychopharmacol. 20, 533–559 (2005). 13. Amado-Boccara, I. et al. Neurosci. Biobehav. Rev. 19, 479–493 (1995). 14. Siegel, J.M. Science 294, 1058–1063 (2001). 15. Vertes, R.P. Neuron 44, 135–148 (2004).

397

ARTICLES

miR-124 regulates adult neurogenesis in the subventricular zone stem cell niche

© 2009 Nature America, Inc. All rights reserved.

Li-Chun Cheng1, Erika Pastrana1, Masoud Tavazoie2 & Fiona Doetsch1–4 The subventricular zone (SVZ) is the largest neurogenic niche in the adult mammalian brain. We found that the brain-enriched microRNA miR-124 is an important regulator of the temporal progression of adult neurogenesis in mice. Knockdown of endogenous miR-124 maintained purified SVZ stem cells as dividing precursors, whereas ectopic expression led to precocious and increased neuron formation. Furthermore, blocking miR-124 function during regeneration led to hyperplasias, followed by a delayed burst of neurogenesis. We identified the SRY-box transcription factor Sox9 as being a physiological target of miR-124 at the transition from the transit amplifying cell to the neuroblast stage. Sox9 overexpression abolished neuronal differentiation, whereas Sox9 knockdown led to increased neuron formation. Thus miR-124–mediated repression of Sox9 is important for progression along the SVZ stem cell lineage to neurons.

The SVZ is the largest germinal region in the adult mammalian brain and harbors stem cells that generate olfactory bulb interneurons. The neural stem cells in this neurogenic niche are specialized astrocytes (type B cells) that give rise to rapidly dividing transit amplifying cells (type C cells)1. The majority of these cells then generate neuroblasts (type A cells) that migrate along the rostral migratory stream (RMS) and differentiate into granule and periglomerular interneurons in the olfactory bulb1. A small number of oligodendrocytes are also generated by the adult SVZ2,3. Although much has been elucidated about the identity and lineage of SVZ stem cells, the regulatory mechanisms underlying in vivo stem cell self-renewal and differentiation are still largely unknown. MicroRNAs (miRNAs) are small noncoding RNAs that are emerging as important post-transcriptional regulators and have been implicated in developmental and disease processes4,5. miRNAs largely act as repressors of gene expression either by guiding the cleavage of their target mRNAs or by inhibiting their translation4,6. Their ability to potentially regulate large numbers of target genes simultaneously suggests that they may be important sculptors of transcriptional networks. As such, they are attractive candidates for regulating stem cell lineage progression. We have identified several miRNAs that are expressed at different stages of the SVZ stem cell lineage, one of which is miR-124, the most abundant miRNA in the adult brain7. We therefore investigated the role of miR-124 in the adult SVZ neurogenic niche. Previous work has shown that overexpression of miR-124 in HeLa cells shifts their mRNA profile toward a brain-enriched pattern8, whereas blocking miR-124 in cultured neurons leads to the upregulation of non-neuronal transcripts9. On the basis of in vitro overexpression in cell lines and in embryonic stem cells, it has been proposed that

miR-124 mediates neuronal differentiation10,11. This is achieved in part by targeting polypyrimidine tract binding protein 1 (PTBP1), a repressor of neuron-specific splicing11, and small C-terminal domain phosphatase 1 (CTDSP1, also known as SCP1), a component of the repressor element 1–silencing transcription factor (REST) transcription repressor complex12. However, probing the role of miR-124 in vivo has been challenging. Two studies in the developing chick spinal cord that investigated the role of miR-124 had differing findings. One reported that miR-124 had no effect on neuronal differentiation13, whereas the other reported that miR-124 had modest effects on promoting neurogenesis12. Thus, the in vivo role of miR-124 in neurogenesis is still unclear. We found a previously unknown role for miR-124 in regulating the temporal progression of neurogenesis in the adult SVZ. miR-124 was first upregulated at the transition between transit amplifying cells and neuroblasts and was then further upregulated as neuroblasts exit the cell cycle. Blocking miR-124 maintained SVZ cells as dividing precursors, whereas ectopic expression of miR-124 promoted precocious neuronal differentiation. Knockdown of endogenous miR-124 during regeneration led to the formation of hyperplasias and a delay in neuronal regeneration. We identified Dlx2, Jag1 (also known as Jagged-1) and Sox9 as being miR-124 targets. We found that Sox9 is an important physiological target of miR-124 in SVZ neuroblasts, where Sox9 mRNA, but not protein, is present. Sox9 overexpression in SVZ cells abolished the production of neurons. In contrast, Sox9 knockdown led to increased neurogenesis and decreased glial formation. Thus, the protein levels of Sox9 must be downregulated for neuronal differentiation and are tightly controlled at the post-transcriptional level by miR-124 as cells progress along the SVZ lineage.

1Departments of Pathology and Cell Biology, 2Neuroscience, and 3Neurology, 4Center for Motor Neuron Biology and Disease, Columbia University, College of Physician and Surgeons, New York, New York, USA. Correspondence should be addressed to F.D. ([email protected]).

Received 8 December 2008; accepted 13 February 2009; published online 15 March 2009; doi:10.1038/nn.2294

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

399

ARTICLES

a

OB

f

Stem cell astrocyte

Transit amplifying

Mature neuron

Neuroblast

SVZ RMS GFAP BLBP EGFR

b

Ki67 mCD24 DCX TuJ1 NeuN

c

d

e RMS

g

i

h

*

6

* 4 2

th

St

er SV em Z a Tr an ce stro c si t a ll as yte m tro s N pl eu ifi cyt ro yi bl ng es as ce ts S- lls G 2M G 0G 1

0

O

© 2009 Nature America, Inc. All rights reserved.

CC

8

OB miR-124 (a.u.)

SVZ

Figure 1 miR-124 expression in the adult SVZ niche. (a–e) In situ hybridization of miR-124 in the adult mouse brain. A sagittal schema showing the SVZ (red) is depicted in a, which extends along the entire lateral ventricle. The image in b corresponds to the area in the box in a. miR-124 was expressed at low levels in the SVZ (arrow) and RMS (arrowhead) and was upregulated in the olfactory bulb (OB, asterisk). Other labeled cells are differentiated neurons (Supplementary Fig. 1). The miR-124 signal in coronal sections of the boxed areas in the SVZ (c, arrow), RMS (d, arrowhead) and OB (e) is shown. miR-124 was not detected in the corpus callosum (CC, c) or other white-matter fiber tracts. Scale bar represents 100 mm. (f) Schema showing the SVZ lineage and markers expressed at each stage. (g,h) In situ hybridization for miR-124 combined with immunostaining in SVZ coronal sections. The corpus callosum is at the top and the striatum is at the bottom. Right, in situ signal alone. miR-124 was expressed in DCX-positive neuroblasts (red, arrow) but not in BLBP-positive astrocytes (green, white arrowheads) (g). Dashed lines outline the lateral ventricle. miR-124 was expressed in DCX-positive neuroblasts (green) and in mature NeuN-positive neurons (blue, black arrowheads) but was not detected in Ki67-positive, DCX-negative transit amplifying cells (red, white arrowheads) (h). Scale bars represent 20 mm. (i) qRT-PCR for miR-124 on FACS-purified SVZ populations. Data represent mean ± s.e.m. normalized to 5S rRNA from five independent experiments. * P o 0.05, two-tailed paired Student’s t test.

RESULTS miR-124 is expressed by neuroblasts in the adult SVZ niche To determine the spatial and cell type–specific distribution of miR-124 in the adult SVZ neural stem cell niche, we carried out in situ hybridization using digoxigenin (DIG)-labeled RNA probes that were targeted to the mature form of miR-124 (Fig. 1). miR-124 was expressed at low levels in both the SVZ and RMS (Fig. 1a–d) and was greatly upregulated in mature granule and periglomerular neurons in the olfactory bulb (Fig. 1b,e) to levels that are seen in mature neurons throughout the brain (Fig. 1h and Supplementary Fig. 1 online). In contrast, miR-124 was not expressed in oligodendrocytes or astrocytes (data not shown). To distinguish which cell types in the adult SVZ express miR-124, we carried out in situ hybridization in combination with immunostaining for markers expressed at different stages of the SVZ lineage (Fig. 1f). miR-124 was not expressed at the earliest stages of the lineage in brain lipid binding protein (BLBP)-positive SVZ astrocytes or in dividing doublecortin (DCX)-negative transit amplifying cells (Fig. 1g,h), but was expressed in DCX-positive neuroblasts (Fig. 1g,h). Notably, miR-124 was expressed in a similar pattern in germinal zones in the developing forebrain. miR-124 was largely absent in the ventricular zone at embryonic day 11 (E11), E14 and E17, with the exception of a few individual cells at E11 and E14 (Supplementary Fig. 1), which probably correspond to the first appearing neurons14,15. At E14 and E17, miR-124 was highly expressed in the

400

SVZ (Supplementary Fig. 1) and was present at high levels in the cortical plate at E17 (Supplementary Fig. 1). To confirm our in situ hybridization expression results, we used another approach to detect miRNAs in the adult SVZ. We recently developed a simple strategy to prospectively isolate cells at each stage of the SVZ stem cell lineage using fluorescence activated cell sorting (FACS) on the basis of the differential expression of glial fibrillary acidic protein (GFAP), epidermal growth factor receptor (EGFR) and mCD2416 (Fig. 1f). This approach makes use of the fact that activated stem cells express EGFR, as do transit amplifying cells17. By sorting from hGFAP-gfp mice, in which SVZ astrocytes express green fluorescent protein (GFP) under the control of the human GFAP promoter, activated stem cell astrocytes (EGFR positive) can be separated from niche astrocytes (EGFR negative) on the basis of binding to a fluorescently complexed EGF ligand. By combining these two markers with mCD24, which labels neuroblasts and ependymal cells with different intensities, but not astrocytes and transit amplifying cells, SVZ-activated stem cell astrocytes (GFAP+ EGFR+ mCD24–), transit amplifying cells (GFAP– EGFR+ mCD24–) and neuroblasts (GFAP– EGFR– mCD24low) can all be isolated in a single sort16. We therefore carried out quantitative RT-PCR (qRT-PCR) for miR-124 on FACS-purified SVZ cells. As seen by in situ hybridization, qRT-PCR confirmed that miR-124 was highly expressed in neuroblasts, as compared with other SVZ cell types (Fig. 1i). In addition, by using

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES Figure 2 Knockdown of miR-124 in vitro. (a–c) Representative micrographs of colonies from SVZ stem cells cocultured on rat astrocyte monolayers in the presence of penetratin (Pen)-conjugated AS-124 or AS-194. M2/M6 (green) distinguished mouse cells on immunonegative rat cells. More BrdU-positive cells (blue) and fewer neurons (TuJ1, red) were present in AS-124–treated cultures. BF, bright field. Scale bar represents 50 mm. (d) Quantification of total neurons (TuJ1+ BrdU–), dividing cells (BrdU+) and other cells (TuJ1– BrdU–). (e) Colony composition of neurons (TuJ1+ BrdU–), neuroblasts (TuJ1+ BrdU+), dividing cells (BrdU+) and other cells (TuJ1– BrdU–) in individual colonies. AS-124 treatment decreased the proportion of neurons (black) derived from single SVZ stem cells by maintaining them as dividing neuroblasts (dark gray) and transit amplifying cells (light gray). (f) Quantification of neurosphere formation of unsorted and FACS-purified cells. Histogram shows the ratio to untreated culture. (g) Quantification of survival of FACS-purified neuroblasts. All data represent mean ± s.e.m. from at least three independent experiments. * P o 0.05, two-tailed paired Student’s t test.

Untreated

a

AS-124

c

*

30 20 10 0

f

2 1.5

Pen AS-194 AS-124

1 0.5 st am T pl ran ify si in t g St as em tro c cy ell te O th as er tro SV cy Z te

bl a ro eu N

ns or SV ted Z

0

U

80 60

* * *

40 20

0 Untreated AS194 AS124

Other

*

*

Dividing Neuron

100

40

Neuron Dividing

Other Neuroblast

g Percentage of neuron survival

*

50

e

Untreated AS-194 AS-124

60

Percentage of cells per colony

Percentage of total M2M6-positive cells

d

Relative no. of neurosphere

© 2009 Nature America, Inc. All rights reserved.

AS-194

b

70

Pen AS-194 AS-124

50 30 10

the DNA dye Vybrant DyeCycle to separate neuroblasts at different stages of the cell cycle16, we found that miR-124 levels were higher in neuroblasts in G0/G1 phase than in S/G2-M phase (Fig. 1i). Thus, miR124 is upregulated as cells transition from transit amplifying cells to neuroblasts and is further upregulated as neuroblasts exit the cell cycle. miR-124 knockdown retains SVZ cells as dividing precursors Our expression analysis suggested that miR-124 is important for lineage progression during adult neurogenesis. To determine the functional role of miR-124, we carried out knockdown studies on FACS-purified SVZ cell populations. Mouse SVZ-activated stem cell astrocytes were plated at clonal density on neonatal (postnatal day 0, P0) rat cortical astrocytes in a coculture assay18,19. We identified the progeny of plated SVZ cells using mouse-specific M2/M6 antibodies19.

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

To assess proliferation, we added BrdU during the final 24 h of culture. After 5 d in vitro, colonies containing postmitotic neurons (TuJ1+ BrdU–), dividing neuroblasts (BrdU+ TuJ1+), dividing precursors (BrdU+ TuJ1–) and M2/M6-only cells were present (Fig. 2a). The majority of M2/M6-only cells had a typical astrocytic morphology (Fig. 2a); we did not detect O4-positive oligodendrocytes in this coculture assay (data not shown), presumably as a result of the proneurogenic environment created by neonatal astrocytes18. To block miR-124 function, we treated cultures with penetratinconjugated 2¢-O-methyl (2¢OMe) antisense miR-124 (AS-124)20–22, which is efficiently taken up by SVZ cells (Supplementary Fig. 2 online) and knocks down endogenous miR-124 (Supplementary Fig. 2). As a control, we used a 2¢OMe antisense RNA to miR-194 (AS-194) (Supplementary Fig. 2), a microRNA that is enriched in liver and kidney23. miR-124 knockdown resulted in a B30% decrease in the total number of postmitotic neurons and an increase in the total number of dividing cells (1.7-fold relative to untreated culture (P ¼ 0.025) and 1.3-fold relative to AS-194 treated culture (P ¼ 0.012); Fig. 2a–d). Analysis of individual colonies revealed that blocking miR-124 led to an increase of both dividing neuroblasts (TuJ1+ BrdU+) and transit amplifying cells (TuJ1– BrdU+) (Fig. 2e). Blocking miR-124 caused an overall increase in the total number of cells (Supplementary Fig. 2), not only by retaining SVZ cells as dividing precursors in each colony, but also by increasing the total number of colonies formed in the coculture (Supplementary Fig. 2). This could be a result of blocking the progression of precursors along the lineage at the transit amplifying stage. Alternatively, miR-124 knockdown may cause differentiating neuroblasts to revert back to an EGF-responsive state. We therefore used the neurosphere assay to directly address this question, as both activated stem cell astrocytes and transit amplifying cells generate neurospheres, but neuroblasts do not16,17. The number of neurospheres formed from unpurified SVZ cells that were treated with AS-124 increased 1.3-fold and 1.5-fold as compared with penetratin-only and AS-194–treated cultures, respectively (Fig. 2f). To identify which SVZ cell types were being affected by miR-124 knockdown, we compared the neurosphere forming capacity of FACS-sorted populations from each stage of the lineage after culture with AS-124. Only the number of neurospheres formed by transit amplifying cells was significantly increased (1.4-fold relative to penetratin only (P ¼ 0.03) and 1.24-fold relative to AS-194–treated culture (P ¼ 0.04)) by miR-124 knockdown (Fig. 2f). Notably, no neurospheres formed in AS-124–treated purified neuroblasts (Fig. 2f), indicating that AS-124 treatment does not cause neuroblasts to revert back to an EGF-responsive state.

401

ARTICLES Figure 3 Ectopic expression of miR-124 in vitro. (a) Schema shows the retroviral construct for miR-124 egfp PgkP H1P 5′LTR miR-124 overexpression. The mouse mir-124-3 3′dLTR miR-124mt ψ packaging genomic locus (hairpin) with B125-bp flanking signal sequences was cloned into the murine stem cell Other Dividing 12 RV-GFP Neuroblast Neuron 100 virus backbone under the control of the H1 * RV-124mt 10 90 promoter. The expression of enhanced GFP was 100 RV-124 80 * 90 8 driven by a second promoter (Pgk1 promoter) 70 80 6 60 placed in the opposite orientation to the miRNA 70 60 50 expression cassette. (b) Sequence of the mutated 4 * 50 * 40 2 40 miR-124 in the retroviral construct (RV-124mt). 30 30 20 Six nucleotides (bold) of the 5¢ seed region (box) 0 20 10 10 were mutated. (c) qRT-PCR of the miR-124 levels 0 0 that were induced by retroviral infection in Neuron Dividing Other RV-GFP RV-124mt RV-124 neurospheres. Neurospheres transduced by RV124 ectopically expressed miR-124 to the physiological levels that were present in purified neuroblasts (mCD24 population). (d) Quantification of GFP-positive neurons (TuJ1+ BrdU–), dividing cells (BrdU+) and other cells (TuJ1– BrdU–) after miR-124 overexpression. (e) Colony composition of neurons (TuJ1+ BrdU–, black), neuroblasts (TuJ1+ BrdU+, dark gray), dividing cells (BrdU+, light gray) and other cells (TuJ1– BrdU–, white) in individual colonies after RV-124 overexpression. All data represent mean ± s.e.m. from six independent experiments. * P o 0.05, two-tailed paired Student’s t test.

a

b

e

Furthermore, miR-124 knockdown in purified neuroblasts did not affect survival (Fig. 2g) or their capacity to mature and extend neurites when cultured on laminin (Supplementary Fig. 3 online) with continuous AS-124 treatment for 5 d. Thus, the onset of miR-124 expression as SVZ cells transition from transit amplifying cells to neuroblasts is important for SVZ cells to proceed from an actively proliferating state into differentiating neuroblasts in the SVZ stem cell lineage. miR-124 overexpression promotes neuronal differentiation To examine the consequences of ectopically expressing miR-124 at earlier stages in the SVZ lineage, we generated a replication incompetent retrovirus (RV-124) that encodes dual promoters driving the expression of miR-124 and GFP (Fig. 3a). As a control, we mutated six nucleotides in the seed region at the 5¢ end of miR-124 (miR-124mt) (Fig. 3b) to disrupt its interaction with miR-124 targets. Notably, the RV-124 retroviral construct produced physiological levels of miR-124 that were comparable to those expressed by G0/G1 SVZ neuroblasts (Fig. 3c). miR-124 overexpression had the opposite effect of miR-124 knockdown, which maintained SVZ cells as dividing precursors at the

expense of neuron formation. miR-124 overexpression caused a substantial decrease in the overall proportion of dividing cells and an increase in the number of postmitotic neurons (RV-124 to RV-GFP: 1.42-fold, P ¼ 0.02; RV-124 to RV-124mt: 1.39-fold, P ¼ 0.009; Fig. 3d,e). Analysis of individual colonies revealed that RV-124–transduced cells generated smaller colonies than RV-GFP– transduced cells (Supplementary Fig. 4 online). Frequently, the colonies comprised only one or two postmitotic TuJ1-positive neurons, suggesting that miR-124 overexpression caused cell cycle exit. In addition, miR-124 overexpression also caused a significant decrease in astrocytic cells (‘other’ cells) (RV-124 to RV-GFP: 0.57-fold, P ¼ 0.029; RV-124 to RV-124mt: 0.51-fold, P ¼ 0.04; Fig. 3d,e). miR-124 regulates neurogenesis in the SVZ stem cell niche To define the function of miR-124 in vivo, we analyzed the effect of miR-124 knockdown and overexpression on SVZ lineage progression under homeostasis (Fig. 4). We stereotactically injected RV-GFP, RV-124mt or RV-124 into the SVZ of adult mice to infect dividing precursors. To block miR-124 in vivo, we implanted a mini-osmotic pump filled with AS-124 into the lateral ventricle immediately after RV-GFP injection. Whole mounts were dissected 3 d after RV injection

Figure 4 miR-124 overexpression and knockdown Pen AS-124 in vivo. (a–c, e–g) Representative micrographs of 70 * * SVZ whole-mount preparations 3 d after retrovirus 50 infection. GFP-positive dividing cells (arrows) and migratory neuroblasts (in chains, TuJ1+ or DCX+, 30 red) were present after RV-GFP (a), RV-GFP and 10 AS-124 (b), RV-124mt (e) and RV-124 (f) infection. Higher-magnification view of boxes in b and f are shown in c and g, respectively. The lower panels show split channel of Ki67 immunostaining. AS-124 maintained SVZ RV-124mt 90 * precursors as clusters of dividing cells (c, RV-124 arrowheads), whereas RV-124 induced cell cycle 70 exit and neuronal differentiation (g, arrowheads * 50 indicate post-mitotic migrating neurons). Scale 30 bar represents 50 mm. (d,h) Quantification of total * neurons (TuJ1+ Ki67– or DCX+ Ki67–), dividing 10 + – – cells (Ki67 ) and other cells (Ki67 DCX ) derived from infected SVZ precursors after miR-124 knockdown (d) and RV-124 overexpression (h). Data represent mean ± s.e.m. from six whole mounts. * P o 0.05, two-tailed paired Student’s t test. A total of n ¼ 1,260 (RV-GFP with saline), n ¼ 492 (RV-GFP with penetratin only), n ¼ 477 (RV-GFP with AS-124), n ¼ 895 (RV-124) and n ¼ 938 (RV-124mt) retrovirally transduced cells were counted.

b

c

d

e

f

g

h

VOLUME 12

[

NUMBER 4

[

er

th

O

in g

id

iv

er

th

O

D

iv

id

ro n

eu

N

402

in g

Percentage of total infected cells

D

N

eu ro n

GFP TuJ1 Ki67

Percentage of total infected cells

a

GFP DCX Ki67

© 2009 Nature America, Inc. All rights reserved.

SV

Z

st e N m b ce G ll R 0/G RV V-G 1 -1 FP 2 RV 4m -1 t 24

Percentage of total infected cells

Percentage of cells per colony

d

Relative quantity of miR-124

c

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

a

2% AraC 6 d Ki67

Figure 5 miR-124 knockdown delays SVZ regeneration. (a) Timeline of experiments. Ara-C–filled mini-osmotic pumps were implanted for 6 d and replaced by AS-124– or penetratin-filled pumps for an additional 5 or 7 d. (b–q) SVZ whole-mount preparations after 5 (b–i) or 7 (j–q) d of regeneration were immunostained for BLBP (b,f,j,n), Ki67 (c,g,k,o), TuJ1 (d,h,l,p) or DCX (e,i,m,q). At 5 d, more dividing cells and small hyperplasias (g, arrowheads) and very few neuroblasts (h,i) were present in AS-124–treated brains. At 7 d, a large number of TuJ1-positive and DCX-positive neuroblasts suddenly appeared (p,q). Scale bar represents 100 mm.

7d

TuJ1

DCX

c

d

e

f

g

h

i

Pen

b

AS-124

In vivo knockdown of miR-124 resulted in a B30% decrease in postmitotic neurons that was accompanied by a B1.5- fold increase in dividing cells (Fig. 4b–d). Conversely, overexpression of miR-124 by RV-124 resulted in a B1.3fold increase in the proportion of postmitotic neurons and a decrease in dividing precursors (Fig. 4f–h). Labeled neurons were present in the granule and periglomerular cell layers of the olfactory bulb 2 weeks after infection (Supplementary Fig. 5 online), suggesting that miR-124 overexpression does not affect neuronal migration. Consistent with the above in vitro findings, miR124 overexpression also led to a significant decrease in astrocyte-like cells in vivo (RV-124 to RV-124mt, P ¼ 0.046). However, ectopic expression of physiological levels of miR-124 did not appear to impair the production of Olig2-positive oligodendrocyte lineage cells (Supplementary Fig. 5).

j

k

l

m

n

o

p

q

7 d regeneration

AS-124

and the lineage progression of GFP-labeled cells was analyzed. Both GFP-positive postmitotic neurons in migratory chains and clusters of dividing cells were present in the SVZ (Fig. 4a,e).

Blocking endogenous miR-124 delays neuronal regeneration The adult SVZ rapidly regenerates after anti-mitotic treatment (Fig. 5). When the SVZ was depleted of transit amplifying cells and neuroblasts using cytosine-b-D-arabinofuranoside (Ara-C), SVZ stem cells

Figure 6 Sox9 is a miR-124 target. (a) Luciferase Dlx2 transcript 124mt 124rev miR-124 reporter assays for DLX2, JAG1 and SOX9, three 120% miR-124 targets. The histogram shows the ratio 100% of Renilla to firefly luciferase activity normalized * 80% Jag1 transcript * to empty vector–transfected cells (baseline). Data 60% * represent mean ± 2 s.e.m. from six independent 40% Sox9 transcript experiments. * P o 0.01, two-tailed paired 20% 0% Student’s t test. 124rev is a control construct DLX2 JAG1 SOX9 containing the reverse sequence of miR-124. (b) Schema of Dlx2, Jag1 and Sox9 transcripts GFP 124mt 124 and miR-124 target sites (red). Lines underneath depict the region of 3¢ UTR that we cloned into luciferase reporter vectors. (c–f) miR-124 overexpression in neonatal astrocytes. Astrocytes were stained for Sox9 (blue, c,e), Vimentin (red, c,d) and GFP (green, c). Sox9 protein was SOX9 downregulated by miR-124 overexpression (c,e, arrow) compared with untransfected cells (c–e, arrowheads), shown in the corresponding fluorescence intensity plot (f, red is bright, blue is dim). (g–i) miR-124 overexpression in undifferentiated adult SVZ adherent cultures. We stained for Sox9 (red), Nestin (blue) and GFP (green). Sox9 protein was downregulated by miR124 overexpression (i, arrows) but not in controls (g,h, arrowheads). (j–m) Sox9 immunostaining in coronal sections of adult mouse SVZ. Sox9 (red, upper and lower panels) was expressed by GFPpositive SVZ astrocytes in hGFAP-gfp mice (j), by a subset of EGFR-positive transit amplifying cells (k, green) and mCD24-positive ependymal cells (m, green). DCX-positive neuroblasts (l, green, outlined) lacked Sox9 protein. (n–p) In situ hybridization of Sox9 (black signal) combined with immunostaining for DCX (green) and NeuN (red). (o,p) High-magnification images of the boxed region in n showing Sox9 mRNA (arrow, p) in a DCX-positive neuroblast (arrow, o). Scale bars represent 20 mm.

b

a

Luciferase activity

© 2009 Nature America, Inc. All rights reserved.

Pen

5 d regeneration

BLBP

AS-124 5 d

c

d

e

f

g

h

i

Intensity 3,500 2,500 1,500 500 0

j

k

200

400

600

800 x 1,0001,200

l

0

200

600 400 y

800

1,200 1,000

m

n

o

p

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

403

ARTICLES

© 2009 Nature America, Inc. All rights reserved.

Untreated AS-124

Pen

AS-124

Untreated

astrocytes that did not divide during Ara-C a c treatment proliferated and regenerated the network of chains of neuroblasts via transit amplifying cells (Fig. 5a–c)1,24. Differentiating neurons first appear in foci after 4.5–5 d of regeneration (Fig. 5d,e), and they formed migratory chains around 6.5–7 d (Fig. 5l,m). Because regeneration only takes about 1 week, d b it is feasible to block miR-124 throughout regeneration and to directly ask whether miR-124 is required for neuronal differentiation itself or whether it regulates the timing of progression along the SVZ lineage. We infused Ara-C for 6 d to eliminate transit amplifying e cells and neuroblasts. After Ara-C treatment, Figure 7 Effect of miR-124 knockdown on Sox9 we implanted a pump containing either pene- protein levels. (a,b) Immunostaining of TuJ1 tratin alone or AS-124 into the lateral ven- (green), Sox9 (red) and BrdU (blue) in 5 d in vitro cocultures. Right, Sox9 channel. In tricle for 5 or 7 d (Fig. 5a). untreated cultures, BrdU-positive dividing Notably, neuroblast foci were almost com- neuroblasts expressed low or undetectable pletely absent after 5 d of miR-124 knock- levels of Sox9 protein (a). AS-124 treatment down. Very few TuJ1-positive (Fig. 5h) or upregulated SOX9 protein expression in dividing DCX-positive (Fig. 5i) cells were present in neuroblasts (b, arrows). Note that the rat astrocyte monolayer underneath also expressed Sox9 whole mounts of AS-124–treated brains. In (a,b, asterisks). (c–e) Immunostaining for DCX (green) and Sox9 (red) in coronal sections of the adult contrast, the number of Ki67-positive cells mouse brain. Right, Sox9 channel in untreated, penetratin only (pen) and AS-124–infused brain. After 6 d of AS-124 infusion, Sox9 protein was ectopically expressed by migrating DCX-positive neuroblasts in increased twofold (AS-124, 2,852 cells mm–2; the RMS (e, outlined by dashed lines). Scale bars represent 20 mm. –2 versus saline, 1,347 cells mm , P ¼ 0.009, Newman-Keul test; versus penetratin only, 1,686 cells mm–2, P ¼ 0.01, Newman-Keul test). Hyperplasias of dividing cells occurred frequently in the SVZ, encoding an SRY-box transcription factor that has been implicated in suggesting that SVZ precursors failed to differentiate into neuroblasts glial cell specification during development31,32 and stem cell mainand were retained as transit amplifying cells (Fig. 5g). After 7 d of tenance in the intestinal epithelium and hair follicle33–35, was also a regeneration with AS-124 treatment, the proliferating SVZ precursors predicted target. Unlike Dlx2 and Jag1, the role of Sox9 in the adult SVZ differentiated en masse into neuroblasts and formed large disorganized is unknown. We validated these three gene candidates as being miR-124 clusters (Fig. 5p,q) that disrupted the migratory chains present in targets by carrying out luciferase assays (Fig. 6a). The 3¢ UTRs of Dlx2, control brains (Fig. 5l,m). Thus, knocking down endogenous miR-124 Jag1 and Sox9 were cloned into luciferase reporter vectors (Fig. 6b) and delays regeneration, but does not permanently block neuronal differ- transfected with miR-124 into HeLa cells. miR-124 significantly entiation. Together, our loss-of-function and gain-of-function analyses, decreased luciferase activity in all three constructs, with the largest both in vitro and in vivo under homeostasis and during regeneration, effect being observed for Sox9 (P o 0.0001; Fig. 6a). indicate that miR-124 is an important regulator of the timing of Given the glial identity of stem cells in the adult brain, progression SVZ lineage progression at the transition from transit amplifying along the SVZ lineage into neurons requires that both glial genes and cells to neuroblasts. those involved in self-renewal be downregulated. As such, Sox9 is an attractive miR-124 target. miR-124 overexpression markedly reduced Sox9 is a direct target of miR-124 in the SVZ lineage Sox9 protein levels in neonatal cortical astrocytes (Fig. 6c–f) and adult Individual miRNAs can regulate numerous target mRNAs at the SVZ cells cultured as adherent cells with EGF (Fig. 6g–i), both of which post-transcriptional level4. To identify potential physiological targets express high levels of Sox9. To investigate whether Sox9 is an in vivo target of miR-124 in the of miR-124 in the SVZ, we performed a Gene Ontology analysis of computationally predicted targets from TargetScan25. We analyzed adult SVZ, we analyzed the expression patterns of Sox9 protein and over-representation of Gene Ontology terms among miR-124 targets, mRNA by immunohistochemistry (Fig. 6j–m) and in situ hybridizacomparing them to the mouse genome, and found that ‘development’ tion (Fig. 6n–p). Sox9 protein was present in ependymal cells and was and ‘cell differentiation’ were significantly enriched (Supplemen- expressed by all SVZ astrocytes and approximately 30% of transit tary Table 1 online). We also compared miR-124 targets with amplifying cells, but not by neuroblasts (Fig. 6j–m). Notably, although those of heart-enriched miR-1 (ref. 26), B cell–enriched miR-181 neuroblasts did not express Sox9 protein, they did have detectable levels (ref. 27), miR-194 (ref. 23) and another brain-enriched miRNA, of Sox9 mRNA (Fig. 6n–p), suggesting that Sox9 is post-transcriptionmiR-9 (ref. 28). Among miR-124 targets, cell differentiation genes ally regulated by miR-124 in vivo. Indeed, knockdown of endogenous were consistently highlighted across all comparisons (Supplementary miR-124 with AS-124 resulted in the upregulation of Sox9 protein in dividing neuroblasts both in vitro (Fig. 7a,b) and in vivo in migratory Table 2 online). Among the predicted cell differentiation targets were several that are chains (Fig. 7c–e). Thus, miR-124 regulates endogenous Sox9 expresknown to be expressed in the SVZ and are involved in the progression sion in the adult SVZ. Sox9 protein was also expressed in neural stem cells in the ventricular of the SVZ stem cell lineage. These included Dlx2, a transcription factor that is involved in interneuron formation17,29, and the Notch ligand zone of the developing brain (Supplementary Fig. 6 online). Notably, Jag1, which is important for self-renewal in the SVZ30. Notably, Sox9, the onset of Sox9 in the forebrain paralleled that of miR-124, but they

404

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES Figure 8 Sox9 knockdown induces neuronal mCherry CMVP 3′dLTR 5′LTR Sox9 orf differentiation. (a) Schema of the Sox9 retroviral ψ IRES construct (RV-Sox9). Sox9 open reading frame 124 124mt Sox9 124 + Sox9 shRan shSox9-C (orf) and the reporter gene (mCherry) are under control of the CMV promoter and are transcribed as bicistronic transcripts. The internal ribosomal entry site (IRES) allows the expression of mCherry. (b–g) Representative micrographs of retrovirally transduced SVZ adherent cultures after 7 d of differentiation following EGF withdrawal. We immunostained for GFAP (blue), TuJ1 (red) and reporter genes (GFP or mCherry, green; except for e, mCherry in red). RV-124 induced neuron production at the expense of astrocyte formation (b), as compared with RV-124mt controls (c). In contrast, Sox9 overexpression eliminated 100 100 Astrocyte neurogenesis and maintains infected cells as Neuron 80 80 GFAP-positive astrocytes (d). Sox9 that lacked a 3¢ UTR dominated the gliogenesis phenotype and 60 60 was not rescued by miR-124 overexpression (e). ** Sox9 knockdown by RV-shSox9-C (g) increased 40 40 * neuronal differentiation compared with controls (f) 20 20 and resembled the phenotype of RV-124– transduced cells. Scale bar represents 20 mm. * * 0 0 (h,i) Quantification of astrocytes (GFAP+) and Uninfected 124mt 124 Sox9 Sox9 + 124 shRan shSox9A shSox9-C + neurons (TuJ1 ) present after retroviral infection. Data represent mean ± s.e.m. from at least three independent experiments. * P o 0.01, two-tailed paired Student’s t test. The percentages of both astrocytes and neurons in RV-124, RV-Sox9 and RV-124/RV-Sox9 were statistically significant compared with the uninfected and RV-124mt controls (h). The percentage of neurons in RV-shSox9-C was significantly higher compared with RV-shRan and RV-shSox9-A (i).

a

c

e

f

GFP TuJ1 GFAP

g

i

Percentage of total infected cells

h

© 2009 Nature America, Inc. All rights reserved.

d

GFP mCherry GFAP

GFP or mCherry TuJ1 GFAP

b

were spatially exclusive (Supplementary Fig. 6). Therefore, it is very likely that Sox9 is also under post-transcriptional control by miR-124 during development. Sox9 downregulation is required for neurogenesis The in vivo expression pattern of Sox9 showed that it is downregulated during neurogenesis. To evaluate the effect of Sox9 overexpression, we generated a replication-incompetent retrovirus encoding only the open reading frame of Sox9 (RV-Sox9) and lacking the 3¢ UTR, which has two miR-124 target sites (Fig. 8). miR-124 overexpression in SVZ adherent cultures led to a twofold increase in the number of neurons and a onethird reduction in the number of GFAP-positive astrocytes that were generated (Fig. 8a–c,h). In contrast, Sox9 overexpression maintained SVZ cells as GFAP-positive astrocytes and completely eliminated neuronal production (Fig. 8d,h). Simultaneous miR-124 overexpression did not rescue neuronal production in Sox9-overexpressing cells (Fig. 8e,h). Neither miR-124 nor Sox9 overexpression affected oligodendrocyte formation. However, this may be a result of the low number of oligodendrocytes generated (data not shown). Conversely, to investigate the effect of Sox9 knockdown on differentiation, we cloned two short hairpin (sh)RNAs, shSox9-C and shSox9-A, targeting Sox9 transcripts into a murine stem cell virus retroviral vector. As a control, we used a shRNA encoding a scrambled sequence (shRan). shSox9-C was more potent than shSox9-A at knocking down endogenous Sox9 and caused a fivefold decrease in Sox9 protein levels (Supplementary Fig. 7 online). In contrast, shSox9A only mildly decreased Sox9 levels (Supplementary Fig. 7). Knockdown of Sox9 by retroviral infection with RV-shSox9-C significantly increased (shSox9C to shRan, P ¼ 0.003; shSox9C to shSox9A, P ¼ 0.03) the proportion of neurons that were generated, mimicking the effect induced by RV-124 (Fig. 8b,f,g,i). These results suggest that Sox9 is an important target of miR-124 in neuronal differentiation. No significant effect (P ¼ 0.6) was observed with either RV-Sox9-A or RV-Sox9-C on the production of GFAP-positive

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

astrocytes (Fig. 8i), suggesting that low levels of Sox9 expression are sufficient for maintaining astrocyte differentiation. Thus, these functional experiments indicate that Sox9 transcripts interact with miR-124 at the transition of SVZ transit amplifying cells to neuroblasts and that downregulation of Sox9 permits neuronal differentiation. Altogether, our results suggest that Sox9 protein levels need to be tightly controlled in the SVZ lineage and that posttranscriptional regulation of Sox9 by miR-124 is an important step in the glial-to-neuron transition along the SVZ stem cell lineage. DISCUSSION We provide here the first evidence, to the best of our knowledge, that miRNAs are important regulators of adult neurogenesis in vivo in the SVZ stem cell niche. We found that the brain-enriched miRNA miR124 was differentially expressed in neuroblasts in the stem cell lineage, with its expression increasing as the neuroblasts exit the cell cycle. Under both homeostasis and regeneration, miR-124 functionally regulated the timing of the transition between transit amplifying cells and neuroblasts by promoting neuronal differentiation and cell cycle exit and repressing genes that are important for stem cell selfrenewal and glial differentiation (Supplementary Fig. 8 online). Moreover, post-transcriptional regulation of Sox9 by miR-124 is a critical step that allows glial stem cells to transition into neurons in adult neurogenesis. miR-124 was one of the first miRNAs that was identified in mammals as a result of its abundant expression in the mature brain7. Its temporal expression profile suggested that it may be involved in both brain development and in mature neuronal function10,36. As previously reported28, we observed miR-124 expression in mature neurons, but not in glial cells. The first evidence that miR-124 may be involved in neuronal differentiation came from P19 carcinoma cells and embryonic stem cells10,11. miR-124 may act in part by antagonizing the REST pathway, which represses neuronal gene expression9,12, and switching on neuron-specific splicing by targeting PTBP1, a repressor

405

© 2009 Nature America, Inc. All rights reserved.

ARTICLES of pre-mRNA splicing in non-neuronal cells11. However, the in vivo function of miR-124 is unclear. One study in the developing chick spinal cord reported that blocking miR-124 had no effect on neurogenesis, whereas another suggested that miR-124 had a modest role in neuronal differentiation and cell cycle exit12,13. The discrepancy between these results may be technical, as transient transfection of antisense oligonucleotides may not be sufficient to knockdown miR-124 levels because progenitor cells proliferate frequently and may dilute their efficacy. Furthermore, electroporation was used to introduce high levels of either miR-124 duplexes or precursors, which may overwhelm the capacity of the endogenous miRNA processing machinery to produce the normal complement of miRNAs and could possibly cause off-target effects. We identified at single-cell resolution the cells that expressed miR124 in vivo in the adult SVZ and quantitatively analyzed miR-124 expression levels in FACS-purified populations. These approaches revealed that miR-124 acts at the transition from transit amplifying cells to neuroblasts. By manipulating the physiological levels of miR124 in purified SVZ cells in vitro and in vivo, both under homeostasis and regeneration, we were able to define the function of miR-124 in adult neurogenesis. We found, consistent with a previous study in the developing spinal cord12, that miR-124 promotes neuronal differentiation and cell cycle exit in the adult SVZ. Our observations further suggest that miR-124 is not involved in neuronal versus glial cell fate decisions. Loss of miR-124 in SVZ cells is not sufficient to cause gliogenesis, probably because neuron/glial fate decisions are made by other factors before endogenous miR-124 is expressed in vivo. Notably, we uncovered a previously unknown role for miR-124, namely the regulation of the timing of neuronal differentiation along the SVZ stem cell lineage. This is most clearly seen during regeneration, where blocking miR-124 led to a temporary arrest at the transit amplifying stage and the formation of hyperplasias, followed by a delayed burst of neuronal production. As such, we propose that miR-124 is important for the temporal regulation of progression along the SVZ stem cell lineage. Stem cells in the adult SVZ are a subset of astrocytes. Progression along the SVZ lineage into neurons therefore requires the downregulation of genes involved in glial function and self-renewal. Consistent with this, miR-124 overexpression reduced the number of astrocytes formed. We focused on Sox9 as a physiological target of miR-124. Sox9 regulates glial fate specification and controls the transcription of glial gene networks in the CNS31,32, as well as self-renewal in other stem cell niches. We validated Sox9 to be a bona fide miR-124 target and found that Sox9 protein levels can be directly controlled by miR-124 through its 3¢ UTR. Notably, Sox9 transcripts were present in neuroblasts, and coexisted with miR-124 both spatially and temporally. Furthermore, when we blocked endogenous miR-124, Sox9 protein was upregulated in neuroblasts, which are still in cell cycle. Thus, Sox9 levels are regulated in neuroblasts post-transcriptionally by miR-124. It will be interesting to determine whether Sox9 directly regulates the cell cycle. Our data further suggest that the downregulation of Sox9 is necessary for neuronal differentiation. In vivo, Sox9 levels decreased as cells progressed along the SVZ lineage and Sox9 knockdown in SVZ adherent cultures led to an increase in the number of neurons that were formed. Because Sox9 is expressed in multiple SVZ cell types, including stem cells and transit amplifying cells, conditional knockout of Sox9 in a cell type–specific manner will be required to dissect its role in vivo in stem cell self-renewal, proliferation and gliogenesis. Two models of miRNA repression of target genes have been proposed: complete elimination of protein expression and balancing of target protein levels37. Most reported miR-124 targets, including

406

Ctdsp1 (ref. 12), Ptbp1 (ref. 11) and Lamc1 (ref. 13), as well as Sox9 and Jag1, two targets that we describe here, show reciprocal protein expression with miR-124, suggesting that miR-124 keeps these genes silenced. In contrast, miR-124 and Dlx-2 protein are simultaneously expressed in SVZ neuroblasts, with Dlx-2 levels being lower in neuroblasts than in transit amplifying cells17. This suggests that miR124 balances Dlx-2 translation to a defined level as cells differentiate. The expression of miR-124 is thought to be regulated by REST9. This zinc-finger repressor negatively regulates many neuronal genes in stem cells, progenitors and non-neuronal cell types38. Potential REST binding sites are located near all three mir-124 loci9 and miR-124 upregulation is probably a result of de-repression of the RESTrepressor complex during neuronal differentiation. Indeed, miR-124 expression is upregulated in Rest+/– embryonic stem cell lines39. One component of the REST repressor complex, SCP1, is among miR-124’s targets and is expressed in the ventricular zone germinal region in the developing spinal cord12. By targeting SCP1, miR-124 may reinforce its own expression in neurons by repressing REST activity. It is important to note that miR-124 continues to be expressed by mature neurons throughout the brain, suggesting that miR-124 probably has other physiological functions in mature neurons, in addition to its functional role in neurogenesis. Here, we show that miRNAs regulate adult neural stem cell lineages in vivo. It will be important to define the full complement of miRNAs that are expressed at each stage in the SVZ lineage to uncover their regulatory networks and to compare them with those involved in other neuronal systems40. miR-124 is important in maintaining SVZ homeostasis, as it regulates the number of progenitors and the timing of neuronal differentiation. Understanding the pleiotropic effects mediated by miR-124 may uncover new approaches toward stimulating neurogenesis in non-neurogenic brain areas and toward inhibiting the progression of brain tumors41. METHODS Animals. All animal care was in accordance with institutional guidelines and was approved by the Institutional Animal Care and Use Committee of Columbia University. Details for immunostaining, FACS, cell culture, retroviral constructs, luciferase reporter assays and western blots are described in the Supplementary Methods online. In situ hybridization. miRNA probes were generated with the mirVana probe construction kit (Ambion). For detecting sox9, we used I.M.A.G.E. clone ID 3666566. In vitro transcription of DIG-labeled RNA probes was performed according to manufacturer’s protocols. Embryonic brains from stages E11, E14 and E17 and 2-month-old CD-1 mice (Charles River Lab) were perfused with 3% paraformaldehyde (PFA) and 0.1% glutaraldehyde (for miRNA, wt/vol) or with 3% PFA alone (for Sox9). Brains were immersed in 30% sucrose (wt/vol), embedded in Tissue-Tek optimal cutting temperature compound and sectioned on the cryostat. We acetylated 12-mm coronal or 25-mm sagittal sections and hybridized them with DIG-labeled RNA probes overnight at 42 1C (miRNA) or at 68 1C (Sox9). Sections were washed extensively and treated with RNase after hybridization. Signals were revealed using alkaline phosphatase–conjugated antibodies to DIG (1:4,000, Roche) and nitroblue tetrazolium chloride/5-bromo-4-chloro-3-indolylphosphate, toluidine salt. FACS. The different cell types of the SVZ lineage were purified as described previously16. See Supplementary Methods for more information. RNA extraction and qRT-PCR. For qRT-PCR of miRNA, the small RNA fraction was isolated with the mirVana miRNA isolation kit (Ambion) and 250 pg were used for each reaction. qRT-PCR was performed in triplicates using SYBR green on a Stratagene MX3000 thermocycler (Stratagene). The threshold cycle (Ct) value was determined using the automatic baseline

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES determination feature on Stratagene MX3000 and the relative expression of miR-124 was normalized to 5S rRNA by 2–DCt. Penetratin 1–mediated delivery. Antisense 2¢Ome RNA oligonucleotides containing a 5¢ thiol group (Dharmacon) were conjugated to penetratin-1 (Qbiogene), as described previously22. Cocultures. Cocultures were based on previously described protocols19,42 and adapted to FACS-purified SVZ populations. Detailed procedures are described in the Supplementary Methods. Neurosphere assays. Neurosphere assays were carried out using FACS-purified SVZ populations as described previously17.

© 2009 Nature America, Inc. All rights reserved.

Neuron survival assays. FACS-purified SVZ neuroblasts (CD24low population) were plated at 20,000 cells per cm2 on 16-well LabTek chamber slides coated with 0.5 mg ml–1 poly-D-Lysine (Sigma) and 0.01 mg ml–1 Laminin (Sigma). Cells were cultured for 5 d in NB/B27 medium with 200 nM penetratin1– conjugated antisense 2¢OMe-RNAs. Cell Tracker (1 mM, Molecular Probes) was added to the cells for 30 min at 37 1C before fixation and staining. Adherent cultures. Adult SVZ cells were dissociated and plated on 16-well LabTek chamber slides coated with 0.01 mg ml–1 Laminin at 10,000 cells per cm2. Cells were grown in DMEM-F12 supplemented with N2, 2 mM glutamine, 20 mg ml–1 insulin, 15 mM HEPES and 20 ng ml–1 EGF. Animal surgery. Retroviruses (0.2 ml with 8 mg/ml polybrene in solution) were stereotaxically injected into the two hemispheres of anaesthetized 2-month-old CD-1 mice (coordinates relative to bregma: [0, 1.4, 1.6], [0.5, 1.1, 1.7] and [1, 1, 2.3] for anterior, lateral and depth, respectively). After 3 d, mice were perfused with saline and SVZ whole mounts were dissected, fixed and processed for immunostaining. To block miR-124 in vivo, we delivered 4 mM of penetratinconjugated 2¢OMe-RNAs in 100 ml saline by micro-osmotic pump (1007D, Durect) into the right ventricle (0, 1, 2.6 for anterior, lateral and depth, respectively) directly after retroviral injection. Only cells in the right SVZ were quantified in the miR-124 knockdown experiment. For the regeneration assay, micro-osmotic pumps filled with 2% Ara-C (Sigma) were implanted onto the surface of the brain, as described previously24. After 6 d of Ara-C infusion, the implant was replaced with a second pump filled with penetratin-conjugated 2¢OMe-RNAs for an additional 5 or 7 d. Quantification and statistical analysis. Images were taken under the 20
objective of the inverted Axiovert200 microscope (Zeiss). Cells were quantified blind using the ImageJ software. A minimum of 20 randomly chosen fields were quantified for the neuroblast survival and adherent culture assays. In the coculture experiments, we counted all of the M2M6-positive or GFP-positive cells that were present in the wells. All experiments were done a minimum of three times. Statistical comparison of data sets was performed by two-tailed paired Student’s t test unless otherwise noted. Computational analysis of miR-124 targets. The Functional Annotation Clustering and the overrepresentation statistics of the predicted miR-124 targets (TargetScan) was performed by DAVID-EASE software (http://david.abcc. ncifcrf.gov/ease/ease.jsp). Potential miR-124 targets with the highest EASE scores (Cell Differentiation) were further confirmed by their expression pattern in Allen Brain Atlas (http://mouse.brain-map.org/welcome.do). Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank K. Gordon for assistance with FACS at the Herbert Irving Comprehensive Cancer Center at Columbia University. We thank members of the Doetsch and Wichterle laboratories for critical reading of the manuscript. We are grateful to C. Troy for advice on penetratin 1–mediated delivery, N. Heintz for antibody to BLBP and R. Tsien for pCMV-Cherry plasmid. The nestin, M2 and M6 antibodies developed by S. Hockfield and C. Lagenaur, respectively, were obtained from the Developmental Studies Hybridoma Bank developed under the auspices of the National Institute of Child Health and Human Development and maintained by the University of Iowa. This work was supported by a US National Institutes of Health National Institute of Neurological Disorders and Stroke grant to F.D. E.P. was supported by a US National Institutes of Health

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

training grant and a grant from the Spanish Ministerio de Educacion y Ciencia. M.T. was supported by a Medical Scientist Training Program fellowship from the US National Institutes of Health, F.D. is a Packard Foundation Fellow and an Irma T. Hirschl Fellow, and this work was partially supported by the Jerry and Emily Spiegel Laboratory for Cell Replacement Therapies and the Anne and Bernard Spitzer Fund for Cell Replacement Therapy. We dedicate this article to the memory of Emily Spiegel. AUTHOR CONTRIBUTIONS L.-C.C. and F.D. initiated the project. L.-C.C. generated the pSUPERretro (pSR)GFP, pSR-124mt and pSR-shRNA constructs. L.-C.C. and M.T. generated the pSR-124 construct. M.T. and E.P. generated the Sox9 overexpression construct. L.-C.C. performed the in situ hybridization, qRT-PCR, neurosphere assays, neuron survival assays, Sox9 expression assay, retroviral production, penetratin conjugation, in vivo delivery and Ara-C experiments. L.-C.C. and E.P. carried out the FACS sorting, cocultures, adherent cultures and in vivo injections. E.P. performed the Gene Ontology studies. M.T. carried out the luciferase assays and Sox9 immunohistochemistry. F.D. was involved in the study design, data collection, quantification and data analysis. The manuscript was written by F.D., L.-C.C. and E.P. All authors performed data quantification, discussed the results and commented on the manuscript. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/

1. Doetsch, F., Caille, I., Lim, D.A., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. Subventricular zone astrocytes are neural stem cells in the adult mammalian brain. Cell 97, 703–716 (1999). 2. Menn, B. et al. Origin of oligodendrocytes in the subventricular zone of the adult brain. J. Neurosci. 26, 7907–7918 (2006). 3. Lachapelle, F., Avellana-Adalid, V., Nait-Oumesmar, B. & Baron-Van Evercooren, A. Fibroblast growth factor-2 (FGF-2) and platelet-derived growth factor AB (PDGF AB) promote adult SVZ-derived oligodendrogenesis in vivo. Mol. Cell. Neurosci. 20, 390–403 (2002). 4. Bartel, D.P. MicroRNAs: genomics, biogenesis, mechanism and function. Cell 116, 281–297 (2004). 5. Stefani, G. & Slack, F.J. Small non-coding RNAs in animal development. Nat. Rev. Mol. Cell Biol. 9, 219–230 (2008). 6. Flynt, A.S. & Lai, E.C. Biological principles of microRNA-mediated regulation: shared themes amid diversity. Nat. Rev. Genet. 9, 831–842 (2008). 7. Lagos-Quintana, M. et al. Identification of tissue-specific microRNAs from mouse. Curr. Biol. 12, 735–739 (2002). 8. Lim, L.P. et al. Microarray analysis shows that some microRNAs downregulate large numbers of target mRNAs. Nature 433, 769–773 (2005). 9. Conaco, C., Otto, S., Han, J.J. & Mandel, G. Reciprocal actions of REST and a microRNA promote neuronal identity. Proc. Natl. Acad. Sci. USA 103, 2422–2427 (2006). 10. Krichevsky, A.M., Sonntag, K.C., Isacson, O. & Kosik, K.S. Specific microRNAs modulate embryonic stem cell-derived neurogenesis. Stem Cells 24, 857–864 (2006). 11. Makeyev, E.V., Zhang, J., Carrasco, M.A. & Maniatis, T. The microRNA miR-124 promotes neuronal differentiation by triggering brain-specific alternative pre-mRNA splicing. Mol. Cell 27, 435–448 (2007). 12. Visvanathan, J., Lee, S., Lee, B., Lee, J.W. & Lee, S.K. The microRNA miR-124 antagonizes the anti-neural REST/SCP1 pathway during embryonic CNS development. Genes Dev. 21, 744–749 (2007). 13. Cao, X., Pfaff, S.L. & Gage, F.H. A functional study of miR-124 in the developing neural tube. Genes Dev. 21, 531–536 (2007). 14. Noctor, S.C., Flint, A.C., Weissman, T.A., Dammerman, R.S. & Kriegstein, A.R. Neurons derived from radial glial cells establish radial units in neocortex. Nature 409, 714–720 (2001). 15. Haubensak, W., Attardo, A., Denk, W. & Huttner, W.B. Neurons arise in the basal neuroepithelium of the early mammalian telencephalon: a major site of neurogenesis. Proc. Natl. Acad. Sci. USA 101, 3196–3201 (2004). 16. Pastrana, E., Cheng, L.-C. & Doetsch, F. Simultaneous prospective purification of adult subventricular zone neural stem cells and their progeny. Proc. Natl. Acad. Sci. USA (in press). 17. Doetsch, F., Petreanu, L., Caille, I., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. EGF converts transit-amplifying neurogenic precursors in the adult brain into multipotent stem cells. Neuron 36, 1021–1034 (2002). 18. Lim, D.A. & Alvarez-Buylla, A. Interaction between astrocytes and adult subventricular zone precursors stimulates neurogenesis. Proc. Natl. Acad. Sci. USA 96, 7526–7531 (1999). 19. Malatesta, P., Hartfuss, E. & Gotz, M. Isolation of radial glial cells by fluorescentactivated cell sorting reveals a neuronal lineage. Development 127, 5253–5263 (2000). 20. Meister, G., Landthaler, M., Dorsett, Y. & Tuschl, T. Sequence-specific inhibition of microRNA- and siRNA-induced RNA silencing. RNA 10, 544–550 (2004). 21. Hutvagner, G., Simard, M.J., Mello, C.C. & Zamore, P.D. Sequence-specific inhibition of small RNA function. PLoS Biol. 2, E98 (2004).

407

ARTICLES 33. Nowak, J.A., Polak, L., Pasolli, H.A. & Fuchs, E. Hair follicle stem cells are specified and function in early skin morphogenesis. Cell Stem Cell. 3, 33–43 (2008). 34. Bastide, P. et al. Sox9 regulates cell proliferation and is required for Paneth cell differentiation in the intestinal epithelium. J. Cell Biol. 178, 635–648 (2007). 35. Vidal, V.P. et al. Sox9 is essential for outer root sheath differentiation and the formation of the hair stem cell compartment. Curr. Biol. 15, 1340–1351 (2005). 36. Krichevsky, A.M., King, K.S., Donahue, C.P., Khrapko, K. & Kosik, K.S. A microRNA array reveals extensive regulation of microRNAs during brain development. RNA 9, 1274–1281 (2003). 37. Bartel, D.P. & Chen, C.Z. Micromanagers of gene expression: the potentially widespread influence of metazoan microRNAs. Nat. Rev. Genet. 5, 396–400 (2004). 38. Ballas, N., Grunseich, C., Lu, D.D., Speh, J.C. & Mandel, G. REST and its corepressors mediate plasticity of neuronal gene chromatin throughout neurogenesis. Cell 121, 645–657 (2005). 39. Singh, S.K., Kagalwala, M.N., Parker-Thornburg, J., Adams, H. & Majumder, S. REST maintains self-renewal and pluripotency of embryonic stem cells. Nature 453, 223–227 (2008). 40. Choi, P.S. et al. Members of the miRNA-200 family regulate olfactory neurogenesis. Neuron 57, 41–55 (2008). 41. Silber, J. et al. miR-124 and miR-137 inhibit proliferation of glioblastoma multiforme cells and induce differentiation of brain tumor stem cells. BMC Med. 6, 14 (2008). 42. Lim, D.A. et al. Noggin antagonizes BMP signaling to create a niche for adult neurogenesis. Neuron 28, 713–726 (2000).

© 2009 Nature America, Inc. All rights reserved.

22. Davidson, T.J. et al. Highly efficient small interfering RNA delivery to primary mammalian neurons induces MicroRNA-like effects before mRNA degradation. J. Neurosci. 24, 10040–10046 (2004). 23. Sempere, L.F. et al. Expression profiling of mammalian microRNAs uncovers a subset of brain-expressed microRNAs with possible roles in murine and human neuronal differentiation. Genome Biol. 5, R13 (2004). 24. Doetsch, F., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. Regeneration of a germinal layer in the adult mammalian brain. Proc. Natl. Acad. Sci. USA 96, 11619–11624 (1999). 25. Lewis, B.P., Shih, I.H., Jones-Rhoades, M.W., Bartel, D.P. & Burge, C.B. Prediction of mammalian microRNA targets. Cell 115, 787–798 (2003). 26. Zhao, Y., Samal, E. & Srivastava, D. Serum response factor regulates a muscle-specific microRNA that targets Hand2 during cardiogenesis. Nature 436, 214–220 (2005). 27. Chen, C.Z., Li, L., Lodish, H.F. & Bartel, D.P. MicroRNAs modulate hematopoietic lineage differentiation. Science 303, 83–86 (2004). 28. Smirnova, L. et al. Regulation of miRNA expression during neural cell specification. Eur. J. Neurosci. 21, 1469–1477 (2005). 29. Panganiban, G. & Rubenstein, J.L. Developmental functions of the Distal-less/Dlx homeobox genes. Development 129, 4371–4386 (2002). 30. Nyfeler, Y. et al. Jagged1 signals in the postnatal subventricular zone are required for neural stem cell self-renewal. EMBO J. 24, 3504–3515 (2005). 31. Finzsch, M., Stolt, C.C., Lommes, P. & Wegner, M. Sox9 and Sox10 influence survival and migration of oligodendrocyte precursors in the spinal cord by regulating PDGF receptor a expression. Development 135, 637–646 (2008). 32. Stolt, C.C. et al. The Sox9 transcription factor determines glial fate choice in the developing spinal cord. Genes Dev. 17, 1677–1689 (2003).

408

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

Proteoglycan interactions with Sonic Hedgehog specify mitogenic responses

© 2009 Nature America, Inc. All rights reserved.

Jennifer A Chan1–3,5,6, Srividya Balasubramanian1,2,6, Rochelle M Witt1,2,6, Kellie J Nazemi1,2,5, Yoojin Choi1,2,5, Maria F Pazyra-Murphy1,2, Carolyn O Walsh1,2, Margaret Thompson4 & Rosalind A Segal1,2 Sonic Hedgehog (Shh) has dual roles in vertebrate development, promoting progenitor cell proliferation and inducing tissue patterning. We found that the mitogenic and patterning functions of Shh can be uncoupled from one another. Using a genetic approach to selectively inhibit Shh-proteoglycan interactions in a mouse model, we found that binding of Shh to proteoglycans was required for proliferation of neural stem/precursor cells, but not for tissue patterning. Shh-proteoglycan interactions regulated both spatial and temporal features of Shh signaling. Proteoglycans localized Shh to specialized mitogenic niches and also acted at the single-cell level to regulate the duration of Shh signaling, thereby promoting a gene expression program that is important for cell division. Because activation of the Shh pathway is a feature of diverse human cancers, selective stimulation of proliferation by Shh-proteoglycan interactions may also figure prominently in neoplastic growth.

Development of complex tissues requires concomitant growth and cell fate specification. One mechanism for achieving spatial and temporal coordination of size and form is to utilize the same signaling molecules for both processes. The morphogen Hedgehog and its mammalian counterparts, Sonic, Desert and Indian Hedgehog, are critical in the growth and patterning of developing embryos1. One model for morphogen activity postulates that Hedgehog proteins disperse from a localized source and form a gradient that patterns fields of responsive cells2. Controlled ligand distribution may also localize stem cell proliferation to specialized niches3,4. By influencing patterning and proliferation, Hedgehog proteins could coordinate form and size to allow scaling of developing organisms. We asked whether proteoglycans differentially regulate Shhdependent proliferation and patterning. Genetic evidence in Drosophila indicates that proteoglycans are required for Hedgehog dispersal and gradient formation5,6. Biochemical evidence indicates that proteoglycans bind to Hedgehog and the receptor component Ihog to affect Hedgehog activity7,8. In Drosophila, heparan sulfate proteoglycans (HSPGs) are critical for Hedgehog functions and the core protein components are glypicans (glycosylphosphatidylinositol (GPI)-linked proteins)6,9–12. Dally-like glypicans are required for Hedgehog and Wingless dispersal and responses, modulating the signaling of both ligands13. However, the manners in which HSPGs affect Hedgehog signaling and biological responses are not yet understood. Proteoglycans have also been implicated in Hedgehog pathway signaling in mammalian systems11,12,14. Loss of the glypican GPC3

causes an overall growth increase, reflecting changes in Shh and/or insulin-like growth factor signaling11, whereas mutations in HSPGsynthesizing enzymes cause marked defects that may reflect changes in fibroblast growth factor, Wnt and/or Shh responses15,16. To investigate the physiologic role of Shh-proteoglycan interactions, we took the alternative approach of mutating Shh itself. Shh contains an N-terminal Cardin-Weintraub motif that mediates Shh-proteoglycan interactions (KRRHPKK). Mutations in this motif (R34A/K38A, designated ShhAla) reduce high-affinity Shh-proteoglycan interactions without altering Shh’s affinity for its receptor Patched (Ptc)14, allowing us to use this mutation to investigate Shh-HSPG interactions without confounding effects resulting from other growth factors that bind proteoglycans. To identify organismal level responses that require Shh-proteoglycan interactions, we generated mice in which wild-type Shh is replaced with ShhAla. Although Shh is needed for growth and patterning of diverse tissues, we found that proteoglycan interactions selectively affect Shh-induced proliferation rather than Shh-induced patterning. Proteoglycans localized Shh ligand to mitogenic niches in developing brain to promote proliferation of neural stem/precursor cells. Proteoglycans also function at the level of individual responding cells; cell-associated proteoglycans bound Shh and altered ligand perdurance. In this way, Shh-proteoglycan interactions stimulate expression of Bmi1, Ccnd1 (cyclin D1), Ccnd2 (cyclin D2) and other genes that have been implicated in proliferation and neoplastic growth17,18. These studies indicate that HSPGs selectively promote mitogenic responses to Shh.

1Neurobiology Department, Harvard Medical School, Boston, Massachusetts, USA. 2Pediatric Oncology Department, Dana-Farber Cancer Institute, Boston, Massachusetts, USA. 3Pathology Department, Brigham and Women’s Hospital, Boston, Massachusetts, USA. 4Neurology Department, Children’s Hospital, Boston, Massachusetts, USA. 5Present addresses: Department of Pathology & Laboratory Medicine, University of Calgary, Calgary, Alberta, Canada (J.A.C.), Division of Pediatric Hematology/Oncology, Doernbecher Children’s Hospital, Oregon Health & Science University, Portland, Oregon, USA (K.J.N.) and Science Department, Phillips Exeter Academy, Exeter, New Hampshire, USA (Y.C.). 6These authors contributed equally to this work. Correspondence should be addressed to R.A.S. ([email protected]).

Received 30 November 2008; accepted 28 January 2009; published online 15 March 2009; doi:10.1038/nn.2287

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

409

ARTICLES

a

0.8

Optical density (405 nm)

0.7

Shh-AP Ala Shh -AP AP

0.6 0.5 0.4 0.3 0.2 0.1

–4

–5

–6

–7

–8

–9

–1 0

–1 1

–1 2

–1 3

0 log [heparin] (M)

Shh-AP

-AP

ShhC24S

ShhAla

Antibody to Shh

Shh

30

H-Palmitate ShhC24S

50

Ala

Shh

3

ShhAla

MW (kDa)

Shh-AP Pl-PLC

d Shh

Mock

Mock Ala

Shh-AP heparinases

Shh

c

Shh-AP Vehicle

Shh

© 2009 Nature America, Inc. All rights reserved.

b

MW (kDa) 25

25

RESULTS HSPG binding motif mutations in Shh We identified a conserved Cardin-Weintraub motif in Shh that is responsible for the high-affinity binding of Shh to HSPGs14. To determine whether mutations in this motif can be used to identify the functions of Shh-proteoglycan interactions, we quantitatively evaluated Shh binding to heparin. We assessed the binding of equal amounts of wild-type Shh or ShhAla to heparin sulfate–coated plates using alkaline phosphatase–tagged Shh isoforms. We measured binding in the absence or presence of increasing amounts of soluble heparin sulfate. The overall binding of mutant Shh was substantially less than that of wild type. Furthermore, higher concentrations of soluble heparin sulfate were required to abolish the binding of wild-type Shh to immobilized heparin than were needed to abolish binding of ShhAla (Fig. 1a). The binding of ShhAla to endogenous proteoglycans was also markedly impaired. Wild-type Shh, but not ShhAla, bound endogenous proteoglycans in CNS tissue sections14 (Fig. 1b). Pretreatment of sections with heparinases, which remove heparan sulfates, or phosphoinositide-specific phospholipase C (PI-PLC), which removes GPIlinked molecules, prevented the binding of wild-type Shh to tissue proteoglycans (Fig. 1b and Supplementary Fig. 1 online). As glypicans are HSPGs with a GPI-linkage, these results suggest that wild-type Shh binds glypican proteoglycans in the developing cerebellum and that the

410

Figure 1 ShhAla specifically alters proteoglycan binding. (a) ShhAla showed reduced binding to heparin-coated plates. Alkaline phosphatase–tagged Shh (Shh-AP, circles), ShhAla (ShhAla-AP, squares) or alkaline phosphatase (AP, triangles) were incubated with heparin-coated plates in the presence of increasing concentrations of soluble heparin. Error bars represent ± s.e.m. (b) Shh bound GPI-linked proteoglycans in cerebellar sections, but ShhAla did not. P6 sections are shown that have been treated with vehicle control, heparinases or PI-PLC, incubated with Shh-AP, ShhAla-AP or vehicle controls, and then processed for binding of alkaline phosphatase–tagged ligand. Scale bars represent 100 mm. (c) ShhAla was processed to mature isoform. Lysates of HEK293 that were transfected with Shh or ShhAla were analyzed by immunoblot with antibody to Shh. Immature 45-kDa (arrowhead) and mature 20-kDa isoforms (arrow) were seen. (d) ShhAla was palmitoylated. HEK293 cells expressing Shh, ShhAla or ShhC24S (which has a mutation in the palmitoylation site) were labeled with 3H-palmitate, analyzed for palmitoylation and probed with antibody to Shh. Shh and ShhAla were palmitoylated and ShhC24S was not.

ShhAla mutation interferes with the binding of these endogenous glypicans in developing cerebellum. We next introduced the ShhAla mutation into a full-length Shh expression vector and produced protein in HEK cells. The ShhAla precursor underwent proteolytic cleavage to generate a mature isoform of the correct size (Fig. 1c) that, as with wild-type Shh, was palmitoylated (Fig. 1d)19. We previously demonstrated that ShhAla binds Ptc with similar affinity as wild-type ligand14. Together, these data show that ShhAla can be used as a specific reagent to determine functions of Shh-proteoglycan interactions. Using homologous recombination, we generated knock-in mice in which endogenous Shh was replaced with ShhAla (Supplementary Fig. 2 online). With this targeting strategy, a loxP site remained in the intron between exons 1 and 2. Because intronic elements can affect Shh expression, we also generated a line of control animals with an alteration in this loxP site (ShhCtl); ShhCtl/Ctl animals are indistinguishable from wild type. Furthermore, the ShhAla mutation itself did not alter the expression of Shh protein in vivo: Shh levels were indistinguishable in Shh+/+, ShhAla/+ and ShhAla/Ala tissues (Supplementary Fig. 2). This genetic approach enables us to determine the role of Shh binding to proteoglycans without interference from proteoglycandependent modulation of other growth factor pathways. Shh-proteoglycan interactions affect growth, not patterning Shh
/
animals are embryonic lethal and have holoprosencephaly and limb patterning defects20. Hypomorphic alleles of Shh usually alter size and pattern19. In contrast, homozygous ShhAla/Ala mice only showed growth defects and had no patterning defects. Although Shh is critical in generating and patterning diverse organs, mutant animals were viable and fertile, with all organs being present and correctly localized. In contrast, ShhAla/Ala mice had multiple growth abnormalities. Overall size was reduced in mutant animals, with particular differences in the sizes of the brain, spinal cord and eyes. The body weight of ShhAla/Ala mice was 11% less than that of wild-type or ShhCtl/Ctl animals and brain weight was reduced by 13% (P ¼ 0.0018). There was, however, disproportionate hypoplasia of cerebellum and olfactory bulb (Fig. 2a). In contrast with these clear differences in tissue growth, when we examined areas where Shh is critical for pattern formation1,21,22, we found that ShhAla/Ala mice had normal digit number and shape and the spinal cord laminae were correctly formed (Fig. 2a,b and Supplementary Fig. 3 online). Furthermore, although holoprosencephaly and cyclopia are cardinal features of Shh
/
mice20, ShhAla/Ala animals showed normal separation of cerebral ventricles and had two well-spaced, but small, eyes (Fig. 2a). Together, these findings indicate that Shh-proteoglycan interactions mediated by the Cardin-Weintraub motif are specifically required for growth regulation, but not for patterning activities.

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

Shh +/+

Shh Ala/Ala

b

Shh +/+

Shh Ala/Ala

c

Shh +/–

lsl1 Dbx1

© 2009 Nature America, Inc. All rights reserved.

Nkx6.1

Nkx2.2

Shh

a

In mice, one copy of Shh is sufficient for normal development, as Shh+/
mice show normal patterning and growth20. One copy of ShhAla is also sufficient for patterning, as ShhAla/
mice were viable with normal limbs and digits. However, even more notable changes in growth were evident in ShhAla/
than in ShhAla/Ala animals (Fig. 2c). The overall size of ShhAla/
mice, and cerebellar and olfactory bulb size in particular, were greatly reduced compared with Shh+/
animals. Even when one copy of ShhAla was present, however, no defects were seen in tissue patterning (Fig. 2c). The characterization of ShhAla/Ala and ShhAla/
mice underscores the selectivity of the phenotype and indicates that Shh-proteoglycan interactions regulate tissue growth in particular. Shh-proteoglycan interactions promote proliferation To determine why proteoglycan interactions are critical for tissue growth, we focused on the cerebellum, as the cerebella of ShhAla/Ala mice are one-third smaller than those of their wild-type littermates. Because Shh regulates cell division and cell death in the nervous system23,24, we asked whether the decrease in cerebellar size in ShhAla/Ala and ShhAla/
mice reflects abnormal proliferation or apoptosis. In early postnatal life, granule cell precursors (GCPs) undergo extensive proliferation in the external granule cell layer (EGL) before exiting the cell cycle and migrating inwards past the Purkinje cell layer to reside in the internal granule cell layer (IGL). Previous studies have shown that Purkinje cell–synthesized Shh promotes GCP proliferation23,25. In ShhAla/Ala mice, GCP proliferation in the EGL was approximately 30% less than that in wild type, as detected by determining the percentage of EGL cells that were positive for M-phase marker phosphohistone H3 and by S-phase labeling using bromodeoxyuridine (BrdU) (Fig. 3a–c). This decreased proliferation, which was seen throughout the cerebellum (Fig. 3d), indicates that Shh-proteoglycan interactions promote

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

Shh Ala/–

Figure 2 ShhAla/Ala and ShhAla/
mice show defects in growth, but show normal patterning. (a) Skeletal morphology, body and brain patterning were normal in ShhAla/Ala mice, but the size of adult ShhAla/Ala mice was 11% less than that of wild type and sizes of the olfactory bulbs and cerebella were 30 and 31% reduced (arrows), respectively. A sagittal view of cerebellum showed normal patterning with reduced size in ShhAla/Ala. Scale bar represents 1 mm. ShhAla/Ala animals also had well-spaced eyes. (b) Spinal cord patterning was normal in ShhAla/Ala mice. We carried out in situ hybridization for Shh (expressed in notochord and floor plate), Nkx2.2 (motor neuron precursors), Nkx6.1 (ventral spinal cord), Isl1 (dorsal root ganglia and motor neurons) and Dbx1 (V0 interneurons) in embryonic day 10.5 (E10.5) littermates. Scale bar represents 100 mm. (c) Growth defects were seen in ShhAla/
, but not defects in patterning. The olfactory bulb and cerebellum were reduced in size (red arrows). Scale bar represents 1 mm. The eyes of ShhAla/
mice were well spaced.

mitogenesis of cerebellar precursors during early postnatal life across the rostral-caudal axis of the developing hindbrain. Exostoses 1 and 2 are required for HSPG chain elongation and are therefore necessary for generating glycan chains that interact with Shh. Although we previously identified developmental changes in expression of Ext1 and Ext2 in BALB/c mice during the early postnatal period14, the 129/C57BL6 strain that we used here showed consistent expression of Ext1 and Ext2 in the postnatal period when proliferation was reduced in ShhAla/Ala (Supplementary Fig. 4 online). Many other genes required for synthesizing and modifying glycans are also appropriately expressed at this time26. Thus, Shh and proteoglycans are expressed appropriately to regulate GCP proliferation, which is reduced when Shh-proteoglycan interactions are impaired. The subventricular zone (SVZ) adjacent to the lateral ventricles and the subgranular layer (SGL) of the hippocampus represent two additional locations where there is substantial Shh-regulated postnatal proliferation of neural stem/precursor cells3,4,27, whereas the embryonic spinal cord represents a site where Shh regulates prenatal expansion of neural stem/precursors28. Neural precursor proliferation in the spinal cord was reduced in ShhAla/Ala embryos as compared with Shh+/+ littermates (Fig. 4a,b). Proliferation of neural stem/precursors in the SVZ and SGL of adult mutants was also decreased compared with age- and sex-matched ShhCtl/Ctl mice (Fig. 4c,d). As stem/ precursor cells in the SVZ generate olfactory bulb neurons, decreased SVZ proliferation may contribute to the olfactory bulb hypoplasia that we observed (Fig. 2 and Supplementary Fig. 5 online). Together, these findings demonstrate that Shh-proteoglycan interactions promote neural stem/precursor cell proliferation in multiple locations in both the developing and adult CNS, highlighting the widespread importance of proteoglycans in mitogenesis. Apoptosis was decreased in the EGL of ShhAla/Ala mutants compared with wild-type littermates (Supplementary Fig. 6 online). Because the Shh pathway can stimulate or suppress apoptosis under different conditions24,29, it is possible that the pro-apoptotic effects of Shh may require proteoglycan interactions, whereas the anti-apoptotic effects do not. Alternatively, decreased apoptosis in ShhAla/Ala mice may be an indirect consequence of the mutation. In either case, the

411

ARTICLES

Shh Ala/Ala

Shh +/+

a

* 2.0

* 1.0

40 30

*

20

* 10 0

P1

P3

P6

P12

P3

P6

P9

P12

3.0

2.0

1.0

0 Ant

Mid

Post

increased survival did not mitigate the proliferation decrease and the net result was a markedly smaller cerebellum (30% smaller in ShhAla/Ala and 50% smaller in ShhAla/
). Proteoglycans delineate the Shh mitogenic niche in the EGL There are two potential mechanisms to explain the effects of proteoglycans on Shh-induced proliferation. One is that proteoglycan interactions localize Shh to a mitogenic niche. Previous studies have suggested that proteoglycans regulate Hedgehog dispersal5,9,12,13 and localize Indian Hedgehog in developing bone30. Thus, without proteoglycan binding, Shh might not collect in the EGL, resulting in decreased proliferation. A second potential mechanism is that binding of Shh to proteoglycans on the surface of GCPs modulates Shhdependent intracellular signal cascades7,13. Because these models are not mutually exclusive, we investigated both. We previously developed an assay to evaluate the nature and location of mitogenic niches in neural tissues31. In this assay, green fluorescent

Figure 4 Embryonic and adult neural stem/precursor proliferation is reduced in ShhAla/Ala mice. (a) More proliferating neural stem/precursors were seen in the spinal cord of Shh+/+ mice (left) than in their ShhAla/Ala (right) E10.5 littermates. Phosphohistone H3 is shown in red and DAPI in blue. Scale bar represents 100 mm. (b) A reduction in the total number of phosphohistone H3–positive cells in the spinal cord of ShhAla/Ala embryos was seen in comparison with the number in ShhCtl/Ctl littermates (*, P o 0.05). A significant reduction in the total number of phosphohistone H3–positive cells per unit spinal cord area was also seen in ShhAla/Ala E10.5 embryos (data not shown; P o 0.05). Error bars represent ± s.e.m. (c) Fewer phosphohistone H3–positive cells were present in the SVZ of adult ShhAla/Ala mice (*, P o 0.05). Error bars represent ± s.e.m. (d) Fewer BrdU-positive cells were seen in the hippocampal SGL of adult mutant mice than in control mice (*, P r 0.05). Error bars represent ± s.e.m.

412

protein (GFP)-labeled precursors are introduced onto organotypic slices and incorporated into the slices. We incubate the slice culture with BrdU to label proliferating cells and then measure the proliferative index of GFP-positive precursors in distinct microenvironments. Precursors that are exposed to the cerebellar EGL proliferate extensively as a result of the mitogenic effects of Shh31. To determine whether Shhproteoglycan interactions are important in delineating this niche, we introduced wild-type GFP-labeled precursors or 3,3¢-dioctadecyloxacarbocyanine perchlorate (DiO)-labeled ShhAla/Ala precursors onto slices cultured from wild-type or ShhAla/Ala littermates. Although wild-type or ShhAla/Ala precursor proliferation was enhanced when cells were introduced onto a wild-type EGL, the EGL of ShhAla/Ala mice did not provide a mitogenic niche for wild-type precursors (Fig. 5a).

a

Shh Ala/Ala

Shh +/+

b 100

c

80

*

60 40 20

d 400

800

*

600

400

200

Shh +/+ Shh Ala/Ala

VOLUME 12

[

300

*

200

100

0

0

0

Number of BrdU-positive cells

0

d

*

c

Number of pH3-positive cells

Percentage of pH3-positive cells

3.0

Shh +/+ Shh Ala/Ala

Number of pH3-positive cells

4.0

BrdU

Percentage of BrdU-positive cells

b

Percentage of pH3-positive cells

© 2009 Nature America, Inc. All rights reserved.

pH3

Figure 3 Reduced proliferation of ShhAla/Ala cerebellar granule precursors is seen in the EGL of developing mice. (a) Phosphohistone H3 (pH3)-positive or BrdU-positive (both in red) cells were fewer in the EGL of P3 ShhAla/Ala mice compared with Shh+/+ littermates. DAPI staining is shown in gray or blue. Scale bars represent 100 mm (left) and 50 mm (right). (b) Phosphohistone H3 mitotic indices were reduced in ShhAla/Ala mice (white bars) at multiple postnatal ages (Shh+/+, black bars) (*, P o 0.001). We also observed reduced proliferation in ShhAla/Ala mice in one wild-type and mutant littermate pair at P9 (1.37% versus 1.07% for Shh+/+ and ShhAla/Ala, respectively). Error bars represent ± s.e.m. (c) BrdU proliferation indices were less in ShhAla/Ala mice (white bars) compared with wild type (black bars) at multiple postnatal ages (*, P o 0.001). Although no difference was seen at intermediate time points (P6 and P9), this partly reflected the cerebellar size difference between the ShhAla/Ala and Shh+/+ mice. There was still a significant difference in the number of proliferating cells/EGL length at P6 (3.6% in Shh+/+ versus 2.7% in ShhAla/Ala; P o 0.01). Error bars represent ± s.e.m. (d) Phosphohistone H3 mitotic indices in P3 wild-type (black bars) and ShhAla/Ala mice (white bars) in the anterior, middle and posterior cerebellum were not statistically different from their respective totals in b, demonstrating that lobes throughout the cerebellum were affected in the mutant. Error bars represent ± s.e.m.

Shh Ctl/Ctl Shh Ala /Ala

NUMBER 4

[

Shh Ctl/Ctl Shh Ala /Ala

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES Figure 5 ShhAla cannot specify a mitogenic niche. EGLa (a) GCPs from GFP-positive mice were cultured on 7 * EGL * EGLb P6 Shh+/+ or ShhAla/Ala cerebellar slices, or DiO7 lGL 6 labeled ShhAla/Ala GCPs were cultured on wild-type 6 ML 5 slices. The proliferation of introduced GCPs was 5 PCL analyzed by BrdU incorporation. The proliferative 4 4 * ** index is the percent of GFP- or DiO-positive cells IGL 3 3 in each layer that are also BrdU positive (*, P o Shh +/+ Shh Ala/Ala 2 2 0.05 versus EGL of wild-type slice; **, P o 0.05 EGLa 1 versus IGL of wild-type slice; no significant 1 difference between EGL and IGL proliferation on 0 0 EGLb Cont +HS +/+ Ala/Ala ShhAla/Ala slices was seen; P 4 0.05). WT, Shh+/+; ALA, ShhAla/Ala. Error bars represent ± s.e.m. (b) The EGL mitogenic niche in ShhAla/Ala slices (light gray bar) was phenocopied by added exogenous heparan sulfates (HS, dark gray bar) (*, P o 0.05). Error bars represent ± s.e.m. (c) There was reduced Shh staining in the EGLa of ShhAla/Ala P6 cerebella (arrowhead) compared with wild-type cerebella (arrows) (52 ± 11% of wild type, P ¼ 0.01). EGLb, external granule cell layer, inner postmitotic zone; ML, molecular layer; PCL, Purkinje cell layer. Scale bars represent 50 mm.

b

ShhAla/Ala mice results from both altered localization of ligand and changes in intracellular signaling cascades. Shh responses are mediated by the transcription factors Gli1, Gli2 and Gli3 (ref. 36). Therefore, we analyzed the expression of these factors in ShhAla/Ala mice. Although Gli transcription factors can function as either transcriptional activators or repressors37, Gli1 only functions as an activator, and only one isoform of Gli1 has been reported38. Gli1 protein levels were equivalent in postnatal day 3 (P3) ShhAla/Ala and Shh+/+ mice (Fig. 7a). Gli2 can function as activator and repressor, and distinct isoforms have been identified that subserve these different functions38,39. In the cerebella of mutant mice, the ratio of the Gli2 activator (GliAct) to the shorter Gli2 repressor (Gli2Rep) was reduced by 50.9% (Fig. 7b). These data suggest that overall Gli2-dependent transcription is likely to be altered in mutants. Previous studies have

a Normalized proliferation (relative to mock)

To determine whether decreased proliferation results from impaired proteoglycan interactions, we added heparan sulfates to cultures of wild-type precursors on wild-type slices. Excess glycans can compete with, and so diminish, the actions of endogenous proteoglycans30. Excess heparan sulfates phenocopied the ShhAla mutation, indicated by decreased proliferation of cells introduced onto the EGL (Fig. 5b). Together, these data indicate that interactions between Shh’s CardinWeintraub motif and endogenous proteoglycans promote neural precursor proliferation and establish the EGL mitogenic niche. In slice overlay cultures, proliferation of introduced precursor cells was decreased in the EGL, but increased in the IGL, of ShhAla/Ala slices (Fig. 5a). These data suggest that proteoglycan binding enables Shh to accumulate in the EGL rather than in other layers. To test this, we carried out immunohistochemistry for Shh. Consistent with previous studies31,32 in wild-type animals, we detected Shh immunostaining in Purkinje cells and in the IGL and EGL (Fig. 5c and Supplementary Fig. 7 online). In ShhAla/Ala mice, Shh immunostaining was decreased in the outer EGL (EGLa), where proliferation occurs (Fig. 5c). These findings indicate that proteoglycans localize Shh to a mitogenic niche in the EGLa. Electron microscopy studies have identified a proteoglycan matrix in the EGLa, just under the pia and adjacent to proliferating GCPs33. Thus, one mechanism by which Shh-HSPG interactions stimulate precursor proliferation is by localizing ligand to the EGL mitogenic niche.

on AL W Ac T e sl lls ic e

W AL T A ce sl lls ic e

on

W Tc T e sl lls ic e

Proliferative index (%)

W

on

3.0 2.5

Shh Ala Shh

** *

2.0 1.5 1.0 0.5 0

Proteoglycan interactions modulate responses to Shh Although the data above indicate that Shh-proteoglycan interactions are important for localizing Shh to proliferative zones, and thus establishing a mitogenic niche, it is also possible that proteoglycans on neural stem/precursors bind Shh, thereby specifying proliferative responses. To address this possibility, we asked whether proteoglycans on GCPs affect Shh-induced responses at the level of an individual responding cell. Consistent with previous studies14,23,25,34,35, wild-type Shh induced robust GCP proliferation. In dissociated cell cultures, the proliferative response evoked by wild-type Shh was much greater than that evoked by equivalent concentrations of ShhAla (Fig. 6a). In contrast, no consistent effects on survival were seen when dissociated GCP cultures were stimulated with either wild-type Shh or ShhAla (Fig. 6b). Thus, the decreased proliferation that we observed in ShhAla/Ala mice is a direct consequence of the Cardin-Weintraub mutation, whereas decreased apoptosis may be an indirect result of the mutation. These data demonstrate that Shh interacts with proteoglycans on individual GCPs, thereby inducing cells to proliferate, and suggest that the decreased proliferation that we observed in

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

5

15

30

75

150

225

300

–1 Shh concentration (ng ml )

b

1.6 1.4

Normalized survival (relative to mock)

© 2009 Nature America, Inc. All rights reserved.

c

Proliferative index (%)

a

1.2 1.0 0.8 0.6 0.4 0.2 0 1

2.5

5

15

30

75

150

–1 Shh concentration (ng ml )

Figure 6 Shh-proteoglycan interactions promote proliferation in dissociated cell cultures of GCPs, but are not needed for survival. (a) The proliferative response to Shh (gray), as assessed by BrdU incorporation, was greater than that of GCPs to ShhAla (black) (*, P r 0.05 and **, P o 0.01). Error bars represent ± s.e.m. (b) No consistent effects on survival, as assessed by activated caspase 3 immunostaining, were seen in GCPs stimulated with Shh (gray) or ShhAla (black). Error bars represent ± s.e.m.

413

ARTICLES Figure 7 Shh-proteoglycan interactions MW Shh ShhAla Shh +/+ (kDa) – + – + modulate transcriptional activity through the 1.8 Ala/Ala Shh Act 175 regulation of Gli2 isoforms and signaling Gli2 1.5 MW kinetics. (a) Gli1 protein levels in P3 ShhAla/Ala (kDa) 1.2 83 and Shh+/+ cerebella were equivalent. Left, Ala/Ala +/+ 0.9 western blot of Shh and Shh cerebellar 62 Gli2Rep lysates stained with antibody to Gli1. Right, 0.6 1.2 83 quantification of blotting results. Error bars 0.3 1.0 represent ± s.e.m. (b) The ratio of Gli2Act to 0 * 0.8 Gli2Rep was reduced in P3 ShhAla/Ala cerebella MW 0.6 compared with Shh+/+ cerebella. Left, western (kDa) 3.0 0.4 175 blot stained with antibody to Gli2. Right, 0.2 Gli2Act * 2.5 quantification of results (*, P o 0.01). Error 0 2.0 bars represent ± s.e.m. (c) Gli3 repressor Shh ShhAla Ala/Ala protein levels were unchanged in P3 Shh 1.5 83 MW cerebella, as compared with Shh+/+ cerebella. (kDa) 24 32 38 48 56 62 72 1.0 62 Left, western blot stained with antibody to Gli3. 175 Rep Shh Gli2Act 0.5 Gli2 Right, quantification of results. Error bars 47.5 Tubulin 0 represent ± s.e.m. (d) The ratio of Gli2Act to Ala Gli2Rep was greater in cells stimulated for 24 h Gli2Act Shh 1.2 with Shh compared with ShhAla. Top, western Tubulin MW blot stained with antibody to Gli2. Bottom, MW (kDa) 0.9 quantification of results (*, P o 0.01). Error (kDa) 175 24 32 38 48 56 62 72 bars represent ± s.e.m. (e) The kinetics of 62 Gli2Rep Shh 0.6 signaling differed for GCPs stimulated with Tubulin ShhAla as compared with GCPs stimulated 0.3 with Shh. The ligand was applied for 24 h 83 ShhAla Gli2Rep Rep Gli3 and then removed. Top, western blot of 0 Tubulin Shh- or ShhAla-stimulated GCPs stained with antibody to Gli2Act (antibody to tubulin was used as a loading control). The numbers above the blots represent the hours spent in culture. Bottom, western blot of Shh- or ShhAla-stimulated GCPs stained with antibody to Gli2Rep (antibody to tubulin was used as a loading control). Gli2Act/Gli2Rep ratio

Al

a/

Al

a

Sh h

Gli2Act/Gli2Rep ratio

Sh

h

+/

h

Gli1 levels (arbitrary units)

/A l

Al a

+

+/

+

a/

Al

Sh

Sh

Sh

h

h

+/

+

Al

a

b

Sh h

Sh h

Sh h

+/

Al a

+

/A l

a

d

a

a

Rep

levels (arbitrary units)

a

Al

a/

Al

h

Sh

Sh

h

+/

+

a/

Al

h

Sh

Gli3

© 2009 Nature America, Inc. All rights reserved.

Sh

h

+/

+

Al

a

c

e

indicated that Gli3 functions primarily as a repressor40. We did not detect any differences in Gli3Rep levels, and we did not detect Gli3Act in cerebellar tissue of either wild-type or mutant mice (Fig. 7c). Thus, Gli2 isoforms represent the Shh-dependent transcription factor(s) that are clearly altered in ShhAla/Ala cerebellum. As Gli2 activity is required to mediate the expansion of GCPs41, changes in Gli2 might explain the observed ShhAla/Ala phenotype. As the Gli2 activator/repressor ratio is altered in vivo in ShhAla/Ala cerebella and Gli2 has been shown to be important in GCP proliferation41, we examined in greater detail Gli2 protein levels in GCPs that were acutely stimulated with wild-type or mutant Shh. We found that GCPs stimulated with either Shh or ShhAla showed an increase in Gli2Act protein (9.7% and 34.7% increase for Shh- and ShhAla-stimulated GCPs, respectively, compared with unstimulated controls). However, consistent with our in vivo results, the Gli2Act/Gli2Rep ratio in GCPs that were stimulated for 24 h with ShhAla was lower than that seen in GCPs stimulated with Shh (Fig. 7d). Furthermore, the temporal profiles of Gli2Act and Gli2Rep differed in experiments in which GCPs were first stimulated for 24 h with ShhAla or wild-type Shh, ligand was removed, and Gli2 levels were analyzed over time (Fig. 7e and Supplementary Fig. 8 online). These data indicate that Shh interacts with proteoglycans on GCPs to alter the nature and timing of Gli2dependent transcription. Proteoglycans alter Shh-dependent gene expression Changes in the Gli2 activator:repressor ratio probably alter the expression of Shh-responsive genes. A number of Shh-responsive genes have been identified, including genes that are involved in proliferation and tissue patterning17,34,35,42–44. We analyzed Shh-responsive gene expression when dissociated wild-type GCPs were acutely stimulated with equal concentrations of wild-type Shh or ShhAla. We separated Shhdependent genes into three clusters that were differentially modulated by proteoglycan interactions (Fig. 8a). One cluster of genes, which

414

included Gli1 and Ptc1, were similarly induced by ShhAla and wild-type Shh. A second cluster of target genes, Gli2 and Mycn (also known as N-myc), was induced to a greater extent by ShhAla than by wild-type Shh. Notably, expression of Gli2 and Mycn, genes that have been implicated in proliferation, was better induced by ShhAla. It should be noted that Shh regulates both Gli2 and N-myc via transcriptional and post-transcriptional mechanisms39,45, and so Gli2 and N-myc activity need not correspond to RNA levels. Induction of a third cluster, which included Gli3, Ccnd1, Ccnd2 and Bmi1, was diminished when cells were stimulated with ShhAla rather than with wild-type ligand. Cyclins D1 and D2 are cell cycle regulators at the G1/S transition that are induced in response to Shh stimulation35 and have been implicated in Shh-induced stem/precursor cell proliferation35. The polycomb protein Bmi-1 is a Shh target that is critical for self-renewal of stem cells and for cancer cell proliferation17,18,44. Together, these data indicate that the gene expression program induced by Shh is modulated by proteoglycans of responding cells. Moreover, Shh target genes that are altered when cells are stimulated by ShhAla include many that have been implicated in proliferation and oncogenesis. Previous studies have demonstrated that distinct responses to Shh can be elicited depending on ligand concentration22. To determine whether a shift in the dose-response curve explains why ShhAla induces an altered program of gene expression, we carried out dose-response experiments using the Shh and ShhAla proteins. C3H10T1/2 differentiation, as assessed by alkaline phosphatase induction (Supplementary Fig. 9 online), is a standard assay for Shh activity; we found no shift in the ShhAla dose-response curve compared to that of Shh. Furthermore, maximal efficacy of ShhAla was not reduced compared to Shh. We next measured Gli1 and Ccnd2 mRNA levels in GCPs that were stimulated across a wide range of Shh and ShhAla doses (Supplementary Fig. 9). Gli1 expression did not differ significantly (P 4 0.05) in response to equivalent doses of Shh and ShhAla throughout the dose range, whereas Ccnd2 induction by ShhAla was diminished regardless of dose. These

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

a

*

Gene induction with ShhAla relative to Shh stimulation

1.3 1.2

* 1.1

*

*

Ptch1

Gli2

Mycn

Gli3

Bmi1

Ccnd1

Ccnd2

C3H10T1/2 cells MW (kDa) 24 32 38 48 56 62 72 Shh 25

Shh

Actin

Tubulin

Shh

GCPs 24 32 38 48 56 62 72

Ala

ShhAla Tubulin

Actin Shh ShhAla

80 60 40 20

100 Shh (% of level at 24 h)

Shh (% of level at 24 h)

100

80 60 40 20 0

0 24

c In vivo gene expression ShhAla/Ala relative to wild type

© 2009 Nature America, Inc. All rights reserved.

*

0.9 Gli1

b

*

1.0

32

38 48 56 Time (h)

62

24

72

1.2

**

32

38 48 56 Time (h)

62

72

*

1.1

**

1.0

*

*

0.9

0.8

Shh Ccne1 Gli1

Ptch1

Gli2

Mycn

Gli3

Bmi1 Ccnd1 Ccnd2

data indicate that the altered biological activity of ShhAla does not reflect a shift in the dose-response curve. As recent studies have highlighted the importance of Shh signaling kinetics in determining Shh responses46, we asked whether proteoglycan interactions alter the ability of Shh to signal over prolonged periods of time. Equivalent amounts of Shh or ShhAla were added to either C3H10T1/2 or GCP cultures. After 24 h of stimulation, we removed ligand from the media and analyzed Shh perdurance in the stimulated cells over the ensuing 2 d. In both systems, ShhAla levels declined more precipitously than those of wild-type Shh (Fig. 8b). Taken together, these data indicate that proteoglycans alter the kinetics of signaling, promoting a gene expression signature that is important for Shh-dependent precursor proliferation. To determine whether changes in Shh-dependent gene expression explain the in vivo phenotype of ShhAla/Ala mice, we used quantitative RT-PCR to compare Shh-responsive gene expression in wild-type and mutant cerebella. Gli1 and Ptc1 levels were equivalent in wild-type and ShhAla/Ala cerebella. However, expression of a second set of genes, including Ccnd1, Ccnd2 and Bmi1, was significantly reduced in ShhAla/Ala animals compared with wild-type littermates (P o 0.01 for Ccnd1 and Ccnd2, P o 0.05 for Bmi1; Fig. 8c). Decreased expression of Ccnd1, Ccnd2 and Bmi1 do not reflect a generalized decrease in cell cycle-associated proteins, as Ccne1 expression was unchanged. A third set of Shh-regulated genes (Gli2 and Mycn) was expressed at higher levels in ShhAla/Ala than in wild-type mice. The observed changes in vivo

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

Figure 8 Proteoglycan interactions modulate Shh perdurance and differentially affect Shh-dependent gene expression. (a) The gene pattern induced by ShhAla stimulation of wild-type P6 GCPs was different than that of GCPs stimulated by equivalent amounts of Shh (*, P o 0.05). Error bars represent ± s.e.m. (b) GCP proteoglycans modulated Shh ligand perdurance. Top left, western blot of Shh- or ShhAla-stimulated C3H10T1/2 cells using antibody to Shh (antibody to actin was used as a loading control). Top right, western blot of Shh- or ShhAla-stimulated GCPs using antibody to Shh (antibody to tubulin was used as a loading control). The numbers above the blots represent the hours spent in culture. The bottom two panels each show a quantification of results from one representative experiment. (c) Expression of Shh target genes in P1–2 ShhAla/Ala cerebella (P1–3 ShhAla/Ala cerebella for Gli2) relative to Shh+/+ littermates indicates that Shh-proteoglycan interactions differentially affect gene subsets (*, P o 0.01 and **, P o 0.05). Error bars represent ± s.e.m.

do not reflect alterations in the distribution of cells expressing Shhresponsive genes, as indicated by in situ hybridization for Gli1, Gli2 and Gli3 (Supplementary Fig. 10 online). An important feature of these data is that proteoglycan interactions do not uniformly increase or decrease Shh-dependent gene expression in vivo. Furthermore, clusters that are unchanged, increased or decreased in ShhAla/Ala mice show a marked similarity to gene expression patterns in GCPs that are acutely stimulated with wild-type versus mutant Shh. Taken together, our data indicate that Shh-proteoglycan interactions selectively promote neural stem/precursor proliferation by two mechanisms. First, proteoglycans localize the ligand in developing tissue and so establish the mitogenic niche. Second, they alter ligand perdurance, preferentially activating intracellular cascades that culminate in mitogenesis and precursor renewal. DISCUSSION Using a genetic approach, we found that Shh-proteoglycan interactions are required for proliferative, rather than patterning, responses to Shh. We generated mutant mice that express a Shh protein that cannot bind proteoglycans, but can bind to Ptc. These mutant mice had a selective deficit in neural stem/precursor cell proliferation. We identified two distinct activities of proteoglycans in Shh-dependent proliferation: localization of Shh to mitogenic niches and triggering of a gene expression program that is important for cell division and stem cell renewal. Proteoglycans are tremendously diverse; each proteoglycan represents a complex, yet poorly understood, readout of the glycogenes, the genes encoding core proteins, sugar transporters, glycosyltransferases, sulfatases and acetylating enzymes. As proteoglycans interact with many growth factors and other molecules, analyses of mutants that interfere with individual steps of proteoglycan synthesis are difficult to interpret. To identify the functions of proteoglycans for Shh responses in vivo, we took an alternative approach. Mutations in the CardinWeintraub motif of Hedgehog proteins interfere with proteoglycan binding, but do not alter Ptc affinity, lipid modifications or expression level. Therefore, ShhAla/Ala and ShhAla/
mutants provide a unique genetic approach for identifying functions of Shh-proteoglycan interactions. We found that proteoglycans are specifically needed for proliferative responses to Shh, but are dispensable for most of its patterning activities. One intriguing finding is that proteoglycans localize Shh to proliferative zones and function at the single-cell level to determine the nature of the response. Previous studies using mutations in the HSPG-synthesizing enzymes of the Ext gene family, mutations in the glypican Dally-like or deletion of the Cardin-Weintraub motif have indicated that HSPGs are critical for appropriate localization of

415

© 2009 Nature America, Inc. All rights reserved.

ARTICLES Hedgehog proteins11–13,47. We found that Shh proteins that cannot bind to HSPGs do not accumulate in the EGL mitogenic niche. Thus, direct interactions of Shh with proteoglycans are important for Shh dispersal from the Purkinje cell layer and/or its sequestration in the EGL. Altered Shh localization in mutant animals impairs the establishment of mitogenic niches in developing brain. In addition to their function in localizing Shh, we found that proteoglycans on receiving cells modulate the Shh response. Previous studies have indicated that proteoglycans function in Hedgehogresponding cells7,13. Some groups have found that glypicans on receiving cells are needed for full-strength signaling, whereas others have found that responding cell glypicans compete with Ptc1 for Hedgehog interaction, thereby inhibiting Hedgehog activity11,13. Our studies suggest that proteoglycan interactions differentially affect intracellular signaling cascades downstream of Smoothened and thus selectively regulate Shh target gene subsets, modifying the response. Because Shh target gene expression is primarily regulated by the Gli transcription factors36, we examined the expression and isoforms of Gli1, Gli2 and Gli3 proteins in wild-type and mutant mice. Although there were no discernable differences between Gli1 and Gli3 expression in wild-type and mutant cerebella, we did see clear differences in Gli2 protein expression. Gli2 is the major transcription factor driving Shhinduced proliferation41,48, and proliferation is reduced in multiple locations pre- and postnatally in ShhAla/Ala mice. Gli2 proteins can function as either transcriptional activators or repressors; unprocessed Gli2 functions as a weak transcriptional activator37,39, whereas Gli2 cleavage and removal of the C-terminal activator domain generates a transcriptional repressor39. Therefore, both Gli2 protein expression and the ratio between unprocessed and processed forms are probably important for the net Gli2-dependent transcriptional response. In vivo, Gli2 mRNA is lower in wild-type than in ShhAla/Ala cerebella, but the Gli2 activator/repressor ratio is twofold greater in wild type. Similarly, Gli2 mRNA levels are lower in GCPs that were acutely stimulated with Shh than ShhAla, but GCPs that were acutely stimulated with Shh had a larger Gli2 activator/repressor ratio than those stimulated with ShhAla. Gli2 expression and processing are both Shh regulated39,48. The ability of proteoglycans to affect the expression and proportions of the Gli2 isoforms in vivo and in vitro could alter Shh-dependent transcriptional activation and repression, and could therefore account for increases and decreases in Shh-target gene expression. One set of Shh-regulated genes, which includes classical targets such as Ptc1 and Gli1, is similarly induced by Shh proteins that can or cannot bind proteoglycans. A second set, which includes Gli2 and Mycn, is preferentially increased by Shh proteins that cannot bind to proteoglycans. The induction of a third set of Shh-regulated genes, which includes Bmi1, Ccnd1 and Ccnd2, genes that are implicated in stem cell maintenance and in tumor biology, requires proteoglycan interactions. Differences in Shh target genes are observed in vivo and in vitro and are not the results of shifts in dose-response curves, but may instead be explained by changes in ligand signaling kinetics. Taken together, analysis of the mutant phenotype indicates that Shh-HSPG interactions particularly affect overall Gli2 transcriptional activity, thereby promoting proliferative responses. Changes in ligand intracellular localization or ligand perdurance provide two possible mechanisms by which Shh-proteoglycan interactions could selectively modulate signaling. Shh signaling kinetics are important in determining morphogen responses46. We found that Shhproteoglycan interactions affected Shh perdurance in responding cells and altered the temporal profile of Shh signaling. It has been proposed that signal duration is proportional to Shh concentration46; our data

416

indicate that proteoglycan interactions alter that relationship. In this way, proteoglycan interactions can modulate the pattern of gene induction and the biological response to Shh. Proliferation without appropriate patterning is a cardinal feature of tumor biology. The Hedgehog pathway may promote oncogenesis when it stimulates proliferation without patterning. It is interesting to note that glypican overexpression can augment growth in cancers that depend on Shh activity, including rhabdomyosarcoma, prostate cancer and pancreatic carcinoma49,50. Similarly, Ccnd1, Ccnd2 and Bmi1, gene targets that require Shh-HSPG interactions, are important in tumor stem cell biology17,18. In summary, we found that Shh-proteoglycan interactions are selectively critical for Shh-dependent mitogenesis and function in two ways. First, proteoglycans localize Shh to establish mitogenic niches. Second, proteoglycans on responsive cells selectively promote intracellular pathways that lead to precursor proliferation. Both mechanisms contribute to the phenotype of the ShhAla/Ala mice. Thus, Shh-proteoglycan interactions promote proliferation in mitogenic niches by localizing Shh to these proliferative zones and by modulating intracellular signaling cascades and transcriptional programs. METHODS Section-binding assay. To evaluate proteoglycan binding in situ14, cryosections were treated with phosphate-buffered saline, vehicle control, and 500 mU ml
1 of heparinases (Sigma) or 500 mU ml
1 of PI-PLC for 1 h at 37 1C and then left overnight at 4 1C. We added alkaline phosphatase–tagged Shh or ShhAla for 1 h (20–25 1C), washed the sections and then immunostained them with antibodies to alkaline phosphatase. Size measurements. We weighed and killed 6-month-old mice. Total brain and cerebella were weighed. Olfactory bulb dimensions were measured from 10-mm thick coronal cryosections and the overall volume was calculated. We assessed at least three pairs of ShhAla/Ala and age-matched ShhCtl/Ctl mice. Shh staining quantification. Quantification was carried out on mid-sagittal cerebellar sections, stained in parallel for Shh, from three pairs of wild-type and ShhAla/Ala littermates. We analyzed staining in primary and secondary fissures using Image J software (US National Institutes of Health). In each fissure, three lines from the pia through the IGL were drawn perpendicular to the pia and an intensity plot was determined for each line (plot profile function). Proliferation and apoptosis quantification in vivo. Analyses of phosphohistone H3 staining, BrdU labeling and TUNEL staining were carried out on cerebella from 3–5 pairs of matched wild-type and mutant littermates. We calculated the proliferation or apoptotic indices by dividing the number of positive cells by the number of DAPI-stained cells in the EGL. Comparable locations (midpoint of primary, secondary and tertiary fissures in vermis) were assessed. In coronal sections of adults, every ninth section of the SGL or SVZ was stained with phosphohistone H3, BrdU and activated caspase 3. At least three pairs of adult ShhAla/Ala mice and age-matched controls were analyzed. The SGL was the inner half of the thickness of the dentate gyrus, extending two cell widths into the hilus. Additional details are provided in the Supplementary Methods online. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank J. Despinoy for excellent assistance, C. Stiles, D. Rowitch, M. Greenberg and members of the Segal laboratory for helpful discussions, and D. Rowitch, Q. Ma, P. Chuang, D. Paul, S. O’Gorman, P. Silver and A. McMahon for reagents. This work was supported by the US National Institutes of Health (J.A.C., R.A.S. and S.B.), the Dana Farber Cancer Institute Mahoney Center for Neuro-Oncology (J.A.C.), the Musella Foundation (K.J.N.), the Quan Fellowship (R.M.W.), the Children’s Hospital Mental Retardation and Developmental Disabilities Research Center and the Harvard NeuroDiscovery Center.

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES COMPETING INTERESTS STATEMENT The authors declare competing financial interests: details accompany the full-text HTML version of the paper at www.nature.com/natureneuroscience/.

© 2009 Nature America, Inc. All rights reserved.

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

1. Capdevila, J. & Izpisua Belmonte, J.C. Patterning mechanisms controlling vertebrate limb development. Annu. Rev. Cell Dev. Biol. 17, 87–132 (2001). 2. Hooper, J.E. & Scott, M.P. Communicating with Hedgehogs. Nat. Rev. Mol. Cell Bio. 6, 306–317 (2005). 3. Machold, R. et al. Sonic hedgehog is required for progenitor cell maintenance in telencephalic stem cell niches. Neuron 39, 937–950 (2003). 4. Palma, V. et al. Sonic hedgehog controls stem cell behavior in the postnatal and adult brain. Development 132, 335–344 (2005). 5. Bellaiche, Y., The, I. & Perrimon, N. Tout-velu is a Drosophila homologue of the putative tumor suppressor EXT-1 and is needed for Hh diffusion. Nature 394, 85–88 (1998). 6. Desbordes, S.C. & Sanson, B. The glypican Dally-like is required for Hedgehog signaling in the embryonic epidermis of Drosophila. Development 130, 6245–6255 (2003). 7. Lum, L. et al. Identification of Hedgehog pathway components by RNAi in Drosophila cultured cells. Science 299, 2039–2045 (2003). 8. Yao, S., Lum, L. & Beachy, P. The ihog cell-surface proteins bind Hedgehog and mediate pathway activation. Cell 125, 343–357 (2006). 9. Han, C., Belenkaya, T.Y., Khodoun, M., Tauchi, M. & Lin, X. Distinct and collaborative roles of Drosophila EXT family proteins in morphogen signalling and gradient formation. Development 131, 1563–1575 (2004). 10. Bornemann, D.J., Duncan, J.E., Staatz, W., Selleck, S. & Warrior, R. Abrogation of heparan sulfate synthesis in Drosophila disrupts the Wingless, Hedgehog and Decapentaplegic signaling pathways. Development 131, 1927–1938 (2004). 11. Capurro, M.I. et al. Glypican-3 inhibits Hedgehog signaling during development by competing with patched for Hedgehog binding. Dev. Cell 14, 700–711 (2008). 12. Hacker, U., Nybakken, K. & Perrimon, N. Heparan sulphate proteoglycans: the sweet side of development. Nat. Rev. Mol. Cell Bio. 6, 530–541 (2005). 13. Gallet, A., Staccini-Lavenant, L. & Therond, P.P. Cellular trafficking of the glypican Dally-like is required for full-strength Hedgehog signaling and wingless transcytosis. Dev. Cell 14, 712–725 (2008). 14. Rubin, J.B., Choi, Y. & Segal, R.A. Cerebellar proteoglycans regulate sonic hedgehog responses during development. Development 129, 2223–2232 (2002). 15. Pallerla, S.R., Pan, Y., Zhang, X., Esko, J.D. & Grobe, K. Heparan sulfate Ndst1 gene function variably regulates multiple signaling pathways during mouse development. Dev. Dyn. 236, 556–563 (2007). 16. Inatani, M., Irie, F., Plump, A.S., Tessier-Lavigne, M. & Yamaguchi, Y. Mammalian brain morphogenesis and midline axon guidance require heparan sulfate. Science 302, 1044–1046 (2003). 17. Molofsky, A.V. et al. Bmi-1 dependence distinguishes neural stem cell self-renewal from progenitor proliferation. Nature 425, 962–967 (2003). 18. Pardal, R., Molofsky, A.V., He, S. & Morrison, S.J. Stem cell self-renewal and cancer cell proliferation are regulated by common networks that balance the activation of protooncogenes and tumor suppressors. Cold Spring Harb. Symp. Quant. Biol. 70, 177–185 (2005). 19. Chen, M.H., Li, Y.J., Kawakami, T., Xu, S.M. & Chuang, P.T. Palmitoylation is required for the production of a soluble multimeric Hedgehog protein complex and long-range signaling in vertebrates. Genes Dev. 18, 641–659 (2004). 20. Chiang, C. et al. Cyclopia and defective axial patterning in mice lacking Sonic hedgehog gene function. Nature 383, 407–413 (1996). 21. Harfe, B.D. et al. Evidence for an expansion-based temporal Shh gradient in specifying vertebrate digit identities. Cell 118, 517–528 (2004). 22. Briscoe, J. & Ericson, J. The specification of neuronal identity by graded Sonic Hedgehog signaling. Semin. Cell Dev. Biol. 10, 353–362 (1999). 23. Wechsler-Reya, R.J. & Scott, M.P. Control of neuronal precursor proliferation in the cerebellum by Sonic Hedgehog. Neuron 22, 103–114 (1999). 24. Charrier, J.B., Lapointe, F., Le Douarin, N.M. & Teillet, M.A. Anti-apoptotic role of Sonic hedgehog protein at the early stages of nervous system organogenesis. Development 128, 4011–4020 (2001).

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

25. Wallace, V.A. Purkinje cell–derived Sonic hedgehog regulates granule neuron precursor cell proliferation in the developing mouse cerebellum. Curr. Biol. 9, 445–448 (1999). 26. Yabe, T., Hata, T., He, J. & Maeda, N. Developmental and regional expression of heparan sulfate sulfotransferase genes in the mouse brain. Glycobiology 15, 982–993 (2005). 27. Ahn, S. & Joyner, A.L. In vivo analysis of quiescent adult neural stem cells responding to Sonic hedgehog. Nature 437, 894–897 (2005). 28. Jeong, J. & McMahon, A.P. Growth and pattern of the mammalian neural tube are governed by partially overlapping feedback activities of the hedgehog antagonists patched 1 and Hhip1. Development 132, 143–154 (2005). 29. Yamamoto, Y., Stock, D.W. & Jeffery, W.R. Hedgehog signaling controls eye degeneration in blind cavefish. Nature 431, 844–847 (2004). 30. Hilton, M.J., Tu, X., Cook, J., Hu, H. & Long, F. Ihh controls cartilage development by antagonizing Gli3, but requires additional effectors to regulate osteoblast and vascular development. Development 132, 4339–4351 (2005). 31. Choi, Y., Borghesani, P.R., Chan, J.A. & Segal, R.A. Migration from a mitogenic niche promotes cell-cycle exit. J. Neurosci. 25, 10437–10445 (2005). 32. Gritli-Linde, A., Lewis, P., McMahon, A.P. & Linde, A. The whereabouts of a morphogen: direct evidence for short- and graded long-range activity of hedgehog signaling peptides. Dev. Biol. 236, 364–386 (2001). 33. Hausmann, B. & Sievers, J. Cerebellar external granule cells are attached to the basal lamina from the onset of migration up to the end of their proliferative activity. J. Comp. Neurol. 241, 50–62 (1985). 34. Kenney, A.M., Cole, M.D. & Rowitch, D.H. Nmyc upregulation by sonic hedgehog signaling promotes proliferation in developing cerebellar granule neuron precursors. Development 130, 15–28 (2003). 35. Kenney, A.M. & Rowitch, D.H. Sonic hedgehog promotes G(1) cyclin expression and sustained cell cycle progression in mammalian neuronal precursors. Mol. Cell. Biol. 20, 9055–9067 (2000). 36. Ulloa, F. & Briscoe, J. Morphogens and the control of cell proliferation and patterning in the spinal cord. Cell Cycle 6, 2640–2649 (2007). 37. Sasaki, H., Nishizaki, Y., Hui, C., Nakafuku, M. & Kondoh, H. Regulation of Gli2 and Gli3 activities by an amino-terminal repression domain: implication of Gli2 and Gli3 as primary mediators of Shh signaling. Development 126, 3915–3924 (1999). 38. Aza-Blanc, P., Lin, H.Y., Ruiz i Altaba, A. & Kornberg, T.B. Expression of the vertebrate Gli proteins in Drosophila reveals a distribution of activator and repressor activities. Development 127, 4293–4301 (2000). 39. Pan, Y., Bai, C.B., Joyner, A.L. & Wang, B. Sonic hedgehog signaling regulates Gli2 transcriptional activity by suppressing its processing and degradation. Mol. Cell. Biol. 26, 3365–3377 (2006). 40. Blaess, S., Corrales, J.D. & Joyner, A.L. Sonic hedgehog regulates Gli activator and repressor functions with spatial and temporal precision in the mid/hindbrain region. Development 133, 1799–1809 (2006). 41. Corrales, J.D., Rocco, G.L., Blaess, S., Guo, Q. & Joyner, A.L. Spatial pattern of sonic hedgehog signaling through Gli genes during cerebellum development. Development 131, 5581–5590 (2004). 42. Yoon, J.W. et al. Gene expression profiling leads to identification of GLI1-binding elements in target genes and a role for multiple downstream pathways in GLI1-induced cell transformation. J. Biol. Chem. 277, 5548–5555 (2002). 43. Duman-Scheel, M., Weng, L., Xin, S. & Du, W. Hedgehog regulates cell growth and proliferation by inducing Cyclin D and Cyclin E. Nature 417, 299–304 (2002). 44. Leung, C. et al. Bmi1 is essential for cerebellar development and is overexpressed in human medulloblastomas. Nature 428, 337–341 (2004). 45. Kenney, A.M., Widlund, H.R. & Rowitch, D.H. Hedgehog and PI-3 kinase signaling converge on Nmyc1 to promote cell cycle progression in cerebellar neuronal precursors. Development 131, 217–228 (2004). 46. Dessaud, E. et al. Interpretation of the sonic hedgehog morphogen gradient by a temporal adaptation mechanism. Nature 450, 717–720 (2007). 47. Vyas, N. et al. Nanoscale organization of hedgehog is essential for long-range signaling. Cell 133, 1214–1227 (2008). 48. Galvin, K.E., Ye, H. & Wetmore, C. Differential gene induction by genetic and ligandmediated activation of the Sonic hedgehog pathway in neural stem cells. Dev. Biol. 308, 331–342 (2007). 49. Kleeff, J. et al. The cell-surface heparan sulfate proteoglycan glypican-1 regulates growth factor action in pancreatic carcinoma cells and is overexpressed in human pancreatic cancer. J. Clin. Invest. 102, 1662–1673 (1998). 50. Williamson, D. et al. Role for amplification and expression of glypican-5 in rhabdomyosarcoma. Cancer Res. 67, 57–65 (2007).

417

ARTICLES

Paracrine control of oligodendrocyte differentiation by SRF-directed neuronal gene expression

© 2009 Nature America, Inc. All rights reserved.

Christine Stritt1,5, Sina Stern1,5, Kai Harting1,5, Thomas Manke2, Daniela Sinske1, Heinz Schwarz3, Martin Vingron2, Alfred Nordheim4 & Bernd Kno¨ll1 In neurons, serum response factor (SRF)-directed transcription regulates migration, axon pathfinding and synapse function. We found that forebrain-specific, neuron-restricted SRF ablation in mice elevated oligodendrocyte precursors while inhibiting terminal oligodendrocyte differentiation. Myelin gene and protein expression were downregulated and we observed a lack of oligodendrocytes in mixed neuron/glia and oligodendrocyte-enriched cultures derived from Srf 2/2 mutants. Ultrastructural inspection revealed myelination defects and axonal degeneration in Srf 2/2 mutants. Consistent with our finding that neuronal SRF depletion impaired oligodendrocyte fate in a non–cell autonomous manner, neuron-restricted expression of constitutively active SRF-VP16 affected neighboring oligodendrocyte maturation. Genome-wide transcriptomics identified candidate genes for paracrine regulation of oligodendrocyte development, including connective tissue growth factor (CTGF), whose expression is repressed by SRF. Adenovirus-mediated CTGF expression in vivo revealed that CTGF blocks excessive oligodendrocyte differentiation. In vitro, CTGF-mediated inhibition of oligodendrocyte maturation involved sequestration and thereby counteraction of insulin growth factor 1–stimulated oligodendrocyte differentiation.

Functional neuronal networks require that neurons interact with oligodendrocytes, the myelin sheath–producing glia of the CNS1. Oligodendrocyte differentiation from oligodendrocyte precursor cells (OPCs) follows a temporally and spatially well-orchestrated path, which can be traced by consecutive marker acquisition, starting with early markers such as 2¢, 3¢-cyclic nucleotide 3¢ phosphodiesterase (CNPase) and galactocerebroside, which is a major myelin glycolipid produced by UDP galactosyltransferase 8A. Subsequently, mature differentiation markers are acquired, including the main myelin proteins myelin basic protein (MBP), proteolipid protein (PLP) and myelin-associated glycoprotein, followed by late myelin proteins such as myelin oligodendrocyte glycoprotein and myelin-associated oligodendrocytic basic protein1. Oligodendrocyte maturation is in part intrinsically programmed, as shown by cell-autonomous oligodendrocyte development in the absence of neurons2. However, neuronal contact enhances myelin gene expression3. These data suggest that secreted or membrane-associated neuron-derived factors regulate oligodendrocyte differentiation. Some of these have been identified, including insulin growth factors (IGFs) and their antagonists, the insulin growth factor binding proteins (IGFBPs)3–5. These signals regulate gene activity to generate precisely adjusted amounts and compositions of myelin proteins. So far, many transcription factors have been shown to have important functions in oligodendrocytes6,7. In contrast, there is lack of knowledge regarding the

transcription factors that control the neuron-derived signals that are involved in oligodendrocyte maturation. We found that neuron-restricted SRF ablation interfered with oligodendrocyte development in a non–cell autonomous manner. SRF, a MADS box transcription factor, recognizes CArG boxes (CC(AT)6GG) in promoters. SRF has been implicated in both gene activation8 and repression9–11, depending on, for example, cell type– specific cofactors and repressors. SRF elicits the immediate early gene response, resulting in rapid and transient induction of, for example, c-fos, Cyr61 and Egr1. Furthermore, SRF controls cytoskeletal genes (such as Actb and Gsn), which give SRF access to cytoskeletal regulation and in turn exert feedback on SRF activity8. Conditional Srf mouse mutagenesis using Cre recombinase driven by the Camk2a promoter, resulting in neuron-exclusive excision of a loxP-flanked Srf allele, reveals SRF functions in neuronal migration, axonal pathfinding and synaptic transmission12–16. Genome-wide transcriptomics of Srf
/
; Camk2a-iCre mutant forebrains revealed an unanticipated downregulation of a gene set that is associated with oligodendrocyte development. The number of mature oligodendrocytes was reduced in Srf mutants both in vivo and in vitro, whereas the amount of OPC markers that we detected was increased. We used transcriptomics to identify SRF-responsive genes and found Ctgf, whose expression is repressed by wild-type SRF. Using adenoviral CTGF delivery in vivo, which resulted in impaired oligodendrocyte development, we found that CTGF blocked excessive

1Neuronal

Gene Expression Laboratory, Eberhard Karls University Tu¨bingen, Interfaculty Institute for Cell Biology, Department of Molecular Biology, Tu¨bingen, Germany. Planck Institute for Molecular Genetics, Department of Computational Molecular Biology, Berlin, Germany. 3Max Planck Institute for Developmental Biology, Microscopy Unit, Tu¨bingen, Germany. 4Molecular Organ Function Laboratory, Eberhard Karls University Tu¨bingen, Interfaculty Institute for Cell Biology, Department of Molecular Biology, Tu¨bingen, Germany. 5These authors contributed equally to this work. Correspondence should be addressed to B.K. ([email protected]). 2Max

Received 28 October 2008; accepted 20 January 2009; published online 8 March 2009; doi:10.1038/nn.2280

418

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

Actb Adamts4 Arid4b Aspa Atm Cldn11 Cnksr3 Cnp1 Ddit4l Dsp Edg2 Elovl1 Elovl7 Enpp6 Erbb3 Fa2h Fbxo30 Fzd2 Gamt Gdpd3 Gja12 Gsn Hes5 Mag Mal Manba Mbp Mobp Mog Nfasc Pdlim2 Plekhh1 Plp1 Ppp1r14a Selpl Sox11 Srf Tmem10 Tshz2 Tspan2 Ttc14 Ugt8a Wfs1 Xlr3a

2.5 4.1 2.1 2.2 3.4 4.1 3.1 3.4 3.7 2.6 2.2 2.0 2.7 2.7 4.2 3.4 2.4 2.2 2.1 2.2 2.7 2.9 2.4 7.1 11.4 2.5 6.6 15.3 11.1 2.3 2.6 2.8 5.6 4.5 2.4 3.8 2.4 11.9 3.1 3.3 2.1 6.2 2.3 3.1

4 0 0 2 2 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 2 0 0 0 0 0 0 1 0 0 1 2 0 0 0 4 0 0 2 0 0 1 0

0.002 0.510 0.961 0.006 0.301 0.923 0.853 0.033 0.336 0.541 0.420 0.085 0.646 0.401 0.372 0.717 0.508 0.678 0.832 0.946 0.371 0.070 0.495 0.872 0.723 0.954 0.888 0.550 0.078 0.783 0.430 0.306 0.008 0.130 0.313 0.946 0.0001 0.572 0.271 0.201 0.857 0.642 0.037 1

b WT A

Gene

KO B

A

B A2m Adarb2 Arrdc4 Arsb C4 Ccl12 Cirbp Dncl2b Edg3 Gfap Glcci1 Glycam1 Gna13 Irs4 Lgals3 Lsm8 Lyzs Mgst1 Mrpl15 Rcor3 Rshl3 S100a4 Sfrs7 T2bp Thbs4 Tlk2 Txnip Ugt1a2 Xlkd1

c

oligodendrocyte differentiation, which, to the best of our knowledge, is the first demonstration of a function for CTGF in the brain. In vitro, this CTGF-mediated block on oligodendrocyte maturation involved antagonizing IGF1-stimulated oligodendrocyte differentiation. RESULTS To unravel neuronal SRF functions, both us12,14 and other groups13,15 have established neuron-exclusive, forebrain-specific conditional Srf mouse mutants. The Srf mutants that we created suffer from weight loss and motor dysfunction, which results in death at around postnatal day 21 (P21)12,14. Myelin gene expression is impaired in Srf mutants To identify genome-wide SRF target genes in the brain, we used GeneChip technology to examine RNA that was isolated from P14 hippocampi (Fig. 1 and Supplementary Table 1 and Supplementary Methods online). We found that 44 out of 39,000 transcripts were downregulated (Fig. 1a) and 29 were upregulated (Fig. 1b) in our Srf mutants12,14, which were used throughout the study. The downregulated genes included the well-known SRF target genes Actb, Gsn and Srf17,18. Notably, an entire gene group (20 out of 44) encoded myelin structural proteins or products that are associated with oligodendrocyte differentiation (Fig. 1a). In fact, 16 of these 20 were amongst the 40 genes that were most specific for oligodendrocytes19. In particular, structural myelin genes (Plp1, Mbp, Mag, Mog, Mobp and Mal) were the most affected by SRF deficiency. In addition, the expression of genes with established oligodendrocyte function (Ugt8a, Cldn11 and Hes5) and of oligodendrocyte-enriched genes of unknown function (such as Cnp, Tspan2, Tmem10 (also known as Opalin), Aspa, Edg2 (also known as Bud31), Fa2h and Plekhh1) was impaired19,20. The upregulated transcripts did not include genes with obvious oligodendrocyte annota-

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

Fold change (up) 2.9 2.2 2.7 2.4 5.5 3.4 2.4 2.7 2.2 2.9 3.1 4.5 2.4 4.5 3.0 5.7 2.3 2.3 3.4 2.4 2.2 4.7 2.2 3.1 2.3 2.2 2.4 2.7 2.1

SRF Conserved affinity hits (P value) (#) 0.416 0.676 0.282 0.214 0.954 0.805 0.013 0.901 0.413 0.452 0.347 1 0.657 0.443 0.491 0.931 0.206 0.289 0.921 0.569 0.986 0.274 0.939 0.955 0.848 0.317 0.766 0.655 0.144

0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Figure 1 Genome-wide profiling of SRF target genes in the brain. We hybridized RNA prepared from P14 hippocampi of control and Srf mutant mice (each in doublet A, B) to Affymetrix GeneChips. Only genes with at least twofold change and a statistical significance of r0.05 are depicted. The relative SRF affinity (P value) to given CArG box sequences is indicated with the lowest values predicting highest affinity. In addition, the number of species-conserved SRF binding sites is listed (see also Supplementary Table 1). (a) Depicted are 44 transcripts that are downregulated in Srf mutants, with the respective fold change. Myelin structural genes or genes associated with myelin function are highlighted in red. KO, knockout; WT, wild type. (b) We found 29 transcripts that were upregulated in Srf mutants as compared with control. (c) The scale indicating the relative signal strength normalized to internal standards of the color-coded bars is shown. High expression levels are indicated by red colors, whereas blue colors represent lower expression levels.

tion. However, the astrocyte marker gene Gfap (glial fibrillary acidic protein) was upregulated. To distinguish between direct and indirect SRF target genes, we ranked the affinity of SRF for each gene’s promoter21,22 (lower values predict highest SRF affinity) and determined species-conserved SRF binding sites (Fig. 1 and Supplementary Table 1). Indeed, SRF target gene promoters (Actb and Srf) had the strongest SRF affinity. We validated myelin gene expression in independent RNA and protein samples from hippocampi (n Z 3 mice, each genotype; Supplementary Fig. 1 online) and cortices (data not shown), and found reduced expression of all major myelin structural genes following SRF ablation. As ablation of SRF in the present mouse model is exclusive to neurons, downregulation of oligodendrocyte-specific transcripts indicates that a previously unknown non–cell autonomous regulation of oligodendrocyte development by neuronal SRF is occurring. To ascertain the exclusive nature of neuronal Srf deletion, we examined the expression of SRF and Cre recombinase driven by the Camk2a promoter. Consistent with previous reports12,13,15,23, Cre recombinase was absent from white-matter areas (Supplementary Fig. 2 online). In vitro, SRF levels decreased following Cre-mediated recombination in Srf mutant neurons but not in oligodendrocytes (Supplementary Fig. 2). Overall, our data argue that Srf
/
; Cam2a-iCre mice are well suited to investigate the effect of impaired neuronal signaling on oligodendrocyte development caused by SRF deficiency.

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0

B

SRF Conserved affinity hits (P value) (#)

1.2

A

Fold change (down)

1.5

B

Gene

2.0

© 2009 Nature America, Inc. All rights reserved.

A

KO

2.5

WT

3.0

a

SRF controls astrocyte number and axon integrity To examine axon integrity and the surrounding myelin, we used immunohistochemistry in the corpus callosum and cortex (n Z 3, each genotype; Fig. 2). MBP (Fig. 2a,b) and PLP (Fig. 2c,d) were reduced tenfold (10.6 ± 4.4, P o 0.0001) and 2.5-fold (2.45 ± 0.9, P ¼ 0.04), respectively, in Srf mutants. Next, we analyzed OPC markers that either continued to be expressed in oligodendrocytes (OLIG1, OLIG2 and SOX10)7,19,24 or remained restricted to OPCs (NG2 and platelet-derived growth factor receptor a)1,24,25. At P5, the expression of all OPC markers was elevated in Srf mutants (Supplementary Fig. 3 online). At P16, OLIG1, OLIG2 and SOX10 expression were not altered, whereas PDGFRa (Supplementary Fig. 4 online) and particularly NG2 (wild type, 10,500 ± 7,700 pixels in the NG2-positive

419

ARTICLES Control

–/–

; Camk2a-iCre

b

MBP/DAPI

a

Srf

SVZ

d

e

f

© 2009 Nature America, Inc. All rights reserved.

NG2/DAPI

PLP/NFAP

c

g

h

NFAP

CC HC

j

k

l

GFAP/DAPI

APP

i

area; Srf
/
, 86,000 ± 29,000; P ¼ 0.01; Fig. 2e,f) remain upregulated in Srf mutants. Neuronal SRF ablation impinges on axon formation14 and myelination defects might therefore be a result of axonal target deprivation. We stained for neurofilament-associated protein (NFAP; Fig. 2g,h) and bIII-tubulin (Supplementary Fig. 5 online) expression and found no gross alteration in corpus callosum diameter (diameter for NFAP: wild type, 234 ± 42 mm; Srf
/
, 208 ± 50 mm). We used markers that are associated with axonal pathologies to further assess axonal degeneration. Amyloid precursor protein (APP) accumulation is associated with disturbed axonal transport and results in axonal swellings and degeneration (for example, see ref. 26). Almost no APP was observed

420

Figure 2 SRF depletion affects oligodendrocyte differentiation in vivo. (a,b) P14 control (a) and Srf mutant (b) forebrains were stained for MBP and DAPI. MBP expression in the corpus callosum was strongly reduced in Srf mutants. As previously reported12, the SVZ, as shown by accumulation of DAPI-positive cells, was widened in Srf mutants. (c,d) PLP expression was markedly reduced in Srf mutants (d) compared with controls (c). The inserts indicate the presence of NFAP-positive axons. (e,f) At P16, NG2-positive OPCs were increased in Srf mutants (f) compared with controls (e). The inserts show NG2 staining of individual cells. (g,h) We used NFAP to label all axons. The diameters of the corpus callosum (CC) and hippocampal commissure (HC) were not altered between controls (g) and Srf mutants (h). (i,j) APP, a marker for axonal swellings, was induced in the P14 corpus callosum (dashed lines) of Srf mutants (j), but not in controls (i). Note that APP in Srf mutants (j) was more prominent in the corpus callosum next to the typically enlarged ventricle (black arrows) than in the neighboring corpus callosum region (white arrows). Inserts are higher magnifications of boxes showing APP-positive signals in mutants (j), but not control (i). (k,l) GFAPpositive astrocytes spread throughout almost all of the layers of the Srf mutant cortex (l), whereas only few cells were present in controls at P14 (k). Scale bars represent 200 mm (a,b,e–h,k,l), 50 mm (c,d) and 100 mm (i,j), and represent 5 mm (e,f) and 20 mm (i,j) in the inserts.

in wild-type mice, whereas a 100-fold increase was seen in Srf mutants (wild type, 3.6 ± 6.3 cells per 0.5 mm2; Srf
/
, 427.5 ± 159 cells per 0.5 mm2; P ¼ 0.0023; Fig. 2i,j). In addition, nonphosphorylated neurofilament levels, which are upregulated in many axonal pathologies (for example, see ref. 27), were increased in Srf mutants (Supplementary Fig. 5). We used periodic acid–Schiff staining, which visualizes carbohydrate-rich material, as an indirect marker for phagocytosing cells26. The numbers of periodic acid–Schiff–positive deposits and microglia were elevated in Srf mutants (Supplementary Fig. 5). Finally, we addressed astrocyte fate using GFAP and the recently identified pan-astrocyte marker Aldh1L1 (ref. 19). The amount of each marker that was present at P5 (Supplementary Fig. 3) and P16 (wild type, 12.2 ± 12 cells per 0.5 mm2; Srf
/
, 110 ± 9 cells per 0.5 mm2; P ¼ 0.00037; Fig. 2k,l and Supplementary Fig. 4) was elevated, indicating increased astrocyte numbers in Srf mutants. We inspected myelination by electron microscopy. Because Srf mutants die at BP21, we were limited to a rather early time point and had to compare age-matched animals between P14–19 (n ¼ 5, each genotype; Fig. 3). We focused on two corpus callosum positions: the corpus callosum dorsally to the striatum or the hippocampus. In controls (Fig. 3a,c), 50–60% of axons were myelinated (from a total of 3,202 and 1,124 axons for corpus callosum dorsally to the hippocampus and striatum, respectively), whereas myelination in Srf mutants was decreased to 10–20% (3,276 and 1,052 axons for corpus callosum dorsally to the hippocampus and striatum, respectively; Fig. 3b,d,g). Notably, large-diameter axons (42 mm) of Srf mutants more frequently lacked myelin than smaller axons (wild type, 66 ± 10% myelinated; Srf
/
, 36 ± 18% myelinated; P ¼ 0.01). In addition, we found no obvious alterations in periodicity and spacing between individual myelin layers (Fig. 3e,f). The g ratios (the ratio of the axon diameter to the combined diameter of the axone and myelin) of wild-type axons (0.93 ± 0.03) and the remaining myelinated axons in Srf mutants were comparable (0.92 ± 0.03; Fig. 3h). This suggests that it is the myelination onset that is affected in Srf mutants rather than the degree of myelination in remaining ensheathed axons. In sum, myelination in the Srf mutant corpus callosum was reduced. SRF regulates oligodendrocyte numbers in vitro To obtain mechanistic insight into SRF’s role in oligodendrocyte maturation, we prepared oligodendrocyte-enriched cultures28. We plated OPCs that were derived from control and Srf mutant mice

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES Figure 3 Electron-microscopic analysis of hypomyelination in the Srf mutant corpus callosum. (a–d) Low (a,b) and high (c,d) magnification views of ultrathin sections derived from control (a,c) and Srf mutant (b,d) corpus callosum. In controls (a,c), the majority of axons were surrounded by myelin sheaths (arrows). In contrast, fewer axons were myelinated in Srf mutants (b,d), with large-caliber axons (asterisks) being more affected than smaller ones. (e,f) Images represent individual myelin sheaths of controls (e) and Srf mutants (f). With regard to periodicity and spacing, we did not observe obvious alterations in the remaining Srf mutant myelin. (g) Average percentage of myelinated axons in controls and Srf mutants. Independent specimens from anterior and posterior (dorsal to striatum and hippocampus, respectively) corpus callosum were evaluated. (h) g ratios of the control and remaining myelinated Srf mutant fibers were comparable. Error bars represent s.d. Scale bars represent 5 mm (a,b), 2 mm (c,d) and 75 nm (e,f). * P r 0.05 and *** P r 0.001.

Srf –/–; Camk2a-iCre

Control

a

b *

* * *

c

d *

*

e

f

g

1.05

60

WT KO

1.00

50

0.95 g ratio

Myelinated axons (%)

h 1.10

WT KO

70

40 30

***

20

*

0.90 0.85 0.80

10

0.75

ca

ria

m

tu

pu

s

m

0

0.70 0.00

1.00

2.00 3.00 4.00 5.00 Axon diameter (µm)

6.00

7.00

H

ip

po

St

© 2009 Nature America, Inc. All rights reserved.

*

(n Z 15 mice; each genotype) for 9 d in vitro (DIV) in differentiation medium (Fig. 4). Soon after plating, wild-type cells assembled in clusters (Fig. 4a), where they differentiated into CNPase-positive oligodendrocytes (Fig. 4e). In contrast, Srf mutant cells grew in isolation (Fig. 4b,d) and, although we initially plated an identical number of cells, essentially produced no oligodendrocytes (CNPase/ DAPI ratio for wild type, 0.51 ± 0.22 s.e.m.; Srf
/
, 0.008 ± 0.003 s.e.m.; P ¼ 0.044; Fig. 4f). OLIG2-positive cells were not substantially reduced in Srf mutants (OLIG2/DAPI ratio for wild type, 0.24 ± 0.04 s.e.m.; Srf
/
, 0.16 ± 0.03 s.e.m.; Fig. 4c–f). Notably, GFAP-positive astrocytes were increased fivefold in Srf mutants (550 ± 50%; P ¼ 0.037; Fig. 4g,h). In light of a non–cell autonomous SRF function in oligodendrocyte development, the oligodendrocyte reduction in Srf mutant oligodendrocyte-enriched cultures might appear to be unexpected. However, even in Srf mutant oligodendrocyte-enriched cultures, OPCs were influenced for a considerable time by neurons lacking SRF (that is, in vivo until cortex dissection and in vitro for the first few days until neurons, initially part of the culture, had died). In this period of OPC-neuron interaction, OPCs may have encountered secreted or

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

membrane-associated neuronal factors that were induced by SRF deficiency and which were capable of preventing oligodendrocyte differentiation (for example, CTGF; see below). In addition, astrocyte numbers were elevated in Srf mutant oligodendrocyte-enriched cultures (Fig. 4h), and this increase might have contributed to the inhibition of oligodendrocyte maturation that we observed in Srf mutant oligodendrocyte-enriched culture. We next employed hippocampal mixed cultures that largely conserved the numbers of and interactions taking place between neurons, oligodendrocytes and astrocytes (Fig. 5). After 5 and 12 DIV, we assessed the number of CNPase-, MBP- and OLIG2-positive cells in relation to the number of DAPI-positive cells (n Z 8 mice for each condition and genotype). The number of OLIG2-positive cells was not substantially reduced by SRF deficiency (Fig. 5a,b,e). In contrast, the number of CNPase- and MBP-positive oligodendrocytes was decreased twofold after 5 DIVand fivefold after 12 DIV in Srf mutants (Fig. 5c–e). In addition, the control oligodendrocytes took up twice the area that the Srf mutant oligodendrocytes occupied (data not shown). How are SRF-deficient neurons affecting oligodendrocytes? Because only neurons were SRF deficient in Srf mutant mixed cultures (Supplementary Fig. 2), the results that we obtained can be explained by a paracrine mechanism that involves either disturbed cell-cell communication and/or secretion of signaling molecule(s). To test perturbed cell-cell communication, we cocultured wild-type cells (unlabeled) with labeled wild-type or Srf mutant cells in a 1:3 ratio to ensure extensive cell-cell contact. After 4 DIV, we quantified the effect of 4-(4-(dihexadecylamino)styryl)-N-methylpyridinium iodine (DiAsp)-labeled cells (wild type or mutant) on oligodendrocyte differentiation of wild-type, unlabeled cells. We found that the number of unlabeled, wild-type, CNPase-positive oligodendrocytes was reduced twofold when grown next to Srf mutant cells (Fig. 5f). These data indicate that molecules associated with Srf mutant cells are affecting wild-type oligodendrocyte differentiation. To obtain further evidence for SRF regulating oligodendrocyte differentiation non–cell autonomously, we used a lentivirus29 that drives SRF-VP16–IRES–green fluorescent protein (GFP) expression via the Camk2a promoter in neurons. SRF-VP16 is a constitutively active fusion protein of SRF and the viral protein VP16’s transactivation domain, with the latter replacing the wild-type SRF transactivation domain. As a control, we used SRF-VP16DMADS–IRES-GFP, which partially lacks SRF’s DNA binding domain. The constitutive transcriptional activator SRF-VP16 is highly useful for identifying SRF-regulated genes. Notably, SRF-VP16 is not subject to the endogenous control mechanisms that operate on wild-type SRF and therefore cannot provide insight as to whether such target genes are activated or repressed by wild-type SRF in vivo (see Discussion). SRF-VP16 expression in neurons affected neighboring oligodendrocytes in vitro.

421

ARTICLES

h

In contrast with results obtained in vivo (Figs. 1–3 and Supplementary Fig. 1), SRF-VP16 suppressed oligodendrocyte maturation (SRFVP16DMADS, 0.09 ± 0.009 s.e.m. CNPase positive:DAPI ratio; SRFVP16, 0.03 ± 0.01 s.e.m.; n ¼ 5 experiments, P ¼ 0.039). Similarly, electroporation of a Srf-Vp16 expression vector resulted in a reduction in the number of CNPase-positive cells (SRF-VP16DMADS, 0.19 ± 0.026 s.e.m. CNPase positive:DAPI ratio; SRF-VP16, 0.07 ± 0.007 s.e.m.; n ¼ 3, P ¼ 0.0008). Ctgf is a SRF target gene that is relevant to myelination SRF-VP16 might influence oligodendrocytes by regulating neuron-tooligodendrocyte cell-cell contact and/or by providing a neuron-derived secreted factor that is recognized by oligodendrocytes. To obtain global insight into SRF-regulated neuronal genes, we carried out microarrays using wild-type cortical neurons that expressed SRF-VP16. These data confirmed that well-known target genes of SRF were upregulated (for

Figure 5 Loss of neuronal SRF affects oligodendrocyte number in mixed neuron/glia cultures. (a,b) Hippocampal cultures derived from control (a) and Srf mutant (b) brains were stained for bIII-tubulin (green) to visualize the entire cell population. Staining for OLIG2 (red) revealed a slight reduction in the number of oligodendrocytes in Srf mutants (see e). (c,d) CNPase-labeled oligodendrocytes with well-elaborated protrusions in control cultures (c). CNPase-positive cells were decreased in number and area in Srf mutants (d), with longer culturing times being more severe (see e). (e) Quantification of MBP-, CNPase- and OLIG2-positive cells after 5 and 12 DIV, with control numbers set to 100%. We normalized to the number of DAPI-positive cells. (f) Coculturing Srf mutant, but not wild type (labeled green with DiAsp) cells with unlabeled wild-type cells decreased the number of CNPase-positive wildtype cells (expressed as ratio of DiAsp-negative, CNPase-positive cells:DAPI) compared with wild-type/wild-type cocultures. Error bars represent s.e.m. Scale bar represents 50 mm. * P r 0.05 and ** P r 0.01.

422

Control

Srf

a

b

c

d

–/–

; Camk2a-iCre

e

100 80

WT KO

*

**

60

f

14

–

g

12

+

β tubulin/αOLIG2/DAPI CNPase/αOLIG2/DAPI

f

GFAP/Actin/DAPI

© 2009 Nature America, Inc. All rights reserved.

e

example, Acta1, Egr1 and c-fos), along with many previously unknown (neuron associated) putative SRF target genes (Supplementary Fig. 6 and Table 2 online). As mentioned before, SRF-VP16 is useful for examining the general responsiveness of genes toward SRF; however, it remains to be investigated separately whether a given target gene is activated or repressed by wild-type SRF. Neuronal SRF target genes that are responsible for a paracrine role of SRF in oligodendrocyte development should be either secreted or membrane attached, their localization should be physiologically relevant to oligodendrocyte development and their basal expression should be altered between genotypes. CTGF fulfills all three of these criteria and we therefore examined whether it is indeed a SRF target gene that is potentially involved in oligodendrocyte maturation. CTGF, a CCN (CTGF/CYR61/NOV) family member, is a secreted factor that is associated with the extracellular matrix30–32. Ctgf has been reported as being a SRF target gene outside of the brain17,18,33. Ctgf expression was increased 10.6-fold by SRF-VP16 in our microarray (Supplementary Fig. 6), which we confirmed by quantitative PCR (data not shown). Basal Ctgf levels were altered between genotypes; Ctgf expression was increased in P14 Srf mutant forebrain (wild

CNPase and DiAsp + cells per 0.5 mm

d

β tubulin/OLIG2

c

CNPase/DAPI

b

β tubulin/DAPI

a

Figure 4 Absence of oligodendrocytes in Srf mutant oligodendrocyte-enriched cultures. (a–h) Control (a,c,e,g) and Srf mutant (b,d,f,h) OPCs were cultured in differentiation medium followed by immunocytochemistry. Low (a,b) and high (c,d) magnification views are shown of control (a,c) and Srf mutant (b,d) cultures stained for bIII-tubulin (a–d) to visualize all cells and OLIG2 (c,d). Typically, control oligodendrocytes (a,c) assembled in clusters (dashed lines in a), whereas cells in Srf mutants (b) remained isolated from each other. DAPI intensity was reduced for better visualization of nuclear OLIG2 localization in c and d. Control (e) and Srf mutant (f) cultures were stained for CNPase and OLIG2. In contrast with controls (e), almost no CNPasepositive oligodendrocytes were present in Srf mutant cultures (f). The insert in e is a higher magnification of the dashed area, revealing nuclear OLIG2 staining. Increased numbers of GFAP-positive astrocytes were seen in Srf mutant (h) cultures compared with controls (g). Scale bars represent 100 mm (a,b), 20 mm (c,d,g,h) and 50 mm (e,f), and 10 mm in the insert.

Positive cells/DAPI (%)

Srf –/–; Camk2a-iCre

Control

**

40

*

20 0 DIV

5

12

MBP

VOLUME 12

[

5

12

NUMBER 4

[

8

**

6 4 2

12

0 Cell A

WT

WT

OLIG2

Cell B

WT

KO

5

CNPase

10

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES Control

Srf

a

–/–

; Camk2a-iCre

b

CTGF

Cortex

CC

d

CTGF

c

CC

e

Control

Srf

CC

f

–/–

; Camk2a-cre

hSRF –

CTGF

GAPDH

*** *

h

**

* * ** 0 0 0 7.5 15 30 7.5 15 30 30 100 300 0 0 0 300 300 300

i 0.06

WT KO

0.05 0.04

*

0.03 0.02 0.01

0 SRF-Eng –

–

Ctgf expression level (rel. units)

5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0 SRF-VP16 (ng) 0 SRF-Eng (ng) 0

Ctgf expression level (rel. units)

g

Ctgf luciferase activity (rel. units)

© 2009 Nature America, Inc. All rights reserved.

CC

+

type, 0.005 ± 0.002; Srf
/
, 0.014 ± 0.004; P ¼ 0.037) and cortical cultures (wild type, 0.026 ± 0.005; Srf
/
, 0.08 ± 0.037; P ¼ 0.045, n ¼ 3 mice; see below). Although CTGF was uniformly expressed in culture (data not shown), localization of CTGF is restricted to cortical layer VI in vivo34, which we confirmed in control mice (Fig. 6). In Srf mutants, the number of CTGF-positive cells in this area was increased and the CTGF-positive cells were more dispersed (wild type, 110 ± 25 cells per 0.5 mm2; Srf
/
, 190 ± 12; P ¼ 0.0089, n ¼ 3; Fig. 6b,d). In addition, we found ectopic CTGF-positive cells in the striatum (data not shown) and upper cortical layers of Srf mutants (wild type, 1.3 ± 0.6 cells per 0.5 mm2; Srf
/
, 21 ± 1.5; P o 0.0001). CTGF was also upregulated in Srf mutant cortex protein lysates (Fig. 6e). These data suggest that SRF represses the Ctgf promoter. Accordingly, in SRF-deficient embryonic stem cells35, Ctgf promoter–driven luciferase activity was increased almost threefold compared with control (2.7 ± 0.4, P ¼ 0.0026, n ¼ 3). In addition to our SRF loss-of-function data (Fig. 6a–e), SRF gain of function

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

Figure 6 SRF represses CTGF in vivo and in vitro. (a–d) P14 brain sections of control (a,c) and Srf mutants (b,d) were stained for CTGF. In Srf mutants (b), we noted an increase in the numbers and dispersion of CTGF-positive cells dorsally to the corpus callosum compared with controls (a). (c,d) Higher magnifications of areas in boxes from a and b. Dashed lines mark the dorsal corpus callosum border. (e) CTGF levels in Srf mutants were higher in P14 brain lysates. (f) Overexpression of wild-type SRF suppressed CTGF levels compared with mock-transfected (
) SRF-deficient embryonic stem cells. (g) SRF fused to the engrailed repressor domain (SRF-Eng) reduced Ctgf reporter gene activity and counteracted SRF-VP16–mediated upregulation of Ctgf activity. (h) SRF-Eng repressed Ctgf levels in Srf mutant cortical neurons, resulting in Ctgf levels that were comparable to those of wild-type controls. (i) Stimulation of wild-type cortical cultures with TGFb resulted in the induction of Ctgf RNA, in contrast with Srf mutant cultures. Note that in the absence of stimulation, Ctgf levels were higher in Srf mutants. Error bars represent s.d. Scale bars represent 50 mm (a,b) and 30 mm (c,d). * P r 0.05, ** P r 0.01 and *** P r 0.001. n.s., not significant (P 4 0.05).

resulted in decreased CTGF expression levels (Fig. 6f). In addition, we constructed a fusion protein of SRF and the D. melanogaster engrailed repressor domain36, which repressed Ctgf (and c-fos; data not shown) CTGF promoter activity and counteracted SRFVP16–mediated Ctgf activation (Fig. 6g). SRF SRF-engrailed–mediated repression was GAPDH abolished by mutating the c-fos CArG box (data not shown). Furthermore, SRFengrailed suppressed Ctgf expression in Srf WT 0.12 n.s. KO mutant cortical neurons, resulting in Ctgf 0.1 levels that were comparable to those of wild 0.08 * type (n ¼ 6 mice; Fig. 6h). Next, we stimu0.06 lated neurons with transforming growth 0.04 factor b (TGFb), a potent Ctgf inducer out0.02 side of the brain30 and a physiologically 0 relevant myelination modulator1. In wildTGFβ – + – + type neurons, TGFb stimulated Ctgf expression. This Ctgf activation was SRF dependent, as TGFb failed to increase Ctgf expression in Srf mutant neurons (n ¼ 4 experiments; Fig. 6i). Without stimulation, Ctgf in Srf mutants was elevated, which is consistent with SRF repressing Ctgf in a basal state (Fig. 6i and above). Together, our data suggest that SRF operates as a Ctgf repressor in wild-type mice. These results superficially appear to be at odds with Ctgf stimulation by SRF-VP16, but it has to be taken into account that promoter occupation by the artificial and constitutively active transcriptional stimulator SRF-VP16 results in unrestrained transcriptional activation. In contrast, wild-type SRF, mediating either gene activation or repression, is subject to the stringent physiological regulation machinery that operates in vivo. CTGF application interferes with myelination in vivo So far, no function has been attributed to CTGF in the brain and Ctgf mutants die perinatally37, precluding analysis on CNS oligodendrocyte

423

ARTICLES

a

Virus injection (P1)

Ad-CTGF

Ad-GFP Histology (P6–12) αCTGF

GFP DAPI

Ad-GFP

Ad-CTGF

c

d

e

GFP

© 2009 Nature America, Inc. All rights reserved.

CTGF

b

MBP

f

g

P6

h

i

P12

development. The potential of SRF to repress Ctgf (Fig. 6) suggests that elevated CTGF expression might inhibit oligodendrocyte development in Srf mutants. To test this in vivo, we used an adenovirus to increase CTGF expression (Fig. 7). We injected mice with an adenovirus expressing CTGF (Ad-CTGF) or GFP as a control (Ad-GFP). We targeted CTGF (or GFP) expression ectopically around the lateral ventricle and ventrally to the corpus callosum; both regions were devoid of endogenous CTGF (Fig. 7a–e; see also Fig. 6). Notably, CTGF (or GFP)-expressing cells reproducibly localized in a Y-shaped pattern ventrally to the midline corpus callosum (Fig. 7a,c,d). This allowed us to monitor the consequences of CTGF application on oligodendrocyte development in this corpus callosum subregion. Ad-CTGF reduced the number of MBP-positive oligodendrocytes by almost 40% compared with controls at 6 d postinfection (Ad-GFP, 47.3 ± 5.2 s.e.m. MBP-positive cells per 0.5 mm2; Ad-CTGF, 28.8 ± 2.5 s.e.m.; P ¼ 0.0067, n ¼ 7 mice, each condition; Fig. 7f,g). At P12, we noticed an almost threefold decrease in MBP-positive oligodendrocytes in Ad-CTGF–injected mice (Ad-GFP, 34.1 ± 5.9 s.e.m. MBP-positive cells per 0.5 mm2; Ad-CTGF, 12 ± 2.3 s.e.m.; P ¼ 0.0025, n Z 8 mice, each condition; Fig. 7h,i). Myelination in brain regions (for example, fimbria) that were not targeted by Ad-CTGF was not obviously altered.

424

Figure 7 CTGF expression in vivo impairs oligodendrocyte development. (a) Experimental set-up for delivery of adenovirus expressing either GFP (control, Ad-GFP) or CTGF (Ad-CTGF). Top, head of a newborn mouse (P1) with approximate injection position of adenovirus solution relative to blood vessels (dotted lines). Bottom left, GFP distribution of an Ad-GFP–injected mouse on a P6 coronal brain section. Bottom right, distribution of endogenous (arrows) and exogenous (arrowheads) CTGF in an Ad-CTGF– injected mouse stained for CTGF expression at P6. The dashed boxes indicate the approximate position of the corpus callosum depicted in b–i. (b,c) We stained Ad-GFP– (b) or Ad-CTGF–injected (c) brains for CTGF expression at P6. Black arrows in b and c point to endogenous CTGF expression dorsally to the corpus callosum, whereas arrowheads point at exogenous, virally delivered CTGF expression (visible in c only; see also a). Notably, this ectopic CTGF expression typically formed a Y-shaped structure underneath the ventral corpus callosum border. (d,e) We visualized Ad-GFP– (d) or Ad-CTGF–injected (e) brains for GFP expression at P6. The pattern and number of GFP-positive cells of Ad-GFP–injected mice (d) were very similar to the CTGF distribution of Ad-CTGF–injected pups (compare with c). (f–i) Coronal sections of Ad-GFP– (f,h) or Ad-CTGF–injected (g,i) mice were stained for MBP at P6 (f,g) or P12 (h,i). The number of MBP-positive oligodendrocytes (arrows point to individual oligodendrocytes) in the corpus callosum was reduced in Ad-CTGF–injected mice (g,i) compared to controls (f,h). Dashed lines in a–i indicate the dorsal and ventral corpus callosum borders. Scale bars represent 1 mm (a) and 100 mm (b–i).

CTGF antagonizes IGF-mediated oligodendrocyte maturation To examine the mechanism of CTGF-mediated oligodendrocyte inhibition (Fig. 7), we prepared oligodendrocyte-enriched cultures (Fig. 8a). In control cultures, oligodendrocytes gathered in clusters, whereas incubation with CTGF interfered with colony formation (control, 38.6 ± 4.5 s.e.m. colonies per 100 mm2; CTGF, 13.8 ± 3.3 s.e.m.; P ¼ 0.0005, n ¼ 4 experiments). In addition, application of CTGF resulted in a concentration-dependent reduction in the number of oligodendrocytes in mixed neuronal/glial cultures but had no effect on astrocytes (n ¼ 4 experiments; Fig. 8b and data not shown). CTGF, also known as IGFBP8, contains an IGFBP domain binding to IGF1 (ref. 38). IGFs promote oligodendrocyte differention39,40, whereas IGFBPs antagonize oligodendrocyte differentiation by sequestering IGFs41,42. Accordingly, IGF1 alone increased oligodendrocyte numbers but did not affect astrocytes (n ¼ 4; Fig. 8c,d). We then asked whether CTGF could use its IGFBP domain to counteract this IGF1mediated stimulation of oligodendrocyte differentiation by controlling the amount of free IGF1 that is available to oligodendrocytes (Fig. 8c,e,f). Pre-incubation of IGF1 with CTGF antagonized IGF1– stimulated oligodendrocyte maturation by reducing oligodendrocyte number (Fig. 8e) and size (n ¼ 3 experiments; Fig. 8f). Astrocytes were not affected, indicating that CTGF specifically influences oligodendrocytes (Fig. 8e). Finally, we assessed the extent of CTGF’s contribution to the decrease in the number of oligodendrocytes in Srf mutants by immunodepleting CTGF with antibodies to it (Fig. 8g). The addition of CTGF antibodies to wild-type culture medium had no effect, whereas the number of oligodendrocytes was substantially elevated in Srf mutants (n ¼ 4 mice for wild type, 8 for Srf
/
). These data indicate that increased CTGF levels are a major source of the inhibition of oligodendrocyte differentiation that we observed in Srf mutants in vitro. SRF regulation of myelin gene expression SRF was expressed in oligodendrocytes in vivo (although weaker than in neurons; Supplementary Fig. 2). Srf
/
; Camk2a-iCre mice provide data indicating that there is non–cell autonomous regulation of oligodendrocyte differentiation by neuronal SRF. This does not exclude an additional SRF function in myelin gene expression in oligodendrocytes.

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

a

b

CTGF

CNPase/DAPI

CNPase-positive cells/DAPI

Control

c

*

*

–

1:100 1:50 1:25

IGF1

1:10

IGF1 + CTGF

CNPase/GFAP/DAPI

Control

0.45 0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0 CTGF

**

e

*

**

150 100 50

0 0 IGF1 (µg ml–1)

0.02

0.1

250

CNPase GFAP

200

***

150 100 50

0 IGF1 CTGFa CTGFb

0.5

f

– – –

+ – –

+ + –

g

CNPase-positive cells per 100 mm2

80 70 60 50 40 30 20 10 0 Anti-CTGF –

+ – +

WT KO n.s.

**

+

–

DISCUSSION Oligodendrocyte development by SRF involves CTGF In the Srf
/
; Camk2a-iCre mice that we used here, SRF ablation only occurred in neurons (Supplementary Fig. 2). Thus, all of the effects on oligodendrocytes involved an SRF function in neurons that was affecting macroglia in a non–cell autonomous manner. How would neuronal SRF mediate paracrine oligodendrocyte development? SRF could regulate secreted and/or membraneassociated signal(s) between neurons and OPCs/ oligodendrocytes. Consequently, SRF ablation could result in insufficient amounts of a stimulatory signal or elevated expression of an inhibitory signal. Ctgf, being transcriptionally repressed by SRF (Fig. 6), represents such an inhibitory signal. CTGF inhibited

[

NUMBER 4

[

***

+ – –

+ + –

+ – +

+

To test this, we performed luciferase assays for Cnp, Plp1, Mog, Tspan2, Hes5 and Mbp (Supplementary Fig. 7 online) in HEK293 cells using constitutively active SRF-VP16. We evaluated the data by comparing it with c-fos, a well-established SRF target gene, whose induction resulted in a 15-fold stimulation. In sum, SRF-VP16 drove luciferase expression from all of the promoter constructs that we used. These data in HEK293 cells reveal general myelin promoter responsiveness to SRF-VP16; however, caution should be taken when extrapolating these results, which were acquired in HEK293 cells, to the situation of myelin promoter regulation exerted by wild-type SRF in oligodendrocytes.

NATURE NEUROSCIENCE VOLUME 12

180 160 140 120 100 80 60 40 20 0 IGF1 – CTGFa – CTGFb –

Oligodendrocyte area (% of control)

200

CNPase GFAP

Percentage pos. cells rel. to control

250

Percentage pos. cells rel. to control

© 2009 Nature America, Inc. All rights reserved.

d

Figure 8 CTGF inhibits oligodendrocyte differentiation in vitro by counteracting IGF1. (a) We treated oligodendrocyte-enriched cultures with control fractions (left) or purified recombinant hCTGF (right). Cultures containing CTGF developed fewer CNPase-positive colonies than control treated cultures. (b) Similar to oligodendrocyte-enriched cultures (a), incubation with CTGF, compared with control fractions, reduced the number of CNPase-positive oligodendrocytes in a dose-dependent manner in mixed neuron/glia cell cultures. (c) Mixed neuron/ glia cultures derived from the hippocampus were incubated with insulin-free medium (left), IGF1containing medium (middle) or in medium with IGF1 that was pre-incubated with CTGF (right). IGF1 strongly increased oligodendrocyte differentiation, as shown by an increase in the number and area of CNPase-positive oligodendrocytes (middle). Conversely, preincubation with CTGF blocked IGF1-mediated stimulation of oligodendrocyte differentiation (right). (d) IGF1 increased the number of CNPasepositive oligodendrocytes, but not astrocytes, in mixed neuron/glia cultures in a concentrationdependent manner. (e,f) Quantification of CNPasepositive oligodendrocyte and GFAP-positive astrocyte numbers (e) or the area taken occupied by individual oligodendrocytes (f) after the various treatments. CTGFa, CTGF that we produced ourselves; CTGFb, CTGF that we obtained commercially (Biomol). (g) CTGF immunodepletion by the addition of antibodies to CTGF to the growth medium increased oligodendrocyte numbers in Srf mutant, but not control mixed neuron/glia cultures. Error bars represent s.d. Scale bars represent 50 mm (a,c). * P r 0.05, ** P r 0.01 and *** P r 0.001.

APRIL 2009

oligodendrocyte development in vivo and in vitro (Figs. 7 and 8). How could CTGF mediate this inhibition? CTGF has IGFBP-like functions that result in IGF1 sequestration38 and we found that IGF1-stimulated oligodendrocyte increase could be antagonized by CTGF (Supplementary Fig. 7). Because CTGF also binds to IGF2 (ref. 38), this mechanism probably also operates on IGF2stimulated oligodendrocyte differentiation. Notably, the IGF/ IGFBP signaling complex (for example, Igf2 and Igfbp6/7) was strongly responsive to SRF (Supplementary Table 2). In Srf mutants, myelination onset appeared to be affected (Fig. 3), indicating that CTGF, which also affects cell adhesion and migration31, might interfere with initial oligodendrocyte-neuron recognition. In wild-type mice, the numbers of oligodendrocytes that are available to contact axons are tightly adjusted by trophic or apoptosis-inducing signals. Neuronally derived CTGF might contribute toward this fine-tuning of oligodendrocyte numbers by constituting a barrier that limits IGF diffusion and availability for oligodendrocyte maturation locally in the corpus callosum area and thereby preventing excess oligodendrocyte numbers. SRF appears to adjust IGF signaling by repressing or activating (via TGFb) the IGF antagonist CTGF (Fig. 6). Our data indicate that CTGF is an important regulator of myelination in the corpus callosum. However, as CTGF is not expressed in all myelination areas, CTGF probably does not account for the full scope of myelination defects that are observed in Srf mutants.

425

© 2009 Nature America, Inc. All rights reserved.

ARTICLES SRF-mediated transcription in oligodendrocyte development SRF’s role in gene expression during oligodendrocyte development appears to be at least twofold. First, SRF might contribute to myelin gene activation in oligodendrocytes (Supplementary Fig. 7). Second, in neurons, SRF’s influence on oligodendrocyte differentiation appeared to at least partially result from a repressive function, as shown for Ctgf (and other genes; see below). In contrast, SRF-VP16 appeared to counteract the repressive function of SRF by inducing rather than by repressing, Ctgf. It should be noted that SRF-VP16, unlike wild-type SRF, is not subject to the control mechanisms operating in vivo. A similar antagonism on oligodendrocyte maturation as that observed for wild-type SRF versus SRF-VP16 was described for OLIG2. OLIG2VP16 blocks oligodendrocyte differentiation43, whereas OLIG mouse mutagenesis44,45 demonstrates their crucial requirement for proper oligodendrocyte development1. SRF has been associated with gene repression in previous reports9–11, which might be caused by SRF itself acting as a repressor, SRF acting in repressor recruitment or by directing micro-RNA synthesis. Notably, the expression of other SRF target genes, such as c-fos and Cyr61 (which also belongs to the CCN family), are likewise increased in Srf mutants (data not shown). This indicates that SRF has a selective repressor function in basal gene transcription, which has to be distinguished from SRF’s role in activating transcription on entry of exogenous stimuli (for example, TGFb), as we found for the Ctgf (Fig. 6h) and c-fos promoters (data not shown). In Srf mutants, the amount of astrocytes was increased and the amount of oligodendrocytes was decreased. Our data suggest that CTGF is not crucially involved in regulating this astrocyte/ oligodendrocyte macroglia balance (Fig. 8). Oligodendrocytes in the brain arise from an early embryonic phase in the ventricular zone and a later phase in the subventricular zone (SVZ)1,46. We previously described an enlarged SVZ that trapped neuronal precursors and thereby prevented migration along the rostral migratory stream12. Similarly, we observed elevated levels of oligodendrocyte progenitors in the corpus callosum and adjacent SVZ in Srf mutants (Fig. 2 and Supplementary Figs. 3 and 4). Thus, proper OPC migration might be impaired in Srf mutants, resulting in accumulation of OPCs that do not differentiate into mature oligodendrocytes. Notably, OPC accumulation has also been described in Yy1 mutants47, a transcription factor that interacts with SRF (for example, see ref. 48). This OPC accumulation in the Srf mutant SVZ might also impinge on the astrocyte-oligodendrocyte equilibrium. Therefore, SRF directs oligodendrocyte development by regulating CTGF/IGF signaling and by controlling the number of OPCs that are available for differentiation and astrocyte numbers. METHODS Conditional Srf mutants. We used Srf
/
; Camk2a-iCre mutant mice and Srf+/–; Camk2a-iCre or Srf+/–; Camk2a-iCre–negative animals as wild types12, in which Cre recombinase expression was induced in neurons by the Camk2a locus. The Srf gene was excised via loxP sites. The Regierungspra¨sidium Tu¨bingen (IM 2/08) approved all of the mouse experiments used in this study. Cell culture. We prepared mixed neuron/glia cultures from hippocampi as previously described14 and cortical cultures from embryonic day 17.5 embryos. For electroporation (Amaxa), we used 3 mg of DNA and plated 104 cells per 12-mm coverslip. Cultures were infected with lentivirus after 1 DIV29. We cultured embryonic stem cells as described previously35. In coculture experiments, we plated 104 wild-type cells with 3  104 wild-type or Srf mutant hippocampal cells, which were visualized with DiAsp (10 mg ml
1 in Hanks Balanced Salt Solution) on 12-mm coverslips. Oligodendrocyte-enriched cultures were prepared on the basis of a previous study28. Briefly, we plated cells from one pup in a 75-cm2 flask and grew them in DMEM with 10% fetal calf

426

serum (vol/vol) for 10 d. The flasks were then shaken vigorously (16–18 h) in the incubator. We filtered supernatants (70-mm sieve) and plated the cells on a bacterial-grade dish (30 min at 37 1C). We centrifuged supernatants (600 g for 10 min) and plated pellets on poly-L-lysine–coated coverslips in differentiation medium (DMEM with 0.5% fetal calf serum (vol/vol) and 10 ng ml
1 ciliary neurotrophic factor, Peprotech) for 9 d. We purified human HIS-tagged CTGF (a gift from M. Goppelt-Struebe, University of Erlangen) from HEK293 supernatants. We added 0.1–10 mg ml
1 of CTGF to hippocampal and oligodendrocyte-enriched cultures for 3 consecutive days. We added IGF1 (R&D systems) at 3 DIV in NMEM medium containing insulin-free B27 supplement (Gibco). For IGF1/CTGF experiments, we pre-incubated 0.02 mg ml
1 IGF1 in a volume of 50 ml for 30 min at 37 1C with 2 mg ml
1 CTGF that we produced ourselves or that was commercially available (Biomol). We stimulated cortical cultures for 1 h with 10 ng ml
1 TGFb1 or 2 (R&D systems). We added antibody to CTGF or control IgG (1 mg per 500 ml of medium) for 3 consecutive days to mixed hippocampal cultures. Virus injection. Ad-GFP and Ad-CTGF (Vector Biolabs) have been described previously49. We anaesthetized newborn mice by isofluorane inhalation and injected 1.5 ml of virus solution into the lateral ventricle using a 33-gauge Hamilton syringe (Fig. 7a). We controlled virus delivery by co-injection of a fast green solution. Only the mice with uniform dye diffusion into the entire lateral ventricle were analyzed. We dissected brains at P6 or P12 for histological inspection. The experimental procedure is in accordance with institutional guidelines (Regierungspra¨sidium Tu¨bingen; IM 2/08). Quantification and statistics. For quantification of histology, we used Slidebook software (Intelligent Imaging Innovations). To compare pictures, we kept segmentation and object size thresholds identical, and measured either the total number of defined objects (reflecting cell number) or the total area (in pixels) of all objects (reflecting area engaged by all cells). We assessed statistical significance by two-tailed Student’s t-test. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We are grateful to M. Schmid for help with GeneChips, J. Berger for electron microscopy, A. Sedlag for excellent student labwork and G. Schu¨tz for the Camk2a-cre mice. We thank M. Jucker, G. Schratt and A. Wizenmann for critically reading the manuscript. B.K. is supported by the DFG Emmy Noetherprogram, Sonderforschungsbereich 446, the Schram-Stiftung and by young investigator grants from Tu¨bingen University. A.N. is supported by the Deutsche Forschungsgemeinschaft (grant NO 120/12-2). AUTHOR CONTRIBUTIONS C.S. performed the experiments in Figures 1, 2, 6 and 7, in Supplementary Figures 1–7 and in Supplementary Tables 1 and 2. S.S. carried out the experiments in Figures 5, 6 and 8, and in Supplementary Figures 2 and 7. K.H. performed the experiments in Figures 4–6. B.K. carried out the experiments in Figures 3, 7 and 8. D.S. provided excellent technical assistance throughout. T.M. and M.V. provided bioinformatical analysis. H.S. supervised the electron microscopy. A.N. supplied Srf mutants and co-designed the microarrays. B.K. supervised the study, designed the experiments and wrote the manuscript. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/

1. Baumann, N. & Pham-Dinh, D. Biology of oligodendrocyte and myelin in the mammalian central nervous system. Physiol. Rev. 81, 871–927 (2001). 2. Durand, B. & Raff, M. A cell-intrinsic timer that operates during oligodendrocyte development. Bioessays 22, 64–71 (2000). 3. Bozzali, M. & Wrabetz, L. Axonal signals and oligodendrocyte differentiation. Neurochem. Res. 29, 979–988 (2004). 4. Chesik, D., De Keyser, J. & Wilczak, N. Insulin-like growth factor system regulates oligodendroglial cell behavior: therapeutic potential in CNS. J. Mol. Neurosci. 35, 81–90 (2008). 5. Simons, M. & Trajkovic, K. Neuron-glia communication in the control of oligodendrocyte function and myelin biogenesis. J. Cell Sci. 119, 4381–4389 (2006). 6. Nicolay, D.J., Doucette, J.R. & Nazarali, A.J. Transcriptional control of oligodendrogenesis. Glia 55, 1287–1299 (2007).

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

© 2009 Nature America, Inc. All rights reserved.

ARTICLES 7. Wegner, M. A matter of identity: transcriptional control in oligodendrocytes. J. Mol. Neurosci. 35, 3–12 (2008). 8. Posern, G. & Treisman, R. Actin’ together: serum response factor, its cofactors and the link to signal transduction. Trends Cell Biol. 16, 588–596 (2006). 9. Ernst, W.H., Janknecht, R., Cahill, M.A. & Nordheim, A. Transcriptional repression mediated by the serum response factor. FEBS Lett. 357, 45–49 (1995). 10. Rivera, V.M., Sheng, M. & Greenberg, M.E. The inner core of the serum response element mediates both the rapid induction and subsequent repression of c-fos transcription following serum stimulation. Genes Dev. 4, 255–268 (1990). 11. Shaw, P.E., Frasch, S. & Nordheim, A. Repression of c-fos transcription is mediated through p67SRF bound to the SRE. EMBO J. 8, 2567–2574 (1989). 12. Alberti, S. et al. Neuronal migration in the murine rostral migratory stream requires serum response factor. Proc. Natl. Acad. Sci. USA 102, 6148–6153 (2005). 13. Etkin, A. et al. A Role in Learning for SRF: Deletion in the adult forebrain disrupts LTD and the formation of an immediate memory of a novel context. Neuron 50, 127–143 (2006). 14. Knoll, B. et al. Serum response factor controls neuronal circuit assembly in the hippocampus. Nat. Neurosci. 9, 195–204 (2006). 15. Ramanan, N. et al. SRF mediates activity-induced gene expression and synaptic plasticity, but not neuronal viability. Nat. Neurosci. 8, 759–767 (2005). 16. Wickramasinghe, S.R. et al. Serum response factor mediates NGF-dependent target innervation by embryonic DRG sensory neurons. Neuron 58, 532–545 (2008). 17. Philippar, U. et al. The SRF target gene Fhl2 antagonizes RhoA/MAL-dependent activation of SRF. Mol. Cell 16, 867–880 (2004). 18. Sun, Q. et al. Defining the mammalian CArGome. Genome Res. 16, 197–207 (2006). 19. Cahoy, J.D. et al. A transcriptome database for astrocytes, neurons and oligodendrocytes: a new resource for understanding brain development and function. J. Neurosci. 28, 264–278 (2008). 20. Dugas, J.C., Tai, Y.C., Speed, T.P., Ngai, J. & Barres, B.A. Functional genomic analysis of oligodendrocyte differentiation. J. Neurosci. 26, 10967–10983 (2006). 21. Manke, T., Roider, H.G. & Vingron, M. Statistical modeling of transcription factor binding affinities predicts regulatory interactions. PLoS Comput. Biol. 4, e1000039 (2008). 22. Roider, H.G., Kanhere, A., Manke, T. & Vingron, M. Predicting transcription factor affinities to DNA from a biophysical model. Bioinformatics 23, 134–141 (2007). 23. Erdmann, G., Schutz, G. & Berger, S. Inducible gene inactivation in neurons of the adult mouse forebrain. BMC Neurosci. 8, 63 (2007). 24. Rivers, L.E. et al. PDGFRA/NG2 glia generate myelinating oligodendrocytes and piriform projection neurons in adult mice. Nat. Neurosci. 11, 1392–1401 (2008). 25. Stallcup, W.B. The NG2 proteoglycan: past insights and future prospects. J. Neurocytol. 31, 423–435 (2002). 26. Lappe-Siefke, C. et al. Disruption of Cnp1 uncouples oligodendroglial functions in axonal support and myelination. Nat. Genet. 33, 366–374 (2003). 27. Hulshagen, L. et al. Absence of functional peroxisomes from mouse CNS causes dysmyelination and axon degeneration. J. Neurosci. 28, 4015–4027 (2008). 28. McCarthy, K.D. & de Vellis, J. Preparation of separate astroglial and oligodendroglial cell cultures from rat cerebral tissue. J. Cell Biol. 85, 890–902 (1980).

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

29. Dittgen, T. et al. Lentivirus-based genetic manipulations of cortical neurons and their optical and electrophysiological monitoring in vivo. Proc. Natl. Acad. Sci. USA 101, 18206–18211 (2004). 30. Chaqour, B. & Goppelt-Struebe, M. Mechanical regulation of the Cyr61/CCN1 and CTGF/ CCN2 proteins. FEBS J. 273, 3639–3649 (2006). 31. Leask, A. & Abraham, D.J. All in the CCN family: essential matricellular signaling modulators emerge from the bunker. J. Cell Sci. 119, 4803–4810 (2006). 32. Perbal, B. CCN proteins: multifunctional signaling regulators. Lancet 363, 62–64 (2004). 33. Muehlich, S. et al. Actin-dependent regulation of connective tissue growth factor. Am. J. Physiol. Cell Physiol. 292, C1732–C1738 (2007). 34. Heuer, H. et al. Connective tissue growth factor: a novel marker of layer VII neurons in the rat cerebral cortex. Neuroscience 119, 43–52 (2003). 35. Schratt, G. et al. Serum response factor is crucial for actin cytoskeletal organization and focal adhesion assembly in embryonic stem cells. J. Cell Biol. 156, 737–750 (2002). 36. Vickers, E.R. & Sharrocks, A.D. The use of inducible engrailed fusion proteins to study the cellular functions of eukaryotic transcription factors. Methods 26, 270–280 (2002). 37. Ivkovic, S. et al. Connective tissue growth factor coordinates chondrogenesis and angiogenesis during skeletal development. Development 130, 2779–2791 (2003). 38. Kim, H.S. et al. Identification of a family of low-affinity insulin-like growth factor binding proteins (IGFBPs): characterization of connective tissue growth factor as a member of the IGFBP superfamily. Proc. Natl. Acad. Sci. USA 94, 12981–12986 (1997). 39. Hsieh, J. et al. IGF-I instructs multipotent adult neural progenitor cells to become oligodendrocytes. J. Cell Biol. 164, 111–122 (2004). 40. McMorris, F.A., Smith, T.M., DeSalvo, S. & Furlanetto, R.W. Insulin-like growth factor I/ somatomedin C: a potent inducer of oligodendrocyte development. Proc. Natl. Acad. Sci. USA 83, 822–826 (1986). 41. Kuhl, N.M., Hoekstra, D., De Vries, H. & De Keyser, J. Insulin-like growth factor-binding protein 6 inhibits survival and differentiation of rat oligodendrocyte precursor cells. Glia 44, 91–101 (2003). 42. Ye, P., Carson, J. & D’Ercole, A.J. In vivo actions of insulin-like growth factor-I (IGF-I) on brain myelination: studies of IGF-I and IGF binding protein-1 (IGFBP-1) transgenic mice. J. Neurosci. 15, 7344–7356 (1995). 43. Zhou, Q., Choi, G. & Anderson, D.J. The bHLH transcription factor Olig2 promotes oligodendrocyte differentiation in collaboration with Nkx2.2. Neuron 31, 791–807 (2001). 44. Lu, Q.R. et al. Common developmental requirement for Olig function indicates a motor neuron/oligodendrocyte connection. Cell 109, 75–86 (2002). 45. Zhou, Q. & Anderson, D.J. The bHLH transcription factors OLIG2 and OLIG1 couple neuronal and glial subtype specification. Cell 109, 61–73 (2002). 46. Rowitch, D.H. Glial specification in the vertebrate neural tube. Nat. Rev. Neurosci. 5, 409–419 (2004). 47. He, Y. et al. The transcription factor Yin Yang 1 is essential for oligodendrocyte progenitor differentiation. Neuron 55, 217–230 (2007). 48. Natesan, S. & Gilman, M. YY1 facilitates the association of serum response factor with the c-fos serum response element. Mol. Cell. Biol. 15, 5975–5982 (1995). 49. Liu, H. et al. Cysteine-rich protein 61 and connective tissue growth factor induce deadhesion and anoikis of retinal pericytes. Endocrinology 149, 1666–1677 (2008).

427

ARTICLES

Trans-synaptic adhesion between NGL-3 and LAR regulates the formation of excitatory synapses

© 2009 Nature America, Inc. All rights reserved.

Jooyeon Woo1,4, Seok-Kyu Kwon1,4, Seungwon Choi1, Seho Kim1, Jae-Ran Lee1, Anthone W Dunah2, Morgan Sheng3 & Eunjoon Kim1 Synaptic adhesion molecules regulate multiple steps of synapse formation and maturation. The great diversity of neuronal synapses predicts the presence of a large number of adhesion molecules that control synapse formation through trans-synaptic and heterophilic adhesion. We identified a previously unknown trans-synaptic interaction between netrin-G ligand–3 (NGL-3), a postsynaptic density (PSD) 95–interacting postsynaptic adhesion molecule, and leukocyte common antigen-related (LAR), a receptor protein tyrosine phosphatase. NGL-3 and LAR expressed in heterologous cells induced pre- and postsynaptic differentiation in contacting axons and dendrites of cocultured rat hippocampal neurons, respectively. Neuronal overexpression of NGL-3 increased presynaptic contacts on dendrites of transfected neurons. Direct aggregation of NGL-3 on dendrites induced coclustering of excitatory postsynaptic proteins. Knockdown of NGL-3 reduced the number and function of excitatory synapses. Competitive inhibition by soluble LAR reduced NGL-3–induced presynaptic differentiation. These results suggest that the transsynaptic adhesion between NGL-3 and LAR regulates excitatory synapse formation in a bidirectional manner.

Synaptogenesis involves a number of molecular processes, including axon-dendrite recognition, formation of nascent synapses and synapse maturation through recruitment of synaptic proteins. Synaptic cell adhesion molecules have been implicated in each of these processes1–6. Adhesion molecules capable of inducing early synaptic differentiation include neuroligin7,8, neurexin7,9, SynCAM10,11, NGL12,13 and EphB receptors14. The heterophilic and trans-synaptic adhesion between postsynaptic neuroligins and presynaptic neurexins is one of the most extensively studied synaptic adhesions7. Neuroligin and neurexin expressed in non-neural cells induce pre- and postsynaptic differentiation, respectively, in a bidirectional manner8,9. The interaction between neuroligin and neurexin is regulated by alternative splicing15–17. Neuroligin interacts with the postsynaptic scaffolding protein PSD-95 (ref. 18), which is thought to couple synaptic adhesion to postsynaptic protein clustering. Consistent with these important characteristics of the neuroligin-neurexin interaction, defects in neuroligin function are associated with human autism19,20. However, considering the great diversity of neuronal synapses, it is probable that additional cell adhesion molecule interactions that regulate synapse formation and functions remain to be discovered. Recent studies have identified the NGL family of PSD-95–interacting postsynaptic adhesion molecules, which contains three known members, NGL-1, NGL-2 and NGL-3 (refs. 12,13). NGL associates with netrin-G/laminet12, a family of glycosylphosphatidylinositol (GPI)anchored adhesion molecules21–23, in an isoform-specific manner;

NGL-1 and NGL-2 associate with netrin-G1 and netrin-G2, respectively12,13. In transgenic mice with netrin-G1 deficiency, NGL-1, but not NGL-2, shows a diffuse dendritic distribution; likewise, NGL-2 is selectively dispersed in netrin-G2–deficient mice24, suggesting that these interactions regulate axon-dependent localization of NGL-1 and NGL-2 in specific segments of dendrites. In support of the role for NGLs in synapse formation, NGL-2 induces presynaptic differentiation in contacting axons when expressed in non-neural cells13. Netrin-Gs show extensive alternative splicing21–23, which may regulate the netrin-G– NGL-1/2 interaction. Mice that are deficient in netrin-G2 or NGL-2 show abnormal auditory responses25. Single-nucleotide polymorphism analysis associated both netrin-G1 and netrin-G2 with schizophrenia26. In contrast with the increasing studies of NGL-1/2 and their ligands, the third member of the NGL family (NGL-3), which does not bind netrinG1 or netrin-G2, has remained an orphan receptor. LAR is well-known for its involvement in axon guidance and presynaptic differentiation27,28. LAR contains three Ig-like domains and eight fibronectin type III domains in the extracellular region and two phosphatase domains in the intracellular region. The membraneproximal phosphatase domain is catalytically active, whereas the membrane-distal phosphatase domain is inactive and associates with various cytoplasmic proteins, including liprin-a29 and b-catenin30. C. elegans and Drosophila homologs of LAR are involved in the regulation of synaptic growth and active zone assembly at the neuromuscular junction (NMJ)31–33. Mammalian LAR has been shown to regulate excitatory synaptic development and maintenance34.

1National

Creative Research Initiative Center for Synaptogenesis and Department of Biological Sciences, Korea Advanced Institute of Science and Technology, Daejeon, Korea. 2MassGeneral Institute for Neurodegenerative Disease, Massachusetts General Hospital and Harvard Medical School, Charlestown, Massachusetts, USA. 3The Picower Institute for Learning and Memory, Howard Hughes Medical Institute, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA. 4These authors contributed equally to this work. Correspondence should be addressed to E.K. ([email protected]). Received 23 December 2008; accepted 22 January 2009; published online 1 March 2009; doi:10.1038/nn.2279

428

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

b ***

c

70

d Intensity of vGlut1 clusters (au)

Intensity of synapsin I clusters (au)

80

60 50 40 30 20

*** **

10

60

*** 50 40 30 20 10 0 N G L- E 3– GF EG P FP

a

N G N L-1 EG G N L-2–EGFP G – L- E FP 3– G EGFP FP

0

g

20

***

16 12 8 4

h

35

***

30 Intensity of SynTag clusters (au)

25 20 15 10 5 0

N

G

L- E 3– GF EG P FP

0

EG N FP G L3

f Intensity of VGAT clusters (au)

© 2009 Nature America, Inc. All rights reserved.

e

Figure 1 NGL-3 expressed in non-neural cells induces functional presynaptic differentiation in contacting axons. (a) Induction of presynaptic differentiation by NGL family proteins. HEK293T cells expressing EGFP alone, NGL-1–EGFP (C-terminally tagged), NGL-2–EGFP or NGL-3–EGFP were cocultured with hippocampal neurons (10–13 DIV) and stained for synapsin I. (b) Quantification of the integrated intensity of synapsin I clusters induced by NGLs (mean ± s.e.m.; n ¼ 32 for EGFP, 29 for NGL-1–EGFP, 32 for NGL-2–EGFP and 29 for NGL-3–EGFP; ** P o 0.01, *** P o 0.001, Student’s t test; au, arbitrary units). (c,d) NGL-3 induced clustering of vGlut1, an excitatory presynaptic marker. HEK293T cells expressing NGL-3–EGFP or EGFP alone were cocultured with neurons (10–13 DIV) and stained for vGlut1 (mean ± s.e.m.; n ¼ 34 for EGFP and 37 for NGL-3, ***P o 0.001, Student’s t test). (e,f) NGL-3 induced clustering of VGAT, an inhibitory presynaptic marker. Cocultured cells prepared as in c and d were stained for VGAT (mean ± s.e.m.; n ¼ 31 for EGFP and 33 for NGL-3, ***P o 0.001, Student’s t test). (g,h) NGL-3 induced functional presynaptic differentiation. Cocultured cells were incubated with antibodies to synaptotagmin I (SynTag) luminal domain to visualize functional presynaptic nerve terminals (mean ± s.e.m.; n ¼ 27 for EGFP and NGL-3, ***P o 0.001, Student’s t test). Scale bars represent 20 mm.

Several extracellular ligands for LAR have been identified. LAR associates with the laminin-nidogen complex, a major component of the extracellular matrix, and regulates extracellular matrix– dependent morphological changes in non-neural cells35. In Drosophila, LAR associates with the heparan sulfate proteoglycans Syndecan and Dallylike at the NMJ36,37. Syndecan is a transmembrane protein that regulates LAR-dependent presynaptic growth and Dallylike is a GPI-anchored protein that inhibits LAR’s ability to regulate active zone size at the NMJ36,37. LAR associates with a small (B11 kDa) ectodomain isoform of LAR, LARFN5C, in a homophilic manner to promote neurite outgrowth in mouse hippocampal neurons38. However, transmembrane ligands of LAR that adhere to LAR in a trans-synaptic manner at interneuronal synapses have not been identified in either invertebrates or vertebrates. Here, we identified a previously unknown interaction between NGL-3 and LAR. NGL-3 and LAR induced pre- and postsynaptic differentiation in contacting axons and dendrites, respectively, when they were expressed in non-neural cells. Data from direct aggregation, knockdown and competitive inhibition experiments suggest that the trans-synaptic adhesion between NGL-3 and LAR regulates excitatory synapse formation.

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

RESULTS NGL-3 in HEK293 cells induces presynaptic differentiation The synaptogenic activity of an adhesion molecule can be studied in a mixed-culture or coculture assay in which an adhesion molecule expressed in non-neural cells is tested for its ability to induce presynaptic differentiation in contacting axons of cocultured neurons8,39. Accordingly, we used this mixed-culture assay to test for possible NGL-3 synaptogenic activity. NGL-3–expressing HEK293T cells that were cocultured with hippocampal neurons induced a strong clustering of synapsin I, a presynaptic vesicle marker, in contacting axons, whereas control cells expressing enhanced green fluorescent protein (EGFP) alone induced minimal synapsin I clustering (Fig. 1a,b). The extent of NGL-3–induced synapsin I clustering was much greater than that induced by NGL-1 or NGL-2 (Fig. 1a,b). This difference was not attributable to differential expression of the three NGL isoforms because all three were expressed on the surface of HEK293T cells at similar levels (Supplementary Fig. 1 online). NGL-3–expressing HEK293T cells also induced the clustering of both vGlut1 and VGAT, which are excitatory and inhibitory presynaptic markers, respectively (Fig. 1c–f). NGL-3 was more efficient in inducing vGlut1 clustering than VGAT clustering; the extent of vGlut1 clustering relative to the EGFP control was about sixfold greater than that of

429

ARTICLES

(kDa)

b

U nt M ran yc s M –N yc GL M –N -1 yc G L S2 –NG -2 LP2 3

a

(kDa) 200

200

E18 1 d 1

2

3

4 6 weeks

116 97

116 97

NGL-3

NGL-3

* *

66

* *

66

45

45 31

Myc

PSD-95

PSD-95

SynPhy

10 2

I

II

III

P3 LP 1 LS 2 LP 2

g

f PSD

P2 (µg)

P2 S3

H NGL-3

NGL-3

e © 2009 Nature America, Inc. All rights reserved.

d

H ea r Br t ai Sp n le e Lu n ng Li ve Sk r .m Ki us dn cle e Te y st is

c

P1 S2

α-tubulin

PNGase F

–

–

+

37°C

–

+

+

NGL-3

(kDa) 116

PSD-95

97

O-glycosidase

–

–

–

+

Neuramidase 37°C (kDa) 116

– –

– +

+ +

+ +

97

VGAT clustering (Fig. 1d,f). NGL-3 did not induce clustering of PSD95 or gephyrin, which are excitatory and inhibitory postsynaptic markers, respectively (Supplementary Fig. 2 online), indicating that the NGL-3–dependent presynaptic protein clustering was not induced by interneural synapses. In contacting axons of cocultured neurons, NGL-3 induced the uptake of antibodies that were directed against the luminal domain of the synaptic vesicle protein synaptotagmin I, which is present during the recycling of presynaptic vesicles (Fig. 1g,h), thus indicating that NGL-3 induces functional presynaptic differentiation. These results suggest that NGL-3 can induce functional presynaptic differentiation at both excitatory and inhibitory synapses. Expression patterns of NGL-3 proteins Previous northern blot and in situ hybridization results have shown that NGL-3 mRNAs are mainly expressed in the brain and are widespread in various brain regions13. The expression patterns of NGL proteins have been studied using an antibody that recognizes all three NGL isoforms (pan-NGL)13. To study the specific expression patterns of NGL-3 proteins, we generated an NGL-3 antibody that does not cross-react with NGL-1 or NGL-2 (Fig. 2a). The NGL-3 antibody detected a single band (B115 kDa) in rat brain (Fig. 2a). The apparent molecular weight (115 kDa) of NGL-3, which falls into the upper end of the size range of the three NGL proteins (95–115 kDa)13, is consistent with the longer length of NGL-3 (709 amino acids in rat) relative to NGL-1 (640 amino acids) and NGL-2 (652 amino acids). NGL-3 expressed in heterologous cells showed two bands (B140 and 100 kDa), which differed from the size of NGL-3 expressed in the brain. This may reflect differential postsynaptic modification (that is, glycosylation) because NGL-3 proteins expressed in heterologous cells (140 and 100 kDa) showed higher and lower levels of glycosylation, respectively, compared with NGL-3 in the brain (Supplementary Fig. 3 online). Expression of NGL-3 proteins was gradually increased during the first 3 weeks of postnatal rat brain development (Fig. 2b) and mainly detected in the brain, but not in other tissues (Fig. 2c). NGL-3 was mainly detected in synaptic fractions, including the crude synaptosomal and synaptic plasma membrane fractions (Fig. 2d). NGL-3 was

430

Figure 2 Expression patterns of NGL-3 proteins in rat brain. (a) NGL-3 antibodies selectively recognized NGL-3, but not NGL-1 or NGL-2 (lanes 1–4), and detected a 115-kDa band (arrow) in brain samples. The immunoblot was also probed with antibody to Myc for normalization. P2, crude synaptosomes; S2, supernatant after P2 precipitation. Asterisks indicate nonspecific bands. (b) NGL-3 protein expression gradually increased during the first 3 weeks of postnatal rat brain development. Whole rat brain extracts from different developmental stages were used. a-tubulin was visualized for normalization. Asterisks indicate nonspecific bands. (c) NGL-3 proteins were mainly expressed in the brain. PSD-95 was visualized for comparison. Sk. muscle, skeletal muscle. (d) Distribution of NGL-3 in subcellular fractions of adult rat brain. Note that NGL-3 proteins were mainly detected in synaptic fractions, including P2 and LP1. PSD-95 and synaptophysin (SynPhy) were probed for comparison. H, homogenates; LP1, synaptosomal membranes; LP2, synaptic vesicle-enriched fraction; LS2, synaptosomal cytosol; P1, crude nuclear fraction; P3, light membranes; S3, cytosol. (e) NGL-3 was detected in PSD fractions of rat brain (3 weeks), with a strong enrichment in the PSD III fraction. (f) N-glycosylation of NGL-3. The crude synaptosomal fraction of adult rat brain was subjected to PNGase F digestion, followed by immunoblot assay. (g) O-glycosylation and sialylation of NGL-3, evidenced by O-glycosidase and neuramidase digestion.

detected in PSD fractions, including the most detergent-resistant PSD III fraction (Fig. 2e), indicative of a tight association of NGL-3 with the PSD. PNGase F, O-glycosidase and neuramidase reduced the apparent molecular weight of NGL-3, indicating that NGL-3 is N- and O-glycosylated and sialylated (Fig. 2f,g). Neuronal NGL-3 overexpression induces presynaptic contacts We next tested the effects of NGL-3 overexpression in cultured neurons. NGL-3 overexpression in cultured hippocampal neurons substantially increased presynaptic contacts, as measured by the integrated intensity of synapsin I clusters on dendrites (Fig. 3a,b). NGL-3 also induced a strong increase in excitatory presynaptic contacts, as measured by vGlut1 clusters (Fig. 3c,d). In contrast, inhibitory presynaptic contacts (VGAT clusters) were not induced by NGL-3 overexpression (Fig. 3e,f). These results suggest that NGL-3 expression in neurons selectively induces excitatory, but not inhibitory, presynaptic differentiation in contact axons, which contrasts with our findings that both excitatory and inhibitory presynaptic differentiation are induced by NGL-3 in the mixed-culture experiments (Fig. 1). However, the number of excitatory synapses, defined as vGlut1positive PSD-95 clusters, was reduced by NGL-3 overexpression, mainly as the result of a decrease in the number of PSD-95 clusters (Supplementary Fig. 4 online). It is possible that NGL-3 overexpression causes the dispersal of NGL-3–associated proteins, including PSD95, to extrasynaptic sites in a dominant-negative manner, which have been observed in neurons overexpressing neuroligin-2 and SALM2 (refs. 9,40). NGL-3 aggregation clusters postsynaptic proteins Because NGL-3 has the ability to induce presynaptic differentiation, we reasoned that it may also promote postsynaptic differentiation by recruiting various postsynaptic proteins. To test this hypothesis, we expressed N-terminally EGFP-tagged NGL-3 in cultured hippocampal neurons and induced direct clustering of EGFP–NGL-3 on the dendritic surface by incubating the neurons with beads coated with antibodies to EGFP. Direct NGL-3 aggregation induced secondary clustering of postsynaptic proteins, including PSD-95, GKAP (a postsynaptic scaffold), Shank (a postsynaptic scaffold), GluR2 (an AMPA receptor subunit) and NR1 (an NMDA receptor subunit), but not gephyrin (Fig. 4). Direct aggregation of a control membrane protein containing EGFP alone in the extracellular region (EGFP-pDis) did not

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

a

Intensity of synapsin I clusters (au)

b

100

80

**

60

40

20

EG N FP G L3

0

c

d

80

Intensity of vGlut1 clusters (au)

60

40

20

EG N FP G L3

0

e 25 20

coclustering of excitatory postsynaptic proteins (Supplementary Fig. 6 online), similar to our results with bead-induced protein clustering (Fig. 4). In quantitative analysis, a large fraction of NGL-3 clusters colocalized with excitatory postsynaptic protein clusters, but minimally colocalized with gephyrin clusters (Supplementary Fig. 6). In addition, the fluorescence intensities of the colocalized excitatory postsynaptic protein clusters, normalized to nearby dendrites, were greater than that of gephyrin (Supplementary Fig. 6). Collectively, these results suggest that NGL-3 is sufficient to induce dendritic clustering of excitatory, but not inhibitory, postsynaptic proteins.

15

NGL-3 knockdown reduces excitatory synapse number We next tested whether knockdown of NGL-3 in neurons would lead to a loss of excitatory 5 synapses by RNA interference. We generated two independent shRNA constructs 0 for NGL-3 knockdown (shNGL-3 #1 and shNGL-3 #2), which reduced NGL-3 expression in HEK293T cells by 61% and 46%, respectively (Supplementary Fig. 7 online). In hippocampal neurons, shNGL-3 #1 and shNGL-3 #2 also reduced the expression of exogenous NGL-3 by 53% and 69%, respectively (Supplementary Fig. 7). Knockdown of endogenous NGL-3 expression in cultured neurons could not be tested because we lacked NGL-3 antibodies that are suitable for immunostaining. In cultured hippocampal neurons, NGL-3 knockdown by the two shNGL-3 constructs reduced the number of excitatory synapses, defined as vGlut1-positive PSD-95 clusters, compared with control neurons transfected with empty shRNA vector (sh-vec) (Fig. 5a,b). In contrast, NGL-3 knockdown did not affect the number of inhibitory synapses, defined as VGATpositive gephyrin clusters (Fig. 5c,d). A variant of shNGL-3 #1 with point mutations that do not induce NGL-3 knockdown in heterologous cells and neurons (shNGL-3 #1*; Supplementary Fig. 7) did not reduce excitatory synapse number relative to sh-vec (Fig. 5a,b). The reduction in excitatory synapse number by shNGL-3 #1 could be rescued by coexpression of a NGL-3 expression construct that is resistant to shNGL-3 #1 (NGL-3*) (Fig. 5a,b and Supplementary Fig. 7), further supporting the specific action of shNGL-3 #1. Functionally, NGL-3 knockdown reduced the frequency, but not the 10

EG N FP G L3

Intensity of of VGAT clusters (au)

© 2009 Nature America, Inc. All rights reserved.

***

f

Figure 3 Overexpression of NGL-3 in cultured neurons increases excitatory, but not inhibitory, presynaptic contacts on dendrites of transfected neurons. (a–f) Cultured hippocampal neurons were transfected with NGL-3 and EGFP or EGFP alone (12–15 DIV) and immunostained for synapsin I (a), vGlut1 (c), VGAT (e) or EGFP (a,c,e). For quantification, integrated fluorescence intensities of presynaptic protein clusters along the dendrites were normalized to the dendrite area (synapsin I: n ¼ 10 for EGFP and 11 for NGL-3, **P o 0.01, Student’s t test, b; vGlut1: n ¼ 12 for EGFP and 10 for NGL-3, ***P o 0.001, d; VGAT: n ¼ 9 for EGFP and 10 for NGL-3, P ¼ 0.3, f). Data are presented as mean ± s.e.m. Scale bar represents 20 mm.

induce PSD-95 coclustering (Supplementary Fig. 5 online). A quantitative analysis showed that a large fraction of NGL-3 clusters colocalized with excitatory postsynaptic protein clusters, but not with gephyrin clusters (95.8 ± 2.7%, n ¼ 41 for PSD-95; 0.0%, n ¼ 38 for gephyrin; 94.6 ± 2.6%, n ¼ 45 for GKAP; 89.6 ± 5.4%, n ¼ 25 for Shank; 83.9 ± 6.8%, n ¼ 30 for GluR2; 98.5 ± 4.5%, n ¼ 44 for NR1; 0.0%, n ¼ 33 for PSD-95 by EGFP-pDis). In addition, the fluorescent intensities of the colocalized proteins normalized to nearby dendrites were 2.36 ± 0.18 for PSD-95 (n ¼ 46), 0.93 ± 0.04 for gephyrin (n ¼ 35), 2.06 ± 0.07 for GKAP (n ¼ 47), 3.26 ± 0.27 for Shank (n ¼ 24), 1.64 ± 0.09 for GluR2 (n ¼ 44), 2.42 ± 0.10 for NR1 (n ¼ 40) and 0.95 ± 0.99 for PSD-95 by EGFP-pDis (n ¼ 33). Postsynaptic protein clusters induced by NGL-3 aggregation were negative for synapsin I or synaptophysin (Fig. 4), indicating that the postsynaptic protein clusters were not induced by interneuronal synapses. These results suggest that NGL-3 clustering on dendrites induces excitatory, but not inhibitory, postsynaptic differentiation. In additional experiments, we induced dendritic NGL-3 clustering by preclustered EGFP antibodies, instead of by EGFP antibody– coated beads. This primary NGL-3 clustering on dendrites induced

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

431

ARTICLES the three NGL isoforms with LAR-ecto-Fc, a soluble LAR protein formed by fusion of the ectodomain of LAR with the immunoglobulin Fc domain. LAR-ecto-Fc selectively bound to NGL-3, but not to NGL-1 or NGL-2 (Fig. 6e). These results suggest that b LAR selectively interacts with NGL-3 in a calcium-independent manner and that this interaction can mediate cell adhesion. Deletion of the leucine-rich repeat (LRR) domain in the extracellular region of NGL-3 (NGL-3DLRR), but not the Ig domain (NGLc 3DIg), abolished the interaction between NGL-3 and LAR in the cell adhesion and soluble LAR binding assays (Supplementary Fig. 9 online). In addition, NGL-3DLRR, but not NGL-3DIg, failed to induce presynaptic d differentiation in contacting axons of cocultured neurons when expressed in heterologous cells (Supplementary Fig. 9). Surface expression levels of these NGL-3 variants were comparable, as determined by biotinylation assays (wild type, 22.0 ± 3.0; DLRR, e 20.1 ± 2.6; DIg, 24.3 ± 1.5 arbitrary units). These results indicate that the LRR domain of NGL-3 is important for the NGL-3–LAR interaction. We further tested the interaction between NGL-3 and LAR in a mixed-culture assay, in f which NGL-3–expressing HEK293T cells were cocultured with neurons expressing exogenous LAR. NGL-3–expressing HEK293T cells induced a strong aggregation of Flagtagged LAR (LAR-Flag) in contacting axons, whereas EGFP-expressing HEK293T cells Figure 4 Direct aggregation of NGL-3 on the surface of dendrites induces coclustering of excitatory (control) induced minimal LAR clustering postsynaptic proteins. (a–f) Cultured hippocampal neurons expressing N-terminally EGFP-tagged NGL-3 (14–16 DIV) were incubated with EGFP-coated beads and visualized at 17 DIV by triple (average fluorescence intensities of LAR clusimmunofluorescence staining for EGFP (NGL-3), synapsin I (a–e) or synaptophysin (f) and the indicated ters normalized to the HEK293T cell area: postsynaptic proteins. sGluR2, surface GluR2. Arrowheads indicate enlarged beads and protein clusters. NGL-3 induced, 24.22 ± 5.98, n ¼ 12; EGFP Scale bar represents 10 mm. induced, 6.34 ± 0.95; n ¼ 14; P o 0.005, Student’s t test; Fig. 6f). Conversely, amplitude, of miniature excitatory postsynaptic currents (mEPSCs) to HEK293T cells expressing LAR, but not EGFP, induced clustering of a greater extent than the morphological reduction (Fig. 5e–g). These Flag-tagged NGL-3 (NGL-3–Flag) in contacting dendrites (average results suggest that NGL-3 is required for the morphological and fluorescence intensities of NGL-3 clusters: LAR induced, 62.27 ± 9.48, n ¼ 11; EGFP induced, 21.37 ± 2.47, n ¼ 11; P o 0.001, Student’s functional maintenance of excitatory synapses. t test; Fig. 6g). These results suggest that the interaction between NGL-3 and LAR may occur in a trans-synaptic manner. NGL-3 interacts with LAR We next determined the binding affinity for the interaction between NGL-3 probably induces presynaptic differentiation in contacting axons by interacting with a presynaptic ligand. To identify a specific NGL-3 and LAR. NGL-3–expressing HEK293T cells were incubated ligand for NGL-3, we systematically screened NGL-3–expressing L cells with increasing amounts of LAR-ecto-Fc and proteins bound to the for their ability to coaggregate with a panel of L cells expressing specific cell surface were quantified by ELISA assays. The Kd value for the synaptic membrane proteins (examples in Supplementary Fig. 8 NGL-3–LAR interaction that we calculated by Scatchard analysis was online). This led us to the identification of LAR as a specific ligand 37.4 ± 2.1 nM (Fig. 6h). for NGL-3. In additional experiments, LAR-expressing cells selectively aggre- Soluble LAR inhibits NGL-3–dependent presynaptic induction gated with NGL-3–expressing cells, but not with those expressing If NGL-3–induced presynaptic differentiation in the mixed-culture NGL-1 or NGL-2 (Fig. 6a,b). Removal of extracellular calcium did assay is dependent on the interaction of NGL-3 with LAR expressed not reduce the aggregation between NGL-3– and LAR-expressing cells on the surface of axons, exogenously added soluble LAR fusion (Fig. 6c,d). There was no homophilic adhesion between cells expressing proteins (LAR-ecto-Fc), which compete with endogenous LAR for NGL-3 or LAR (Supplementary Fig. 8). To further confirm the NGL-3 binding, should inhibit the NGL-3–induced presynaptic differadhesion of NGL-3 and LAR, we incubated HEK293T cells expressing entiation. Indeed, LAR-ecto-Fc reduced NGL-3–induced synapsin I

© 2009 Nature America, Inc. All rights reserved.

a

432

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

sh-vec

shNGL-3 #1

shNGL-3 #2

b

shNGL-3 #1 + NGL-3*

shNGL-3 #1*

vGlut1-positive PSD-95 clusters per 10 µm

a

3.5 3.0 2.5 2.0

**

**

1.5 1.0 0.5 sh s shNG h-v sh NGL-3 ec sh NG L-3 #1 N L- #2 G 3 L- # 3 1 N #1 * G + L3*

0

c shNGL-3 #2

d

shNGL-3 #1 + NGL-3*

shNGL-3 #1*

VGAT-positive gephyrin clusters per 10 µm

shNGL-3 #1

2.5 2.0 1.5 1.0 0.5

1s

0.6 0.4

**

0.2 0

15

10

5

sh

N shG ve L- c 3 #2

0 N shG ve L- c 3 #2

10 pA shNGL-3 #2

g

0.8

sh

f

sh-vec

Amplitude (pA)

e

sh N s sh GLh-v sh NG -3 ec N sh G L-3 #1 N L- # G 3 2 L- # 3 1 N #1 * G + L3*

0

Frequency (Hz)

© 2009 Nature America, Inc. All rights reserved.

sh-vec

Figure 5 Knockdown of NGL-3 leads to decreases in the number and function of excitatory synapses. (a–d) NGL-3 knockdown reduced the number of excitatory synapses (vGlut1-positive PSD-95 clusters) without affecting inhibitory synapses (VGAT-positive gephyrin clusters). Cultured hippocampal neurons transfected with sh-vec, shNGL-3 #1 and #2, shNGL-3 #1* or shNGL-3 #1 + NGL-3* (10–14 DIV) were immunostained with the indicated antibodies (data are presented as mean ± s.e.m.; excitatory synapses: n ¼ 21 for sh-vec, 22 for shNGL-3 #1, 19 for shNGL-3 #2, 17 for shNGL-3 #1* and 22 for shNGL-3 #1 + NGL-3*, *P o 0.005, ANOVA; inhibitory synapses: n ¼ 20 for sh-vec, 21 for shNGL-3 #1, 21 for shNGL-3 #2, 19 for shNGL-3 #1* and 20 for shNGL-3 #1 + NGL-3*). Scale bar represents 10 mm. (e–g) NGL-3 knockdown in cultured neurons (13–16 DIV) reduced the frequency (f), but not amplitude (g), of mEPSCs (n ¼ 12 for sh-vec and 10 for shNGL-3 #2, **P o 0.01, Student’s t test).

clustering in contacting axons of cocultured neurons compared with Fc alone (control) (Fig. 7a,b). However, when we added LAR-ecto-Fc to cultured neurons, it did not reduce the number of excitatory synapses (synapsin I–positive PSD-95 clusters; Supplementary Fig. 10 online). It is possible that other trans-synaptic adhesions unaffected by LAR may be maintaining normal excitatory synapses when the LAR–NGL-3 interaction is inhibited, whereas the synapses formed in the mixed-culture assay, which rely on the LAR–NGL-3 interaction, but not other synaptic adhesions, may be more easily inhibited. These results suggest that the adhesion between NGL-3 and LAR is required for NGL-3–induced presynaptic differentiation. LAR in HEK293T cells induces postsynaptic differentiation If the trans-synaptic interaction between NGL-3 and LAR is involved in the formation of excitatory synapses, presynaptic LAR may also induce postsynaptic differentiation in contacting dendrites. Indeed, in a mixed-culture assay, LAR-expressing HEK293T cells induced the clustering of excitatory postsynaptic proteins PSD-95 and Shank,

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

whereas EGFP alone induced minimal postsynaptic protein clustering (Fig. 8a–d). In addition, LAR expressed in HEK 293T cells did not induce gephyrin clusters in contacting dendrites (Fig. 8e,f). These results suggest that presynaptic LAR induces excitatory, but not inhibitory, postsynaptic protein clustering on contacting dendrites. LAR regulates excitatory synaptic development and maintenance partly through its association with AMPA receptors34, suggesting that LAR is also present on the postsynaptic plasma membrane. We therefore tested whether LAR expressed in HEK293T cells is capable of inducing presynaptic differentiation in contacting axons. LAR-expressing HEK293T cells did not induce synapsin I clustering in contacting axons (Fig. 8g,h). This suggests that postsynaptic LAR has a different role, which is unrelated to the induction of presynaptic differentiation. DISCUSSION Here, we identified a previously unknown trans-synaptic adhesion between NGL-3 and LAR. In support of the synaptogenic function of this interaction, NGL-3 and LAR expressed in non-neural cells induced pre- and postsynaptic differentiation in contacting axons and

433

ARTICLES

NGL-3

5

3 2 1 0

N G N L-1 G N L-2 + L G + A L3 LAR + R LA R

g

h

NGL-3

5 4 3 2 1 0

3

2

1

0.10 0.05 0.00

0 1 2 3 4 Bound (OD 450)

0 0

50

100

150

LAR-ecto-Fc (nM)

Figure 6 NGL-3 interacts with LAR. (a) NGL-3, but not NGL-1 or NGL-2, interacted with LAR in cell-aggregation assays. L cells cotransfected with NGLs and EGFP or LAR (C-terminally cyan fluorescent protein (CFP)-tagged) and dsRed were mixed to induce cell aggregation. Scale bar represents 50 mm. (b) Quantification of the cell aggregation in a. Cell aggregates were defined by four or more clustered cells containing at least one red or green cell (mean ± s.e.m., n ¼ 10, ***P o 0.001, ANOVA). (c,d) Calcium-independent adhesion between NGL-3 and LAR. L cells expressing NGL-3 + EGFP or LAR-CFP + dsRed were mixed in the presence or absence of 10 mM EGTA (mean ± s.e.m., n ¼ 10, ***P o 0.001, Student’s t test). Scale bar represents 50 mm. (e) Selective association of LAR-ecto-Fc with NGL-3, but not with NGL-1 and NGL-2 expressed in HEK293T cells. Scale bar represents 10 mm. (f) NGL-3– expressing HEK293T cells induced LAR-Flag clustering on contacting axons of cocultured neurons. HEK293T cells expressing NGL-3–EGFP or EGFP alone were cocultured (14–15 DIV) with hippocampal neurons expressing LAR-Flag C1522S (C-terminally Flag-tagged, 13–14 DIV), followed by EGFP and Flag staining. The phosphatase-dead LAR C1522S mutant was used to minimize possible effects of phosphatase activity on transfected neurons. (g) LARexpressing HEK293T cells induce NGL-3–Flag clustering on contacting dendrites of cocultured neurons. HEK293T cells expressing LAR-CFP or EGFP alone were cocultured (14–15 DIV) with hippocampal neurons expressing NGL-3–Flag (13–14 DIV). Scale bar represents 20 mm. (h) Saturation curve of LAR-ecto-Fc binding to NGL-3 expressed in HEK293T cells. A Scatchard plot analyzed by linear regression of the data is shown in the inset. The calculated Kd for the interaction was 37.4 ± 2.1 nM.

dendrites, respectively. Neuronal NGL-3 expression increased presynaptic contacts. Dendritic NGL-3 aggregation induced excitatory postsynaptic protein clustering. NGL-3 knockdown reduced excitatory synapse number and function. Soluble LAR inhibited NGL-3–induced presynaptic differentiation. These results suggest that the trans-synaptic adhesion between NGL-3 and LAR regulates excitatory synapse formation in a bidirectional manner. We were unable to determine the ultrastructural synaptic localization of NGL-3 because of the lack of suitable NGL-3 antibodies. However, our previous electron microscopy results, obtained using a pan-NGL antibody, indicate that NGL isoforms are mainly postsynaptically localized at excitatory synapses13. The presynaptic localization of

434

a

b

160

***

140 Fc alone

LAR-ecto-Fc

120 100 80 60 40 20 0 LA Fc R al -e o ct ne oFc

Figure 7 Soluble LAR reduces NGL-3–induced presynaptic differentiation. (a) Reduced NGL-3– induced presynaptic differentiation in the presence of soluble LAR. HEK293T cells expressing NGL-3–EGFP or EGFP alone were cocultured with hippocampal neurons (10–13 DIV) in the presence of LAR-ecto-Fc or Fc alone (10 mg ml
1) and stained for EGFP (for NGL-3), synapsin I and human Fc (for LAR-ecto-Fc). Scale bar represents 20 mm. (b) Quantification of the integrated intensity of synapsin I clusters normalized to the cell area (mean ± s.e.m., n ¼ 20 for Fc alone and 23 for LAR-ecto-Fc, ***P o 0.001, Student’s t test).

LAR in C. elegans and Drosophila has been well characterized28,32,33,36, although the pre- or postsynaptic localization of mammalian LAR remains to be determined34. On the basis of the widespread distribution of both mRNAs, determined by in situ hybridization13,41, the adhesion between NGL-3 and LAR probably occurs in various brain regions. Notably, NGL-3 proteins are mainly expressed in the brain (Fig. 2b), whereas LAR mRNAs, determined by Northern analysis, are expressed in various tissues42, suggesting that the interaction between NGL-3 and LAR occurs primarily in the brain. The three NGL isoforms share an identical domain structure and interact with PSD-95 through their extreme C termini12,13. However, their primary sequences are substantially different (especially in their

Intensity of synapsin I clusters (au)

© 2009 Nature America, Inc. All rights reserved.

Bound (OD 450)

f

6

Bound per free (OD 450 per nM)

4

7

NGL-1

***

LAR-ecto-Fc

NGL-2

7 6

e

N G N L+ GL 3 + 10 -3 L m + AR M L EG AR TA

NGL-3

d

LAR-CFP

NGL-2

LAR-CFP

NGL-1

c

Number of cell aggregates per frame

b Number of cell aggregates per frame

a

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

b 25 20

c

d 25 Intensity of Shank clusters (au)

Intensity of PSD-95 clusters (au)

a

**

15 10 5 0

15 10 5 0

5

LA EG R FP -C FP

g

h 25 Intensity of synapsin I clusters (au)

20

10

20 15 10 5 0 LA EG R FP -C FP

25

15

0

LA EG R FP -C FP

Intensity of gephyrin clusters (au)

f

LA EG R FP -C FP

© 2009 Nature America, Inc. All rights reserved.

e

*** 20

Figure 8 LAR expressed in non-neural cells induces excitatory postsynaptic differentiation in contacting dendrites. (a–f) LAR induced clustering of excitatory postsynaptic proteins in contacting dendrites. HEK293T cells expressing LAR-CFP or EGFP alone were cocultured with neurons (10–13 DIV) and stained for EGFP/CFP (a,c,e), PSD-95 (a), Shank (c, pan-Shank) and gephyrin (e). Scale bar represents 20 mm. We quantified the integrated intensity of postsynaptic protein clusters normalized to the cell area (mean ± s.e.m.; PSD-95: n ¼ 22 for EGFP and 22 for LAR, **P o 0.01, Student’s t test, b; Shank: n ¼ 25 for EGFP and 29 for LAR, ***P o 0.001, Student’s t test, d; gephyrin: n ¼ 24 for EGFP and 20 for LAR, f). (g) LAR did not induce the clustering of synapsin I in contacting axons. HEK293T cells expressing LAR-CFP or EGFP alone were cocultured with neurons (10–13 DIV) and stained for EGFP/CFP and synapsin I (g). Scale bar represents 20 mm. (h) Quantification of the integrated intensity of presynaptic protein clusters normalized to the cell area (mean ± s.e.m.; synapsin I: n ¼ 28 for EGFP and 27 for LAR, P ¼ 0.59, Student’s t test).

cytoplasmic domains), suggesting that NGL isoforms have different functions. In support of this notion, NGLs interact with netrin-Gs in an isoform-specific manner: NGL-1 and NGL-2 interact with netrin-G1 and netrin-G2, respectively, whereas NGL-3 binds neither netrin-G1 nor netrin-G2 (ref. 13). Our data indicate that NGL-3, which to date has remained an orphan receptor, selectively associates with LAR, whereas NGL-1 and NGL-2 do not bind LAR. This suggests that the extracellular domains of NGLs, despite their close amino acid sequence identity relative to the cytoplasmic region, are functionally distinct. Consistently, NGL-3 was more efficient than NGL-1 and NGL-2 in inducing presynaptic differentiation in mixed-culture assays, although they had comparable surface expression levels. Taken together with previous results, our data indicate that the two different types of NGL ligands, netrin-G1/2 and LAR, have both similar and contrasting features. Extensive alternative splicing is observed in both netrin-Gs21–23 and LAR43,44, suggesting that alternative splicing may regulate the NGL-3–LAR adhesion, as shown in the splicingdependent LAR interaction with nidogen35. NGL-3 expression, however, is probably not regulated by alternative splicing, as the NGL-3 gene (LRRC4B) has only two coding exons. Netrin-Gs are unique in that they are GPI-anchored proteins21,22. In addition, netrin-G2 does not induce postsynaptic differentiation in contacting dendrites when it is expressed in non-neural cells13. These results suggest that netrin-Gs may require an additional co-receptor for its functional interaction with NGL-2. In contrast, LAR is a transmembrane protein with the ability to induce postsynaptic differentiation in contacting dendrites, suggesting that LAR alone may be sufficient to function as a presynaptic NGL-3 receptor.

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

The NGL-3–LAR interaction is similar to the well-known adhesion between neuroligin and neurexin in that a heterophilic trans-synaptic adhesion mediates synapse formation in a bidirectional manner7–9. In addition, when it is expressed in non-neural cells, NGL-3 induced both excitatory and inhibitory presynaptic differentiation in contacting axons, similar to neuroligins9, which induce excitatory and inhibitory presynaptic differentiation by interacting with neurexins that are present in both glutamatergic and GABAergic axons9. Localization of LAR at excitatory synapses is supported by the role of LAR in the development and maintenance of excitatory synapses34. Whether LAR is present at inhibitory synapses remains to be determined. It should be noted that NGL-3 that is expressed in neurons selectively induces excitatory presynaptic contacts, contrary to the results from mixed cultured assays. This might be attributable to the interaction of NGL-3 with postsynaptic proteins that promote excitatory presynaptic differentiation in contacting axons. LAR expressed in non-neural cells selectively induced excitatory postsynaptic protein clustering in contacting dendrites, unlike neurexin 1b, which induces both excitatory and inhibitory postsynaptic protein clustering9. Neurexin 1b induces excitatory and inhibitory postsynaptic protein clustering by interacting with both neuroligin 1 and neuroligin 2 (ref. 9), which have specific excitatory and inhibitory synaptic localizations, respectively4,5. The selective excitatory postsynaptic protein clustering by LAR may arise from LAR’s association with NGL-3 at excitatory synapses. This is consistent with our observation that direct dendritic clustering of NGL-3 selectively induced excitatory postsynaptic protein clustering, and suggests that it is unlikely that there are unknown LAR-binding postsynaptic adhesion molecules at inhibitory synapses.

435

© 2009 Nature America, Inc. All rights reserved.

ARTICLES How might NGL-3 and LAR lead to post- and presynaptic differentiation, respectively? One possible mechanism by which synaptic adhesion molecules contribute to synapse formation is by interacting with and promoting the synaptic localization of specific membrane and cytoplasmic proteins. PSD-95, which interacts with NGL-3, is an abundant postsynaptic protein that interacts with a variety of membrane, signaling and scaffolding/adaptor proteins of excitatory synapses. Therefore, PSD-95 may couple NGL-3–dependent synaptic adhesion to the localization of various postsynaptic proteins. Liprin-a, which binds to the membrane-distal phosphatase domain of LAR, may act together with LAR to mediate NGL-3–dependent presynaptic differentiation. Liprin-a has been implicated in the regulation of active zone assembly and presynaptic development through its interactions with synaptic proteins, including ERC/ELKS, RIM and CASK28. Studies on liprin-a homologs in C. elegans (SYD-2) and Drosophila (Dliprin-a) have firmly established the roles of liprin-a in presynaptic differentiation31,32,45,46. Another LAR-interacting cytoplasmic protein that may contribute to LAR-dependent presynaptic differentiation is b-catenin, a component of the cadherin-catenin complex30. Presynaptic b-catenin localizes the reserve pool of vesicles to presynaptic sites through mechanisms involving interaction of its C terminus with synaptic PDZ proteins47. Notably, brain-derived neurotrophic factor activation of the TrkB receptor tyrosine kinase regulates tyrosine phosphorylation of b-catenin, which disrupts the cadherin–b-catenin interaction and promotes synaptic vesicle splitting and synapse formation48. Considering that LAR dephosphorylates b-catenin30,34, LAR and TrkB may reciprocally regulate b-catenin–dependent presynaptic vesicle clustering. One possibility is that the binding of NGL-3 to LAR may regulate the tyrosine phosphatase activity of LAR. Receptor tyrosine phosphatases can be negatively regulated by dimerization49. However, a study on the crystallographic structure of the phosphatase domain of LAR has suggested that the phosphatase activity of LAR is probably not regulated by dimerization50. In conclusion, our results suggest that the trans-synaptic interaction between NGL-3 and LAR regulates excitatory synapse formation in a bidirectional manner. Future investigations will aim to explore possible functional interactions between NGL-1/2–netrin-Gs and NGL-3–LAR adhesions. METHODS DNA constructs and antibodies. Full-length human NGL-1 (NM_020929, amino acids 1–641) and rat NGL-3 (XM_218615, amino acids 1–709) were subcloned into pEGFP-N1 (Clontech). Human LAR (amino acids 1–1881) was subcloned into pECFP-N1 (Clontech). For small interfering RNA knockdown, nucleotides 359–377 of rat NGL-3 (GCA AGA ATC TGG TGC GCA A), its point mutant (GCA AGT CTC TTG TGC GCA A) and nucleotides 1,254–1,272 (GCA CGA TGG CAC ACT CAA T) were subcloned into pSuper.gfp/ neo (OligoEngine) to generate shNGL-3 #1, shNGL-3* and shNGL-3 #2, respectively. The extracellular domain of LAR (amino acids 1–1,235) was subcloned into pEGFP-N1, in which EGFP was replaced with human Fc. Other constructs, antibodies and reagents are described in the Supplementary Methods online. Mixed-culture assay. Mixed-culture assays were carried out as described39. Briefly, primary hippocampal neuron cultures at 10 d in vitro (10 DIV) prepared from embryonic day 18–19 (E18–19) rats were cocultured with HEK293T cells expressing NGLs, LAR or EGFP, followed by immunostaining at 13 DIV. For the synaptotagmin I antibody uptake assay, neurons were incubated with antibodies to the synaptotagmin luminal domain at the dilution of 1:10 in isotonic depolarizing solution for 5 min. For competitive inhibition,

436

LAR-ecto-Fc proteins (10 mg ml
1) were added to NGL-3–expressing HEK293T cells that were cocultured with neurons for 3 d (10–13 DIV). Transfection of neurons and immunocytochemistry. Cultured neurons were transfected using a mammalian transfection kit (Clontech) and fixed with 4% paraformaldehyde/4% sucrose (vol/vol), permeabilized with 0.2% Triton X-100 (vol/vol) in phosphate-buffered saline, incubated with primary antibodies and then incubated with Cy3-, Cy5- or FITC-conjugated secondary antibodies (Jackson ImmunoResearch). For the inhibition of endogenous synapses with soluble LAR, neurons were treated with LAR-ecto-Fc for 3 d (20 mg ml
1, 7/8–10/11 DIV) and transfected with pEGFP-N1 for visualization (9/10–10/11 DIV). Bead and antibody aggregation assays. Bead aggregation assays were performed as described previously13. Briefly, neutravidin-conjugated FluoSphere beads (Molecular Probes) were pre-incubated with biotin-conjugated antibodies to EGFP. The antibody-coated beads were added onto neurons expressing EGFP–NGL-3 (14–16 DIV) and cultured for 24 h. For antibody aggregation, antibodies to EGFP were clustered by FITC-conjugated antibodies to guinea pig in complete neurobasal medium and added to neurons at 16 DIV, followed by 24-h culture and immunofluorescence staining at 17 DIV. Image acquisition and quantification. All z-stacked images were randomly acquired by confocal microscopy (LSM510, Zeiss) and analyzed with MetaMorph image analysis software (Universal Imaging). The density of synaptic protein clusters was measured from 15–30 neurons; the primary dendritic lengths of B50 mm from the cell body were measured for each neuron. For quantification of images from coculture assays, all captured images were thresholded and the integrated intensity of the clusters on transfected HEK293T cells was normalized to the cell area. All values are presented as mean ± s.e.m. and analyzed by Student’s t test or ANOVA Tukey’s test. Quantitative cell surface–binding assay. HEK293T cells transfected with NGL-3 were transferred to 96-well plates and grown for 24 h. After fixation in 4% paraformaldehyde/4% sucrose, cells were incubated with increasing concentrations of LAR-ecto-Fc for 1 h, incubated with horseradish peroxidase– conjugated antibodies to human Fc (Sigma, 1:10,000) and color reacted with TMB (Sigma). Electrophysiology. Cultured pyramidal neurons from the hippocampus transfected with shNGL-3 (13–16 DIV) were whole-cell voltage clamped at
60 mV using an Axopatch 200B amplifier (Axon Instruments). The extracellular solution contained 145 mM NaCl, 2.5 mM KCl, 10 mM HEPES, 1.25 mM NaH2PO4, 2 mM CaCl2, 1 mM MgCl2, 10 mM glucose and 0.4 mM sodium ascorbate. The intracellular solution contained 100 mM potassium gluconate, 20 mM KCl, 10 mM HEPES, 8 mM NaCl, 4 mM magnesium ATP, 0.3 mM sodium GTP and 0.5 mM EGTA. For mEPSC measurement, tetrodotoxin (10 nM, Tocris) and bicuculline (10 mM, Tocris) were added into the extracellular solution. Synaptic currents were analyzed using a custom-written macro in Igor Pro (WaveMetrics). Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We would like to thank A.M. Craig for neurexin 1a, 2a, 3a and 1b cDNAs, T. Biederer for SynCAM 1, Y.-P. Hsueh for Syndecan-2, P. Maness for NCAM-140, J. Ko for quantitative analysis, J. Nam for SALM4, M.-H. Kim for the mini analysis program, and Y.-G. Oh, M.-S. Baek, M. Ryu and J.-G. Jung for the help with antibody generation. This work was supported by the National Creative Research Initiative Program of the Korean Ministry of Science and Technology (E.K.). AUTHOR CONTRIBUTIONS J.W. and S.-K.K. carried out the experiments, analyzed the data and wrote the manuscript. S.C. helped with the shRNA knockdown experiments. S.K., J.-R.L., A.W.D. and M.S. contributed reagents. E.K. supervised the project and wrote the manuscript. All authors discussed the results and commented on the manuscript.

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

© 2009 Nature America, Inc. All rights reserved.

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

1. Yamagata, M., Sanes, J.R. & Weiner, J.A. Synaptic adhesion molecules. Curr. Opin. Cell Biol. 15, 621–632 (2003). 2. Dalva, M.B., McClelland, A.C. & Kayser, M.S. Cell adhesion molecules: signaling functions at the synapse. Nat. Rev. Neurosci. 8, 206–220 (2007). 3. Washbourne, P. et al. Cell adhesion molecules in synapse formation. J. Neurosci. 24, 9244–9249 (2004). 4. Akins, M.R. & Biederer, T. Cell-cell interactions in synaptogenesis. Curr. Opin. Neurobiol. 16, 83–89 (2006). 5. Craig, A.M. & Kang, Y. Neurexin-neuroligin signaling in synapse development. Curr. Opin. Neurobiol. 17, 43–52 (2007). 6. McAllister, A.K. Dynamic aspects of CNS synapse formation. Annu. Rev. Neurosci. 30, 425–450 (2007). 7. Ichtchenko, K. et al. Neuroligin 1: a splice site–specific ligand for beta-neurexins. Cell 81, 435–443 (1995). 8. Scheiffele, P., Fan, J., Choih, J., Fetter, R. & Serafini, T. Neuroligin expressed in nonneuronal cells triggers presynaptic development in contacting axons. Cell 101, 657–669 (2000). 9. Graf, E.R., Zhang, X., Jin, S.X., Linhoff, M.W. & Craig, A.M. Neurexins induce differentiation of GABA and glutamate postsynaptic specializations via neuroligins. Cell 119, 1013–1026 (2004). 10. Biederer, T. et al. SynCAM, a synaptic adhesion molecule that drives synapse assembly. Science 297, 1525–1531 (2002). 11. Fogel, A.I. et al. SynCAMs organize synapses through heterophilic adhesion. J. Neurosci. 27, 12516–12530 (2007). 12. Lin, J.C., Ho, W.H., Gurney, A. & Rosenthal, A. The netrin-G1 ligand NGL-1 promotes the outgrowth of thalamocortical axons. Nat. Neurosci. 6, 1270–1276 (2003). 13. Kim, S. et al. NGL family PSD-95–interacting adhesion molecules regulate excitatory synapse formation. Nat. Neurosci. 9, 1294–1301 (2006). 14. Kayser, M.S., McClelland, A.C., Hughes, E.G. & Dalva, M.B. Intracellular and transsynaptic regulation of glutamatergic synaptogenesis by EphB receptors. J. Neurosci. 26, 12152–12164 (2006). 15. Boucard, A.A., Chubykin, A.A., Comoletti, D., Taylor, P. & Sudhof, T.C. A splice code for trans-synaptic cell adhesion mediated by binding of neuroligin 1 to alpha- and betaneurexins. Neuron 48, 229–236 (2005). 16. Chih, B., Gollan, L. & Scheiffele, P. Alternative splicing controls selective trans-synaptic interactions of the neuroligin-neurexin complex. Neuron 51, 171–178 (2006). 17. Graf, E.R., Kang, Y., Hauner, A.M. & Craig, A.M. Structure function and splice site analysis of the synaptogenic activity of the neurexin-1 beta LNS domain. J. Neurosci. 26, 4256–4265 (2006). 18. Irie, M. et al. Binding of neuroligins to PSD-95. Science 277, 1511–1515 (1997). 19. Tabuchi, K. et al. A neuroligin-3 mutation implicated in autism increases inhibitory synaptic transmission in mice. Science 318, 71–76 (2007). 20. Jamain, S. et al. Reduced social interaction and ultrasonic communication in a mouse model of monogenic heritable autism. Proc. Natl. Acad. Sci. USA 105, 1710–1715 (2008). 21. Nakashiba, T. et al. Netrin-G1: a novel glycosyl phosphatidylinositol–linked mammalian netrin that is functionally divergent from classical netrins. J. Neurosci. 20, 6540–6550 (2000). 22. Nakashiba, T., Nishimura, S., Ikeda, T. & Itohara, S. Complementary expression and neurite outgrowth activity of netrin-G subfamily members. Mech. Dev. 111, 47–60 (2002). 23. Yin, Y., Miner, J.H. & Sanes, J.R. Laminets: laminin- and netrin-related genes expressed in distinct neuronal subsets. Mol. Cell. Neurosci. 19, 344–358 (2002). 24. Nishimura-Akiyoshi, S., Niimi, K., Nakashiba, T. & Itohara, S. Axonal netrin-Gs transneuronally determine lamina-specific subdendritic segments. Proc. Natl. Acad. Sci. USA 104, 14801–14806 (2007). 25. Zhang, W. et al. Netrin-G2 and netrin-G2 ligand are both required for normal auditory responsiveness. Genes Brain Behav. 7, 385–392 (2008).

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

26. Aoki-Suzuki, M. et al. A family-based association study and gene expression analyses of netrin-G1 and -G2 genes in schizophrenia. Biol. Psychiatry 57, 382–393 (2005). 27. Johnson, K.G. & Van Vactor, D. Receptor protein tyrosine phosphatases in nervous system development. Physiol. Rev. 83, 1–24 (2003). 28. Stryker, E. & Johnson, K.G. LAR, liprin alpha and the regulation of active zone morphogenesis. J. Cell Sci. 120, 3723–3728 (2007). 29. Serra-Pages, C. et al. The LAR transmembrane protein tyrosine phosphatase and a coiled-coil LAR-interacting protein colocalize at focal adhesions. EMBO J. 14, 2827–2838 (1995). 30. Kypta, R.M., Su, H. & Reichardt, L.F. Association between a transmembrane protein tyrosine phosphatase and the cadherin-catenin complex. J. Cell Biol. 134, 1519–1529 (1996). 31. Zhen, M. & Jin, Y. The liprin protein SYD-2 regulates the differentiation of presynaptic termini in C. elegans. Nature 401, 371–375 (1999). 32. Kaufmann, N., DeProto, J., Ranjan, R., Wan, H. & Van Vactor, D. Drosophila liprin-alpha and the receptor phosphatase Dlar control synapse morphogenesis. Neuron 34, 27–38 (2002). 33. Ackley, B.D. et al. The two isoforms of the Caenorhabditis elegans leukocyte-common antigen related receptor tyrosine phosphatase PTP-3 function independently in axon guidance and synapse formation. J. Neurosci. 25, 7517–7528 (2005). 34. Dunah, A.W. et al. LAR receptor protein tyrosine phosphatases in the development and maintenance of excitatory synapses. Nat. Neurosci. 8, 458–467 (2005). 35. O’Grady, P., Thai, T.C. & Saito, H. The laminin-nidogen complex is a ligand for a specific splice isoform of the transmembrane protein tyrosine phosphatase LAR. J. Cell Biol. 141, 1675–1684 (1998). 36. Johnson, K.G. et al. The HSPGs Syndecan and Dallylike bind the receptor phosphatase LAR and exert distinct effects on synaptic development. Neuron 49, 517–531 (2006). 37. Fox, A.N. & Zinn, K. The heparan sulfate proteoglycan syndecan is an in vivo ligand for the Drosophila LAR receptor tyrosine phosphatase. Curr. Biol. 15, 1701–1711 (2005). 38. Yang, T. et al. Leukocyte antigen-related protein tyrosine phosphatase receptor: a small ectodomain isoform functions as a homophilic ligand and promotes neurite outgrowth. J. Neurosci. 23, 3353–3363 (2003). 39. Biederer, T. & Scheiffele, P. Mixed-culture assays for analyzing neuronal synapse formation. Nat. Protoc. 2, 670–676 (2007). 40. Ko, J. et al. SALM synaptic cell adhesion–like molecules regulate the differentiation of excitatory synapses. Neuron 50, 233–245 (2006). 41. Zhang, J.S., Honkaniemi, J., Yang, T., Yeo, T.T. & Longo, F.M. LAR tyrosine phosphatase receptor: a developmental isoform is present in neurites and growth cones and its expression is regional- and cell-specific. Mol. Cell. Neurosci. 10, 271–286 (1998). 42. Pulido, R., Serra-Pages, C., Tang, M. & Streuli, M. The LAR/PTP delta/PTP sigma subfamily of transmembrane protein-tyrosine- phosphatases: multiple human LAR, PTP delta and PTP sigma isoforms are expressed in a tissue-specific manner and associate with the LAR- interacting protein LIP.1. Proc. Natl. Acad. Sci. USA 92, 11686–11690 (1995). 43. O’Grady, P., Krueger, N.X., Streuli, M. & Saito, H. Genomic organization of the human LAR protein tyrosine phosphatase gene and alternative splicing in the extracellular fibronectin type-III domains. J. Biol. Chem. 269, 25193–25199 (1994). 44. Zhang, J.S. & Longo, F.M. LAR tyrosine phosphatase receptor: alternative splicing is preferential to the nervous system, coordinated with cell growth and generates novel isoforms containing extensive CAG repeats. J. Cell Biol. 128, 415–431 (1995). 45. Patel, M.R. et al. Hierarchical assembly of presynaptic components in defined C. elegans synapses. Nat. Neurosci. 9, 1488–1498 (2006). 46. Dai, Y. et al. SYD-2 Liprin-alpha organizes presynaptic active zone formation through ELKS. Nat. Neurosci. 9, 1479–1487 (2006). 47. Bamji, S.X. et al. Role of beta-catenin in synaptic vesicle localization and presynaptic assembly. Neuron 40, 719–731 (2003). 48. Bamji, S.X., Rico, B., Kimes, N. & Reichardt, L.F. BDNF mobilizes synaptic vesicles and enhances synapse formation by disrupting cadherin-beta-catenin interactions. J. Cell Biol. 174, 289–299 (2006). 49. Majeti, R. & Weiss, A. Regulatory mechanisms for receptor protein tyrosine phosphatases. Chem. Rev. 101, 2441–2448 (2001). 50. Nam, H.J., Poy, F., Krueger, N.X., Saito, H. & Frederick, C.A. Crystal structure of the tandem phosphatase domains of RPTP LAR. Cell 97, 449–457 (1999).

437

ARTICLES

Altered chloride homeostasis removes synaptic inhibitory constraint of the stress axis

© 2009 Nature America, Inc. All rights reserved.

Sarah A Hewitt1,2,6, Jaclyn I Wamsteeker1,2, Ebba U Kurz3,4 & Jaideep S Bains1,5 In mammals, stress elicits a stereotyped endocrine response that requires an increase in the activity of hypothalamic parvocellular neuroendocrine neurons. The output of these cells is normally constrained by powerful GABA-mediated synaptic inhibition. We found that acute restraint stress in rats released the system from inhibitory synaptic drive in vivo by down-regulating the transmembrane anion transporter KCC2. This manifested as a depolarizing shift in the reversal potential of GABAA-mediated synaptic currents that rendered GABA inputs largely ineffective. Notably, repetitive activation of GABA synapses after stress resulted in a more rapid collapse of the anion gradient and was sufficient to increase the activity of neuroendocrine cells. Our data indicate that hypothalamic neurons integrate psychological cues to mount the endocrine response to stress by regulating anion gradients.

The brain responds to perceived stress by invoking neural and hormonal changes that have widespread effects throughout the body. In mammals, stress activates corticotrophin-releasing hormone neurons in the paraventricular nucleus of the hypothalamus (PVN), which in turn increase circulating levels of corticosteroids. The activity of these neurons is tightly regulated by GABAergic synaptic input1,2 that originates from interneuron populations located adjacent to the nucleus1–3. This inhibition is critical, as manipulations that decrease GABA drive to these cells increase neural activity and circulating corticosteroids. Given that synaptic inhibition actively restrains the output of this system, it has been proposed that release from inhibition is necessary for the initiation of the neuroendocrine response to stress4. However, the cellular mechanisms through which this inhibitory release may be accomplished have not been resolved. In the CNS, fast GABA transmission relies on the influx of Cl– through the GABAA receptor. Influx of Cl– is exquisitely sensitive to the ionic electrochemical gradient5,6. Anionic Cl– homeostasis in adults is primarily maintained by KCC2 and previous work has demonstrated that disruption of transporter activity via pharmacological or physiological manipulations strongly affects the efficacy of synaptic inhibition7–11. We hypothesized that alterations in postsynaptic Cl– homeostasis, which ultimately reduce synaptic inhibition, may contribute to the disinhibition and increased excitability of neuroendocrine neurons following acute stress. Using an in vivo restraint stress protocol combined with in vivo microinjections, hormone measurements and patch clamp recordings from brain slices prepared from control and stressed animals, we found that a single bout of acute stress neutralized inhibitory signaling via a loss of KCC2 function and a resulting decrease in Cl– extrusion capacity.

RESULTS Acute stress decreases synaptic GABAA-mediated inhibition We first established that a 30-min restraint stress protocol was sufficient to increase circulating corticosteroids. Immediately following the protocol, we measured a robust and reliable increase in circulating corticosteroids (control ¼ 28.9 ± 7.3 ng ml–1, n ¼ 6; 30-min protocol ¼ 117.9 ± 20.1 ng ml–1, n ¼ 6; Fig. 1a). Because a 60-min restraint elicited a similar elevation of corticosteroids (122.9 ± 9.7, n ¼ 6, P o 0.01 compared with control), we used a 30-min restraint stress for all subsequent experiments. We then conducted in vivo experiments to examine the effect of synaptic GABAergic inhibition after stress. Pharmacological blockade of GABAA receptors in PVN elicits a robust increase in circulating corticosteroids4 in control conditions. Because an effective stress response requires a disinhibition of neuroendocrine neurons, we hypothesized that synaptic GABAA-mediated inhibition may be compromised following stress. To test this possibility, we injected the GABAA antagonist bicuculline (2 ml, 0.1 nM) directly into PVN and measured the corticosteroid response at different times postinjection (Fig. 1b). Consistent with a previous report4, this manipulation elicited a robust increase in circulating corticosteroids in control animals (1 min ¼ 183.3 ± 30.6%, 5 min ¼ 238.5 ± 54.4%, 10 min ¼ 267.6 ± 58.1%; all values reported as a percentage of control). In contrast, bicuculline microinjection had no effect on circulating corticosteroids in stressed animals (1 min ¼ 105.0 ± 8.2%, 5 min ¼ 110.5 ± 6.3%, 10 min ¼ 112.4 ± 6.8%; P o 0.05 for each time point compared with identical time points in control animals). This observation is consistent with our hypothesis that, following stress, the ability of inhibitory synaptic GABAA to dampen the activity of neuroendocrine cells and curb corticosteroid release is markedly reduced.

1Hotchkiss

Brain Institute, Departments of 2Neuroscience, and 3Pharmacology and Therapeutics, 4Southern Alberta Cancer Research Institute, 5Department of Physiology and Biophysics, University of Calgary, Calgary, Alberta, Canada. 6Present address: Department of Chemical and Biological Science, Mount Royal College, Calgary, Alberta, Canada. Correspondence should be addressed to J.S.B. ([email protected]). Received 16 October 2008; accepted 14 January 2009; published online 1 March 2009; doi:10.1038/nn.2274

438

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

b

200

**

**

100

0

Percentage of baseline

Corticosterone (ng ml–1)

a

*

Bicuculline

*

300

*

200

Control Stress

100

0

30 60 0 Duration of restraint (min)

–5

0

5 10 Time (min)

15

10 µM bicuculline

c Control

Stress

e

2

Percentage of control firing rate

Normalized spike frequency

© 2009 Nature America, Inc. All rights reserved.

d

1

0

10 µM bicuculline Control (n = 7) Stress (n = 5) 0

5 Time (min)

*** 300

** 200

100

0

10

Control Stress

To directly test whether this was the result of a loss of inhibitory drive to parvocellular neuroendocrine cells (PNCs), we carried out tight-seal cell-attached recordings, which do not perturb intracellular Cl–, and assessed the effects of bicuculline (10 mM) on spike activity. Bicuculline increased the activity of neurons in controls (170 ± 20%, n ¼ 7; Fig. 1c–e) but had no effect on neurons recorded from slices obtained from stressed animals (91.8 ± 17%, n ¼ 5; P ¼ 0.019 versus effects of bicuculline on control cells; Fig. 1c–e). Collectively, these observations indicate that GABAA-mediated inhibition is absent following stress. Acute stress causes a depolarizing shift in EGABA We next tested for a possible mechanism that would explain this loss of synaptic inhibition following stress. A number of observations demonstrate that the strength of GABA drive can be modified by changes in transmembrane ionic gradients9,12–15. Signaling through the GABAA receptor depends on the electrochemical gradient for Cl–, which is determined in turn by the activity of transmembrane transporters, specifically the Na/K/2Cl co-transporter NKCC1 and the K/Cl cotransporter KCC213,14,16–18. Developmental studies indicate that there is a decline in NKCC1 expression following birth and a concomitant increase in KCC2 expression, which leaves KCC2 as the predominant determinant of Cl– homeostasis in the mature nervous system19–21. By harnessing the potassium electrochemical gradient to extrude Cl– ions, KCC2 ensures an inward driving force for Cl– at resting membrane potentials5,12,17,18,22. Acute depolarizing shifts in EGABA have previously been observed following injury7,23, cellular oxidative stress11 and seizure activity8–11,24. In contrast, immature neurons undergo a slower shift from a depolarizing to a hyperpolarizing Cl– flux, reflecting a change from high to low intracellular Cl– concentrations, during the course of development19–21. In each case, these shifts are coupled to changes in the Cl– extrusion capacity of KCC2. Therefore, we hypothesized that the apparent disinhibition of neuroendocrine cells after stress mirrors a loss of Cl–

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

Figure 1 Acute stress reduces the strength of synaptic GABAA-mediated inhibition. (a) Acute restraint stress (30 or 60 min) significantly increased circulating corticosteroids (n ¼ 6 for each group, P o 0.01 compared with control). (b) Summary of the percentage change in circulating corticosteroids following microinjection of the GABAA antagonist bicuculline into PVN. Corticosteroids levels in naive, unstressed rats (n ¼ 10) were significantly increased compared with those in stressed rats (n ¼ 7) (P o 0.05 compared with vehicle injection). (c) Cell-attached recordings showed effects of bath application of bicuculline (10 mM) in controls (left) and after stress (right). (d) Summary data showing that blockade of GABAA receptors caused an increase in neuronal activity in control, but had no effect following stress. (e) Effect of bicuculline on spike rate in each cell tested under control and stress conditions. The firing rate was calculated 5 min after onset of bicuculline (control, 170 ± 20% of baseline; stress, 92 ± 17% of baseline; * P o 0.05; ** P o 0.01 versus control baseline, *** P o 0.01 versus bicuculline in stress). All values are mean ± s.e.m.

extrusion capacity through a downregulation in the activity of KCC2. To directly test for changes in Cl– homeostasis following stress, we measured EGABA in slices obtained from an acutely stressed animal. Using gramicidin (40 mM) perforated patch recordings25, we first examined the effect of postsynaptic membrane voltage on evoked inhibitory postsynaptic currents (eIPSCs) in control slices (Fig. 2a). When this experiment was repeated following stress, we observed a depolarizing shift in EGABA (control ¼
70 ± 3.4 mV, n ¼ 11; stress ¼
55 ± 2.3 mV, n ¼ 7; P o 0.01; Fig. 2a). A depolarizing shift in EGABA was also observed when the whole-cell recording configuration was used (control ¼ –54.6 ± 1.6 mV, n ¼ 17; stress ¼
47.9 ± 1.8 mV, n ¼ 21; P o 0.01; Supplementary Fig. 1 online). This is consistent with previous reports showing that EGABA is a reliable reporter for changes in Cl– homeostasis, even in whole-cell conditions12. Consequently, unless otherwise specified, we used the whole-cell recording configuration for the remainder of the experiments in which we assessed EGABA. Both chronic stress and prolonged application of corticosteroids are reported to decrease the release probability at GABA synapses in PVN26 and may also decrease the expression of GABAA receptors27. We found no changes in release probability (as assessed by calculating the paired pulse ratio, PPR) following acute stress (PPRcontrol ¼ 0.84 ± 0.1, n ¼ 21; PPRstress ¼ 0.93 ± 0.1, n ¼ 20; P 4 0.05). Our observations indicate that stress causes a depolarizing shift in EGABA and suggest that mechanisms that control the transmembrane anion gradient, such as KCC2, may be important in promoting synaptic disinhibition in vivo. EGABA is determined by KCC2 To determine the contributions of KCC2 to setting EGABA under these conditions, we first used a pharmacological inhibitor of KCC2, furosemide (200 mM), and tested its effects on EGABA. Consistent with previous reports12, this manipulation caused a depolarizing shift in EGABA (whole-cell recording: control =
59.1 ± 1.4 mV, n = 21; furosemide ¼
48.0 ± 2.2mV, n ¼ 16, P o 0.05 compared with control; Fig. 2b). To demonstrate that KCC2 inhibition is responsible for the stress-induced shift in EGABA, we examined the effects of furosemide on slices obtained from stressed animals. In these conditions, furosemide elicited no further changes in EGABA (whole-cell recording: paired data, stress ¼
50.0 ± 3.0 mV, stress plus furosemide ¼
49.1 ± 2.3 mV, n ¼ 6, P 4 0.05; Supplementary Fig. 2 online). Because furosemide can also partially inhibit NKCC1 (refs. 5,14,19,28), we conducted additional experiments using the specific NKCC1 inhibitor bumetanide (10 mM). Bumetanide had no effect on EGABA (whole-cell recording: control ¼ –54.6 ± 1.6 mV, n ¼ 17; bumetanide ¼
56.6 ± 2.5, n ¼ 4; P 4 0.05), thereby ruling out any involvement for NKCC1. These data demonstrate that KCC2 is an important determinant of the intracellular

439

ARTICLES

Control

25

Furosemide –50

c 300

0

–25 Stress –50 –80

© 2009 Nature America, Inc. All rights reserved.

0

–100 –80 –60 –40 –20 Holding potential (mV)

–60 Holding potential (mV)

–40

Percentage of baseline

IPSC amplitude (pA)

50

50

Furosemide

*

*

200

* 100

0 –5

0

5 10 15 20 Time (min)

Cl– concentration in PNCs and ensures that Cl– extrusion is adequate to maintain an inward driving force for the anion at rest. Finally, we hypothesized that if corticosteroid release during stress results from a loss of synaptic inhibition due to compromised KCC2 activity, then inhibition of KCC2 in vivo should also increase circulating corticosteroids. Furosemide (200 mM) was microinjected directly into PVN of control rats (n ¼ 7–10; Fig. 2c) and blood corticosteroid levels were sampled at different time points. This elicited a robust increase in circulating corticosteroids (1 min ¼ 144 ± 9.4%, 10 min ¼ 220 ± 19.2%, 15 min = 216 ± 29%,; P o 0.05 for each time point compared with identical time points in control animals). Taken together, these observations indicate that stress decreases KCC2 activity and that KCC2 activity is necessary to maintain synaptic inhibition of the hypothalamic-pituitary-adrenal (HPA) axis in vivo. Mechanisms for KCC2 downregulation Changes in Cl– homeostasis can be the result of either changes in the expression of KCC2 (refs. 7,29) or changes in transporter activity30,31. To determine whether the expression of KCC2 was decreased following stress, we prepared whole-cell extracts from dissected hypothalamic tissue and assessed KCC2 expression levels by immunoblotting with an antibody to KCC2. In extracts from both control and stressed animals, this revealed two prominent bands at B140 kDa and B270 kDa, which is consistent with the previously reported presence of monomeric and dimeric KCC2 protein16,32 (Fig. 3a). There was no difference in the expression of total protein between control and stress samples (n ¼ 6; Fig. 3a). Because the oligomerization of KCC2 appears to be necessary for conferring activity32, we tested whether there may be a change in the relative ratio of the dimeric (active) band to the monomeric (inactive) band between control and stress samples. Densitometric analysis revealed no difference in the dimer/monomer ratio between the two conditions (control ¼ 0.98 ± 0.16, stress ¼ 0.88 ± 0.25, n ¼ 6, P ¼ 0.6; Fig. 3b). Figure 3 Stress-induced shift in EGABA is not associated with a change in protein expression and can be mimicked by a1 adrenoceptor activation. (a) Immunoblots of KCC2 from PVN extracts in control and stress. Bands are B140 kDa and B270 kDa. (b) Quantification of optical density measurements of ratio of dimer/monomer bands showed no change in relative ratio of oligomers after stress (P 4 0.05). (c) I-V plot, obtained using wholecell recordings, showing that incubation of control slices with phenylephrine (100 mM for 30 min) resulted in a depolarizing shift in EGABA (n ¼ 8, P o 0.01 versus control). All values are mean ± s.e.m.

440

Figure 2 KCC2 regulates Cl
homeostasis. (a) Gramicidin recordings show IPSCs from a cell in a control slice (black) and a cell in a slice obtained from a stressed animal (green). I–V plot showed a reversal of IPSCs in control (black, n ¼ 11) and after stress (green, n ¼ 7). Holding potential for traces corresponds with each point in the I-V plot. (b) Whole-cell recordings showed that bath application of furosemide (200 mM) caused a depolarizing shift in the I–V relationship (EGABA control ¼ –59.1 ± 1.4 mV, n ¼ 21; EGABA furosemide ¼
48.0 ± 2.2 mV, n ¼ 16; P o 0.05; control shown in black, furosemide shown in orange). Inset traces are from holding potentials corresponding to the I–V plot. (c) Summary of the percentage change in circulating corticosteroids following microinjection of furosemide (200 mM) into PVN (n ¼ 7–10). Scale bars represent 50 pA and 10 ms. All values are mean ± s.e.m. * P o 0.05.

The activation of the HPA axis and the release of corticosteroids in response to stress both require the release of noradrenaline into PVN33. Furthermore, changes in KCC2 activity have been reported to be dependent on the activation of second messengers, such as protein kinase C, and an increase in intracellular Ca2+ (ref. 30). Because PNCs express functional a1 adrenoceptors34 that are coupled to protein kinase C and can increase the release of Ca2+ from intracellular stores, we hypothesized that their stimulation by the exogenous a1 ligand phenylephrine may be sufficient to decrease KCC2 activity. To test this idea, we incubated slices with phenylephrine (100 mM, 30 min) and evoked IPSCs in whole-cell configuration to determine EGABA. In the phenylephrine-treated slices, EGABA was
46 ± 2.5 mV (n ¼ 8, whole cell recording; Fig. 3c), which was significantly more depolarized than controls (P o 0.01 versus control; Supplementary Fig. 1). Together, these results indicate that KCC2 protein expression is unaffected by stress, but the activity of the transporter can be attenuated by activation of a1 adrenoceptors. Cl– gradient collapse during repetitive synaptic activity Computational analyses of changes in Cl– extrusion capacity indicate that shifts in EGABA strongly affect GABA inhibition6. Furthermore, sustained Cl– extrusion is necessary to prevent the collapse of the Cl– gradient during repetitive synaptic activity15,30,35,36. To examine the contribution of KCC2-mediated Cl– extrusion during repetitive activity in control slices, we conducted experiments in conditions that either favored Cl– influx (cell membrane voltage clamped to 0 mV) or efflux (cell membrane voltage clamped to
80 mV) through the GABAA receptor (Supplementary Fig. 3 online). This allowed us to isolate the relative contribution of KCC2-mediated Cl– extrusion to the development of activity-dependent synaptic depression. To maintain

a

Control

c

Stress

250

100

50

KCC2

Control

150

IPSC amplitude (pA)

IPSC amplitude (pA)

a 75

100 Control

100 Actin

b Dimer/monomer ratio

b

1.5

0

–50

–100

1.0

Phenylephrine

–150 0.5 0.0

Control

VOLUME 12

–200 –80

Stress

[

NUMBER 4

[

–40 –60 Holding potential (mV)

APRIL 2009 NATURE NEUROSCIENCE

a

1–3

IPSC Control

13–15

23–25

Vm = 0 mV Stress Vm = 0 mV

Percentage of IPSC event 1

ARTICLES 100 75 50

0 mV control

25 0 mV stress

0

0.0 0.1 0.2 0.3 0.4 0.5 Time (s)

12 10 8 6 4 2 0

2

4

6

8

6 2 4 Time (s)

0

8

20 Hz 12 10 8 6 4 2 0 12 10 8 6 4 2 0

10

–2

0

*

1.25 1.00

Stress Control

0.75 –2

0

6 2 4 Time (s)

8

2

0

1.50

1.50

Control

4

6

2 4 6 Time (s)

10

8

10

* **

1.25 1.00 Stress Control

0.75

10

8

Stress

–2

10

Normalized spike frequency

d

0

Stress

–2

Normalized spike frequency

© 2009 Nature America, Inc. All rights reserved.

–2

c

Control

Spike frequency (Hz)

Spike frequency (Hz) Spike frequency (Hz)

10 Hz 12 10 8 6 4 2 0

Spike frequency (Hz)

b

–2

0

2 4 6 Time (s)

8

10

synaptic strength during a burst of presynaptic activity under conditions of high Cl– influx, KCC2 must rapidly extrude Cl–. At 0 mV, a brief train (six pulses) resulted in the development of frequencydependent synaptic depression (Supplementary Fig. 3). In contrast, the activation of GABAA receptors at
80 mV results in Cl– efflux and imposes no load on KCC2. Consequently, the collapse of the Cl– gradient should not contribute to the development of depression. Under these conditions, depression was less robust and was only partly frequency dependent (Supplementary Fig. 3). Thus, Cl
extrusion is a critical determinant of synaptic depression and the activity of the cell would be highly affected by alterations in its ability to extrude Cl
. Because stress downregulates KCC2 function and compromises Cl
extrusion capacity, one prediction is that depression during high-frequency synaptic stimulation should be more pronounced following stress. We tested this idea by using a longer, 500-ms 50-Hz train to elicit robust depression at holding potentials favoring the influx of Cl
(Ehold ¼ 0 mV) in slices obtained from control and stressed animals. Sample traces from eIPSC 1–3, 12–15 and 23–25 during the train were plotted, which revealed that activity-dependent depression was more robust following stress (Fig. 4a). In contrast, no difference in synaptic depression was observed between control and stress when the Cl
driving force was outward (Ehold ¼ –80 mV, data not shown), suggesting that the effects of acute stress on short-term depression are derived from changes in the efficacy of Cl
transport. These observations indicate that the decrease in KCC2-mediated Cl– efflux following stress compromises the ability of GABA synapses to sustain inhibition when activated repetitively.

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

Figure 4 Effects of repetitive synaptic stimulation on GABA IPSCs and postsynaptic activity. (a) Left, IPSCs recorded at 0 mV during 500-ms, 50-Hz pulse trains under control and stress conditions. Sample IPSCs are shown for events 1–3, 13–15 and 23–25. Right, summary of synaptic depression under stress (filled squares) and control (open circles) conditions. Scale bars represent 200 pA (top trace) or 50 pA (bottom trace) and 20 ms. (b) Top, effect of 10-Hz stimulation on firing in a single control cell (cell-attached recording). Each of seven trials is shown. The solid black line represents the average of all trials in this cell. Bottom, 10-Hz stimulation in a cell from a stressed rat. (c) Top, effect of 20-Hz stimulation on firing in a single control cell as in b. The solid black line represents the average of all eight trials. Bottom, 20-Hz stimulation in a cell from a stressed rat. (d) Left, summary of six control cells and nine stressed cells in response to a 10-Hz synaptic stimulation. The increase in firing after synaptic stimulation in stress was significantly different from the firing rate before stimulation (P o 0.05). Right, summary of six control cells and nine stressed cells in response to a 20-Hz stimulation. The increase in firing after synaptic stimulation in stress was significantly different from the firing rate before stimulation (P o 0.05) and from the firing rate at the same time point following stimulation in control slices (P o 0.05). All values are mean ± s.e.m. * P o 0.05, ** P o 0.01.

GABA signaling is conditionally excitatory after stress Previous reports indicate that epileptiform activity8–11 and injuryinduced decreases in spinal KCC2 (ref. 7) result in GABA-mediated depolarizations and an increase in neuronal excitability. To examine the consequences of impaired Cl– extrusion at high rates of activity, we carried out experiments using the cell-attached recording mode and stimulated synaptic GABA inputs at 10 or 20 Hz for 1 s. Seven to nine trials were conducted at each frequency in each cell and the data from all trials at each frequency were averaged for a given cell (Fig. 4b–d). Under control conditions, stimulation of GABA inputs (10 Hz, 1 s) had no effect on firing (prestimulation, 0.99 ± 0.01; poststimulation, 1.00 ± 0.02; n ¼ 6 cells, P 4 0.05; Fig. 4b,d). Under stress conditions, the same stimulation protocol elicited a transient, but significant, increase in spike activity (prestimulation, 1.01 ± 0.02; poststimulation, 1.2 ± 0.07; n ¼ 9, P o 0.05; Fig. 4b,d). In response to 20-Hz stimulation, control cells showed a decrease in spiking immediately after the synaptic stimulation (prestimulation, 1.00 ± 0.02; poststimulation, 0.88 ± 0.04; n ¼ 6, P o 0.05; Fig. 4c,d). Under stress conditions, 20-Hz stimulation caused an increase in spike activity (prestimulation, 1.00 ± 0.01; poststimulation, 1.22 ± 0.09; n ¼ 9, P o 0.05; Fig. 4c,d). Increasing the rate of synaptic activation to 50 Hz elicited a more robust inhibition of firing (prestimulation, 1.01 ± 0.02; poststimulation, 0.25 ± 0.05; n ¼ 7, P o 0.01). Following stress, however, the same stimulation protocol failed to modify firing rate (prestimulation, 0.99 ± 0.01; poststimulation, 1.06 ± 0.04; n ¼ 7, P 4 0.05). These observations demonstrate that there is a loss of synaptic inhibition at 50 Hz, but not an accompanying GABA-mediated excitation, as was observed at 10- and 20-Hz stimulation, indicating that additional mechanisms, probably resulting from the spillover of GABA and the activation of extrasynaptic GABAB receptors, contribute to the prolonged inhibition following 50-Hz stimulation. Collectively, these experiments indicate that repetitive activation of GABA inputs following stress causes a collapse of the Cl– gradient that, under specific conditions, is sufficient to promote GABA-mediated excitations. DISCUSSION Our data demonstrate that neural pathways activated during stress target KCC2 in neuroendocrine command cells in the PVN that control the output of the HPA axis. By compromising KCC2 activity, stress increases intracellular Cl–, which is reflected as a depolarizing shift in the reversal potential for GABAA-mediated synaptic events, and

441

© 2009 Nature America, Inc. All rights reserved.

ARTICLES profoundly weakens synaptic inhibition. In vivo, this translates into a complete absence of GABA inhibitory control of the HPA axis following stress. Decreased KCC2 activity allows for the buildup of intracellular Cl– during repetitive synaptic activation, which reverses the ionic flux through the receptor and promotes the depolarization and resultant excitation of neuroendocrine cells. Our findings provide, to the best of our knowledge, the first demonstration of ionic plasticity in the mature nervous system in response to acute physiological stimuli. This and other reports7,37 demonstrate that subtle changes in the function of transmembrane ion transport mechanisms profoundly affect the output of neural networks in the adult CNS. Although reciprocal changes in the expression of the Cl– transporters NKCC1 (ref. 24) and KCC2 (ref. 19–21) are important for the hyperpolarizing shift in EGABA during development, pathological conditions in the adult CNS that are associated with depolarizing shifts in EGABA all point to an essential role for KCC2 (refs. 7–11,24). Using both cell-attached recordings that do not perturb intracellular Cl– and in vivo microinjections and hormone measurements, we found that inhibition mediated by the activation of GABAA receptors was absent following stress. When GABA synapses were repetitively activated, however, GABA conditionally became excitatory. This observation is supported by computational modeling work indicating that the depolarizing shift in the reversal potential of the anion (Eanion) markedly reduces the ability to inhibit firing until some threshold shift is reached, after which point hyperexcitability results6. We provide evidence that the change in EGABA is a result of altered KCC2-mediated Cl
extrusion capacity. Collapse of the Cl
gradient as a result of downregulation of KCC2 occurs in response to sustained neuronal activity15,30,35,36. Here we directly examined the activity-dependent effect of a stress-induced decrease in KCC2 activity using high-frequency stimulation procedures in situations designed to favor either the influx or efflux of Cl–. This experimental design allowed us to examine the contribution of Cl– extrusion independently from the development of synaptic depression. Adequate Cl– extrusion is a requirement for the maintenance of a sustained Cl– gradient favoring the influx of Cl–. We found that, even under control conditions, the extrusion capacity of KCC2 was saturated at high rates of activity and was important in the depression throughout the pulse train. Under stress conditions, however, this contribution was greatly enhanced. The lack of parallel changes in activity-dependent synaptic depression is consistent with a downregulation of KCC2 activity31. Our data demonstrating that the activation of a1 adrenoceptors causes a depolarizing shift in EGABA provide a cellular mechanism by which ascending noradrenergic fibers33 can cause a sustained release of pituitary hormones. This probably occurs through a mechanism that is not dissimilar to one in which bouts of enhanced neuronal activity downregulate KCC2 through the recruitment of protein kinase C and intracellular Ca2+ (ref. 30). Whether this is the only means by which KCC2 can be affected in this system is not known. Additional mechanisms, such as coincident and repetitive activation of pre- and postsynaptic elements38 or the actions of different chemical messengers23,29, certainly cannot be ruled out. On the basis of our evidence that neither the expression nor the oligomerization of KCC2 is altered after stress, we favor a scenario whereby activity is compromised independent of change in the expression of the KCC2 protein31. In both physiological and pathological states, inhibitory plasticity appears to be extremely diverse. Our results support the hypothesis that changes in membrane anion transporter activity provide a powerful mechanism for altering the output of neural networks in response to behavioral stimuli.

442

METHODS Slice preparation. PVN hypothalamic slices (300 mm thick) were prepared from postnatal day 21 (P21) to P28 male Sprague Dawley rats as described previously39. Our animal experiments were approved by the University of Calgary Animal Care and Use Committee. The control group was comprised of rats that were anaesthetized with an intraperitoneal injection of sodium pentobarbital (70 mg per kg of body weight) prior to decapitation. The stress group was comprised of animals placed in a plexiglass restrainer for 30 min, and then quickly anaesthetized and decapitated as described above. Slices (300 mm) were superfused with artificial cerebrospinal fluid (ACSF) containing 126 mM NaCl, 2.5 mM KCl, 26 mM NaHCO3, 2.5 mM CaCl2, 1.5 mM MgCl2, 1.25 mM NaHPO4 and 10 mM glucose (bubbled with 95% O2/5% CO2). We added 10 mM 6,7-dinitroquinoxaline-2,3-dione and 100 mM DL-2-amino5-phosphonovaleric acid to specifically measure GABAA currents. Recordings. Whole-cell recordings were obtained from PNCs that were visually identified using an upright microscope (Olympus Optical) that was fitted with infrared differential interference contrast optics. PNC cell type was verified on the basis of the electrical fingerprint in current-clamp configuration40. All recordings were obtained at 30–32 1C using borosilicate glass microelectrodes (tip resistance of 3–7 MO) that were filled with intracellular solution containing 123 mM potassium gluconate, 2 mM MgCl2, 8 mM NaCl, 1 mM EGTA, 4 mM ATP and 0.3 mM GTP. For some experiments, gramicidin (40 mM) was added to the intracellular solution. In these experiments, a high Cl– intracellular solution was used (150 mM KCl and 10 mM HEPES, buffered to pH 7.2 with KOH) to detect rupture of the patch. In other whole-cell experiments, intracellular Cl– was altered from 12 mM to 4, 8 or 24 mM. Because varying intracellular Cl
concentration resulted in predictable shifts in EGABA, we subsequently used whole-cell recordings (12 mM Cl–) to measure EGABA. Cellattached recordings used 12 mM Cl– intracellular solution and whole-cell configuration was achieved following the experiment to determine cell type and to ensure that synaptic stimulation activated GABA synapses. Data acquisition and analysis of postsynaptic currents were carried out as previously described39. Recordings were accepted when access resistance changes were o15%. Data analysis. Whole-cell and gramicidin-perforated patch data were analyzed using pCLAMP 10 (Molecular Devices). Cell-attached data were analyzed using MatLab 7.1 (Mathworks). In Figure 4a, points were fit with a monoexponential decay curve. When comparing two datasets, statistical significance was tested using Student’s t test. For more than two datasets, statistical significance was determined with a one-way ANOVA with a Newman-Keuls post hoc test. All measurements are given as mean ± s.e.m. Corticosterone assay. Blood was collected into heparinized syringes, centrifuged (15 min at 9,500g) and plasma was stored at
80 1C until assayed. Plasma corticosterone was determined using an enzyme immunoassay kit (Correlate-EIA, Assay Designs) and expressed as ng ml–1 of plasma. The intraassay coefficient of variation was 6.6% and the sensitivity was 27.0 pg ml–1 according to the manufacturer’s protocol. PVN microinjection. Rats (P21–28) were anesthetized with ketamine (1 mg per kg) and the femoral artery was cannulated for blood collection. A control blood sample was collected (250–300 ml) approximately 1 min before microinjection. The animal was then placed into a stereotaxic unit and secured. The coordinates used for PVN microinjection were 0.8 mm posterior to bregma and 0.3 mm lateral to midline. A small hole was drilled and the cannula was lowered into position (8 mm ventral from dura). We injected bicuculline (2 ml, 0.1 nM), furosemide (2 ml, 200 mM) or vehicle into PVN with 2% (wt/vol) Evans Blue dye. Subsequent blood samples (250–300 ml) were collected at 1, 5 and 10 min after microinjection. At the termination of the experiment, the animal was killed by transcardial perfusion with formalin and the brain was collected for post hoc analysis of cannula placement. Protein extraction and immunoblotting. Rats were anesthetized and decapitated, and the hypothalamus rapidly excised and homogenized in lysis buffer (20 mM MOPS, 150 mM KCl, 4.5 mM magnesium acetate and 1% (vol/vol) Triton X-100) containing protease inhibitors (2 mg ml–1 aprotinin (Roche Applied Science), 2 mg ml–1 leupeptin (Roche Applied Science), 2 mM

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES phenylmethylsulfonyl fluoride (Sigma) and 1 mg ml–1 pepstatin A (Roche Applied Science)). Protein concentrations of cleared lysates were determined using a detergent-compatible protein assay (Bio-Rad) with bovine serum albumin as a standard. Samples (50 mg of protein) were resolved on an 8% SDS-polyacrylamide gel, transferred to nitrocellulose and blocked for 1 h in 25% nonfat dry milk in blocking buffer (50 mM Tris, 150 mM NaCl, 0.1% (vol/vol) Tween 20) before overnight incubation at 4 1C in rabbit antibody to KCC2 (1:3,000 in blocking buffer, Millipore). After washing in blocking buffer, blots were incubated for 1 h in horseradish peroxidase–conjugated goat antibody to rabbit IgG (1:3,000 in blocking buffer, Bio-Rad). Antibody binding was determined by chemiluminescent detection (ECL, GE Healthcare-Life Sciences) and exposure to Fuji SuperRX film. Densitometric analysis was performed using ImageQuant software (GE Healthcare-Life Sciences).

© 2009 Nature America, Inc. All rights reserved.

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

ACKNOWLEDGMENTS We thank members of the Bains laboratory, Q.J. Pittman and W.H. Mehaffey for comments and thoughtful discussion regarding the manuscript. We also thank C. Sank for assistance with microinjections and corticosterone assays. S.A.H. was supported by a studentship from the Alberta Heritage Foundation for Medical Research. J.I.W. is supported by a T. Chen Fong Scholarship from the Hotchkiss Brain Institute and scholarships from the Faculty of Graduate Studies, University of Calgary and the Government of Alberta. J.S.B. is an Alberta Heritage Foundation for Medical Research Senior Scholar. This work is funded by an Operating Grant from the Canadian Institutes for Health Research. AUTHOR CONTRIBUTIONS S.A.H. conducted the in vivo stress and corticosteroids measurements, the EGABA, PPR and repetitive synaptic activation experiments, analyzed the data and wrote the manuscript. J.I.W. conducted the cell-attached experiments in Figure 1, as well as performing some gramicidin recordings. E.U.K. designed and performed immunoblot experiments and analyses. J.S.B. designed the experiments, analyzed the data, prepared the manuscript and supervised the project. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/ 1. Decavel, C. & Van den Pol, A.N. GABA: a dominant neurotransmitter in the hypothalamus. J. Comp. Neurol. 302, 1019–1037 (1990). 2. Roland, B.L. & Sawchenko, P.E. Local origins of some GABAergic projections to the paraventricular and supraoptic nuclei of the hypothalamus in the rat. J. Comp. Neurol. 332, 123–143 (1993). 3. Boudaba, C., Szabo, K. & Tasker, J.G. Physiological mapping of local inhibitory inputs to the hypothalamic paraventricular nucleus. J. Neurosci. 16, 7151–7160 (1996). 4. Cole, R.L. & Sawchenko, P.E. Neurotransmitter regulation of cellular activation and neuropeptide gene expression in the paraventricular nucleus of the hypothalamus. J. Neurosci. 22, 959–969 (2002). 5. Payne, J.A., Rivera, C., Voipio, J. & Kaila, K. Cation-chloride co-transporters in neuronal communication, development and trauma. Trends Neurosci. 26, 199–206 (2003). 6. Prescott, S.A., Sejnowski, T.J. & DeKoninck, Y. Reduction of anion reversal potential subverts the inhibitory control of firing rate in spinal lamina I neurons: towards a biophysical basis for neuropathic pain. Mol. Pain 2, 32 (2006). 7. Coull, J.A. et al. Trans-synaptic shift in anion gradient in spinal lamina I neurons as a mechanism of neuropathic pain. Nature 424, 938–942 (2003). 8. Jin, X., Huguenard, J.R. & Prince, D.A. Impaired Cl– extrusion in layer V pyramidal neurons of chronically injured epileptogenic neocortex. J. Neurophysiol. 93, 2117–2126 (2005). 9. Khazipov, R. et al. Developmental changes in GABAergic actions and seizure susceptibility in the rat hippocampus. Eur. J. Neurosci. 19, 590–600 (2004). 10. Pathak, H.R. et al. Disrupted dentate granule cell chloride regulation enhances synaptic excitability during development of temporal lobe epilepsy. J. Neurosci. 27, 14012–14022 (2007). 11. Wake, H. et al. Early changes in KCC2 phosphorylation in response to neuronal stress result in functional downregulation. J. Neurosci. 27, 1642–1650 (2007).

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

12. DeFazio, R.A., Keros, S., Quick, M.W. & Hablitz, J.J. Potassium-coupled chloride co-transport controls intracellular chloride in rat neocortical pyramidal neurons. J. Neurosci. 20, 8069–8076 (2000). 13. DeFazio, R.A. & Hablitz, J.J. Chloride accumulation and depletion during GABAA receptor activation in neocortex. Neuroreport 12, 2537–2541 (2001). 14. Payne, J.A. Functional characterization of the neuronal-specific K-Cl co-transporter: implications for [K+]o regulation. Am. J. Physiol. 273, C1516–C1525 (1997). 15. Rivera, C. et al. Mechanism of activity-dependent downregulation of the neuron-specific K-Cl co-transporter KCC2. J. Neurosci. 24, 4683–4691 (2004). 16. Williams, J.R., Sharp, J.W., Kumari, V.G., Wilson, M. & Payne, J.A. The neuron-specific K-Cl co-transporter, KCC2. Antibody development and initial characterization of the protein. J. Biol. Chem. 274, 12656–12664 (1999). 17. Mount, D.B. et al. The electroneutral cation-chloride co-transporters. J. Exp. Biol. 201, 2091–2102 (1998). 18. Payne, J.A., Stevenson, T.J. & Donaldson, L.F. Molecular characterization of a putative K-Cl co-transporter in rat brain. A neuronal-specific isoform. J. Biol. Chem. 271, 16245–16252 (1996). 19. Rivera, C. et al. The K+/Cl– co-transporter KCC2 renders GABA hyperpolarizing during neuronal maturation. Nature 397, 251–255 (1999). 20. Ganguly, K., Schinder, A.F., Wong, S.T. & Poo, M. GABA itself promotes the developmental switch of neuronal GABAergic responses from excitation to inhibition. Cell 105, 521–532 (2001). 21. Cordero-Erausquin, M., Coull, J.A., Boudreau, D., Rolland, M. & DeKoninck, Y. Differential maturation of GABA action and anion reversal potential in spinal lamina I neurons: impact of chloride extrusion capacity. J. Neurosci. 25, 9613–9623 (2005). 22. Mercado, A., Mount, D.B. & Gamba, G. Electroneutral cation-chloride co-transporters in the central nervous system. Neurochem. Res. 29, 17–25 (2004). 23. Coull, J.A. et al. BDNF from microglia causes the shift in neuronal anion gradient underlying neuropathic pain. Nature 438, 1017–1021 (2005). 24. Dzhala, V.I. et al. NKCC1 transporter facilitates seizures in the developing brain. Nat. Med. 11, 1205–1213 (2005). 25. Ebihara, S., Shirato, K., Harata, N. & Akaike, N. Gramicidin-perforated patch recording: GABA response in mammalian neurones with intact intracellular chloride. J. Physiol. (Lond.) 484, 77–86 (1995). 26. Verkuyl, J.M., Karst, H. & Joels, M. GABAergic transmission in the rat paraventricular nucleus of the hypothalamus is suppressed by corticosterone and stress. Eur. J. Neurosci. 21, 113–121 (2005). 27. Cullinan, W.E. & Wolfe, T.J. Chronic stress regulates levels of mRNA transcripts encoding beta subunits of the GABAA receptor in the rat stress axis. Brain Res. 887, 118–124 (2000). 28. Yamada, J. et al. Cl– uptake promoting depolarizing GABA actions in immature rat neocortical neurones is mediated by NKCC1. J. Physiol. (Lond.) 557, 829–841 (2004). 29. Rivera, C. et al. BDNF-induced TrkB activation down-regulates the K+-Cl– co-transporter KCC2 and impairs neuronal Cl– extrusion. J. Cell Biol. 159, 747–752 (2002). 30. Fiumelli, H., Cancedda, L. & Poo, M.M. Modulation of GABAergic transmission by activity via postsynaptic Ca2+-dependent regulation of KCC2 function. Neuron 48, 773–786 (2005). 31. Vale, C., Schoorlemmer, J. & Sanes, D.H. Deafness disrupts chloride transporter function and inhibitory synaptic transmission. J. Neurosci. 23, 7516–7524 (2003). 32. Blaesse, P. et al. Oligomerization of KCC2 correlates with development of inhibitory neurotransmission. J. Neurosci. 26, 10407–10419 (2006). 33. Pacak, K. et al. Effects of various stressors on in vivo norepinephrine release in the hypothalamic paraventricular nucleus and on the pituitary-adrenocortical axis. Ann. NY Acad. Sci. 771, 115–130 (1995). 34. Daftary, S.S., Boudaba, C. & Tasker, J.G. Noradrenergic regulation of parvocellular neurons in the rat hypothalamic paraventricular nucleus. Neuroscience 96, 743–751 (2000). 35. Grob, M. & Mouginot, D. Heterogeneous chloride homeostasis and GABA responses in the median preoptic nucleus of the rat. J. Physiol. (Lond.) 569, 885–901 (2005). 36. Thompson, S.M. & Gahwiler, B.H. Activity-dependent disinhibition. I. Repetitive stimulation reduces IPSP driving force and conductance in the hippocampus in vitro. J. Neurophysiol. 61, 501–511 (1989). 37. Laviolette, S.R., Gallegos, R.A., Henriksen, S.J. & van der Kooy, D. Opiate state controls bi-directional reward signaling via GABAA receptors in the ventral tegmental area. Nat. Neurosci. 7, 160–169 (2004). 38. 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). 39. Hewitt, S.A. & Bains, J.S. Brain-derived neurotrophic factor silences GABA synapses onto hypothalamic neuroendocrine cells through a postsynaptic dynamin-mediated mechanism. J. Neurophysiol. 95, 2193–2198 (2006). 40. Luther, J.A. et al. Neurosecretory and non-neurosecretory parvocellular neurones of the hypothalamic paraventricular nucleus express distinct electrophysiological properties. J. Neuroendocrinol. 14, 929–932 (2002).

443

ARTICLES

Tuning of synapse number, structure and function in the cochlea

© 2009 Nature America, Inc. All rights reserved.

Alexander C Meyer1,6, Thomas Frank1,6, Darina Khimich1,6, Gerhard Hoch1, Dietmar Riedel2, Nikolai M Chapochnikov1,3, Yury M Yarin1,4, Benjamin Harke5, Stefan W Hell5, Alexander Egner5 & Tobias Moser1,3 Cochlear inner hair cells (IHCs) transmit acoustic information to spiral ganglion neurons through ribbon synapses. Here we have used morphological and physiological techniques to ask whether synaptic mechanisms differ along the tonotopic axis and within IHCs in the mouse cochlea. We show that the number of ribbon synapses per IHC peaks where the cochlea is most sensitive to sound. Exocytosis, measured as membrane capacitance changes, scaled with synapse number when comparing apical and midcochlear IHCs. Synapses were distributed in the subnuclear portion of IHCs. High-resolution imaging of IHC synapses provided insights into presynaptic Ca21 channel clusters and Ca21 signals, synaptic ribbons and postsynaptic glutamate receptor clusters and revealed subtle differences in their average properties along the tonotopic axis. However, we observed substantial variability for presynaptic Ca21 signals, even within individual IHCs, providing a candidate presynaptic mechanism for the divergent dynamics of spiral ganglion neuron spiking.

After processing by the mammalian ear’s exquisite micromechanics and mechanoelectrical transduction, acoustic information is encoded at the afferent synapses of IHCs with high temporal precision1,2. Presynaptic active zones of IHCs contain a synaptic ribbon, a multiprotein structure that tethers synaptic vesicles3,4 and ensures a large pool of readily releasable vesicles5–8. Stimulus–secretion coupling is governed by CaV1.3 L-type Ca2+ channels9,10 that tightly control the release of glutamate from nearby fusion-competent vesicles2,8,11 onto postsynaptic AMPA receptors12,13 on the unbranched peripheral axon of the bipolar spiral ganglion neuron (SGN)14. As a result, each SGN receives input from only one IHC active zone, while each IHC drives several SGNs. Whether and how the number and properties of afferent synapses of the cochlea are ‘tuned’ for optimal sound encoding remains an important question (refs. 15,16, for example). The frequency selectivity of SGNs is primarily determined by the location of the innervated IHC on the cochlea’s tonotopic axis, providing a place code for frequency (refs. 17,18, for example). In addition, studies of small samples of synapses from distinct cochlear regions have indicated that the innervation density varies along the length of the cochlea19–21. Moreover, it has been shown that SGNs covering a narrow frequency range differ markedly in spontaneous and evoked firing rates, sound threshold and dynamic range (for example, refs. 22,23) and that they collectively encode a large range of sound pressures. It is generally believed, but not yet

directly proven, that each IHC makes contact with such physiologically diverse SGNs. If true, the heterogeneity of SGN dynamics could be caused by pre- and postsynaptic mechanisms7,19,24,25. Pioneering work on the cat cochlea suggested that lowspontaneous-rate SGNs preferentially contact active zones with large or even multiple synaptic ribbons at the neural side of IHCs (toward the modiolus), whereas high spontaneous rate SGNs are driven by small, ‘simple’ synapses at the abneural IHC side (toward the outer hair cells)24. Here we have used patch-clamp, confocal imaging of IHC presynaptic Ca2+ signals; confocal, 4Pi26,27 and stimulated emission depletion (STED)28 microscopy of immunolabeled synapses; and electron microscopy to characterize the distribution of afferent synapses as well as their structure and function at different tonotopic regions of the cochlea. Having investigated thousands of synapses in hundreds of IHCs, we provide a continuous representation of synapse number per IHC along the entire mouse cochlea, and we show that synapse density parallels the neuronal population audiogram. Using STED microscopy, we provide optical, nanometer-scale measurements of individual clusters of presynaptic Ca2+ channels and postsynaptic AMPA receptors. Whereas average structural and functional synapse properties varied only slightly along the cochlea’s tonotopic axis, we found considerable heterogeneity of presynaptic Ca2+ signals among the synapses in IHCs in a given region.

1InnerEarLab, Department of Otolaryngology and Center for Molecular Physiology of the Brain, University of Go ¨ttingen, Go¨ttingen, Germany. 2Laboratory of Electron Microscopy, Max-Planck-Institute for Biophysical Chemistry, Go¨ttingen, Germany. 3Bernstein Center for Computational Neuroscience, University of Go¨ttingen, Go¨ttingen, Germany. 4Clinic of Otorhinolaryngology, Department of Medicine, Technical University of Dresden, Dresden, Germany. 5Department of NanoBiophotonics, Max-PlanckInstitute for Biophysical Chemistry, Go¨ttingen, Germany. 6These authors contributed equally to the work. Correspondence should be addressed to T.M. ([email protected]) or A.E. ([email protected]).

Received 6 October 2008; accepted 12 February 2009; published online 8 March 2009; doi:10.1038/nn.2293

444

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES Figure 1 The number of afferent synapses per 53% 3% IHC co-varies with ABR threshold along the 500 µm 10 µm tonotopic axis. (a) Projections of confocal stacks of immunostained mouse IHC afferent synapses at different tonotopic locations (red, anti-CtBP2/ RIBEYE; green, anti-GluR2/3; blue, Hoechst Hoechst GluR2/3 CtBP2/RIBEYE 34580 nuclear stain). Percentages indicate the normalized location of the imaged IHCs in the 95% 9% cochlea (0%, apical end; 100%, basal end). Center: montage of the low-magnification view of fragments of the full explanted organ of Corti. Red circles, locations of the confocal images. (b) Synaptic cochleograms and tone-burst ABR audiograms of NMRI and C57BL/6 mice Mouse Gerbil Frequency Frequency (together, more than 15,000 synapses) overlaid 0.7 1 2 4 6 8 1216 24 32 48 64 kHz 6 8 12 16 24 32 48 64 kHz by relating distance to cochlear apex to the 30 30 0 20 CAP threshold 18 tonotopic map of mice of strain CBA (top axis ). (ref. 29) 40 25 ABR C57BL/6 NMRI mice: red open circles, number of synapses 20 50 15 20 per IHC (26 ears, P15–P20); red filled circles, 60 ABR threshold (average ± s.e.m., ten ears from six 40 15 70 10 mice). C57BL/6: black open squares, number of ABR NMRI C57BL/6 P30 10 synapses per IHC (two ears, P30); connected Gerbil P28 80 60 NMRI P15-P20 2nd-order polynomial fit black filled squares, ABR threshold. Continuous 2nd-order polynomial fit 90 5 0 5 0 2 4 6 8 10 11.2 mm 1 2 3 4 5 6.2 mm black line: fit of a quadratic function to the 0 20 40 60 80 100% 0 20 40 60 80 100% collective data set of both mouse lines. Apex Base Apex Base Position Position (c) Synaptic cochleograms (circles): number of synapses per IHC (more than 10,000 synapses in six ears from four P28 gerbils), plotted against a tonotopic map (taken from ref. 29) and frequencydependent compound action potentials (CAP) thresholds (from ref. 29) (red line). Continuous black line: fit of a quadratic function to the data.

a

b

c

b

Synapses per IHC

Apical

c

Abneural

z-distance from nucleus center (µm)

r




0

n = 373 synapses from 23 midcochlear IHCs

–5

–10 –15

2 4 6 8 µm

NUMBER 4

.5 1

0.0 0.5 1.0 1.5 2.0 2.5 –1 Synapse density (µm )

f

Neural

1.5

2

Normalized frequency

e

Neural

Basal

d

Apical

[

Neural

z

0.15

0.10

0.05

0.00 Abneural

NATURE NEUROSCIENCE VOLUME 12

Basal

5

Basal

© 2009 Nature America, Inc. All rights reserved.

Synapses per IHC

a

Apical

Figure 2 Spatial distribution of afferent synapses within IHCs of one tonotopic region. (a) Cartoon illustrating the analysis and the use of cylindrical coordinates for descriptions of synapse position. (b) Four en face views of six midcochlear IHCs and their afferent synapses (red dots), overlaid after normalization of synapse coordinates in the z axis (from the center of the nucleus to the basal end) and in the radius, r (according to the width at nuclear level) of each IHC. Scale bars, 5 mm. (c) Histogram shows synapse number as a function of axial distance from the center of the cell’s nucleus (z) for a total of twenty-three midcochlear IHCs. (d) Polar scatter plot of 373 synapses mapped in cylindrical coordinates (radius (r) and angle (y) are displayed). (e) Polar histogram of synapse density in 361 sectors. The sum of all 10 sectors equals the average synapse number of 16.2 per cell. (f) Histogram of the nearest neighbor distance distribution of 373 synapses.

CAP threshold (dB)

ABR threshold (dB SPL)

count by the number of IHC nuclei6. The study used hearing mice (as determined by auditory brainstem responses (ABR)) of two wild-type laboratory strains (C57BL/6N, postnatal day (P) 30; and NMRI, P15–P20). We approximated their collective synaptic cochleograms by a quadratic function (Fig. 1b) to describe synapse number per IHC along the tonotopic axis (see Supplementary Table 1 online for fit coefficients). In gerbils (P28), we counted only synaptic ribbons because we were not able to stain postsynaptic glutamate receptor clusters reliably (Fig. 1c; Supplementary Fig. 1 and Supplementary Table 1). The synaptic cochleograms were then related to hearing thresholds as estimated by ABR recordings (Fig. 1b; mice were thereafter used for immunohistochemistry) or compound action potentials24 (Fig. 1c) using published place-frequency maps18,29. For both species, the synaptic cochleograms peaked within

RESULTS Tonotopic synapse density map and subcellular distribution First, we assessed the number of ribbon synapses per IHC throughout the entire cochlea (creating a synaptic cochleogram) of mice (Fig. 1a,b; high-frequency hearing) and gerbils (Fig. 1c and Supplementary Fig. 1 online; low-frequency hearing) by confocal microscopy of immunolabeled, whole-mount organs of Corti. The microdissected parts of the organ of Corti were aligned for measuring the distance of a given synapse to the apex of the cochlea (see Methods; Fig. 1a). We identified IHC ribbon synapses as colocalized spots of presynaptic ribbons (using antibody to transcription factor CtBP2 sharing homologous domain with RIBEYE, the main protein component of the synaptic ribbon) and postsynaptic GluR2/3 (labeling the AMPA receptor clusters) immunofluorescence in stacks of confocal sections, and divided their

[

APRIL 2009

Abneural

0

1 2 3 4 5 6 Distance to nearest neighbor (µm)

445

ARTICLES

d

RIBEYE

RIBEYE

e

700 600

Confocal STED

500 400 300 200

0 0 100 200 300 400 500 600 700 Short object axis (nm)

g

700

Object long axis (nm)

Confocal

Ribbons Apical (n = 144) Midcochlear (n = 136)

100

Midcochlear

Midcochlear

Apical

600 500 400 300 200 100

Ca2+-channel clusters Apical (n = 84) Midcochlear (n = 82)

STED

0 0 100 200 300 400 500 600 700 Object short axis (nm)

Apical

Midcochlear

i

Fluoresc. intensity (a.u.)

CaV1.3

GluR2/3 RIBEYE

h

4

Apical Midcochlear (from ref. 12)

Confocal

STED

8

3

6

2

4

1

2

0 –1.0

0 –0.5 0.0 0.5 1.0 Distance from center (µm)

j

Merge

the most sensitive region of the cochlea at B17 (mice) or B24 (gerbils) synapses per IHC and declined toward the cochlear apex and base. Next, we studied the synapse distribution within apical (distance to cochlear apex 200 ± 100 mm, n ¼ 2 mice; Supplementary Fig. 2 online) and midcochlear (1,700 ± 400 mm, n ¼ 4 mice; Fig. 2) IHCs. Stacks of confocal images obtained from organs of Corti immunolabeled for CtBP2/RIBEYE and calbindin-28k, a Ca2+-binding protein marking the IHC cytosol, were aligned according to the tonotopic axis (indicated by the row of IHCs) to identify synaptic ribbons and IHC nuclei (see Methods). Synapses were assigned to IHCs based on their distance to the center of the nearest IHC’s nucleus. Data were discarded if assignments were ambiguous. The position of each synapse was expressed in cylindrical coordinates (z-axial position, with z ¼ 0 at the center of the nucleus; radius r and angle y as illustrated in Fig. 2a) after normalizing the distance between the center of the nucleus and the basal end of each IHC to the respective mean value of all analyzed IHCs. We observed a strong base-to-apex decline in synapse abundance from the base of the IHC to its apex and a rather uniform distribution

446

Figure 3 Ribbon synapse morphology in the apical cochlea (100–400 mm) and mid-cochlea (1,300–2,100 mm) of the mouse (P16–P21). (a–c) Representative electron micrographs of midcochlear (a,b) and apical (c) synapses, after either high-pressure rapid freeze and freezesubstitution (HPF, a) or aldehyde fixation (chem., b and c). Scale bar, 100 nm. (d) Representative confocal (top) and STED (bottom) images of 40-nm beads (left) and fluorescently labeled synaptic ribbons of apical (middle) and midcochlear (right) IHCs. Scale bar, 2 mm. (e) Long versus short axes for apical (blue) and midcochlear (red) ribbons. Grey bar: resolution limit of the STED microscope in front (lower boundary) and behind (upper boundary) the sample. (f) Representative confocal (top) and STED (bottom) images of immunolabeled CaV1.3 clusters of an apical and a midcochlear IHC imaged 15–20 mm deep in the sample. Scale bar, 1 mm. The double cluster in the midcochlear IHC was associated with a very large RIBEYE signal (not shown). (g) Long versus short axes for apical (blue) and midcochlear CaV1.3 clusters (red). Expected density (Supplementary Fig. 5) overlaid as contour plot. (h–j) Imaging of postsynaptic AMPA receptor clusters (green, GluR2/3; red, RIBEYE) contacting apical and midcochlear IHCs. (h) Ring-like appearance of clusters oriented en face. Scale bar, 2 mm. (i) Fluorescence (fluoresc.) profiles from STED en face views of apical (blue) and midcochlear (red) clusters. Profiles were centered at half the distance of the two side-peaks for alignment. Gray bars, counts of immunogold particles as a function of distance from synapse center (rat IHC, taken from ref. 12); a.u., arbitrary units. (j) Two leftmost panels: representative deconvolved xy-confocal and STED sections of clusters. Middle panel: same cluster overlaid with confocal image of ribbon. Scale bar, 500 nm. Two rightmost panels: more synapses.

Apical, chem.

Number of particles

© 2009 Nature America, Inc. All rights reserved.

f

Apical

c

Midcochlear, chem.

Long object axis (nm)

40 nm beads

b

PSF range

Midcochlear, HPF

PSF range

a

along y. This is exemplified in views from all four sides of six overlaid midcochlear IHCs (Fig. 2b). We further described the synapse distribution as functions of z-position (Fig. 2c) and angle y (Figs. 2d,e). The distribution of the three-dimensional nearest neighbor distance had a mean of B2 mm and did not show obvious higher-order peaks (Fig. 2f, estimated before normalization of cell dimensions). The synapse distribution was similar for IHCs in the apex of the cochlea, except for a tendency of synapses to accumulate in the apical and basal IHC sectors (Supplementary Fig. 2). Synapse morphology as a function of tonotopic position The size of the synaptic ribbon largely determines the vesicle complement of each synapse (reviewed in ref. 3) and was previously used to approximate the area of the active zone over which Ca2+ channels are distributed30,31. The size of postsynaptic AMPA receptor clusters is a key determinant of synaptic strength (for example, refs. 32,33). Thus, we explored IHC synapse morphology in apical and midcochlear regions using high-resolution fluorescence microscopy of immunolabeled IHC synapses, as well as electron microscopy. Electron microscopy was performed on ultrathin sections of first apical turns of cochleae. First, we compared the synaptic ultrastructure after (i) chemical immersion fixation or (ii) high-pressure rapid freeze

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES dimensions of point spread function (PSF), B150
150 nm; axial dimension, Apical (mean ± s.d.) Midcochlear (mean ± s.d.) P-value B500 nm at the position of the synapses in the tissue for Fig. 3d–g). The resolution of CaV1.3 cluster long axis (STED) 327 ± 95 nm (n ¼ 84) 320 ± 97 nm (n ¼ 82) P ¼ 0.40 the STED microscope decreases with the 228 ± 59 nm (n ¼ 84) 208 ± 55 nm (n ¼ 82) P o 0.01 CaV1.3 cluster short axis (STED) depth of penetration into the tissue and Ribbon long axis (STED) 379 ± 62 nm (n ¼ 144) 357 ± 63 nm (n ¼ 136) P o 0.001 was controlled by measuring the PSF with Ribbon long axis (EM) 228 ± 60 nm (n ¼ 29) 223 ± 59 nm (n ¼ 37) P ¼ 0.74 Ribbon short axis (STED) 299 ± 46 nm (n ¼ 144) 273 ± 39 nm (n ¼ 136) P o 0.001 100-nm fluorescent beads in front and behind the sample (Fig. 3e,g; SupplemenRibbon short axis (EM) 118 ± 27 nm (n ¼ 29) 117 ± 36 nm (n ¼ 37) P ¼ 0.86 tary Fig. 4). The short and long axes of Ribbon cross-section (EM) 21,337 ± 8,609 nm2 (n ¼ 29) 20,495 ± 9,795 nm2 (n ¼ 37) P ¼ 0.28 fluorescence spots in the STED sections 15.1 ± 3.5 (n ¼ 27) 12.7 ± 3.7 (n ¼ 34) P ¼ 0.01 NumSV o50 nm from ribbon (EM) 5.5 ± 4.2 (n ¼ 27) 0.8 ± 1.0 (n ¼ 34) P o 0.001 were approximated as the full widths at NumSV 450 nm from ribbon (EM) SV diameter (EM, CF) 41.4 ± 5.7 nm (n ¼ 411 SVs) 44.6 ± 5.6 nm (n ¼ 431 SVs) P o 0.001 half maximum (FWHM) of two orthogonal SV diameter (EM, HPF) – 48.7 ± 8.2 nm (n ¼ 156 SVs) gaussian functions (Fig. 3e,g). They were GluR2/3 cluster (STED) ‘size’ 884 ± 15 nm (n ¼ 16) 751 ± 11 nm (n ¼ 19) P o 0.01 slightly but significantly larger for apical ‘Width’ 180 ± 0.71 nm 193 ± 0.89 nm P ¼ 0.39 ribbons (Table 1 and Supplementary 2.5 ± 0.9 2.3 ± 0.7 P ¼ 0.45 Ratiomax/min Fig. 4). On average, the mean apparent Postsynaptic density (EM) 819 ± 154 nm (n ¼ 29) 888 ± 200 nm (n ¼ 30) P ¼ 0.74 axes ((short + long)/2) differed by 25 nm between the two locations. The ratios of EM, electron microscopy; NumSV, number of synaptic vesicles; SV, synaptic vesicle; CF, chemical fixation; HPF, high pressure freeze; ‘size’ of GluR2/3 clusters, sum of peak-to-peak distance and average FWHM of the two peaks in a fluorescence profile of the cluster long to short apparent axes were statistically (Fig. 3i); ‘width’, average of FWHM of the two fluorescence peaks; ratiomax/min, ratio of peak fluorescence (average over one-half FWHM around the peak location) and minimal fluorescence (average over one-half FWHM around the minimum location). indistinguishable (apical, 1.27 ± 0.16, n ¼ 144; midcochlear, 1.31 ± 0.19, n ¼ and subsequent freeze substitution (Supplementary Fig. 3 and Sup- 136; P ¼ 0.08) and were consistent with an ellipsoid ribbon structure. CaV1.3 Ca2+ channels cluster at active zones of hair cells11,35. Using plementary Methods online) because the latter method may better 34 preserve synapse morphology . We did not observe obvious differ- STED microscopy, we studied size and shape of immunolabeled CaV1.3 ences in the complement of vesicles; the mean vesicle diameter estimate clusters in IHCs at the two different tonotopic positions (Fig. 3f,g). The in the chemically fixed tissue was smaller by less than 10% (Fig. 3a,b; size of the observed fluorescent spots (two-dimensional FWHM) Table 1). Therefore, we used chemical fixation for further analyses. We ranged between 140 nm and 650 nm; hence, one of the object axes did not detect significant differences in ribbon size and shape (short was typically above the resolution limit of this STED microscope but in and long axes, cross-sectional area), nor in the length of the post- many cases too small for estimation by confocal microscopy. Nearby synaptic density (PSD) between midcochlear (Fig. 3a,b) and apical Ca2+ channel clusters (Fig. 3f, right panel), which are readily resolved (Fig. 3c) IHC synapses (Table 1). We quantified the abundance by STED microscopy but not discernable by confocal microscopy, of synaptic vesicles at the active zone for ribbon-associated (o50 nm existed rarely (o10% of all analyzed synapses). The average size and from ribbon) and unassociated (450 nm from ribbon) vesicles. There shape of the synaptic CaV1.3 clusters were similar for apical and were slightly more ribbon-associated vesicles and substantially midcochlear IHCs (Fig. 3g, Table 1 and Supplementary Fig. 5 online) more unassociated ones in apical synapses compared to midcochlear (mean apparent axis: apical 278 ± 71 nm, n ¼ 84; midcochlear 264 ± 69 ones (Table 1). nm, n ¼ 82, P ¼ 0.10; axis ratio: apical 1.45 ± 0.30; midcochlear 1.56 ± Optical microscopy of immunolabeled subcellular structures enables 0.42; P ¼ 0.06). We modeled the two-dimensional projection of high throughput analysis, providing a robust basis for statistical randomly oriented objects (after convolution with the point spread comparison. IHC ribbons, Ca2+ channel clusters and aspects of post- function of the STED microscope; Supplementary Fig. 5) because the synaptic AMPA receptor clusters (see below) are at or below the real three-dimensional shape of the clusters cannot be readily deduced resolution limit of confocal microscopy (B250 nm). Therefore, we from the data. We simulated variably sized objects with several used high-resolution 4Pi microscopy (one-dimensional axial resolu- geometric shapes aiming to match the experimentally observed distion B100 nm; Supplementary Fig. 4 online) and STED microscopy tributions of short and long axes (see Supplementary Fig. 5). The data (two-dimensional lateral resolution B50–150 nm, Fig. 3; three- could be reasonably well described by assuming a flat, oblate ellipsoid dimensional spherical resolution B150 nm, Supplementary Movie 1 with diameter 420 ± 130 nm, short axis below the resolution limit online) in addition to confocal microscopy to compare those structures (Fig. 3g). We cannot exclude the existence of subclusters in mice, as between apical and midcochlear synapses. Note that these measure- proposed for frog hair cells35, but, if present, those must be separated by ments report apparent rather than absolute object size because of less than 150 nm. immunolabeling and fluorescence imaging, and thereby overestimate AMPA receptor clusters, detected as GluR2/3 immunofluorescent true object size. However, this does not impede the comparison spots, showed a ring-like shape when oriented in parallel with the xy between synapses of different tonotopic regions or sectors of IHCs. plane (Fig. 3h–j), indicating a gradient of receptor density in the plane To study ribbons, we analyzed 4Pi images stacks as described of the PSD with an off-center maximum. The ring-like fluorescence previously6. The analysis revealed overlapping distributions of apparent pattern was confirmed by high-resolution three-dimensional STED axial diameter of ribbons between apical and midcochlear IHCs, microscopy (Supplementary Movie 1). We cannot entirely rule out indicating similar ribbon size and shape (Supplementary Fig. 4). We lower accessibility to antibody labeling of AMPA receptors in the center fitted gaussian functions to the distributions and found that the of the synapse. However, we consider this highly unlikely, as a similar means and s.d. values (for apical, 323 ± 57 nm, n ¼ 193 ribbons; gradient of receptor density was found in a previous immunoelectron for midcochlear, 324 ± 62 nm, n ¼ 168 ribbons) were indistinguish- microscopy study of rat IHC afferent synapses12, where antigen accesable. Compared to confocal microscopy implemented on the same sibility should not be a concern owing to postembedding immunogold microscope, STED yielded superior resolution (Fig. 3d,f; lateral labeling of AMPA receptors in ultrathin sections. En face views of AMPA

© 2009 Nature America, Inc. All rights reserved.

Table 1 Morphology of ribbons synapses in apical and midcochlear IHCs

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

447

ARTICLES (DCm) in IHCs from those apical and midcochlear regions (Fig. 4a) that had also been 50 40 morphologically investigated (Figs. 1–3). 40 20 IHCs at these two locations had 14.0 ± 0.3 Midcochlear 30 0 DTA: 1,400 µm Apical (midcochlear) and 9.7 ± 0.7 (apical) synapses, 20 0 DTA: respectively, and, as judged from their resting 10 300 µm –50 0 membrane capacitance (midcochlear IHCs –100 30 8.7 ± 1.1 pF, n ¼ 11; versus apical IHCs 7.7 20 –150 10 ms 50 ms 100 ms 200 ms 10 ± 0.7 pF, n ¼ 14), they differed slightly in size. Apical 0 We recorded their Ca2+ currents and memd e Midcochlear 0 50 100 150 200 Vm (mV) brane capacitance increments in the perfoDuration of depolarization (ms) 0 –80 –60 –40 –20 0 20 40 60 rated-patch configuration in response to step –100 depolarizations to –14 mV (Fig. 4b,c). For Apical 200 f Midcochlear 100 -200 most stimuli, we identified larger exocytic 250 0 0.4 DCm for midcochlear IHCs. Exocytosis of 200 0 –400 –0.4 midcochlear and apical IHCs could be –0.8 150 –1.2 100 roughly matched when scaling the responses –600 200 Apical 100 (n = 19) 50 of apical IHCs by the ratio of the number of 0 Midcochlear I (pA) –800 Ca 0.4 synapses (1.44; Fig. 4c). (n = 14) 0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 –0.4 Mean ICa (nA) –0.8 Midcochlear IHCs tended to show larger –1.2 peak Ca2+ currents (ratio of peak currents; 0 5 10 15 20 Time (ms) under non-augmenting conditions (2 mM [Ca2+], perforated-patch): 1.1, augmenting Figure 4 Exocytosis scales with the number of afferent synapses per IHC. (a) Distance-to-apex (DTA) conditions (10 mM extracellular [Ca2+], measurement of representative tonotopic regions used for patch-clamp recordings, identified by locally 5 mM BayK8644, whole-cell): 1.2, Fig. 4d removed rows of outer hair cells and pillar cells (ellipses). Scale bar, 200 mm. (b) Average membrane capacitance (Cm) traces (baseline subtracted) and calcium current (ICa) of apical (n ¼ 11; gray) and and Supplementary Table 2 online). The midcochlear (n ¼ 14; black) IHCs in response to 10-, 50-, 100- and 200-ms depolarizations. (c) Mean difference in Ca2+ current integrals (Fig. 4c) exocytic DCm (top) and Ca2+ current integrals (bottom) of the same cells as a function of stimulus did not reach statistical significance, probably duration. Scaled apical (dotted line) is Cm of apical IHCs multiplied by 1.44 (ratio of the number of because of the more pronounced Ca2+ current synapses for midcochlear and apical IHCs). (d) Average current–voltage relationships (thick traces with inactivation in midcochlear IHCs (Fig. 4b). symbols) of apical (n ¼ 19, thin gray traces) and midcochlear (n ¼ 14, thin black traces) IHCs in Neither were voltage dependence and kinetics augmenting conditions (10 mM extracellular [Ca2+], 5 mM BayK8644). (e,f) Analysis of Ca2+ tail current of Ca2+ current activation different (Supplefluctuations in augmenting conditions: (e) Pulse protocol and typical examples of current ensemble variance and mean of an apical and a midcochlear IHC, respectively. (f) Variance–mean relationships of mentary Fig. 6 online). Assuming an excluall apical and midcochlear IHCs (light and dark thin traces, respectively) and grand mean for both cell sively synaptic localization of Ca2+ channels populations (thick traces with symbols). Error bars, s.e.m. and given the roughly comparable size of synaptic clusters of Ca2+ channels (Fig. 3), receptor clusters were analyzed by fitting the STED (Fig. 3i) fluorescence one would have expected the peak Ca2+ currents in midcochlear and profiles with a sum of two gaussians. As expected, the limited resolution apical IHCs to scale with the number of synapses, as seen for exocytosis. of confocal microscopy led to an overestimation of the peak width The observed discrepancy could be due to differences in channel open (confocal B350 nm; STED B180 nm (lateral resolution o 80 nm for probability or single-channel current among IHCs at the two positions. this microscope)), and thus we based further analysis solely on STED When analyzing fluctuations among repetitively evoked Ca2+ tail microscopy. Whereas the width of the peaks was statistically indistin- currents under augmenting conditions (Fig. 4e,f), we found more guishable between apical and midcochlear clusters, we found slightly Ca2+ channels (1.34:1, midcochlear/apical), a slightly higher open larger peak-to-peak distances (apical 683 ± 84 nm, n ¼ 16; midco- probability and a somewhat smaller single-channel current in the chlear 562 ± 85 nm, n ¼ 19) and total sizes for clusters of apical synapses midcochlear cells (see Supplementary Table 2). (Table 1). The outer diameter of the cluster roughly matched the length Conclusions from whole-cell recordings on synaptic Ca2+ channels of the PSD as measured by electron microscopy (Table 1), indicating would be confounded by the presence of extrasynaptic Ca2+ channels. that AMPA receptors populate most of the PSD. On average, the To directly compare synaptic Ca2+ signaling at different tonotopic fluorescence peaks were about twofold brighter than the center, for places, we used fast confocal imaging of IHCs loaded with Fluo-5N. both apical and midcochlear AMPA receptor clusters (Table 1). Depolarization caused the rapid appearance of spatially restricted fluorescence hotspots in the basolateral compartment of IHCs Tonotopy of IHC presynaptic physiology from hearing mice (Fig. 5a) that were mediated by voltage-gated Recent studies have used patch-clamp recordings of Ca2+ currents and Ca2+ influx and localized with ribbons40, as previously described in membrane capacitance changes (DCm) to study the presynaptic prop- nonmammalian hair cells36,41,42. Once such Ca2+ microdomains were erties of hair cells of various species (for review, see ref. 3). It remains to identified in confocal sections, we used spot detection (continuous be clarified whether this technique reports exclusively synaptic Ca2+ read-out of fluorescence from the maximum-intensity location current and exocytosis in IHCs35–37 or also a significant extrasynaptic inside the Ca2+ microdomain (Fig. 5b,c) and perpendicular line fraction of channels and/or vesicles5,6,11,38,39. We also wondered scans (Fig. 5d,e) to study the kinetics (Fig. 5b), voltage dependence whether, on top of the number of synapses, their function may (Fig. 5c) and FWHM (Fig. 5e) of these synaptic Ca2+ signals also vary along the cochlea’s tonotopic axis15,16,30,31,37. We first studied during 20-ms depolarizations). On average, these properties were whole-cell Ca2+ currents and exocytic membrane capacitance changes indistinguishable between apical and midcochlear IHCs (Table 2).

c

60

Midcochlear (n = 11) Apical (n = 14) Apical, scaled

∆Cm (fF)

Apical Midcochlear

QCa (pC)

ICa (pA)

© 2009 Nature America, Inc. All rights reserved.

ICa (nA) Var. (pA2) ICa (nA) Var. (pA2) Vm (mV)

448

60 200 ms

Variance (pA2)

b

∆Cm (fF)

a

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

b

40

F – F0 (a.u.)

a

30 20 10 Apical (n = 25) Midcochlear (n = 25)

0 –10 40

c –80

50

–60

60 70 Time (ms)

–40

–20

80

20

5

Vm (mV)

90

40

10 15 20

0

200

400

600

800

25

F – F0 (a.u.)

Apical (n = 21) Midcochlear (n = 21) F – F0 (a.u.)

30 35

20 ms

150

100

FWHM (µm)

e

200

F – F0 (a.u.)

0.8 0.6 0.4

NUMBER 4

[

APRIL 2009

(y )

ar

al

le

ic

ch

co

M

id

Ap

(n = 15)

(n = 12)

(n = 15)

)

(y

(x

(x )

ar

al

[

le

ic

ch

co

id

M

NATURE NEUROSCIENCE VOLUME 12

)

(n = 17)

and starting xy-position in repetitive xy-scans, followed by (ii) selection of the brightest pixel 50 0.0 identified during lateral spot displacement along one dimension (from the starting position). Moreover, we obtained a second esti0 mate by fitting gaussian functions to the line scans, and we found the variability comparable (Supplementary Table 3). The scatter of fluorescence amplitude Variability among afferent synapses of individual IHCs Although Ca2+ microdomains were on average similar in populations (see Methods) for 65 individual synapses in 20 apical and midcochlear from the two tonotopic positions, they varied greatly among individual IHCs responding to 20 ms depolarizations to –7 mV (Fig. 6a) had a synapses, regardless of cell location. When comparing Ca2+ signals coefficient of variation (CV) of 0.63. Similar variability was also found measured at the two tonotopic locations, their amplitudes showed the for the background-normalized data (DF/F0, CV ¼ 0.61), arguing largest variance, but substantial differences among Ca2+ microdomains against a contribution of variation in dye concentration. were also observed for the FWHM and voltage of half activation Even synapses of an individual IHC differed up to tenfold in their (Supplementary Table 3 online). For spot detection, we carefully fluorescence amplitude (Fig. 6b; maximum/minimum DF on average, searched for the maximum-intensity location by (i) adjusting focus 4.5-fold; average CV ¼ 0.57 for seven apical IHCs). The variance among the fluorescence changes was not Table 2 Properties of synaptic Ca2+ signals as a function of tonotopic position systematic with time (Fig. 6b), arguing against Ca2+ current run-down, indicator Apical (mean ± s.e.m.) Midcochlear (mean ± s.e.m.) P-value bleaching, or buffer saturation being sizeable contributors. The trial-to-trial variaSpot detection 2+ Peak DF (F – F0, a.u.) 39.1 ± 5.3 (n ¼ 25) 39.3 ± 4.5 (n ¼ 25) P ¼ 0.97 bility of DF of the same Ca microdomain 1.0 ± 0.1 (n ¼ 25/25) 1.0 ± 0.1 (n ¼ 25/25) P ¼ 0.97 was small (CV ¼ 0.09). We found only Aonset,fast (ms) 17 ± 8 (n ¼ 8/25) 60 ± 38 (n ¼ 12/25) P ¼ 0.16 weak, although significant, correlations Aonset,slow (ms) 2.3 ± 0.7 (n ¼ 8/25) 1.5 ± 0.4 (n ¼ 12/25) P ¼ 0.16 between fast-onset kinetics and fluoresAonset,fast/Aonset,slow 1.1 ± 0.1 (n ¼ 25/25) 1.0 ± 0.1 (n ¼ 25/25) P ¼ 0.24 cence amplitude (Fig. 6c; linear correlation Adecay,fast (ms) 8 ± 1 (n ¼ 13/25) 28 ± 10 (n ¼ 13/25) P ¼ 0.11 coefficient Pr ¼ 0.28, P o 0.05), and Adecay,slow (ms) 5.1 ± 0.8 (n ¼ 13/25) 3.9 ± 0.7 (n ¼ 13/25) P ¼ 0.27 FWHM of Ca2+ microdomains and their Adecay,fast/Adecay,slow fluorescence amplitude (Fig. 6d; Pr ¼ –26.1 ± 3.8 (n ¼ 13) –26.1 ± 2.0 (n ¼ 13) P ¼ 0.28 –0.23, P o 0.05). Taken together, these V1/2 (mV) Slope (mV) 7.3 ± 1.4 (n ¼ 13) 6.3 ± 0.6 (n ¼ 13) P ¼ 0.65 correlations argue against a substantial contribution of defocus from the Ca2+ microLine scan domain’s center to the observed variance. x FWHM (mm) 0.6 ± 0.1 (n ¼ 17) 0.7 ± 0.1 (n ¼ 15) P ¼ 0.09 We therefore interpret these differences to y FWHM (mm) 0.6 ± 0.1 (n ¼ 12) 0.6 ± 0.1 (n ¼ 15) P ¼ 0.35 mostly reflect genuine variance among Exponential fitting was used to estimate the on- and off-kinetics of the synaptic Ca2+ microdomain signals. A double-exponential fit was Ca2+ signals of the individual synapses. accepted if the time constants differed by at least a factor of 3 and if both components were sizeable. A, amplitude of the fluorescence Next, we asked whether active zones component indicated in the subscript; V1/2, voltage of half-maximal fluorescence increase; slope, steepness of the voltage-dependent increase of the fluorescence change. V1/2 and slope were obtained by fitting Boltzmann functions to the fluorescence–voltage with different properties segregated along 2+ relationships (Fig. 5c). FWHM, full width at half-maximum of the time-averaged Ca microdomain fluorescence. x and y designate the the perimeter of the IHC, as previously directions of the orthogonal line scans; a.u., arbitrary fluorescence intensity units. 0.2

Ap

© 2009 Nature America, Inc. All rights reserved.

d

Figure 5 Synaptic Ca2+ signals are comparable at different tonotopic locations. (a) xy scan reveals hotspots of Fluo-5N fluorescence in the base of a P14 IHC (loaded with 400 mM Fluo-5N and 2 mM EGTA; dotted line, IHC border) during membrane depolarization to –7 mV (red bar, time of depolarization). Resting fluorescence (F0) was subtracted. Scale bar, 2 mm; a.u., arbitrary fluorescence intensity units. (b) Fluorescence time courses for Ca2+ microdomains in apical (gray) and midcochlear (black) IHCs. Data were obtained at high temporal resolution at the brightest pixel of the hotspot (spot detection; see Methods) (dotted crosshairs in a), on which the laser beam was parked. Red bar, depolarization to –7 mV. (c) Ca2+ microdomain amplitude as a function of membrane potential. The average DF amplitude over the last 15 ms of the respective spot detection response to 20 ms depolarizations is plotted. (d) Line scan. Dashed line in left panel, IHC border. (e) Spatial extent of Ca2+ microdomains. FWHM values were obtained by fits of gaussian functions to time-averaged (20 ms) line scans (d). Error bars, s.e.m.

449

ARTICLES

b

F – F0 (a.u.)

80 60 40 20 0

60 40

d

20

80

0

f

1.6

© 2009 Nature America, Inc. All rights reserved.

(F – F0) / F0

1.4

Neural (n = 10) Abneural (n = 14)

1.2 1.0 0.8 0.6 0.4 0.2 0.0 60

70 Time (ms)

80

Neural (n = 46) Abneural (n = 43)

0.8 0.4

15

g

10

5

90

15

10

Figure 6 Intracellular variability of synaptic Ca microdomains. (a) Time course of fluorescence at the center of Ca2+ microdomains during 20-ms depolarization to peak Ca2+ current potential (–7 mV, as described in Fig. 5c; n ¼ 65 Ca2+ microdomains in 20 IHCs); a.u., arbitrary fluorescence intensity units. (b) Ca2+ microdomain amplitude within individual IHCs as a function of time of acquisition. (c,d) Ca2+ microdomain amplitude (DF amplitude over the last 15 ms of data in a) correlates weakly with fast onset kinetics (c, linear correlation coefficient Pr = 0.28, n = 65) and with its FWHM (d; Pr = 0.23, n = 87). DF amplitude and FWHM were obtained by fitting gaussian functions to time-averaged line scans. (e) Differences in average peak Ca2+ microdomain amplitude between neural and abneural synapses (P o 0.05, for peak amplitude). Gray bands, s.e.m. (f,g) Overlapping distributions of short and long ribbon axes of neural and abneural synapses of apical IHCs (distance to apex, 100–300 mm) as estimated by STED microscopy.

5

the mouse, this region encodes the range of B10–20 kHz, which is important, for examLong object axis (nm) ple, for the perception of wriggling calls of mouse pups46. Each SGN conveys information transmitted by one hair cell synapse to several neurons of the cochlear nucleus, which in turn integrate information from several SGNs. The reliability and acuity of afferent information provided by a region of the cochlea will increase with the number of innervating SGNs. Here we took advantage of the tonotopic gradient of synapse number but uniform average presynaptic morphology in the mouse cochlea to ask whether whole-cell Ca2+ current and exocytic DCm scaled with the number of ribbon synapses, as would be expected if both occurred exclusively at the synapse. This was indeed found to be the case for exocytosis, as previously described for turtle hair cells37. Although we cannot exclude some contribution of vesicles fusing outside the active zone to exocytosis seen during prolonged stimulation3,47, we suppose that much of this sustained release occurs at the synapse. This view is supported by paired pre- and postsynaptic recordings from rat IHC synapses8. These data revealed AMPA receptor current integrals that were compatible with the notion of sustained exocytosis—as reported by capacitance measurements—reflecting synaptic release. Our STED data suggest that, at the IHC synapse, AMPA receptors have a peripheral, ‘ring-like’ density maximum, which correlates with electron microscopy data on the rat cochlea12. The observed receptor distribution seems well suited for the efficient detection of glutamate release, in particular if this release occurs preferentially at the circumference of the ribbon35. It deviates from the uniform receptor distribution of glutamatergic CNS synapses suggested by immunoelectron microscopy (ref. 12; see review in ref. 48). The Ca2+ current integral was not significantly different between apical and midcochlear IHCs. In part, this can be attributed to the stronger Ca2+ current inactivation in IHCs in the higher frequency region, which is consistent with findings in the gerbil49. Still, even the initial Ca2+ current was somewhat less correlated to synapse number than exocytosis. Similar findings were obtained in the turtle cochlea, where low-frequency synapses were assumed to mediate larger Ca2+ influx37. Unlike that study, we conclude, based on analysis of Ca2+ current, synaptic Ca2+ signals and Ca2+ channel cluster size that, except for inactivation, Ca2+signaling of the average mouse IHC synapse is structurally and functionally similar

0 200 300 400 500 600 700

Short object axis (nm)

suggested in the cat24. We used confocal Ca2+ imaging and STED microscopy of immunolabeled ribbons in the neural and abneural 601 sectors to test for potential functional and morphological differences of the respective synapses. Abneural synapses on average showed a smaller fluorescence increment (Fig. 6e; DF/F0 neural/abneural ratio, 1.6). The distributions of short (Fig. 6f) and long axes of ribbons (Fig. 6g) were indistinguishable, suggesting that the average ribbon size did not differ among neural and abneural synapses in mouse IHCs. DISCUSSION Our findings demonstrated that tonotopic variations in the sensitivity of the mammalian cochlea are paralleled by the density of its afferent innervation. We present nanoscale estimates of the size of ribbons, Ca2+ channel and AMPA receptor clusters at the IHC synapse for two different tonotopic regions. Although substantially changing in number, IHC synapses in these cochlear regions differ only slightly in structure. Exocytosis of IHCs scaled with the number of synapses, which, together with the comparable average properties of presynaptic Ca2+ signals, suggests a fairly uniform average synapse function in these two regions and also, presumably, further along the cochlea’s tonotopic axis. However, Ca2+ signals showed substantial heterogeneity among synapses of an individual IHC, providing a potential presynaptic substrate for the divergent spiking properties of the SGNs that are driven by a given IHC. Quantitative light microscopy of whole-mounted organs of Corti allowed assembly of continuous synaptic cochleograms for mouse (high-frequency hearing) and gerbil (low-frequency hearing) with a sample size and frequency resolution exceeding those of previous electron microscopy studies17,20,21,43. Synaptic cochleograms were well described by quadratic functions, providing a simple and useful tool for future studies on cochlear neurotransmission. The shape of the mouse synaptic cochleogram roughly followed that of the mouse behavioral audiogram44, that of the neural population audiogram (for example, Fig. 1) and that of the distortion product otoacoustic emission audiogram (for example, ref. 45). We interpret these findings as better neural sampling from the most sensitive cochlear regions. In

450

1.6 1.2

100 200 300 Peak intensity (F – F0)

0 50

20 40 60 80 100 Peak intensity (F – F0)

10 20 30 40 Experiment time (min)

Absolute frequency

e

0

2.5 2.0 1.5 1.0 0.5

Absolute frequency

20 ms

c τonset, fast (ms)

100

Individual IHCs (n = 16) Average trace 100

FWHM (µm)

Peak fluorescence intensity (F – F0)

a

200 300 400 500 600 700

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

© 2009 Nature America, Inc. All rights reserved.

ARTICLES in the two frequency regions. The existence of extrasynaptic Ca2+ channels, as suggested by single channel recordings from bullfrog hair cells39, is likely to explain the small remaining mismatch between the scaling of number of Ca2+ channels (1.34) and number of synapses (1.44). Using various approaches, we compared, at two different frequency regions, key morphological determinants of IHC synaptic transmission, including synaptic ribbons and vesicle complement, synaptic Ca2+ channel clusters and postsynaptic density. We found only subtle differences (Fig. 3, Table 1), all indicative of slightly larger IHC synapses in the apex of the cochlea. Testing the functional relevance of these tonotopic differences requires more sensitive detection, such as in paired pre- and postsynaptic recordings8. However, we found considerable variability for all of the investigated morphological and functional synapse parameters (Fig. 3, Tables 1 and 2, Supplementary Table 3 and Supplementary Fig. 4 and 5). Although this limits detection of small systematic changes along the tonotopic axis, it probably relates to the important question in auditory research of the mechanisms behind the different functional properties of SGNs. Pioneering work primarily in the cat cochlea has identified morphological correlates for high and low spontaneous firing rate SGNs (for example, active zone morphology, abneural versus neural insertion; diameter and mitochondrial content of the peripheral axon)14,22,24,43. In the present study, we observed slightly lower synaptic Ca2+ signals in the abneural sector of IHCs and found ribbon sizes to be comparable for neural and abneural synapses. These findings seem hard to reconcile with a systematically higher activity of abneural synapses in mouse IHCs, for which one might have expected larger ribbons providing more releasable vesicles3 and more Ca2+ channels. Nevertheless, our experiments revealed substantial heterogeneity between Ca2+ signals at active zones within individual IHCs (Fig. 6b and Supplementary Table 3). Furthermore, we found a large variation of CaV1.3 channel cluster size (CV ¼ 0.3), and, assuming a constant channel density in the cluster, we propose that differences in channel number contribute to this heterogeneity. Placement of synapses with different properties seems rather random in mouse IHC. Future studies combining imaging of synaptic Ca2+ signals with readouts of transmitter release and/or postsynaptic response7,8 will test how this translates into differences of transmitter release among the synapses of an individual mouse IHC. METHODS Animals. C57BL/6 and NMRI (Naval Medical Research Institute) mice aged 2–4 weeks and 4-week-old gerbils were used for experiments. Animal experiments complied with national animal care guidelines and were approved by the University of Go¨ttingen Board for animal welfare and the animal welfare office of the state of Lower Saxony. Auditory brainstem responses. See Supplementary Methods. Patch-clamp and confocal Ca2+ imaging. IHCs from apical coils of freshly dissected organs of Corti from NMRI and C57BL/6 mice (P14–18) were patchclamped as described5. The pipette solution for perforated-patch recordings contained (in mM) 140 cesium gluconate, 13 tetraethylammonium (TEA)-Cl, 10 CsOH-HEPES buffer, 1 MgCl2, and 250 mg/ml amphotericin B, pH 7.2. The pipette solution for whole cell recordings contained (in mM) 135 cesium glutamate, 13 TEA-Cl, 20 CsOH-HEPES, 1 MgCl2, 2 Mg-ATP, 0.3 Na-GTP, 2 EGTA (10 for biophysical analysis of Ca2+ currents), 0.4 Fluo-5N (penta-K+ salt, Invitrogen; for confocal imaging), pH 7.0. The extracellular solution contained (in mM) 105 NaCl, 35 TEA-Cl, 2.8 KCl, 2 CaCl2 (10 for fluctuation analysis, 5 for confocal imaging, balanced by NaCl), 1 MgCl2, 10 NaOHHEPES, 10 D-glucose and 0.005 BayK8644 (Tocris, for fluctuation analysis), pH 7.2 (7.3 for whole-cell recordings). EPC-9 amplifiers controlled by Pulse or

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

Patchmaster software (HEKA Elektronik) were used for measurements. All voltages were corrected for liquid-junction potentials. Currents were low-pass filtered at 14 kHz and 2 kHz and sampled at 100 kHz and 10 kHz for whole-cell recordings and perforated-patch recordings, respectively. Cells that showed a holding current exceeding 50 pA were discarded from analysis. Ca2+ currents were further isolated using a P/n protocol. Series resistance was required to be below 30 MO for perforated-patch experiments and averaged 10.0 ± 0.5 MO (n ¼ 59) in the fluctuation analysis experiments. Fluctuation analysis was performed similarly to that previously described11. To account for channel gating-related and filter-induced correlations between neighboring currentvariance data points, we fitted the variance-over-mean data by an estimated generalized least-squares method (Supplementary Table 2). The first 600 ms of the tail current routinely were discarded. To avoid errors introduced by remaining, uncancelled capacitive transients, we subtracted a scaled version of the transient that was evident at the beginning of the depolarization step (see Fig. 4e) from the mean current trace (after calculation of the ensemble variance). Membrane capacitance increments (DCm) were measured as previously described5. Interstimulus periods were 2–3 s between sweeps, 1–2 min between ensembles for confocal imaging, and 30–70 s for exocytosis measurements. Confocal Ca2+ imaging was performed as described40, using a Fluoview 300 confocal scanner mounted on an upright microscope (BX50WI) equipped with a 0.9 numerical aperture (NA),
60, water immersion objective (all Olympus) and a 50-mW, 488-nm, solid-state laser (Cyan, Newport-Spectraphysics). Fluorescent hotspots were identified in xy-scans during 200-ms depolarizations (0.5% of maximum laser intensity) and further characterized using spot detection (‘point scan’ mode of the confocal scanner, temporally averaged to yield an effective sampling rate of 1.85 kHz, 0.05% of maximum laser intensity) and line scans (at a rate of 0.74 kHz, 0.25% of maximum laser intensity). Spot detection measurements and line scans were repeated 5 and 10 times, respectively, to improve signal-to-noise ratio. Peak DF estimates of spot detection measurements were obtained after repetitive boxcar smoothing (2-ms box). On average, we characterized 3.1 and 1.7 spots per IHC for apical-basal and neural-abneural comparisons, respectively. The average Ca2+ current rundown at the end of the experiment was 30% of the maximum current. For investigating intracellular differences, Ca2+ microdomain characterization was followed by acquisition of a z-stack of the indicator-filled cell so that its location could be retrieved. Immunohistochemistry. Immunostaining was performed as previously described6. Briefly, the freshly dissected apical cochlear turns were fixed with 4% paraformaldehyde for 1 h on ice, with 2% formaldehyde for 10 min at 20–22 1C (for staining with antibody to GluR2/3 (anti-GluR2/3)), or for 25 min in 99% methanol at –20 1C (for anti-CaV1.3). For harvesting the fulllength organs of Corti, cochleae were fixed by cochlear perfusion with 2% formaldehyde for 10 min. The following antibodies were used: mouse IgG1 anti-CtBP2 (also recognizing the ribbon protein RIBEYE; BD Biosciences, 1:150), rabbit anti-GluR2/3 (Chemicon, 1:200), rabbit anti-calbindin (Swant, 1:400), rabbit anti-CaV1.3 (Alomone Labs, 1:75) and secondary Alexa Fluor 488– and Alexa Fluor 568–labeled antibodies (Molecular Probes, 1:200) as well as Atto-647N (AttoTech, 1:60 dilution in PBS with addition of 2% normal goat serum) for STED microscopy. In some experiments, nuclei were stained with Hoechst 34580 (Molecular Probes, 1:1,000). Confocal, 4Pi and STED microscopy. Confocal morphological images were acquired using a laser scanning confocal microscope (Leica TCS SP5, Leica Microsystems) with 405-nm (diode), 488-nm (argon) and 561-nm (diodepumped solid state) lasers for excitation and a
63 oil immersion objective (NA ¼ 1.4). Whole-mount preparations of the organ of Corti allowed us to analyze several IHCs in a row, as previously described6. For three-dimensional reconstructions of the specimen, z-axis stacks of two-dimensional images were taken with a step size of 0.049 mm, 0.2 mm or 0.3 mm. Multifocal 4Pi microscopy with water immersion lenses (NA 1.2) at a two-photon excitation wavelength of 870 nm (average power, 1.5 mW for each of the 4Pi foci) was performed at a custom microscope as described6,26,27. For STED imaging, two different microscopes were used: (i) a Leica TCS STED microscope (Fig. 3d–g) using two pulsed lasers for excitation (diode laser, 635 nm,

451

ARTICLES o90 ps) and stimulated emission (Ti:sapphire, 750 nm, B300 ps), with both lasers running at a repetition rate of 80 MHz and synchronized to each other to ensure optimal STED efficiency in the focal plane; and (ii) a custom microscope50 (Fig. 3h–j). This microscope used identical lasers but allowed for higher STED powers and therefore exhibited a resolution of B50 nm and B150 nm for two-dimensional (x,y) and three-dimensional (x,y,z) imaging, respectively. Single confocal and STED images of ribbons and Ca2+ channel clusters were acquired after adjusting the focus of the
100 oil immersion lens (NA ¼ 0.7 or 1.4) to the fluorescence maximum of an object of interest, as found in a xz-scan.

© 2009 Nature America, Inc. All rights reserved.

Electron microscopy. See Supplementary Methods. Data analysis. Data was analyzed using Matlab (MathWorks), Igor Pro (Wavemetrics) and ImageJ software. Figures were assembled for display in Adobe Photoshop and Illustrator software. Mean DCm and Ca2+ current estimates are grand means calculated from the mean estimates of individual IHCs. Means were expressed ± s.d. (or s.e.m. when noted). If applicable (that is, as determined by normality of the distribution (Jarque-Bera test) and equal variances between (F-test) the two samples), an unpaired, two-tailed t-test was used to compare the two samples. In all other cases, a Mann-WhitneyWilcoxon test was used. Image analysis. For synaptic cochleograms, CtBP2/RIBEYE immunofluorescence spots in the basolateral portion of IHCs (up to the apical end of the CtBP2-stained nucleus) were counted in z-stacks and divided by the number of IHCs (taken as the quantity of nuclei in the field of view). Estimation of the cellular synapse distribution was performed using custom-written MATLAB routines (available at http://www.innerearlab.uni-goettingen.de/) that included (i) alignment of the image stacks with the tonotopic axis, (ii) image segmentation into ribbons (positive, if more than four connected voxels were above threshold) and nuclei, (iii) center of mass calculation for both types of structures (by a three-dimensional gaussian fit in the case of nuclei), (iv) assignment of ribbons to the closest IHC nucleus, (v) alignment of the individual IHC’s z-axis to a common z-axis, (vi) normalization of the cell’s z-extent (measured from the center of the nucleus to the basal end of the IHC) to the population average and (vii) vector calculation. For Figure 2b, we also normalized the radial extent of the IHCs for improved superposition. The FWHM in confocal and STED images was estimated using gaussian functions (two-dimensional for morphology, one-dimensional for functional imaging). The two-dimensional data in Figure 3j was linearly deconvolved with a two-dimensional gaussian PSF (FWHM of 70 nm for the STED and 250 nm for the confocal). Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank E. Neher, M. Go¨pfert, R. Nouvian, N. Strenzke, M. Mu¨ller and A. Lysakowski for comments on the manuscript; K. Wadel and C. Henrich for participation in an early stage of the project; A. Neef for image analysis support; J. Hegerman and S. Eimer for support with high-pressure rapid freeze and freeze substitution; and C. Ru¨diger and M. Ko¨ppler for technical assistance. This work was supported by grants from the Deutsche Forschungsgemeinschaft (Center for Molecular Physiology of the Brain; T.M., A.E. and S.W.H), a Lichtenberg Fellowship from the state of Lower Saxony (T.F.), the European Commission (Eurohear, T.M.), the Max-Planck-Society (Tandemproject, T.M.), BMBF (Bernstein Center for Computational Neuroscience Go¨ttingen, T.M.) and an intramural grant from the University of Go¨ttingen Medical School (A.C.M.). AUTHOR CONTRIBUTIONS The study was designed by T.M., A.C.M., A.E. and T.F. The experimental work was performed by A.C.M., T.F., D.K., D.R., G.H., N.M.C., Y.M.Y. and B.H. S.W.H. co-developed the super-resolution microscopes. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Fuchs, P.A. Time and intensity coding at the hair cell’s ribbon synapse. J. Physiol. 566, 7–12 (2005).

452

2. Moser, T., Neef, A. & Khimich, D. Mechanisms underlying the temporal precision of sound coding at the inner hair cell ribbon synapse. J. Physiol. (Lond.) 576, 55–62 (2006). 3. Nouvian, R., Beutner, D., Parsons, T.D. & Moser, T. Structure and function of the hair cell ribbon synapse. J. Membr. Biol. 209, 153–165 (2006). 4. Sterling, P. & Matthews, G. Structure and function of ribbon synapses. Trends Neurosci. 28, 20–29 (2005). 5. Moser, T. & Beutner, D. Kinetics of exocytosis and endocytosis at the cochlear inner hair cell afferent synapse of the mouse. Proc. Natl. Acad. Sci. USA 97, 883–888 (2000). 6. Khimich, D. et al. Hair cell synaptic ribbons are essential for synchronous auditory signalling. Nature 434, 889–894 (2005). 7. Griesinger, C.B., Richards, C.D. & Ashmore, J.F. Fast vesicle replenishment allows indefatigable signalling at the first auditory synapse. Nature 435, 212–215 (2005). 8. Goutman, J.D. & Glowatzki, E. Time course and calcium dependence of transmitter release at a single ribbon synapse. Proc. Natl. Acad. Sci. USA 104, 16341–16346 (2007). 9. Platzer, J. et al. Congenital deafness and sinoatrial node dysfunction in mice lacking class D L-type Ca2+ channels. Cell 102, 89–97 (2000). 10. Brandt, A., Striessnig, J. & Moser, T. CaV1.3 channels are essential for development and presynaptic activity of cochlear inner hair cells. J. Neurosci. 23, 10832–10840 (2003). 11. Brandt, A., Khimich, D. & Moser, T. Few CaV1.3 channels regulate the exocytosis of a synaptic vesicle at the hair cell ribbon synapse. J. Neurosci. 25, 11577–11585 (2005). 12. Matsubara, A., Laake, J.H., Davanger, S., Usami, S. & Ottersen, O.P. Organization of AMPA receptor subunits at a glutamate synapse: a quantitative immunogold analysis of hair cell synapses in the rat organ of Corti. J. Neurosci. 16, 4457–4467 (1996). 13. Glowatzki, E. & Fuchs, P.A. Transmitter release at the hair cell ribbon synapse. Nat. Neurosci. 5, 147–154 (2002). 14. Liberman, M.C. Morphological differences among radial afferent fibers in the cat cochlea: an electron-microscopic study of serial sections. Hear. Res. 3, 45–63 (1980). 15. Rutherford, M.A. & Roberts, W.M. Frequency selectivity of synaptic exocytosis in frog saccular hair cells. Proc. Natl. Acad. Sci. USA (2006). 16. Johnson, S.L., Forge, A., Knipper, M., Munkner, S. & Marcotti, W. Tonotopic variation in the calcium dependence of neurotransmitter release and vesicle pool replenishment at mammalian auditory ribbon synapses. J. Neurosci. 28, 7670–7678 (2008). 17. Liberman, M.C. The cochlear frequency map for the cat: labeling auditory-nerve fibers of known characteristic frequency. J. Acoust. Soc. Am. 72, 1441–1449 (1982). 18. Muller, M., von Hunerbein, K., Hoidis, S. & Smolders, J.W. A physiological placefrequency map of the cochlea in the CBA/J mouse. Hear. Res. 202, 63–73 (2005). 19. Liberman, M.C., Dodds, L.W. & Pierce, S. Afferent and efferent innervation of the cat cochlea: quantitative analysis with light and electron microscopy. J. Comp. Neurol. 301, 443–460 (1990); erratum 304, 341 (1991).. 20. Slepecky, N.B., Galsky, M.D., Swartzentruber-Martin, H. & Savage, J. Study of afferent nerve terminals and fibers in the gerbil cochlea: distribution by size. Hear. Res. 144, 124–134 (2000). 21. Francis, H.W., Rivas, A., Lehar, M. & Ryugo, D.K. Two types of afferent terminals innervate cochlear inner hair cells in C57BL/6J mice. Brain Res. 1016, 182–194 (2004). 22. Liberman, M.C. Single-neuron labeling in the cat auditory nerve. Science 216, 1239–1241 (1982). 23. Taberner, A.M. & Liberman, M.C. Response properties of single auditory nerve fibers in the mouse. J. Neurophysiol. 93, 557–569 (2005). 24. Merchan-Perez, A. & Liberman, M.C. Ultrastructural differences among afferent synapses on cochlear hair cells: correlations with spontaneous discharge rate. J. Comp. Neurol. 371, 208–221 (1996). 25. Ruel, J. et al. Dopamine inhibition of auditory nerve activity in the adult mammalian cochlea. Eur. J. Neurosci. 14, 977–986 (2001). 26. Hell, S. & Stelzer, E.H.K. Properties of a 4Pi-confocal fluorescence microscope. J. Opt. Soc. Am. A 9, 2159–2166 (1992). 27. Egner, A., Jakobs, S. & Hell, S.W. Fast 100-nm resolution three-dimensional microscope reveals structural plasticity of mitochondria in live yeast. Proc. Natl. Acad. Sci. USA 99, 3370–3375 (2002). 28. Hell, S.W. & Wichmann, J. Breaking the diffraction resolution limit by stimulated emission: stimulated emission depletion microscopy. Opt. Lett. 19, 780–782 (1994). 29. Muller, M. The cochlear place-frequency map of the adult and developing Mongolian gerbil. Hear. Res. 94, 148–156 (1996). 30. Wu, Y.C., Tucker, T. & Fettiplace, R. A theoretical study of calcium microdomains in turtle hair cells. Biophys. J. 71, 2256–2275 (1996). 31. Martinez-Dunst, C., Michaels, R.L. & Fuchs, P.A. Release sites and calcium channels in hair cells of the chick’s cochlea. J. Neurosci. 17, 9133–9144 (1997). 32. Masugi-Tokita, M. et al. Number and density of AMPA receptors in individual synapses in the rat cerebellum as revealed by SDS-digested freeze-fracture replica labeling. J. Neurosci. 27, 2135–2144 (2007). 33. Tokuoka, H. & Goda, Y. Activity-dependent coordination of presynaptic release probability and postsynaptic GluR2 abundance at single synapses. Proc. Natl. Acad. Sci. USA 105, 14656–14661 (2008). 34. Rostaing, P., Weimer, R.M., Jorgensen, E.M., Triller, A. & Bessereau, J.L. Preservation of immunoreactivity and fine structure of adult C. elegans tissues using high-pressure freezing. J. Histochem. Cytochem. 52, 1–12 (2004). 35. Roberts, W.M., Jacobs, R.A. & Hudspeth, A.J. Colocalization of ion channels involved in frequency selectivity and synaptic transmission at presynaptic active zones of hair cells. J. Neurosci. 10, 3664–3684 (1990). 36. Tucker, T. & Fettiplace, R. Confocal imaging of calcium microdomains and calcium extrusion in turtle hair cells. Neuron 15, 1323–1335 (1995).

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES 43. Tsuji, J. & Liberman, M.C. Intracellular labeling of auditory nerve fibers in guinea pig: central and peripheral projections. J. Comp. Neurol. 381, 188–202 (1997). 44. Ehret, G. Age-dependent hearing loss in normal hearing mice. Naturwissenschaften 61, 506–507 (1974). 45. Liberman, M.C. et al. Prestin is required for electromotility of the outer hair cell and for the cochlear amplifier. Nature 419, 300–304 (2002). 46. Ehret, G. Common rules of communication sound perception. in Behavior and Neurodynamics for Auditory Communication (ed. J.S. Kanwal & G. Ehret) 85–114 (Cambridge University Press, Cambridge, 2006). 47. Fuchs, P.A., Glowatzki, E. & Moser, T. The afferent synapse of cochlear hair cells. Curr. Opin. Neurobiol. 13, 452–458 (2003). 48. Nusser, Z. AMPA and NMDA receptors: similarities and differences in their synaptic distribution. Curr. Opin. Neurobiol. 10, 337–341 (2000). 49. Johnson, S.L. & Marcotti, W. Biophysical properties of CaV1.3 calcium channels in gerbil inner hair cells. J. Physiol. (Lond.) 586, 1029–1042 (2008). 50. Harke, B. et al. Resolution scaling in STED microscopy. Opt. Express 16, 4154–4162 (2008).

© 2009 Nature America, Inc. All rights reserved.

37. Schnee, M.E., Lawton, D.M., Furness, D.N., Benke, T.A. & Ricci, A.J. Auditory hair cellafferent fiber synapses are specialized to operate at their best frequencies. Neuron 47, 243–254 (2005). 38. Beutner, D., Voets, T., Neher, E. & Moser, T. Calcium dependence of exocytosis and endocytosis at the cochlear inner hair cell afferent synapse. Neuron 29, 681–690 (2001). 39. Rodriguez-Contreras, A. & Yamoah, E.N. Direct measurement of single-channel Ca2+ currents in bullfrog hair cells reveals two distinct channel subtypes. J. Physiol. (Lond.) 534, 669–689 (2001). 40. Frank, T., Khimich, D., Neef, A. & Moser, T. Mechanisms contributing to synaptic Ca2+ signals and their heterogeneity in hair cells. Proc. Natl. Acad. Sci. USA (in the press). 41. Issa, N.P. & Hudspeth, A.J. Clustering of Ca2+ channels and Ca2+-activated K+ channels at fluorescently labeled presynaptic active zones of hair cells. Proc. Natl. Acad. Sci. USA 91, 7578–7582 (1994). 42. Zenisek, D., Davila, V., Wan, L. & Almers, W. Imaging calcium entry sites and ribbon structures in two presynaptic cells. J. Neurosci. 23, 2538–2548 (2003).

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

453

ARTICLES

Phosphodiesterase 1C is dispensable for rapid response termination of olfactory sensory neurons

© 2009 Nature America, Inc. All rights reserved.

Katherine D Cygnar & Haiqing Zhao In the nose, odorants are detected on the cilia of olfactory sensory neurons (OSNs), where a cAMP-mediated signaling pathway transforms odor stimulation into electrical responses. Phosphodiesterase (PDE) activity in OSN cilia has long been thought to account for rapid response termination by degrading odor-induced cAMP. Two PDEs with distinct cellular localization have been found in OSNs: PDE1C in the cilia and PDE4A throughout the cell but absent from the cilia. We disrupted both of these genes in mice and carried out electro-olfactogram analysis. Unexpectedly, eliminating PDE1C did not prolong response termination. Prolonged termination occurred only in mice that lacked both PDEs, suggesting that cAMP degradation by PDE1C in cilia is not a rate-limiting factor for response termination in wild-type mice. Pde1c2/2 OSNs instead showed reduced sensitivity and attenuated adaptation to repeated stimulation, suggesting that PDE1C may be involved in regulating sensitivity and adaptation. Our observations provide new perspectives on the regulation of olfactory transduction.

In vertebrates, OSNs use a cAMP second messenger–mediated signal transduction pathway to convert odor stimulation into electrical signals. Olfactory signal transduction takes place on the cilia extending from the dendritic knob of the OSN1,2. Odor exposure leads to elevated activity of adenylyl cyclase III (ACIII), resulting in a rapid rise of the cAMP concentration in the cilia. cAMP directly opens olfactory cyclic nucleotide–gated (CNG) ion channels, resulting in an influx of Ca2+ and Na+ and membrane depolarization3,4. Ca2+ can then open Ca2+activated Cl
channels, which results in an efflux of Cl
and further depolarization, contributing to the generation of action potentials5–8. In cyclic nucleotide–mediated signal cascades, the magnitude and duration of second messenger signals are determined by the activities of two opposing enzymes: the cyclase that produces cyclic nucleotides and the PDE that degrades them. In olfactory transduction, PDE activity in the cilia has long been thought to account for rapid termination of the OSN response by degrading odor-induced cAMP9,10, leading to the closure of CNG channels. Two distinct cyclic nucleotide PDEs, PDE1C11,12 and PDE4A13,14, have been identified in mammalian OSNs. PDE1C is a Ca2+/calmodulin-stimulated PDE11 and PDE4A is Ca2+ insensitive but has substrate specificity for cAMP15. Immunohistochemical staining revealed that the PDE1C protein is selectively enriched in the cilia, whereas the PDE4A protein is distributed throughout the cell, including in the dendritic knob from which the cilia emanate, but is absent from the cilia13,14,16. PDE1C was therefore hypothesized to be critical for rapid termination of the OSN response as a result of its cilial localization and Ca2+ dependency, whereas PDE4A was not expected to affect OSN responses, as it is excluded from the cilia. To assess the function of PDE1C and PDE4A in olfactory transduction, we disrupted both the Pde1c and the Pde4a genes in mice and

conducted electrophysiological analysis of OSN responses by electroolfactogram (EOG). A loss of PDE1C was predicted to prolong response termination. Contrary to this expectation, eliminating PDE1C alone resulted in accelerated termination of the EOG response. Prolonged response termination occurred only in mice that lacked both PDE1C and PDE4A, indicating that the activity of either PDE1C in the cilia or PDE4A outside of the cilia is sufficient to allow rapid termination of the EOG response. These results suggest that cAMP degradation by PDE1C in the cilia is not a rate-limiting factor for response termination in wild-type OSNs. We found that Pde1c
/
OSNs instead showed reduced sensitivity to odors and attenuated adaptation to repeated stimulation. RESULTS Generation of Pde1c
/
and Pde4a
/
mice We used a loss-of-function approach to examine the role of PDEs in olfactory transduction. We generated mouse lines in which the Pde1c and Pde4a genes were disrupted individually and in combination (Supplementary Fig. 1 online). This genetic approach allows specific and complete elimination of the targeted PDE activity, a condition that is not achievable with current pharmacological inhibitors. Both individual knockout mice and Pde1c
/
; Pde4a
/
double knockout mice showed apparently normal growth rate and feeding, nursing and mating behaviors. We confirmed the loss of PDE1C and PDE4A proteins in the relevant knockout strains by immunohistochemistry (Fig. 1a). In wild-type olfactory epithelium, PDE1C was detected primarily in the cilial layer, whereas PDE4A was detected in OSN cell bodies, dendrites and axons. PDE1C and/or PDE4A immunoreactivity was absent in the relevant mutant mice. The thickness of the olfactory epithelium at similar nasal

Department of Biology, the Johns Hopkins University, Baltimore, Maryland, USA. Correspondence should be addressed to H.Z. ([email protected]). Received 15 December 2008; accepted 2 February 2009; published online 22 March 2009; doi:10.1038/nn.2289

454

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

Pde1c
/
OSNs show smaller responses and faster termination To determine the function of PDE1C in olfactory transduction, we recorded EOG signals from Pde1c
/
and wild-type mice in response to 100-ms odor pulses (Fig. 2). The EOG has been widely used as a reliable and convenient means to assess OSN responses17,18 and is thought to result primarily from the summed generator potentials of OSNs, although it may also have other components. We used two common odorants, amyl acetate and heptaldehyde, and analyzed four metrics of the EOG signal: the amplitude, the response latency, the rise time and the rate of response termination. Similar results were obtained for both odorants (quantification of the responses to heptaldehyde is shown in the Supplementary Table 1 online). Eliminating PDE1C, the only PDE known to be expressed in the cilia, was expected to prolong response termination. It may also allow for larger responses with quicker onset, as cAMP concentrations could increase more rapidly when the degradation of cAMP is reduced. Contrary to expectations, Pde1c
/
mice had faster response termination compared with the wild-type mice (Fig. 2b). The rate of response termination was quantified by calculating the time constant (t) via fitting the decay phase of the EOG signal with a single exponential equation. The termination t’s of Pde1c
/
mice were smaller than those

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

–

4a

–/

T 1C 4A D K

W

W

T 1C 4A D K

CNGB1b level (a.u.)

W

W

T 1C 4A D K

ACIII level (a.u.)

© 2009 Nature America, Inc. All rights reserved.

positions was comparable between all of the genotypes and all of the genotypes showed similar immunostaining for other olfactory transduction components, including ACIII and the CNG channel subunit CNGB1b, and for GAP43 (immature OSN marker) and OMP (mature OSN marker) (Fig. 1a). The loss of one of the PDEs did not affect the localization of the other PDE, that is, the localization of PDE4A protein was normal in Pde1c
/
mice and vice versa (Fig. 1a). However, we found that there was a significant reduction in the amount of ACIII protein in Pde1c
/
and Pde1c
/
; Pde4a
/
olfactory epithelium, as determined by western blotting (Fig. 1b,c). The level of CNGB1b protein in these mice appeared to be unchanged.

Pd e

e

1c

ty p

T 1C 4A D K

PDE4A level (a.u.)

Pd e

W

PDE1C level (a.u.)

c

–/

–

b

ild

a

P Pd de1 e4 c – / – a –/ ; –

Figure 1 Molecular characterization of Pde1c
/
, Pde1c –/–; Pde4a
/
and Pde1c
/
; Pde4a
/
mice. –/– Wild type Pde4a –/– Pde4a –/– Pde1c (a) Immunostaining of sections of olfactory C S epithelium. Sections were counterstained with PDE1C O DAPI (blue) to label cell nuclei. C, cilial layer; BL, PDE1C BL basal lamina; O, olfactory sensory neuron layer; S, PDE4A supporting cell layer. Scale bar represents 20 mm. (b,c) Western blot analysis of total olfactory ACIII epithelium proteins. Example western blots are PDE4A shown for PDE1C, PDE4A, ACIII and CNGB1b on CNGB1b total olfactory epithelium proteins from all four α tubulin genotypes (b). a tubulin was used as a loading control. Quantification of the western blotting is ACIII 2.5 shown in c. Values are normalized to a tubulin 2.0 2.0 staining and shown relative to wild type in 1.5 arbitrary units (a.u.). PDE1C was detected at 1.5
/
1.0 similar levels in wild type and Pde4a mice 1.0 CNGB1b (1.55 ± 0.79), but was undetectable in Pde1c
/
0.5 0.5 and Pde1c
/
; Pde4a
/
mice. PDE4A was ** ** ** ** 0.0 0.0 detected at similar levels in wild type and
/
Pde1c mice (1.32 ± 0.73), but was un2.5 GAP43 1.0 detectable in Pde4a
/
and Pde1c
/
; Pde4a
/
2.0 0.8 * * mice. ACIII levels were significantly reduced in 1.5 0.6 Pde1c
/
(0.52 ± 0.22; P ¼ 0.05) and Pde1c–/–; –/– 1.0 0.4 Pde4a mice (0.47 ± 0.26; P ¼ 0.05), but not OMP 0.5 0.2 in Pde4a–/– mice (0.72 ± 0.28; P o 0.19). CNGB1b levels were similar in all genotypes 0.0 0.0 (Pde1c
/
, 0.97 ± 0.26; Pde4a
/
, 1.55 ± 1;
/

/
Pde1c ; Pde4a , 0.85 ± 0.31). Data are the average of three independent sets of mice, each set consisting of a mouse of each genotype. Error bars are 95% confidence intervals. Statistical significance was determined by one-sample t test. * P ¼ 0.05, ** P o 0.01. WT, wild type; 1C, Pde1c
/
; 4A, Pde4a
/
; DK, Pde1c
/
; Pde4a
/
.

of wild types for all odorant concentrations tested (Fig. 2c and Table 1, which lists data for 10
4, 10
3 and 10
2 M amyl acetate). The responses of Pde1c+/
heterozygous mice were no different from those of wild types (Supplementary Fig. 2 online). Pde1c
/
mice also had significantly reduced EOG amplitudes with slower onset kinetics. The maximum EOG amplitude of Pde1c
/
mice (9.8 ± 2.5 mV, n ¼ 5 mice) was approximately half of that of wild-type mice (21.4 ± 3.9 mV, n ¼ 7 mice) (Fig. 2a). Both the response latency, defined as the time between the initiation of the odor pulse and the start of the response (1% of EOG peak value), and the rise time, defined as the time from the start of the response to the peak, were significantly longer in Pde1c
/
mice (Fig. 2d,e and Table 1). The reduced response magnitude, slower onset kinetics and faster response termination that we observed in Pde1c
/
mice are reminiscent of OSNs in a state of cellular adaptation. PDE1C knockout eliminates all PDE activity from OSN cilia Because Pde1c
/
mice did not show the expected prolonged response termination, we asked whether there is additional PDE activity in the Pde1c
/
cilia that are responsible for removal of cAMP. Our initial characterization of Pde1c
/
mice showed that PDE4A, the only other PDE found in OSNs, remained localized in the dendrite and cell body but not in the cilial layer (Fig. 1a). We further confirmed this observation by immunofluorescence, comparing PDE4A staining with that of the cilia marker acetylated tubulin. Antibodies to PDE4A and acetylated tubulin labeled distinct spatial domains in the olfactory epithelium of both wild-type and Pde1c
/
mice (Fig. 3a). We also carried out a PDE catalytic activity assay on preparations of OSN cilia (Fig. 3b). PDE activity in OSN cilia is known to be stimulated by Ca2+ and calmodulin (CaM)16, which has been attributed to the Ca2+/CaM sensitivity of PDE1C11. We measured PDE activity in wild-type and Pde1c
/
cilia preparations under three conditions: in a standard reaction buffer containing no Ca2+ (0 mM

455

ARTICLES

a

c

Wild type

25

Terminatiom
(s)

Pde1c–/–

EOG (mV)

20 15 10

3.0 2.5

1.5

0.5

Latency (ms)

0

0 Wild type

–0.2

120 100 80 60 40 20 0

Pde1c–/– –0.4

–4

–0.5 –0.8 –1.0 0

–1.0 0

2

–3

4 Time (s)

0.5

1.0 6

8

400

–2

10 10 10 Amyl acetate (M)

**

*

**

10–2 10–4 10–3 Amyl acetate (M)

e 500

0 –0.6

Rise time (ms)

EOG (V/Vmax)

**

d 140 –7 –6 –5 –4 –3 –2 –1 Amyl acetate (log M)

** **

**

300 200 100 0

10–2 10–4 10–3 Amyl acetate (M)

Ca2+ and 0.1 mM EGTA), in the reaction buffer supplemented with 200 mM Ca2+ and 50 nM CaM, and in the reaction buffer supplemented with a broad-range PDE inhibitor 3-isobutyl-1-methylxanthin (IBMX, 0.5 mM). Wild-type preparations showed PDE activity in the absence of Ca2+ (1.5 ± 0.33 nmol mg
1 min
1), reflecting the basal PDE activity in the cilia. When the buffer was supplemented with Ca2+ and CaM, wild-type cilial PDE activity was increased B2.3–fold (3.4 ± 0.60 nmol mg
1 min
1), consistent with a previous report16. IBMX blocked all cilial PDE activity (0.17 ± 0.50 nmol mg
1 min
1). In Pde1c
/
cilia preparations, no basal PDE activity was detected in the standard reaction buffer (0.08 ± 0.21 nmol mg
1 min
1) and the addition of Ca2+/CaM did not elicit any PDE activity (0.002 ± 0.16 nmol mg
1 min
1). In both conditions, the activity levels were similar to those when IBMX was added (0.02 ± 0.23 nmol mg
1 min
1 for Pde1c
/
cilia). We concluded that knocking out PDE1C eliminated all PDE activity from OSN cilia.

stimulation protocols as for Pde1c
/
mice. No differences were observed between Pde4a
/
; Pde4a+/
and wild-type mice in any aspect of the EOG signal (Fig. 4 and Supplementary Fig. 2). These results are consistent with the idea that because PDE4A is excluded from the cilia, its effect on olfactory transduction may be minimal under most circumstances. Pde1c
/
; Pde4a
/
OSNs show prolonged response termination Previous electrophysiological experiments showed that treatment of OSNs with IBMX prolonged OSN responses to odors, indicating that degradation of cAMP by PDE activity is essential for rapid response termination9. However, neither Pde1c
/
nor Pde4a
/
mice showed prolonged termination of EOG signals, suggesting that the rate of cilial cAMP removal in these mutant mice is still adequate to allow rapid response termination. Because IBMX is a broad-range PDE inhibitor, treatment with it would simultaneously inhibit both PDE1C and PDE4A, which should mirror the phenotype of a Pde1c
/
; Pde4a
/
double knockout mouse. We generated Pde1c
/
; Pde4a
/
mice and carried out EOG recordings. These mice showed a significantly prolonged response

a

Acetyl tubulin

PDE4A C

C

K

Merge C

K

K

Wild type 5 µm

Pde4a
/


OSNs show no aberrant response properties To address whether PDE4A influences OSN responses, we measured EOG signals in Pde4a
/
mice using the same odorants and

Figure 3 Knockout of PDE1C eliminates all PDE activity from OSN cilia. (a) Co-immunostaining for PDE4A (red) and the cilia marker acetylated tubulin (green) on sections of olfactory epithelium from wild-type and Pde1c
/
mice. Shown are apical portions of the olfactory epithelium under high magnification. In both wild-type and Pde1c
/
mice, immunoreactivity for PDE4A and acetylated tubulin appeared in distinct spatial domains. C, cilial layer; K, dendritic knobs; S, supporting cell layer. Sections are counterstained with DAPI. Scale bar represents 5 mm. (b) PDE catalytic activity assay in cilia preparations from wild-type and Pde1c
/
mice. PDE activity in wild-type cilia was increased by addition of Ca2+/CaM and inhibited by addition of IBMX. Activity from Pde1c
/
cilia under all conditions was virtually undetectable, similar to wild-type cilia treated with IBMX. Error bars are 95% confidence intervals. Data for each genotype are the average of five independent preparations (mice). Each preparation was assayed in duplicate. **P o 0.01; n.s., not significant.

456

C S

Pde1c–/–

b

** 4.0 Cilial PDE activity (nmol mg–1 min–1)

© 2009 Nature America, Inc. All rights reserved.

*

1.0

0

b

**

2.0

5 0

Wild type –/– Pde1c

Figure 2 Pde1c
/
OSNs show reduced EOG amplitude, faster response termination and slower onset kinetics. (a) Dose-response relations of EOG responses to amyl acetate from wild type (n ¼ 7) and Pde1c
/
(n ¼ 5) mice. Amyl acetate was delivered as 100-ms pulses. Concentrations on the x axis are those of the liquid solution. Data points are linked with straight lines and error bars represent s.d. (b) EOG responses to a single 100-ms pulse of 10
3 M amyl acetate. Responses were normalized and averaged for comparison of response kinetics (wild type, n ¼ 15; Pde1c
/
, n ¼ 10). The inset shows the traces plotted on an expanded time axis. (c) Termination time constants (t), determined by a single exponential fit to the decay phase of the EOG signal (for 10
4 M, 10
3 M and 10
2 M, P ¼ 9.7  10
4, 0.011 and 8.4  10
4, respectively; wild type, n ¼ 20; Pde1c
/
, n ¼ 14). (d) Response latencies, defined as the time between the initiation of odor pulse and the start of the response (1% of peak amplitude; for 10
4 M, 10
3 M, and 10
2 M, P ¼ 4.5  10
5, 0.011 and 5.1  10
5, respectively; wild type, n ¼ 15; Pde1c
/
, n ¼ 10). (e) Response rise times, defined as the time from the start of the response to the peak (for 10
4 M, 10
3 M and 10
2 M, P ¼ 1.1  10
6, 9.0  10
7 and 5.8  10
8, respectively; wild type, n ¼ 15; Pde1c
/
, n ¼ 10). Error bars in c–e are 95% confidence intervals. * P o 0.05, ** P o 0.01.

Wild type Pde1c

3.5

–/–

3.0 2.5 2.0

**

1.5 n.s.

1.0 0.5 0 2+ Ca /CaM

EGTA

VOLUME 12

[

NUMBER 4

[

IBMX

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES Table 1 Analysis of EOG responses (amyl acetate stimulation) [Odorant] Response to single pulse

Maximum amplitude (mV) Latency (ms)

Pde1c
/
; Pde4a
/


21.4 ± 3.9 (7) 100.6 ± 3.4 (15) 108.8 ± 4.7 (15)

9.8 ± 2.5 (5)** 115.1 ± 4.7 (10)** 123.0 ± 10 (10)*

20.2 ± 2.1 (5) 102.6 ± 4.2 (14) 110.7 ± 5.2 (14)

8.7 ± 1.4 (8)** 116.4 ± 3.9 (17)** 125.0 ± 4.5 (17)**

10
2 M 10
4 M

87.5 ± 2.8 (15) 257 ± 11 (15)

111.2 ± 11 (10)** 341 ± 27 (10)**

87.8 ± 2.3 (14) 264 ± 10 (14)

118.9 ± 8.7 (17)** 405 ± 13 (17)**

10
3 M

287 ± 15 (15)

376 ± 23 (10)**

289 ± 15 (14)

441 ± 19 (17)**

10
2 M 10
4 M

306 ± 20 (15) 0.62 ± 0.07 (20)

445 ± 30 (10)** 0.45 ± 0.04 (14)**

307 ± 14 (14) 0.65 ± 0.09 (14)

562 ± 25 (17)** 1.02 ± 0.14 (17)**

10
3 M 10
2 M

0.97 ± 0.10 (20) 2.39 ± 0.27 (20)

0.80 ± 0.06 (14)* 1.70 ± 0.20 (14)**

0.99 ± 0.13 (14) 2.27 ± 0.30 (14)

1.79 ± 0.22 (17)** 6.78 ± 2.43 (17)**

(Paired-pulse: ratio of second pulse) to first pulse

10
4 M 10
3 M

0.48 ± 0.04 (13) 0.39 ± 0.04 (13)

0.61 ± 0.03 (9)** 0.51 ± 0.04 (9)**

0.48 ± 0.05 (8) 0.39 ± 0.04 (8)

0.44 ± 0.06 (8) 0.34 ± 0.04 (8)

(Sustained pulse: percentage decline)

10
2 M 10
4 M

0.20 ± 0.03 (13) 61.3 ± 4.6 (15)

0.29 ± 0.04 (9)** 56.8 ± 3.5 (10)

0.22 ± 0.05 (8) 69.2 ± 2.6 (6)

0.19 ± 0.03 (8) 32.8 ± 12.2 (9)**

10
3 M 10
2 M

54.7 ± 4.9 (15) 43.5 ± 4.1 (15)

53.4 ± 5.9 (10) 44.7 ± 4.9 (10)

56.4 ± 3.7 (6) 42.6 ± 6.7 (6)

26.9 ± 9.2 (9)** 13.1 ± 6.1 (9)**

Termination t (s)

at 10 s

Odorant concentrations are for amyl acetate. For maximum amplitude, values are ± s.d.; for all others, values are ± 95% confidence interval. The number of mice (n) is indicated in parentheses. * P o 0.05 in an unpaired t test compared to wild type, ** P o 0.01.

Pde1c
/
mice showed only a trend of increased baseline noise and the difference from wild type was not significant. OSN adaptation is affected differently in mutant mice Adaptation, the phenomena of decreasing responses either to repeated stimuli or during sustained stimuli, is a characteristic feature of OSN responses. Both forms of adaptation are thought to rely on different, but overlapping, sets of mechanisms19,20. Because of its Ca2+/CaM sensitivity, PDE1C has long been hypothesized to contribute to OSN adaptation11,12,20. Ca2+, which enters cilia as part of the OSN response, can bind CaM and subsequently elevate PDE1C activity. If the

c

25 Wild type

20

–/–

Pde4a

15 10 5

–5 –4 –3 –2 –1 Amyl acetate (log M)

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

0

1.5 1.0 0.5 –2 10–4 10–3 10 Amyl acetate (M)

100 Latency (ms)

–6

Wild type –/– Pde4a

2.0

d 120 –7

0

80 60 40 20

–0.2

Wild type

0

–/–

Pde4a

–0.4

e

0 –0.6

10–2 10–4 10–3 Amyl acetate (M)

350 300

Rise time (ms)

Figure 4 Pde4a
/
OSNs show no aberrant EOG response properties. Data for wild-type mice is replotted from Figure 2 for reference. (a) Dose-response relationship of EOG responses to amyl acetate from Pde4a
/
mice (n ¼ 5). Error bars represent s.d. (b) Normalized and averaged EOG responses from Pde4a
/
mice (n ¼ 14) to a single 100-ms pulse of 10
3 M amyl acetate. Inset shows the traces plotted on an expanded time axis. (c) Termination t for responses to three amyl acetate concentrations ( 10
4 M, P ¼ 0.70; 10
3 M, P ¼ 0.82; 10
2 M, P ¼ 0.57). (d) Response latencies (10
4 M, P ¼ 0.47; 10
3 M, P ¼ 0.61; 10
2 M, P ¼ 0.87). (e) Response rise times (10
4 M, P ¼ 0.39; 10
3 M, P ¼ 0.81; 10
2 M, P ¼ 0.97). Error bars in c–e represent 95% confidence intervals. n ¼ 14 for Pde4a
/
. No significant differences (P o 0.05) were found in any parameters.

2.5

0

0

b

3.0

Terminatiom
(s)

a EOG (mV)

termination for all of the odorant concentrations that we tested (Fig. 5a,b and Table 1). This result suggests that removal of cAMP from the cilia is substantially impaired when both PDEs are not functional, consistent with previous studies using IBMX to eliminate all PDE activity9. In addition, Pde1c
/
; Pde4a
/
mice showed significantly reduced EOG amplitudes and slower onset kinetics than wild types (Fig. 5c). The reduction in EOG amplitudes and the extended latency of responses in the double knockout mice were comparable to those of Pde1c
/
mice. The increase in the response rise time in double knockout mice was even greater than that of Pde1c
/
mice (Supplementary Fig. 3 online). Given the additive effect of the double knockout on response kinetics, we further examined Pde1c
/
; Pde4a+/
, Pde1c+/
; Pde4a
/
and Pde1c+/
; Pde4a+/
mice. We did not observe any significant genetic dosage effects, as Pde1c
/
; Pde4a+/
mice showed responses that were similar to those of Pde1c
/
mice and both Pde1c+/
; Pde4a
/
and Pde1c+/
; Pde4a+/
mice showed responses that were similar to that of the wild type (Supplementary Fig. 2 and Supplementary Table 2 online). Pde1c
/
; Pde4a
/
mice also showed significantly larger baseline noise of the EOG signal than the wild type (Fig. 5d,e). The baseline noise was quantified as the s.d. in the voltage during 2 s of baseline recording (2-kHz sampling rate). The increase in baseline noise probably reflects an increase in sporadic opening of CNG channels as a result of spontaneously generated cAMP, which cannot be quickly removed from the cilia of double knockout mice.

EOG (V/Vmax)

© 2009 Nature America, Inc. All rights reserved.

Pde4a
/


10
4 M 10
3 M

Rise time (ms)

Adaptation

Pde1c
/


Wild type

–0.5 –0.8 –1.0 0

–1.0 0

2

4 Time (s)

0.5

1.0 6

8

250 200 150 100 50 0

10–2 10–4 10–3 Amyl acetate (M)

457

ARTICLES Figure 5 Pde1c
/
; Pde4a
/
double knockout Wild type 0 mice show prolonged response termination and Pde1c –/– Pde1c –/–; Pde4a –/– increased baseline noise. Data for wild type and –0.2
/
Wild type Pde1c mice are replotted from Figure 2 for 1.2 ** ** ** 2.0 9.0 Pde1c –/– reference. (a) Normalized and averaged EOG –0.4 8.0 1.0 Pde1c –/– ; Pde4a –/–
/

/
1.6 responses from Pde1c ; Pde4a (n ¼ 17) 7.0 0 0.8 6.0 mice to a single 100-ms pulse of 10
3 M amyl –0.6 1.2 5.0 0.6 acetate. Inset shows the traces plotted on an * 4.0 ** –0.5 0.8 0.4 3.0 –0.8 expanded time axis. (b) Termination t for ** 2.0 0.4 0.2 responses to three amyl acetate concentrations. –1.0 1.0 –1.0 0 0 0 Responses from Pde1c
/
; Pde4a
/
mice have 0 0.5 1.0 1.5 10–4 10–3 10–2 significantly longer termination t compared with 2 4 6 8 0 Amyl acetate (M) wild type. Error bars are 95% confidence intervals Time (s) (10
4 M, P ¼ 8.3  10
6; 10
3 M, P ¼ 3.5 
8
2
4 Wild type 25 10 ; 10 M, P ¼ 5.2  10 ; all unpaired Pde1c –/– * t tests with the wild type, n ¼ 17 for Pde1c
/
; 25 Pde1c –/– ; Pde4a –/– Wild type 20 Pde4a
/
). (c) Dose-response relationship 20 for amyl acetate from Pde1c
/
; Pde4a
/
15 Pde1c –/– 15 mice (n ¼ 8). Error bars represent s.d. 10 10 (d) Representative baseline traces (filtered DC – 1 Pde4a –/– 5 kHz) from each genotype. (e) Baseline noise was 5 0 quantified as the s.d. in the voltage from 2 s of Pde1c –/– ; Wild type recording (sampling rate 2 kHz) without Pde4a –/– 0 Pde1c –/– stimulation. Pde1c
/
; Pde4a
/
mice showed Pde4a –/– 50 µV –2 –1 0 –7 –6 –5 –4 –3 Pde1c –/– ; Pde4a –/– significantly increased baseline noise compared 0.5 s Amyl acetate (log M) with wild types. Error bars represent 95% confidence intervals (compared with wild type: Pde1c
/
, P ¼ 0.144; Pde4a
/
, P ¼ 0.278; Pde1c
/
; Pde4a
/
, P ¼ 0.048; all unpaired t tests, n ¼ 12 for wild type, n ¼ 10 for Pde1c
/
, n ¼ 7 for Pde4a
/
, n ¼ 11 for Pde1c
/
; Pde4a
/
). * P o 0.05, ** P o 0.01.

b

Termination
(s)

EOG (V/Vmax)

a

d

e

stimulation continues or a subsequent stimulation occurs while PDE1C activity is elevated, then the amount of cAMP that is available to open the CNG channel would be reduced, leading to smaller responses (that is, adaptation). However, it has been previously reported that OSNs show a similar extent of adaptation to repeated stimulation when stimulated with intracellular delivery of either cAMP or nonhydrolyzable cAMP analogs, suggesting that the combined activity of all PDEs is not involved in adaptation to repeated stimulation21. The function of individual PDEs in OSN adaptation has not been directly examined.

To assess the role of PDE1C and PDE4A in OSN adaptation to repeated odor exposure, we recorded EOG signals from mice with single or combined mutations of Pde1c and Pde4a using a pairedpulse protocol, which consisted of two identical 100-ms odor pulses separated by a 1-s interpulse interval (Fig. 6a). The extent of adaptation is quantified as the ratio of the second ‘net’ peak amplitude (see Methods for details) to the first peak amplitude18. Pde1c
/
mice still showed adaptation but with significant attenuation. The ratios of the second to first pulse responses were significantly larger compared with those of the wild type for all concentrations tested

a

b

EOG (V/Vmax)

Second/first response ratio

Figure 6 OSN adaptation is differently affected in Wild type –/– Pde1c
/
and Pde1c
/
; Pde4a
/
mice. Pde1c 0.7 Pde4a–/– ** (a) Normalized and averaged EOG responses from 0 –/– 0.6 ** Pde1c –/–; wild type (n ¼ 13), Pde1c
/
(n ¼ 9), Pde4a
/
−0.2 Pde4a 0.5
/

/
(n ¼ 8) and Pde1c ; Pde4a (n ¼ 8) mice to −0.4 0.4 ** two 100-ms pulses of 10
3 M amyl acetate 0.3 −0.6 separated by a 1-s interval. The contribution of 0.2 −0.8 Wild type Pde4a–/– Pde1c–/– ; 0.1 the residual first response was removed from the −1.0 Pde1c–/– Pde4a–/– 0 recorded second response to obtain the net –4 10–3 10–2 10 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 second response peak amplitude (see Methods), Amyl acetate (M) Time (s) Time (s) Time (s) as indicated by the arrows on the Pde1c
/
and Pde1c
/
; Pde4a
/
traces. (b) Ratio of second 70 0 net peak amplitude to the first peak amplitude 60 −0.2
/
(for Pde1c compared with wild type: P ¼ 50 ** −0.4 2.6  10
4, 3.9  10
4 and 0.0011 for 10
4, 40 **
3
2
/
−0.6 30 10 and 10 M, respectively; for Pde4a ** 20 −0.8 compared with wild type: P ¼ 0.95, 0.77 and Wild type Pde4a–/– Pde1c–/– ; 10 −1.0 0.42; for Pde1c
/
; Pde4a
/
compared with wild Pde1c–/– Pde4a–/– 0 type: P ¼ 0.23, 0.071 and 0.85). (c) Normalized 10–4 10–3 10–2 5 10 15 0 5 10 15 0 5 10 15 0 Amyl acetate (M) Time (s) Time (s) Time (s) and averaged EOG responses to a 10-s pulse of 10
3 M amyl acetate from wild type (n ¼ 15), Pde1c
/
(n ¼ 10), Pde4a
/
(n ¼ 6) and Pde1c
/
; Pde4a
/
(n ¼ 9) mice. (d) Response decline during stimulation, quantified by the percent reduction in the peak amplitude at the 10-s time point (for Pde1c
/
compared with wild type: P ¼ 0.18, 0.74 and 0.73 for 10
4, 10
3 and 10
2 M, respectively; for Pde4a
/
compared with wild type: P ¼ 0.056, 0.67 and 0.80; for Pde1c
/
; Pde4a
/
compared with wild type: P ¼ 4.5  10
5, 8.3  10
6 and 1.7  10
8). In b and d, error bars are 95% confidence intervals. * P o 0.05, ** P o 0.01.

c

d

458

Percentage decline at 10 s

EOG (V/Vmax)

© 2009 Nature America, Inc. All rights reserved.

EOG (mV)

Baseline s.d. (µV)

c

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

Cilium (20-µm × 0.15-µm diameter)

Partial dendritic knob (0.86-µm diameter)

cAMP concentration (µM)

12 10 8 6 4 2 0

© 2009 Nature America, Inc. All rights reserved.

0

0.5 1.0 1.5 2.0 2.5 Time (s)

0

0.5 1.0 1.5 2.0 2.5 Time (s)

0

0.5 1.0 1.5 2.0 2.5 Time (s)

Both PDEs

Without PDE4A

Without PDE1C

No PDE

(Fig. 6b). These data suggest that PDE1C probably contributes to OSN adaptation to repeated stimulation. Paired-pulse protocols with interpulse intervals of 2, 4, 6 and 8 s were also examined (Supplementary Fig. 4). As the interpulse interval lengthened, the response amplitude to the second pulse progressively recovered. When the interpulse interval was 4 s or longer, there were no differences in the ratios of second to first pulse responses between Pde1c
/
and wild-type mice, indicating that the influence of PDE1C on adaptation is limited to a short period following stimulation. Pde4a
/
mice showed adaptation that was indistinguishable from that of wild-type mice (Fig. 6a,b). Notably, Pde1c
/
; Pde4a
/
mice also did not show a deficit in adaptation, despite the fact that Pde1c
/
mice had attenuated adaptation. This observation in double knockouts is consistent with a previous report that OSNs did not have apparent deficits in adaptation to repeated stimulation when all PDE activity was bypassed21. To assess the role of PDE1C and PDE4A in OSN adaptation during sustained odor exposure, we recorded EOG signals in response to a 10-s odor pulse (Fig. 6c). During the pulse, the EOG amplitude progressively declined in all genotypes. The extent of adaptation was quantified as the percentage of amplitude decline relative to the peak amplitude at the 10-s time point. Neither Pde4a
/
nor Pde1c
/
showed deficits in adaptation during sustained stimulation, although the longer response latency and rise time were still evident in Pde1c
/
mice (Fig. 6c,d). In contrast, however, Pde1c
/
; Pde4a
/
double knockout mice had significantly less decline during the course of stimulation (Fig. 6c,d). For 10
3 M amyl acetate (Fig. 6c), the response of Pde1c
/
; Pde4a
/
mice declined approximately 27%, whereas the responses of wild-type and single knockout mice declined by approximately 55%. This decreased response decline is consistent with the idea that cilial cAMP removal is substantially impaired when both PDE1C and PDE4A are eliminated. DISCUSSION Rapid response termination after a stimulus is critical for allowing sensory neurons to represent the temporal information of the stimulus and to be prepared for subsequent stimulation. In OSNs, rapid response termination has long been thought to rely substantially on degradation of cAMP in the cilia. However, Pde1c
/
OSNs, despite lacking all cilial PDE activity, do not show prolonged termination. This result suggests that other mechanism(s) must be able to efficiently

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

0

0.5 1.0 1.5 2.0 2.5 Time (s)

Figure 7 Computer modeling of degradation and diffusion of cilial cAMP. A cilium is attached to a ‘mini-knob’, which has approximately 1/12 of the volume of a 2-mm diameter dendritic knob (an estimation of the volume of the knob that would be available to a single cilium). The cilium is initially filled with 12 mM cAMP, a concentration that saturates the olfactory CNG channels. Four scenarios were modeled: with activities of both PDEs (modeling wild type), without PDE4A activity, without PDE1C activity, and without any PDE activity (modeling Pde1c
/
; Pde4a
/
OSNs). Diffusion of cAMP is included in all scenarios. The change in cAMP concentrations is shown at four points along the model cilium for each scenario. Note that the black lines (wild-type modeling) are obscured by the light gray lines (modeling without PDE4A). For further details see the Supplementary Note.

remove cAMP from cilia in the absence of cilial degradation. The most reasonable mechanism is that cAMP diffuses from the cilia into the dendritic knob and dendrite, where it can be degraded by PDE4A (Supplementary Fig. 5 online). Indeed, cAMP diffusion in olfactory cilia has been measured to be nearly as fast as diffusion in water (2.7  10
6 cm2 s
1)22. Computer modeling using this value together with published values of PDE enzyme activities showed that diffusion of cAMP from the cilia followed by degradation in the knob can occur in the time scale of termination of EOG responses (Fig. 7 and Supplementary Note online). cAMP degradation (or sequestration) in the dendritic knob is required for the efficient flux of cAMP away from the cilia, as the dendritic knob is not large enough to be an effective sink for cAMP (given that the knob volume is approximately equal to the total volume of all the cilia of an OSN). In the absence of both cilial and dendritic degradation (as in Pde1c
/
; Pde4a
/
OSNs), removal of cilial cAMP is severely hindered, leading to prolonged response termination. In OSNs, rapid response termination depends on the closing of both CNG channels and Ca2+-activated Cl
channels. Removal of odorevoked cAMP, in addition to feedback inhibition by Ca2+/CaM18,23, leads to the closing of CNG channels, whereas removal of cilial Ca2+ leads to the closing of the Cl
channels24. The slowest of these processes would be the rate-limiting factor for response termination. When cAMP can be degraded in the cilia (as in wild-type or Pde4a
/
OSNs), cAMP removal will be faster than when cAMP can only be removed by diffusion to and degradation in the dendrite (as in Pde1c
/
OSNs). Our modeling suggests the difference between these rates can be several-fold (Fig. 7 and Supplementary Note). It is then notable that in Pde1c
/
OSNs the termination of EOG responses is even accelerated (Fig. 2). The olfactory transduction system appears to tolerate a range in the rate of cilial cAMP removal, suggesting that it is not a rate-limiting factor in the termination of OSN responses under normal circumstances and that other termination mechanisms, namely Ca2+/CaM inhibition of CNG channels18,23 and Ca2+ extrusion24, predominantly determine the kinetics of response termination. Only when cilial cAMP removal is severely hampered, as when the activities of both PDE1C and PDE4A are eliminated, does cAMP removal become rate limiting. In wild-type OSNs, the majority of cAMP generated in the cilia, either spontaneously or from odor stimulation, is probably degraded by PDE1C. This scenario is consistent with the fact that Pde4a
/
mice show no response deficits. In this case, PDE4A in the dendritic knob functions as a fail-safe for response termination. PDE4A may also

459

© 2009 Nature America, Inc. All rights reserved.

ARTICLES prevent odor-evoked cAMP from escaping the dendrite and influencing other cellular processes. The fact that PDE4A may function to spatially restrict cAMP to the cilia is reminiscent of the role of PDE4 enzymes in maintaining cAMP signaling domains in cardiac myocytes (for a review, see ref. 15). Techniques to image cAMP in the living cells25 may be useful to further examine cAMP regulation in OSNs. Although no other PDEs have been reported in canonical OSNs (PDE2 has been found in guanylyl cyclase D–positive neurons in the olfactory epithelium14,26,27), OSNs might express other PDE(s) that may account for eventual termination of OSN responses in Pde1c
/
; Pde4a
/
mice. Our cilial PDE activity assays, and those of others16, indicate that the activity of unknown PDEs, if any, is unlikely to have a substantial effect on olfactory signal transduction. Pde1c
/
OSNs showed an unexpected reduction in sensitivity to odor, a paradoxical phenotype for the removal of an enzyme that negatively regulates signal transduction. The reduced sensitivity in Pde1c
/
OSNs comes, at least in part, from the decrease in the levels of ACIII protein (Fig. 1b), as mice that are heterozygous for Adcy3 (coding for ACIII) show reduced EOG amplitudes28. The cause of the decreased ACIII level in Pde1c
/
OSNs remains unclear. The reduced sensitivity in Pde1c
/
OSNs could also be caused by negative regulation of the activities of signal transduction components. This negative regulation of transduction components probably occurs even in the absence of odor stimulation. It is well established that the PDE inhibitor IBMX can induce responses in a large portion of OSNs, including 40–60% of OSNs in rats and mice7,29, suggesting that there is continued production and degradation of cAMP in the cilia in resting OSNs, even though they often show low spontaneous electrical activity7,30,31. PDE1C, presumably as a result of its basal catalytic activity, probably accounts for degradation of spontaneously produced cAMP. In Pde1c
/
OSNs, the lack of cAMP degradation in cilia would transiently increase cilial cAMP and, subsequently, cilial Ca2+ levels in resting OSNs. Excess Ca2+ is expected to negatively regulate the signal transduction pathway20, targeting ACIII19,32 and the CNG channel23. Regardless of the exact mechanism(s), reduced ACIII levels, negative feedback on transduction components or both, the Pde1c
/
phenotype nevertheless suggests that cilial PDE activity may indirectly aid in maintaining the signal transduction machinery in a state that is primed for odor response. It is worth noting that several previously unexplained features of odor responses from IBMX-treated OSNs9 can be interpreted after considering the PDE knockout phenotypes. When bathed in low concentrations of IBMX, OSNs did not show prolonged response termination but instead showed smaller response amplitudes and slower onset kinetics9, reminiscent of the Pde1c
/
phenotype and indicating that a basal level of PDE activity may be necessary to maintain OSN sensitivity. When bathed in high concentrations of IBMX, OSNs showed prolonged response termination with reduced response amplitude9, reminiscent of the Pde1c
/
; Pde4a
/
double knockout phenotype. We also found that PDE1C probably contributes to OSN adaptation to repeated stimulation. Until recently, adaptation to repeated stimulation was attributed mainly to the Ca2+/CaM feedback inhibition on the CNG channel33–36. However, our recent work suggested that Ca2+/ CaM-mediated desensitization of the CNG channel has little effect on adaptation to repeated stimulation but rather contributes to rapid response termination18. An increase in cilial Ca2+ as part of the OSN response, in addition to feedback on the CNG channel, probably elevates PDE1C activity by binding with CaM, potentially leading to smaller responses to subsequent stimuli. Pde1c
/
OSNs showed

460

attenuated adaptation in a paired-pulse stimulation procedure, consistent with the lack of elevated PDE activity from Ca2+/CaM feedback. It is worth noting that, as Pde1c
/
OSNs showed a reduced sensitivity to odors that is reminiscent of already adapted OSNs, the phenotype of attenuated adaptation might also result from a lower ability to further adapt. Notably, adaptation to repeated stimulation in Pde1c
/
; Pde4a
/
double knockout mice was not substantially different from that of wild type. This result is consistent with a previous report showing that when the effects of all PDEs were bypassed pharmacologically, OSNs adaptation was not affected21. The fact that eliminating PDE4A masks the adaptation deficit from the loss of PDE1C suggests that the mechanisms by which adaptation is achieved when both PDEs are eliminated (either by genetic or pharmacological means) are substantially different from the wild-type situation, even though the apparent adaptation phenotype is similar. When both PDEs are eliminated, greater recruitment of the remaining adaptation mechanisms, potentially as a result of additional Ca2+ entering the cilia during the prolonged response termination phase, may account for the apparently normal adaptation, whereas in wild-type mice, Ca2+/CaM stimulated PDE1C activity is probably a substantial contributor to adaptation. Response to a sustained stimulation can be viewed as an integration of responses to a series of short stimulations, with the final response being determined by a composite of adapted activation and termination mechanisms. Pde1c
/
OSNs show a response decline that is comparable to that of wild-type OSNs during a sustained stimulation. This result could be superficially understood as the attenuated adaptation to repeated stimulation (that is, larger subsequent responses) and the faster termination counteracting during sustained stimulation, producing an intermediate response. Therefore, the apparently normal phenotype of adaptation during sustained stimulation in Pde1c
/
OSNs is probably determined by a different composition of mechanisms than in wild-type OSNs. It is interesting to compare the phenotype of the Pde1c
/
mice to that of the CNGB1DCaM mice, which carry a mutation that renders the olfactory CNG channel resistant to Ca2+/ CaM-mediated desensitization18. In CNGB1DCaM OSNs, the response decline during sustained stimulation is attenuated, consistent with the reduced ability of these OSNs to terminate the response in combination with the normal adaptation to repeated stimulation. Pde1c
/
; Pde4a
/
OSNs do show a profound deficit in adaptation during sustained stimulation, again consistent with the response being a composite of the phenotypes of prolonged response termination and wild-type–like adaptation to repeated stimulation in these neurons. METHODS Animals. For all experiments, mice were handled and killed with methods approved by the Animal Care and Use Committees of the Johns Hopkins University. Gene targeting. For Pde1c knockout, the targeting vector was designed to delete exons 5–8 (6.7 kb of genomic DNA on chromosome 6), which encodes the N-terminal portion of the predicted catalytic domain, and to insert a stop codon in the remainder of exon 5 through homologous recombination in embryonic stem cells. For Pde4a knockout, the targeting vector was designed to delete exons 9–14 and part of exon 15 (9.9 kb of genomic DNA on chromosome 9), which encode the entire predicted catalytic domain. Further details of the generation of these knockout mice are provided in Supplementary Figure 1. Immunohistochemistry. Anesthetized mice were killed by transcardial perfusion with 1 phosphate-buffered saline (PBS) followed by 4% (wt/vol) paraformaldehyde. Olfactory tissues were dissected and postfixed for 2 h at 4 1C, followed by cryoprotection in 30% (wt/vol) sucrose overnight at 4 1C. We

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

© 2009 Nature America, Inc. All rights reserved.

cut 14-mm-thick cryosections and stored them at
80 1C. Tissue sections were incubated overnight at 4 1C with primary antibodies in 1 PBS containing 0.1% (vol/vol) Triton X-100 and 1% (vol/vol) normal donkey serum. After washing, the sections were incubated with secondary antibodies conjugated to either Alexa-488, Alexa-546 or Alexa-647 (Molecular Probes) and imaged by confocal microscopy (LSM 510 Meta, Zeiss) using either a 63 (Fig. 1a) or 100 (Fig. 3a) objective. Images were acquired using the Zeiss LSM software v. 4.2 at 1,024  1,024 pixels resolution, 12-bit depth, and cropped using Adobe Photoshop. No adjustments to contrast or brightness were made. For primary antibodies, we used antibody to PDE1C2 (1:500, kindly provided by J. Beavo, University of Washington)13,14, antibody to PDE4A (1:200, kindly provided by J. Cherry, Boston University)13, antibody to CNGB1b18,23 (1:200), antibody to ACIII (1:200, Santa Cruz SC-588), antibody to OMP (1:1,000, kindly provided by F. Margolis, University of Maryland), antibody to Gap43 (1:500, Chemicon MAB347) and antibody to acetylated tubulin (1:500, Sigma T7451). Western blotting. Olfactory epithelium was dissected and homogenized in 2 Laemmli buffer and stored at
80 1C. Tissue homogenates were subjected to SDS-PAGE, followed by transfer onto PVDF membrane. Membranes were incubated with blocking buffer (5% (wt/vol) nonfat dry milk in 20 mM Tris, 150 mM NaCl and 0.1% (vol/vol) Tween 20) for 1 h at room temperature (20– 25 1C) and then incubated overnight at 4 1C with primary antibodies at proper dilutions in blocking buffer. Membranes were then washed with TBST buffer (20 mM Tris, 150 mM NaCl and 0.1% Tween 20) followed by incubation with secondary antibodies conjugated to horseradish peroxidase in blocking buffer for 1 h at room temperature. The blot was visualized with ECL Plus reagent (GE Life Sciences) and detected on a Typhoon 9410 Variable Mode Imager. For primary antibodies, we used antibody to PDE1C (1:5,000), antibody to PDE4A (1:5,000), antibody to CNGB1b (1:2,000), antibody to ACIII (1:1,000) and antibody to alpha tubulin (1:10,000, Sigma). Preparation of OSN cilia and PDE assay. Cilia from olfactory epithelium were isolated by the calcium-shock method37 with slight modifications. Anesthetized mice were killed by transcardial perfusion with 1 PBS to minimize blood contamination. The olfactory turbinates were dissected into a buffer containing 120 mM NaCl, 5 mM KCl, 1.2 mM MgCl2, 10 mM HEPES (pH 7.4) and 0.75 mg ml
1 dithiothreitol (DTT) on ice. CaCl2 stock solution was then added to a final concentration of 10 mM Ca2+. The samples were gently rocked at 4 1C for 20 min, followed by centrifugation at 500g for 5 min to pellet de-ciliated epithelia. The supernatant was centrifuged at 18,500g for 20 min to pellet cilia. The cilia pellet was resuspended in BHB buffer (50 mM BES, 0.1 mM EGTA, pH 7.2) with 0.08 mg ml
1 DTT and 1 Complete EDTA-free Protease Inhibitor Cocktail (Roche). Protein concentration was determined by the BCA Protein Assay (Pierce). Our PDE activity assay was adapted from a previous study16. We incubated 4 mg of OSN cilial protein with 80,000 cpm 3H-cAMP (GE Life Sciences) for 20 min at 37 1C in each of three conditions: in the reaction buffer alone (50 mM BES, 5 mM MgCl2, 0.3 mg ml
1 BSA, 0.1 mM EGTA, 10 mM unlabeled cAMP), in the reaction buffer supplemented with 200 mM Ca2+ and 50 nM CaM, and in the reaction buffer supplemented with 0.5 mM IBMX (ICN Biomedicals). The reaction was stopped by heating to 100 1C for 1.5 min, followed by immediate cooling on ice. Subsequently, 1 U of 5¢ nucleotidase (Sigma) was added to the reaction (to convert the PDE product AMP to adenosine) and the reaction was further incubated at 37 1C for 30 min. The reaction was then applied to a 0.5  4-cm column containing 1 ml of QAE– Sephadex A25 (Sigma) pre-equilibrated with BHB buffer. The flowthrough from two 1.5-ml washes with BHB buffer was collected and mixed with 10 ml of Ultima Gold XR scintillation fluid (Packard Bioscience) and we assessed the radioactivity by scintillation counting. EOG. EOG recording was carried out essentially as described previously38. The mouse was killed by CO2 asphyxiation and decapitated. The head was cut sagittally to expose the medial surface of the olfactory turbinates. The recording electrode, a Ag-AgCl wire in a capillary glass pipette filled with Ringer solution (135 mM NaCl, 5 mM KCl, 1 mM CaCl2, 1.5 mM MgCl and 10 mM HEPES, pH 7.4) containing 0.5% agarose, was placed on the surface of the olfactory epithelium and connected to a differential amplifier (DP-301, Warner

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

Instruments). EOG signals were recorded from the surface of turbinate IIB and acquired and analyzed with AxoGraph software (Axon Instruments) on a Macintosh computer. The signals were low-pass filtered at 1 kHz and recorded at a sampling rate of 1 kHz. The recorded signals were further low-pass filtered at 25 Hz during analysis. Vapor-phase odorant stimuli were generated by placing 5 ml of odorant solution in a sealed 60-ml glass bottle. This vapor was delivered by a Picospritzer (Parker Hannifin) as a pulse injected into a continuous stream of humidified air flowing over the tissue sample. All EOG recordings were conducted at room temperature and in mice older than 6 weeks. For analysis of paired-pulse responses, because the response to the first odor pulse has not decayed to baseline at the time the second pulse is given, the recorded second response reflects a sum of the residual first response and the ‘net’ response to the second pulse. The net second response peak amplitude was determined by first fitting a trace to the decay phase of the first response and subsequently subtracting the value of this trace at the peak time of the second response from the second pulse peak amplitude18. Statistical analysis. All statistical significance was determined by unpaired t test. Computational modeling. Computational modeling was performed using the Virtual Cell software from the National Resource for Cell Analysis and Modeling (http://www.vcell.org) (see Supplementary Note for details). The model is available at vcell.org under the username ‘cygnar’. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank J. Beavo for antibody to PDE1C2, J. Cherry for antibody to PDE4A and F. Margolis for antibody to OMP. We also thank L. Brand, R. Cone, S. Hattar, R. Kuruvilla, T. Leinders-Zufall, R. Reed, J. Reisert and Y. Song for suggestions and comments on experiments and the manuscript, and members of the Hattar, Kuruvilla, Zhao laboratory for discussion. This work was supported by US National Institutes of Health National Institute on Deafness and other Communications Disorders grant DC007395. AUTHOR CONTRIBUTIONS K.D.C. and H.Z. designed the experiments, K.D.C. collected the data, and both authors wrote the manuscript. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/

1. Firestein, S., Shepherd, G.M. & Werblin, F.S. Time course of the membrane current underlying sensory transduction in salamander olfactory receptor neurones. J. Physiol. (Lond.) 430, 135–158 (1990). 2. Lowe, G. & Gold, G.H. The spatial distributions of odorant sensitivity and odorantinduced currents in salamander olfactory receptor cells. J. Physiol. (Lond.) 442, 147–168 (1991). 3. Firestein, S. How the olfactory system makes sense of scents. Nature 413, 211–218 (2001). 4. Ma, M. Encoding olfactory signals via multiple chemosensory systems. Crit. Rev. Biochem. Mol. Biol. 42, 463–480 (2007). 5. Kurahashi, T. & Yau, K.W. Co-existence of cationic and chloride components in odorantinduced current of vertebrate olfactory receptor cells. Nature 363, 71–74 (1993). 6. Kleene, S.J. Origin of the chloride current in olfactory transduction. Neuron 11, 123–132 (1993). 7. Lowe, G. & Gold, G.H. Nonlinear amplification by calcium-dependent chloride channels in olfactory receptor cells. Nature 366, 283–286 (1993). 8. Reisert, J., Bauer, P.J., Yau, K.W. & Frings, S. The Ca-activated Cl channel and its control in rat olfactory receptor neurons. J. Gen. Physiol. 122, 349–363 (2003). 9. Firestein, S., Darrow, B. & Shepherd, G.M. Activation of the sensory current in salamander olfactory receptor neurons depends on a G protein–mediated cAMP second messenger system. Neuron 6, 825–835 (1991). 10. Boekhoff, I. & Breer, H. Termination of second messenger signaling in olfaction. Proc. Natl. Acad. Sci. USA 89, 471–474 (1992). 11. Yan, C. et al. Molecular cloning and characterization of a calmodulin-dependent phosphodiesterase enriched in olfactory sensory neurons. Proc. Natl. Acad. Sci. USA 92, 9677–9681 (1995). 12. Yan, C., Zhao, A.Z., Bentley, J.K. & Beavo, J.A. The calmodulin-dependent phosphodiesterase gene PDE1C encodes several functionally different splice variants in a tissuespecific manner. J. Biol. Chem. 271, 25699–25706 (1996).

461

© 2009 Nature America, Inc. All rights reserved.

ARTICLES 13. Cherry, J.A. & Davis, R.L. A mouse homolog of dunce, a gene important for learning and memory in Drosophila, is preferentially expressed in olfactory receptor neurons. J. Neurobiol. 28, 102–113 (1995). 14. Juilfs, D.M. et al. A subset of olfactory neurons that selectively express cGMP-stimulated phosphodiesterase (PDE2) and guanylyl cyclase–D define a unique olfactory signal transduction pathway. Proc. Natl. Acad. Sci. USA 94, 3388–3395 (1997). 15. Conti, M. & Beavo, J. Biochemistry and physiology of cyclic nucleotide phosphodiesterases: essential components in cyclic nucleotide signaling. Annu. Rev. Biochem. 76, 481–511 (2007). 16. Borisy, F.F. et al. Calcium/calmodulin-activated phosphodiesterase expressed in olfactory receptor neurons. J. Neurosci. 12, 915–923 (1992). 17. Scott, J.W. & Scott-Johnson, P.E. The electroolfactogram: a review of its history and uses. Microsc. Res. Tech. 58, 152–160 (2002). 18. Song, Y. et al. Olfactory CNG channel desensitization by Ca2+/CaM via the B1b subunit affects response termination, but not sensitivity to recurring stimulation. Neuron 58, 374–386 (2008). 19. Leinders-Zufall, T., Ma, M. & Zufall, F. Impaired odor adaptation in olfactory receptor neurons after inhibition of Ca2+/calmodulin kinase II. J. Neurosci. 19, RC19 (1999). 20. Zufall, F. & Leinders-Zufall, T. The cellular and molecular basis of odor adaptation. Chem. Senses 25, 473–481 (2000). 21. Boccaccio, A., Lagostena, L., Hagen, V. & Menini, A. Fast adaptation in mouse olfactory sensory neurons does not require the activity of phosphodiesterase. J. Gen. Physiol. 128, 171–184 (2006). 22. Chen, C., Nakamura, T. & Koutalos, Y. Cyclic AMP diffusion coefficient in frog olfactory cilia. Biophys. J. 76, 2861–2867 (1999). 23. Kaupp, U.B. & Seifert, R. Cyclic nucleotide–gated ion channels. Physiol. Rev. 82, 769–824 (2002). 24. Reisert, J. & Matthews, H.R. Na+-dependent Ca2+ extrusion governs response recovery in frog olfactory receptor cells. J. Gen. Physiol. 112, 529–535 (1998). 25. Willoughby, D. & Cooper, D.M. Live-cell imaging of cAMP dynamics. Nat. Methods 5, 29–36 (2008).

462

26. Leinders-Zufall, T. et al. Contribution of the receptor guanylyl cyclase GC-D to chemosensory function in the olfactory epithelium. Proc. Natl. Acad. Sci. USA 104, 14507–14512 (2007). 27. Hu, J. et al. Detection of near-atmospheric concentrations of CO2 by an olfactory subsystem in the mouse. Science 317, 953–957 (2007). 28. Wong, S.T. et al. Disruption of the type III adenylyl cyclase gene leads to peripheral and behavioral anosmia in transgenic mice. Neuron 27, 487–497 (2000). 29. Ma, M., Chen, W.R. & Shepherd, G.M. Electrophysiological characterization of rat and mouse olfactory receptor neurons from an intact epithelial preparation. J. Neurosci. Methods 92, 31–40 (1999). 30. Kurahashi, T. Activation by odorants of cation-selective conductance in the olfactory receptor cell isolated from the newt. J. Physiol. (Lond.) 419, 177–192 (1989). 31. Firestein, S. & Werblin, F. Odor-induced membrane currents in vertebrate-olfactory receptor neurons. Science 244, 79–82 (1989). 32. Wei, J. et al. Phosphorylation and inhibition of olfactory adenylyl cyclase by CaM kinase II in neurons: a mechanism for attenuation of olfactory signals. Neuron 21, 495–504 (1998). 33. Chen, T.Y. & Yau, K.W. Direct modulation by Ca2+-calmodulin of cyclic nucleotide– activated channel of rat olfactory receptor neurons. Nature 368, 545–548 (1994). 34. Kurahashi, T. & Menini, A. Mechanism of odorant adaptation in the olfactory receptor cell. Nature 385, 725–729 (1997). 35. Bradley, J., Reuter, D. & Frings, S. Facilitation of calmodulin-mediated odor adaptation by cAMP-gated channel subunits. Science 294, 2176–2178 (2001). 36. Munger, S.D. et al. Central role of the CNGA4 channel subunit in Ca2+-calmodulin– dependent odor adaptation. Science 294, 2172–2175 (2001). 37. Anholt, R.R., Aebi, U. & Snyder, S.H. A partially purified preparation of isolated chemosensory cilia from the olfactory epithelium of the bullfrog, Rana catesbeiana. J. Neurosci. 6, 1962–1969 (1986). 38. Zhao, H. et al. Functional expression of a mammalian odorant receptor. Science 279, 237–242 (1998).

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

Traveling waves in developing cerebellar cortex mediated by asymmetrical Purkinje cell connectivity

© 2009 Nature America, Inc. All rights reserved.

Alanna J Watt1, Hermann Cuntz1, Masahiro Mori1,2, Zoltan Nusser3, P Jesper Sjo¨stro¨m1 & Michael Ha¨usser1 Correlated network activity is important in the development of many neural circuits. Purkinje cells are among the first neurons to populate the cerebellar cortex, where they sprout exuberant axon collaterals. We used multiple patch-clamp recordings targeted with two-photon microscopy to characterize monosynaptic connections between the Purkinje cells of juvenile mice. We found that Purkinje cell axon collaterals projected asymmetrically in the sagittal plane, directed away from the lobule apex. On the basis of our anatomical and physiological characterization of this connection, we constructed a network model that robustly generated waves of activity that traveled along chains of connected Purkinje cells. Consistent with the model, we observed traveling waves of activity in Purkinje cells in sagittal slices from young mice that require GABAA receptor–mediated transmission and intact Purkinje cell axon collaterals. These traveling waves are absent in adult mice, suggesting they have a developmental role in wiring the cerebellar cortical microcircuit.

The cerebellar cortex is one of the best-characterized circuits of the CNS and is important in the precise timing of motor control1. Purkinje cells form the sole output of the cerebellar cortex and project to the deep nuclei of the cerebellum (DCN), where they form GABAergic synapses. Purkinje cells also have local axon collaterals, which typically bifurcate in the granule cell layer and project back up to the Purkinje cell layer and occasionally into the molecular layer2. The synaptic target(s) and function of the Purkinje cell local axon collaterals have long been the subject of controversy, although some studies since the pioneering work of Cajal have suggested that they form synapses onto other Purkinje cells2–4 (but see refs. 5,6). Direct evidence for functional synaptic connections between Purkinje cells has only recently been described7. Because this local recurrent pathway is generated by the output neurons of the network, it is probably important in controlling activity patterns of the cerebellar cortex. Indeed, a recent modeling study suggested that Purkinje-Purkinje connections could both enhance temporal integration and synchronize neurons in the cerebellar cortex8. Given that Purkinje cells are among the earliest neurons to migrate into the cerebellar cortex (as early as embryonic day 15)9, they are in the right place at the right time to orchestrate the development of the synaptic connections in the cerebellar cortex. It has recently been demonstrated that the axon collaterals of juvenile Purkinje cells are particularly exuberant and are pruned to a mature distribution by the third week of postnatal development10. This suggests that PurkinjePurkinje synapses may be particularly important during early stages of development, at a time when basket and stellate cell synaptic inputs onto Purkinje cells have not yet been established11. In several CNS regions, including the visual system, the hippocampus and the spinal cord, spontaneous traveling waves of activity early in

development are critical for establishing the accurate synaptic connectivity of mature circuits12–14. However, wave-like activity has not previously been described in the developing cerebellum. To investigate the properties of monosynaptic connections between Purkinje cells and to probe their contribution to network activity in cerebellar cortex, we used two-photon laser scanning microscopy to guide targeted patchclamp recordings from connected pairs in slices from transgenic mice expressing green fluorescent protein (GFP) in Purkinje cells. We found that the asymmetrically projecting Purkinje-Purkinje synaptic connections provide a robust substrate for propagating waves of activity in the developing, but not adult, cerebellum. RESULTS Functional synapses between juvenile Purkinje cells To study the synaptic targets of Purkinje cell axon collaterals, we used transgenic mice that express the fusion protein tau-GFP under the control of the Purkinje cell–specific L7 (also known as Pcp2) promoter15. In these mice, the tau component of the chimera results in GFP enrichment in axons15. Because the L7 promoter specifically drives expression in Purkinje cells, we were able to use GFP as a marker for Purkinje cell axon collaterals. To maximize the chance of obtaining synaptically connected pairs, we made targeted patch-clamp recordings from putatively connected pairs that were visualized using a customdesigned two-photon laser-scanning microscope in juvenile slices from L7–tau (also known as Mapt)-gfp transgenic mice or, in a few cases, Gad65 (also known as Gad2)-egfp mice16,17. We visualized individual Purkinje cell axon collaterals and traced them to their putative postsynaptic targets (Fig. 1a; see Methods), which were also visualized simultaneously using laser-scanning Dodt contrast imaging. We then

1Wolfson Institute for Biomedical Research and Department of Neuroscience, Physiology and Pharmacology, University College London, London, UK. 2Department of Physiology and Biological Information, Dokkyo Medical University, Shimotsuga-gun, Tochigi, Japan. 3Laboratory of Cellular Neurophysiology, Institute of Experimental Medicine, Hungarian Academy of Sciences, Budapest, Hungary. Correspondence should be addressed to A.J.W. ([email protected]).

Received 11 August 2008; accepted 27 January 2009; published online 15 March 2009; doi:10.1038/nn.2285

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

463

ARTICLES

a

b

c

100

Control

SR

Wash

2 1 Cell 1

PSC amplitude (pA)

SR95531 0 –100 SR

–300

Control Wash

–400

Cell 2

GFP

–200

1

2 Cell 1

PSC amplitude (pA)

0

5

10 15 Time (min)

20

25

–150 –100 –50 0

Cell 2

d

e

Control

SR

Wash

f 1.0

4 Cell 4 20 Hz

PPR

© 2009 Nature America, Inc. All rights reserved.

Alexa 594

0.5

3 0.0

Cell 4 Alexa 594

0

50 Hz

50

100

Frequency (Hz)

Figure 1 Unitary synaptic connections between neighboring Purkinje cells. (a) Top, two-photon image of Purkinje cells from a P9 L7-tau-gfp mouse. Bottom, quadruple whole-cell recordings of Purkinje cells selected from the GFP image above, imaged with Alexa 594 in the internal solution. A synaptic connection was found between cells 1 and 2. Scale bar represents 20 mm. (b) Unitary synaptic connection between Purkinje cells. Top, a spike in cell 1 evoked a PSP in cell 2 (red, individual responses; black, average response; Vhold ¼
80 mV, with symmetrical chloride internal solution). Scale bars represent 25 mV (top), 2 mV (bottom) and 5 ms. Bottom, the same connection is shown, expanded to illustrate the latency. Scale bars represent 25 mV (top), 1 mV (bottom) and 2 ms. (c) Top, sample experiment showing that synaptic currents were reversibly blocked by SR95531 (SR; open circles ¼ individual responses; filled circles ¼ 2-min averages). Inset, superimposed average PSCs (times indicated by the dashed lines). Scale bars represent 50 pA and 4 ms. Bottom, summary data (gray symbols, individual experiments; black symbols, mean ± s.e.m.; n ¼ 5 for control and SR95531, n ¼ 4 for wash). (d) Recording configuration in e. Cell 3 was presynaptic to cell 4. Scale bar represents 20 mm. (e) Responses in cell 4 to 20 Hz (top) and 50 Hz (bottom) spike trains in cell 3 showed shortterm depression (red, individual responses; black, averages). Scale bars represent 0.5 ms and 20 ms above, and 0.5 ms and 10 ms below. (f) PPR varied with presynaptic firing frequency (3–13 data points per bin, n ¼ 20 connections). All error bars represent ± s.e.m.

made simultaneous triple or quadruple whole-cell recordings from candidate pre- and postsynaptic neurons. To confirm the identity of the neurons (Fig. 1a), Alexa 594 and biocytin were included in the internal solution, permitting the imaging and subsequent reconstruction of the neurons (see below). Purkinje cells were hyperpolarized with constant direct current injection (presynaptic Vm ¼ –69.1 ± 2.1 mV, postsynaptic Vm ¼ –74.6 ± 2.3 mV, n ¼ 20) to prevent spontaneous spiking and brief current pulses were injected to elicit spikes to test for connections. Synaptic connections between pairs of Purkinje cells were observed in 26% of attempts (or 23 pairs out of 88 tested, postnatal days 4–14 (P4–14)). This was a 20-fold improvement on the nontargeted connectivity rate, which was 1.3% (2 pairs out of 154 tested). The Purkinje-Purkinje cell synaptic connections (Fig. 1b) were depolarizing in our recordings because of the symmetrical chloride internal solution that we used to maximize synaptic driving force and thus the signal-to-noise ratio. The postsynaptic responses had short latencies and little temporal jitter (latency ¼ 0.60 ± 0.05 ms, n ¼ 20; Fig. 1b and Supplementary Table 1 online), consistent with monosynaptic connections. The mean postsynaptic potential (PSP) rise time was 2.2 ± 0.24 ms (n ¼ 20), with a tdecay of 20 ± 1.9 ms (n ¼ 17; Supplementary Table 1). The mean peak amplitude of connected pairs was 1.9 ± 0.63 mV (n ¼ 14, including failures), with a high degree of

464

variability across connections (Supplementary Table 1). The trial-totrial variability in PSP amplitude was also considerable, with a coefficient of variation (s.d. divided by the mean) of 0.9 ± 0.13 (n ¼ 20; Supplementary Table 1). Consistent with the large range of coefficient of variations, the failure rates were also variable (Supplementary Table 1). Synaptic currents measured in voltage clamp showed relatively rapid kinetics (rise time ¼ 0.8 ± 0.13 ms, tdecay ¼ 5.8 ± 0.99 ms, n ¼ 11; Supplementary Table 1), consistent with a perisomatic location of the synaptic contacts. Although the Purkinje-Purkinje synaptic properties were characterized over an age range during which the cerebellar circuit changes markedly (P4–14), when comparing data from mice in the P4–6 and P7–14 age groups, we found no significant differences in PSP amplitude (n ¼ 7 out of 7 P4–6/P7–14, P ¼ 0.90), latency (n ¼ 9 out of 12, P ¼ 0.89) or tdecay (n ¼ 9 out of 12, P ¼ 0.63), which allowed us to pool the data from these two age groups (Supplementary Fig. 1 online). We applied the selective GABAA receptor antagonist SR95531 to investigate the receptors underlying the synaptic connection between Purkinje cells. SR95531 completely and reversibly abolished synaptic responses (Fig. 1c), indicating that these synapses are GABAergic. We next investigated the short-term plasticity of the PurkinjePurkinje cell synapses with presynaptic spike trains and measured the paired-pulse ratio (PPR) of the responses (see Methods; Fig. 1d–f).

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

a

f PC soma

PCAT

g

h

PCAT

PC soma

© 2009 Nature America, Inc. All rights reserved.

b

c

AIS

d

i

e

PC soma

Basket AT AIS

Figure 2 Purkinje cell local axon collaterals establish synapses on other Purkinje cells. (a) Confocal laser scanning microscopic image of the cerebellar cortex of an L7-tau-gfp mouse. Scale bar represents 10 mm. (b,c) High-magnification single optical section images from the boxed area in a. Arrows indicate the site of potential synaptic interactions. Scale bar represents 2 mm. (d) A single Purkinje cell soma could be surrounded by a large number of Purkinje cell axon terminals. (Fluorescent image is superimposed on a differential interference contrast image). Scale bar represents 2 mm. (e) Double immunofluorescence labeling for GFP (green) and VIAAT (red) indicated that GFP-containing Purkinje cell axon varicosities (arrows) were VIAAT-positive (yellow). Some VIAATpositive, but GFP-negative, boutons (arrowheads) also contacted Purkinje cell somata. Scale bar represents 5 mm. (f) An electron micrograph showing a Purkinje cell axon terminal (PCAT) establishing a synaptic junction (arrow) with a Purkinje cell (PC) soma. The presence of gold particles indicates immunoreactivity for GFP. For illustrative purposes, the edges of the synaptic junction are marked with small arrows. Scale bar represents 0.2 mm. (g,h) An axon initial segment (AIS) emerging from a Purkinje cell soma received a synapse (arrow) from a GFP-positive axon terminal (PCAT; g). The boxed area in g is shown at a higher magnification in h. Scale bars represent 0.5 mm (g) and 0.2 mm (h). (i) Purkinje cell somata also received synapses (arrow) from GFPnegative boutons, which probably originate from basket cells (Basket AT). Scale bar represents 0.2 mm.

Paired-pulse depression was observed at frequencies above 10 Hz (Fig. 1f), with increasing depression at higher frequencies, which reached a plateau of B40% at B90 Hz. Paired-pulse depression was developmentally regulated; it was strong in the first postnatal week (P4– 6, PPR ¼ 0.49 ± 0.06) and was significantly reduced in the second week (P7–14, 0.97 ± 0.15, P ¼ 0.009; Supplementary Fig. 1). In summary, synaptic connections between young Purkinje cells are GABAergic and show short-term synaptic depression and high trial-to-trial and cell-tocell variability in the amplitude and reliability of the response. Ultrastructure of synapses between Purkinje cells We further studied the distribution of synaptic contacts between Purkinje cells using immunolabeling at both light microscopic and electron microscopic levels. Confocal imaging (Fig. 2a) showed that Purkinje cell axons formed presynaptic varicosities that were often directly apposed to Purkinje cell somata (Fig. 2a–d). To confirm that these varicosities were axon terminals containing vesicles, we carried out double immunolabeling for vesicular inhibitory amino acid

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

transporter (VIAAT) and GFP (Fig. 2e). These two markers showed a colocalization in many varicosities that were seemingly in direct contact with Purkinje cell somata (Fig. 2e). To investigate further, we examined these varicosities at the ultrastructural level by carrying out electron microscopic immunogold labeling of GFP. Our electron microscopy analysis confirmed the existence of symmetrical synaptic junctions made by GFP-positive axon terminals onto GFP-positive somata (Fig. 2f) and axon initial segments (AIS; Fig. 2g,h). We also found perisomatic synaptic junctions that were made by immunonegative axon terminals (Fig. 2i), which probably originated from basket cells that were beginning to innervate Purkinje cells at P8 (ref. 11). Taken together, our results indicate the existence of perisomatically located Purkinje-Purkinje cell synapses in juvenile cerebellum. Anatomical organization of the Purkinje-Purkinje pathway To help determine the effect of the Purkinje-Purkinje connection on the cerebellar cortical network, we examined the anatomical organization of

465

ARTICLES

a Ba nk Ban k

Apex

Mo laye lecula r r PC laye r Gra cell nule laye r Wh ite m atte r

GFP

c

0.06 0.04 0.02 0

Purkinje cell collaterals following the reconstruction of the axons of biocytin-filled Purkinje cells. The main axon of Purkinje cells runs in the sagittal plane, projecting away from the tip (or apex) of a cerebellar lobule along the white matter toward the DCN3. Although there was considerable heterogeneity in the projection pattern of individual Purkinje cell axon collaterals (Fig. 3a), a motif emerged when a large number of collaterals were examined. We found a peak density of collaterals (and thus of postsynaptic partners) that was centered B60 mm basally from the parent Purkinje cell, corresponding to B1–5 Purkinje cell bodies (Fig. 3b). This argues that Purkinje-Purkinje cell connections are not randomly distributed in the cerebellar cortex, but instead form directed chains of connected cells in the sagittal plane. These chains begin at the apex of the cerebellar lobules and project basally (Supplementary Movie 1 online). Because these data were obtained from cells lying at different locations in many lobules in the cerebellar vermis (Supplementary Fig. 2 online), we suggest that this asymmetry may be a general feature of Purkinje axon collaterals. The anatomical asymmetry was confirmed in our functional connectivity data. In 20 out of 23 connected pairs (87%), the postsynaptic Purkinje cell was further away from the apex of the lobule than the presynaptic cell. In addition, although we always tested for reciprocal connections, we never observed one. We were able to visually identify sites of putative synaptic contact at the light microscopic level in a subset of connected pairs (n ¼ 7) that were digitally reconstructed. In most pairs, the presynaptic axon collateral branched substantially in the upper granule cell layer and Purkinje cell layer and then appeared to contact the postsynaptic Purkinje cell with several axon collaterals (Fig. 3c). In the majority of reconstructed pairs, the presynaptic axon collateral remained in the Purkinje cell layer without extending appreciably into the molecular layer, which implies that most synaptic contacts were not on dendrites. However, the occasional Purkinje cell collateral did enter the molecular layer (Figs. 2a and 3b). On average, there were 3.7 ± 0.8 putative synaptic contacts made between each connected Purkinje cell pair (n ¼ 7 reconstructed pairs; Supplementary Table 1). Of these contacts, B90% were made onto the soma and the remaining B10% onto the AIS (Fig. 3c), consistent with our electron microscopic localization of Purkinje-Purkinje cell synapses (Fig. 2g,h) and with the rapid kinetics of PSC (Supplementary Table 1).

466

Synaptic contacts (%)

Away from apex

100 80 60 40 20 0

Ax S on D om en a dr ite

Density of collaterals

© 2009 Nature America, Inc. All rights reserved.

b

Figure 3 Anatomical distribution of Purkinje cell axon collaterals and Purkinje-Purkinje synapses. (a) Image of a lobule from a P9 mouse (left) and a high-magnification image of the region indicated by the blue dashed box (right) with two axon collaterals highlighted with blue arrows. Scale bars represent 100 mm and 50 mm (inset). (b) Density plot of Purkinje cell axon collaterals. Purkinje cell somata and axons (and some dendrites) were reconstructed and superimposed (n ¼ 39; see Methods), and oriented with their axons projecting away from the apex of the lobule (left) and toward the DCN (right), as indicated. The density of collaterals (collateral cable length (mm) per area (mm2)) is shown using a color scale, where red indicates highest density and blue lowest density. Scale bar represents 50 mm. (c) Neurolucida reconstruction of connected Purkinje cells (presynaptic cell, black; postsynaptic cell, blue) and a corresponding light microscopic single optical section (inset). The presynaptic axon (black arrows) made a putative presynaptic bouton onto the postsynaptic cell (white arrowhead). A summary of the subcellular location of putative synaptic contacts determined from biocytin-filled monosynaptically connected pairs is shown (n ¼ 7 pairs). Scale bar represents 50 mm.

Entrainment of spiking by Purkinje-Purkinje synapses Cerebellar Purkinje cells intrinsically generate regular spontaneous firing in the absence of synaptic input18,19. The effect of a synaptic input to a Purkinje cell can thus be measured by the extent to which it shifts the phase of spiking20, rather than by the resulting membrane potential deflection at the soma. To assess the influence of PurkinjePurkinje cell synaptic connections on Purkinje cell spiking, we used dynamic clamp circuitry to inject a conductance that mimicked this synaptic connection while allowing the postsynaptic cell to spike freely. Conductance kinetics and amplitude were determined on the basis of our paired recording data (Supplementary Table 1 and Supplementary Methods online). Because the GABAergic reversal potential changed with development over the age range that we have studied21 (Supplementary Fig. 1), we also benefited from the use of dynamic clamp, as it allowed us to vary the synaptic reversal potential while investigating the effect on postsynaptic spiking in the same Purkinje cell. As Purkinje-Purkinje synapses were primarily located perisomatically (Figs. 2g,h and 3c), dynamic clamp with a somatic pipette accurately simulated this synaptic input (Fig. 4a). When we injected a simulated depolarizing synaptic conductance in a freely spiking neuron (Fig. 4b), a structured firing pattern emerged in the postsynaptic cell. A similar, but opposite, effect was seen when the synaptic conductance was hyperpolarizing (Fig. 4c). As shown by cross-correlograms between spiking in the mock presynaptic neuron and the postsynaptic neuron for depolarizing (Fig. 4b) and hyperpolarizing reversal potentials (Fig. 4c), the simulated Purkinje cell connection entrained the postsynaptic cell. We quantified the degree and the phase of entrainment by fitting sine waves to the cross-correlograms of pre- and postsynaptic spiking; the degree of entrainment was reflected by the sine wave amplitude and the phase (f) by the location of the sinusoid peak. The degree of entrainment is smallest for reversal potentials around –50 mV, when inhibition is shunting22. Entrainment increased progressively in both directions from this point (Fig. 4d). The phase depended on the direction of the reversal potential deflection; depolarizing reversal potentials promoted relatively in-phase synchronization, whereas hyperpolarization favored anti-phase firing (Fig. 4e). Although the sign and magnitude of the change in average postsynaptic spiking

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

a

Dynamic clamp

b

Depolarizing

Synapse off Frequency (%)

6 4 2 0 6 4 2 0

Synapse on

–0.2 0 0.2 Time (s) Hyperpolarizing

Synapse off Frequency (%)

6 4 2 0 6 4 2 0

Synapse on

–0.2 0 0.2 Time (s)

d

e

3π/2 Phase ( , rad)

1.5 Sine amplitude

© 2009 Nature America, Inc. All rights reserved.

c

1.0 0.5

π π/2 0

0.0 –80 –60 –40 –20 0 Synaptic reversal potential (mV)

–80 –60 –40 –20 0 Synaptic reversal potential (mV)

frequency was determined by the reversal potential, the firing rate was overall only weakly affected by turning on the dynamic clamp synapse22 (change in frequency in all cases o 1 Hz; data not shown). Taken together, these results suggest that Purkinje-Purkinje connections can promote synchronization of Purkinje cells, with the phase of entrainment being dependent on the driving force at GABAergic synapses. For a given sign, however, the phase of entrainment was independent of the actual value of EGABA. Waves of activity in model Purkinje cell chains The ability of the Purkinje-Purkinje synapse to produce entrainment of Purkinje cell spiking in combination with the asymmetric nature of the connectivity suggests that this synaptic connection may be involved in coordinating waves of activity in the Purkinje cell network. To investigate this possibility, we built a model network of 50 Purkinje cells, including the known set of active conductances in Purkinje cells23. The neurons were connected in a chain-like manner on the basis of the directional asymmetry that we observed (Fig. 3a,b); each neuron was connected with the following five neurons in the chain (Fig. 5a). The model Purkinje-Purkinje synapse was calibrated on the basis of data obtained from electrophysiological experiments (Fig. 4, Supplementary Fig. 3 and Supplementary Methods online). Two synaptic reversal potentials were studied: one for depolarizing synapses (EGABA ¼
40 mV) and one for hyperpolarizing synapses (EGABA ¼
80 mV; see Supplementary Fig. 1). The model network of connected Purkinje cells showed spontaneous propagation of waves of activity (Fig. 5b,c). When the connection was

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

Figure 4 Purkinje cells synchronize in different phases depending on synaptic reversal potential. (a) Schematic illustration of recording configuration where a computer-generated dynamic clamp synaptic conductance replaces the input from a presynaptic Purkinje cell. (b) After a 4-s-long baseline period (last 0.5 s is shown, gray), a train of 40 dynamicclamp synaptic inputs (synaptic conductance, gsyn ¼ 1.5 nS; synaptic reversal potential, Erev ¼ 0 mV, see Methods) were delivered at 5 Hz (first seven indicated by arrows), which led to the emergence of correlated activity and Purkinje cell entrainment (top, 20 overlaid sweeps; bottom, spike histogram; spike rate ¼ 5.8 Hz). Sinusoid fits to the cross-correlograms (inset right, dashed lines; compare top and bottom) were used to assess the degree and phase of entrainment. Scale bars represent 10 mV (above) or 10 action potentials (APs; below), and 200 ms. (c) Switching on a similar, but hyperpolarizing, dynamic-clamp synapse (gsyn ¼ 1 nS, Erev ¼ –80 mV, see Methods) in another Purkinje cell also entrained structured spiking (top, 20 overlaid sweeps; bottom, spike histogram; rate ¼ 5.4 Hz) but with a different phase of the postsynaptic firing relative to the input (compare bottom right cross-correlogram with that in b). Scale bars represent 10 mV (above) or 10 APs (below), and 200 ms. (d) The amount of entrainment (b,c) was greatest for strongly hyperpolarizing and strongly depolarizing synaptic reversal potentials (blue, Erev o –50 mV; red, otherwise). (e) The average phase (f) of the entrainment was significantly different for hyperpolarizing and depolarizing synaptic reversal potentials (P o 0.001, n ¼ 9 for hyperpolarizing synaptic potentials and n ¼ 10 for depolarizing).

excitatory, the first cell in the network was leading, causing the waves to propagate from the tip of the lobule down into the cerebellum (Fig. 5b and Supplementary Movie 2 online). For the hyperpolarizing case, the last cell led the wave, which propagated from the interior of the cerebellum toward the apex of the lobule (Fig. 5c and Supplementary Movie 3 online). These computer simulations thus provide a proof of principle that the connection between Purkinje cells could be the substrate for propagating waves of activity along the sagittal plane in the cerebellum. To quantify the propagation of the waves, we performed a twodimensional Fourier transformation on Purkinje cell binary spike trains from the network model, resulting in an angular spectrum of the raster plot. This decomposed the network spike train into waves of a given direction and velocity, represented as one distinct point on the spatial versus temporal frequency axes. When unconnected, however, the network produced a band at the intrinsic firing frequencies of the cells without any distinct pattern on the spatial frequency axis (Fig. 5d). The occurrence of waves in the Purkinje cell network model was validated by the peaks at either side of the 0 mm–1 spatial frequency line for the depolarizing and hyperpolarizing connections (Fig. 5e,f). The velocity of the wave was obtained from this; in these examples, the waves traveled at B30 mm s–1 for the depolarizing network and B2 mm s–1 for the hyperpolarizing network. Note that in all cases, higher frequency harmonics were also visible. In addition to the angular spectrum decomposition approach described above, we calculated spike delay histograms to quantify the waves of activity that were generated by the network (Supplementary Fig. 4 online), which provided further verification of these traveling waves. Next, we assessed the robustness of the propagating waves, as they are unlikely to exist in the brain if they are not robust in the network model. The waves appeared across a wide range of parameter values (Methods and Supplementary Fig. 5 online), including halved synaptic strength, reduced number of postsynaptic partners, sparse connectivity and increased firing rates. Changing the synaptic reversal potential to shunting inhibition, however, abolished the waves22 (Supplementary Fig. 5). Although a switch from asymmetric to symmetric connectivity did not eliminate waves, it transformed them into standing waves of similar frequency (Supplementary Fig. 5). In

467

ARTICLES significant (Supplementary Methods). Significant correlation was apparent in B40% of paired extracellular recordings (17 of 41, 5 60 P r 0.05). Considering that many of the axon 40 0 collaterals are probably cut in the slice pre20 paration, this B40% occurrence of waves –5 0 probably represents a lower bound and 5 10 15 20 25 Temporal frequency (Hz) waves may be more prevalent in the intact brain. Notably, even at this young age, the b e Depolarizing 400 asymmetric projection of collaterals was pre5 50 300 sent (n ¼ 10; data not shown). Similar to what Depolarizing 40 0 200 30 has been observed in other brain regions 20 100 during early development24, cerebellar waves –5 10 0 were intermittent, waxing and waning with 1 5 10 15 20 25 –500 0 500 1,000 1,500 2,000 Temporal frequency (Hz) time (Supplementary Fig. 6 online). To test whether the waves depend on Hyperpolarizing c f GABAergic synapses (Fig. 1c), we applied the 50 120 5 40 selective GABAA receptor antagonist SR95531. Hyperpolarizing 30 80 Perfusion of SR95531 significantly reduced the 0 20 10 occurrence of correlated activity from B40% 40 –5 1 in control conditions to B5% of paired 0 –500 0 500 1,000 1,500 2,000 5 10 15 20 25 Time (ms) recordings (10 out of 25 in controls, 1 out of Temporal frequency (Hz) 15 during SR95531, P ¼ 0.012; Fig. 6b,c), Figure 5 Waves of activity in a network model of Purkinje cells. (a) Schematic illustration of the Purkinje demonstrating that GABAergic synapses were cell axon collateral network model (cells numbered starting at the apex of the folium). Each Purkinje cell critical for this form of correlated activity. was connected to the basally located nearest neighboring five cells (for illustrative purposes, only two Although Purkinje cell firing rates (6.1 ± connections are shown and cells have been colored in alternating colors). (b) Raster plot from the 0.40 Hz, n ¼ 48) were not altered by network model showing action potentials (individual dots) of Purkinje cells versus time (cells numbered SR95531 (5.43 ± 0.40 Hz, n ¼ 28, P ¼ 0.48), as in a). Activation of depolarizing synapses (gray/red border, arrow) triggered waves of activity that one explanation for the absence of correlated traveled from the apex to the base of the folium (black arrow). (c) Activation of hyperpolarizing synapses (gray/blue border, arrow) triggered waves of activity that traveled in the opposite direction (black arrow). firing in GABA blockade is that it arose Note that the connectivity of the two networks in b and c was identical; only EGABA differed. (d–f) Twoindirectly as a result of an increase in the dimensional Fourier transformation contour plots corresponding to the angular spectrum of the raster coefficient of variation of firing in individual plots. When the synapse was off (d), the two-dimensional Fourier transformation showed a flat band in Purkinje cells during SR95531 application. the temporal frequency without structure in the spatial frequency dimension. In contrast, the twoHowever, GABA blockade did not significantly dimensional Fourier transformation of the spike trains obtained with depolarizing (e) or hyperpolarizing regularize firing at these ages (coefficient of connections (f) showed peaks that corresponded to traveling waves. variation ¼ 0.26 ± 0.04 for control and 0.35 ± summary, the traveling waves are a robust phenomenon and are 0.05 for SR95531, P ¼ 0.12, n ¼ 28). In addition, although peaky crosscorrelograms were absent in SR95531, auto-correlograms for both preprobably physiologically relevant. and postsynaptic firing were indistinguishable following this treatment (Supplementary Fig. 6 and data not shown). These results demonstrate Traveling waves in juvenile sagittal cerebellar slices To test our model’s prediction that waves of activity travel across that the intrinsic firing of Purkinje cells was not altered by SR95531 at Purkinje cells arranged in the sagittal plane of lobules, we recorded these ages. We conclude that GABAA receptors are important in the from Purkinje cells in sagittal slices from young mice (P4–6). We correlated firing of neighboring Purkinje cells. Given that SR95531 monitored Purkinje cell firing patterns non-invasively with extra- blocks traveling waves at a developmental time point at which other cellular recording electrodes. As basket and stellate cell inhibition is inhibitory inputs are not yet established11, these results argue that the not yet established at this age11, we were able to specifically isolate the asymmetrically projecting connections between Purkinje cells are probeffects of Purkinje cell synapses using GABAA receptor antagonists. ably the key substrate for these waves. We hypothesized that correlated firing between recorded cells arose Because the EGABA that we found is depolarizing at these ages (Supplementary Fig. 1), our model predicted that waves of activity because of waves of activity that travel along chains of Purkinje cells would travel from the apex toward the base of a cerebellar lobule away from the apex of the lobule and toward its base. If this hypothesis was correct, the correlations between pairs of cells should depend on (Fig. 5b and Supplementary Movie 2). We used two-photon imaging to guide our electrodes to a region their spacing, as this would determine their relative position in the with intact axon collaterals and recorded activity from Purkinje cells chain of connected Purkinje cells. To examine this possibility, we lying in the same sagittal plane of the slice, typically two to three cell measured the phase difference, f, between recorded cells (Supplemenlayers deep with a separation of 50–350 mm. Simultaneous extracellular tary Methods) and plotted it as a function of the number of Purkinje recordings were made from two or three Purkinje cells (Fig. 6a,b). We cells separating the electrodes. This produced a linear relationship: the looked for correlated activity between cells using a cross-correlation phase difference was smaller for nearby cells and larger for more distant analysis of spike trains (Fig. 6b). Evidence for traveling waves of activity cells (R ¼ 0.82, P ¼ 0.0001; Fig. 6d), as predicted by the model when in Purkinje cells was determined by measuring the amplitude of a sine waves traveled away from the lobule apex. Furthermore, we found that wave fit to the cross-correlogram; a bootstrap method was then waves moved at a speed of roughly 40 Purkinje cells per cycle, or employed to determine whether the cross-correlation was statistically B3 mm s–1, which was within the predicted range of the model. This Bank

a

...50

d

Spatial frequency (mm–1)

Cell num be r 2 3.. .

Power

468

Spatial frequency (mm–1) Spatial frequency –1 (mm )

Cell number

© 2009 Nature America, Inc. All rights reserved.

Cell number

1

Apex

Synapse off

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

a Apex

Bank

P4 1 3 2

Frequency (%)

b 1

2

3

6

4

4

2

2

0

0 2

6

0.2

–0.2

3

4

4

2

2

d

0.2 3

0 0 0.2 Time (s)

–0.2

0 0.2 Time (s)

2π

100 Phase ( , rad)

Wave occurrence (%)

*

0 2

6

0

c

SR95531 1 2

0 –0.2

–0.2

50

π

0 0 Control

SR

0

10 20 30 Distance between electrodes (Purkinje cell bodies)

lends further weight to our argument that the correlated firing arose from the Purkinje-Purkinje pathway, as this speed of propagation directly depends on the anatomical spacing of connected nodes in the chain of Purkinje cells. To provide additional confirmation that the traveling waves were mediated by Purkinje axon collateral synapses, we carried out twophoton optical lesions of axon collaterals using a Ti:sapphire laser25,26 or, in a subset of cases (n ¼ 2), we lesioned with surgical scissors,

providing an orthogonal approach to pharmacological blockade of the connection. To target the axon collateral connections, we took advantage of the laminar structure of the cerebellar cortex and carried out repeated scanning of the laser in a zone perpendicular to the Purkinje cell layer across visualized axon collaterals (Fig. 7). This approach enabled us to locally sever the majority of Purkinje cell axon collaterals. The tissue around the optical lesion appeared to be healthy both during and after the procedure (examined afterwards for up to 2 h). The morphology of Purkinje cells on either side of the lesion appeared to be normal (Fig. 7a,b) and Purkinje cell firing rates were unaffected (6.61 ± 0.34 Hz, before the optical lesion, n ¼ 34; 6.03 ± 0.37 Hz after, n ¼ 36; P ¼ 0.25). The two-photon optical lesion abolished traveling waves of activity in Purkinje cells across the lesion. In some cases, we were able to maintain continuous extracellular recordings from Purkinje cells on either side of the lesion (Fig. 7a,b). In control experiments, waves of activity were observed in B40% of recordings made in the same, but unlesioned, tissue (7 out of 17 recordings). These were performed either before cutting (Fig. 7a) or ‘upstream’ of the cut, on the side closest to the lobule apex. When the Purkinje cell axon collaterals were

Before optical lesion Figure 7 Optical lesion of Purkinje axon 1 1 collaterals abolishes traveling waves. (a) Left, ** 1 two-photon image (left) and corresponding laser100 scanning Dodt contrast image (right) of L7-tau-gfp 2 Purkinje cells from a P6 mouse illustrating the 4 recording configuration, with the position of 1 2 50 extracellular electrodes 1 and 2 shown. Scale bar represents 50 mm. Top right, sample traces from 2 electrodes in a showing wave-like activity. Scale bars represent 200 pA (top trace), 500 pA 0 0 2 2 (bottom trace) and 200 ms. Bottom right, peaks –0.4 0 0.4 Before After Time (s) After optical lesion in cross-correlograms indicated that firing was lesion lesion 1 1 correlated. (b) Images showing same location as a after optical lesioning of collaterals in the granule 1 cell layer. The lesion appeared both as a bright fluorescent band (left) and a faint scar (middle). 2 Scale bar represents 50 mm. Top right, sample 4 traces from the same cells after the optical lesion, 1 2 when wave-like activity is abolished. Scale bars 2 represent 500 pA (top trace), 750 pA (bottom trace) and 200 ms. Bottom right, absence of 2 2 0 peaks in cross-correlogram after optical lesioning. –0.4 0 0.4 (c) Waves were seldom observed in Purkinje cells Time (s) following optical lesioning of axon collaterals. For control data in the same slice, waves were seen at the same rates as were seen previously, B40% (7 out of 17 recordings). However, across the lesion, waves were seen significantly less often (1 out of 18 recordings, ** P ¼ 0.0047).

c

Frequency (%)

Wave occurrence (%)

a

b

Frequency (%)

© 2009 Nature America, Inc. All rights reserved.

2

1

6

Figure 6 Traveling waves in sagittal cerebellar slices. (a) Cerebellar folium of a P4 L7-tau-gfp mouse illustrating the recording configuration in b. The positions of extracellular electrodes 1, 2 and 3 are indicated. Scale bar represents 50 mm. (b) Left, sample extracellular recording traces from electrodes in a showing waves of activity that traveled away from the lobule apex. Scale bars represent 1 nA (top trace), 100 pA (bottom two traces), and 100 ms. Right, cross-correlograms for pairs 1–2 and 2–3 indicated that firing was correlated. When SR95531 was perfused, the correlations were eliminated. (c) Summary data showing that the waves of activity that we observed in control conditions were abolished by SR95531. The perfusion of SR95531 (SR) significantly reduced the occurrence of wave-like activity between pairs of recordings from B40% to B5% (control waves, 10 out of 25 recordings; SR, 1 out of 15 recordings; * P ¼ 0.012). (d) Waves traveled away from the lobule apex, as the phase (f) of the sinusoid fit to the cross-correlograms between cells was significantly correlated with the number of somata separating recorded cells and the slope was positive (R ¼ 0.81, P ¼ 0.0001). The propagation velocity was B40 Purkinje cells per period, or B3 mm s–1. The closed circle at zero distance shows the auto-correlogram data (n ¼ 4).

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

469

ARTICLES

a

0.01

0

d

e

PCAT d

d

**

100

100 Wave occurrence (%)

*

Connection probability (%)

*** PCAT

f

**

50

50

0

0 Young

Old

100

100 Wave occurrence (%)

PC soma

PC soma

Wave occurrence (%)

Density of collaterals

0.02

c

© 2009 Nature America, Inc. All rights reserved.

b

Away from apex

50

0

0 Young

50

Old

Control

SR

Before lesion

After lesion

Figure 8 Purkinje cell-Purkinje cell connectivity and traveling waves are absent in older mice. (a) Density plot of Purkinje cell axon collaterals from older mice revealed that collaterals projected asymmetrically away from lobule apex (n ¼ 14, P17–22; see Fig. 3b and Methods). Scale bar represents 50 mm. (b) Left, confocal image from a P18 mouse. Scale bar represents 10 mm. Right, high-magnification single optical section images from the boxed area. Arrows point to gaps between Purkinje cell axon terminals and the soma. Scale bars represent 2 mm. (c) Electron micrographs showing that Purkinje cell axon terminals were close to, but did not form synapses with, Purkinje cell somata. Often, thin layers of membrane (left, asterisks) or a gap (right, asterisk) separated the structures. Purkinje cell axon terminals established a synaptic junction (arrows) with nonlabeled, non–Purkinje cell dendrites (d). Scale bars represent 0.2 mm (left) and 0.5 mm (right). (d) The connectivity rate of monosynaptic Purkinje-Purkinje cell connections was significantly reduced in older mice (P4–14, 23 of 88 tested connections; P17–25, 0 of 25 tested connections; ** P o 0.001). (e) The occurrence of waves was reduced in older mice to chance levels (P4–6, 17 out of 41 pairs of recordings; P17–25, 5 out of 81; ** P o 0.001). (f) Neither SR95531 (left; matched control, 3 out of 34 pairs of recordings; SR, 1 out of 20 recordings; P ¼ 0.31) nor optical lesioning of Purkinje cell collaterals (right; before lesion, 2 out of 48 recordings; after lesion, 3 out of 32 recordings; P ¼ 0.18) affected waves in older mice.

ablated, cross-correlations were observed in only 1 out of 18 recordings, which was a significant reduction from control conditions (P ¼ 0.0047; Fig. 7b,c). The occasional presence of correlated activity across an optical lesion was not surprising, as the full depth of the slice could not readily be cut with the laser. Given that optical lesioning abolished correlated activity between pairs of Purkinje cells spanning the lesion, but not between pairs of Purkinje cells located distal to the lesion, these data suggest that the ablated axon collaterals were the critical substrate for these traveling waves. Taken together, our data provide strong evidence in favor of the view that the Purkinje-Purkinje cell axon collateral pathway is the substrate for traveling waves of activity in chains of connected Purkinje cells in the juvenile cerebellar cortex. Traveling waves are ontogenetically transient Although many Purkinje cell axon collaterals are pruned during development, at least some Purkinje cells retain some of their local axon collaterals10. We investigated whether traveling waves were still present in older mice. Because traveling waves critically depend on the asymmetrical projection pattern of Purkinje cell axon collaterals, we first asked whether this persists in older mice. Although subtle differences in the axonal projection pattern may emerge during development, the general asymmetrical projection motif that we found in juveniles (Fig. 3b) was maintained in older mice (P17–24, n ¼ 14; Fig. 8a). This suggested that Purkinje cell axon collaterals might maintain their capacity to generate traveling waves in the mature brain. We next asked whether Purkinje cell axon collaterals made monosynaptic connections with other Purkinje cells in older mice. To address

470

this question, we first examined the target of these collaterals in P18 mice at both light microscopic and electron microscopic levels. To our surprise, confocal imaging showed that although Purkinje cell axon collaterals entered the Purkinje cell layer, they appeared very close to, but not directly apposed to, nearby Purkinje cell somata (Fig. 8b; compare with Fig. 2a–d). A small gap between the presynaptic bouton and Purkinje cell soma was discernable in most cases (Fig. 8b). Our results suggest that these collaterals no longer directly target Purkinje cells in older mice. To further address this possibility, we next examined the target of Purkinje cell axon collaterals at the ultrastructural level. To visualize Purkinje cell axon terminals, we performed immunogold reactions to GFP. Our electron microscopic analysis revealed that although Purkinje cell axon collaterals were found within a few hundred nanometer of Purkinje cell somata (Fig. 8c), they did not form synapses on them. Between Purkinje cell axons and somata we typically found thin layers of presumed glial membrane (Fig. 8c). In addition, we could not find any presynaptic specializations or postsynaptic densities at these appositions, thus suggesting the absence of functional Purkinje-Purkinje cell synapses. However, Purkinje cell axon terminals did form synapses onto nearby dendrites of unlabeled nonPurkinje cells (Fig. 8c), perhaps arising from basket cells27, which argues that the examined Purkinje cell axons were indeed fully functional and were able to form synapses. To investigate further the existence of monosynaptic PurkinjePurkinje cell connections in relatively mature mice, we carried out targeted paired recordings as described earlier. Consistent with our light microscopic and electron microscopic results, we found no

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

© 2009 Nature America, Inc. All rights reserved.

ARTICLES monosynaptically connected Purkinje cell pairs in 3-week-old mice (P17–25, 0 pairs out of 25 tested; P4–14, 23 pairs out of 88 tested; significantly different, P ¼ 0.002; Fig. 8d). Although we cannot rule out the existence of sparse Purkinje-Purkinje synapses in older mice, our results suggest that the Purkinje-Purkinje pathway is pruned following postnatal development and that Purkinje cells are not the main target of Purkinje cell local axon collaterals in adult mice. We next investigated whether traveling waves of activity exist in sagittal slices from older mice using the same methods that we employed in young mice. Although correlated activity was sometimes observed and at times appeared to be quite strong, bootstrap analysis showed that these correlations were usually not statistically significant (P 4 0.05), but were instead spurious, arising as a result of the much more regular firing and overall higher firing frequencies of mature Purkinje cells28,29. Indeed, significantly correlated activity was rarely observed in older mice and significantly less often than in young mice (young, 17 of 41 recordings; old, 5 of 81; P o 0.0001; Fig. 8e). Finally, the occasional incidence of correlated activity in mature cerebellar cortex was not significantly altered following the blockade of GABAA receptors (SR95531: 1 of 20 recordings; control: 3 of 34; P ¼ 0.31; Fig. 8f) nor by optical lesioning of Purkinje cell axons (before lesion: 2 of 46 recordings; after: 3 of 32; P ¼ 0.18; Fig. 8f), again suggesting that the rare occurrence of cross-correlations in older mice was spurious in nature and not mediated by Purkinje axon signaling. Taken together, our anatomical, electrophysiological and modeling results suggest that Purkinje-Purkinje cell monosynaptic connections are the substrate of traveling waves in juvenile mice. The fact that we did not find direct monosynaptic connections or substantial traveling waves in older mice adds weight to our conclusion that the Purkinje-Purkinje pathway underpins these waves. Furthermore, the observation that these traveling waves are ontogenetically transient suggests that they may be important in the development of the cerebellar circuit.

An asymmetric synaptic connection between Purkinje cells Our combined anatomical and physiological results indicate that the Purkinje-Purkinje synapses represent an important feature of the connectivity of the developing cerebellar cortex. Nearly one third of pairs of neighboring Purkinje cells in the sagittal plane formed functional connections, as shown by targeted recordings. Purkinje cell collaterals can extend to more than 200 mm from the parent soma (Fig. 3b, and see refs. 2,27,30 for examples), indicating that a single Purkinje cell could target as many as ten neighboring Purkinje cells, although most of our functional connections were made within 100 mm of the parent soma. The connectivity rate was developmentally regulated. It was highest in young mice and appeared to decrease during the second postnatal week7, and by the third postnatal week, functional synaptic connections were very sparse or even absent (Fig. 8). This finding may explain why functional connectivity among Purkinje cells was not observed in many previous slice experiments. We found that a defining feature of this synaptic connection was a marked asymmetry in the projection of the Purkinje cell axon collateral, which showed a strong preference to follow the direction of projection of the main axon in the sagittal plane. This was predominantly away from the apex of the folium and toward the white matter and the DCN at the center of the cerebellum. We did not observe reciprocal connections between Purkinje cells, in stark contrast with the neocortex, where circular and reciprocal connectivity motifs among neighboring pyramidal neurons are common31. This asymmetry should also hold for any functional connections made by the Purkinje cell collaterals with other potential targets in the cerebellar cortex, such as GABAergic interneurons27. It will be interesting to determine the molecular and cellular mechanisms that are responsible for establishing and maintaining this directionality, particularly as there is evidence for some pruning of axon collaterals during development10 and the collaterals appear to respect the zebrin compartmentalization of the cerebellar cortex30.

DISCUSSION We provide anatomical and physiological evidence that Purkinje cells are monosynaptically connected via GABAergic synapses in juvenile mice. These Purkinje-Purkinje synapses have either a potent excitatory or inhibitory effect on postsynaptic spiking, depending on EGABA and developmental stage. The connection shows a marked asymmetry along the sagittal axis of a cerebellar folium. A realistic model that is based on the experimentally determined parameters demonstrates that this connection can mediate traveling waves of activity in the cerebellar cortex. We tested this model, directly demonstrated for the first time, to the best of our knowledge, the presence of waves of activity in juvenile sagittal cerebellum, and used pharmacological blockers and optical lesions to show that these waves arise predominantly from Purkinje cell local axon collaterals. These waves are ontogenetically transient, as they were not observed in more mature slices, in which our anatomical and physiological results indicate that the majority of monosynaptic connections between Purkinje cells have been pruned. This is consistent with some previous reports5,6, although others have found evidence for Purkinje-Purkinje synapses in adults2,4; however, the criteria for identifying Purkinje cell axons in these studies were not as specific as the immuno–electron microscopic labeling that we used. As the proper development of mature circuits in other juvenile brain regions is known to be critically dependent on early and ontogenetically transient wave activity12–14, we suggest that waves of activity that travel along chains of connected Purkinje cells in immature cerebellum may have a similar role in cerebellar development.

Strength and synaptic dynamics of the connection We found that the synaptic connection between Purkinje cells was mediated by activation of GABAA receptors, indicating that the same transmitter was released from the Purkinje cell collaterals as is released by synapses made by the main Purkinje cell axons in the DCN32,33. Our dynamic clamp data indicate that unitary Purkinje-Purkinje synapses can have a substantial effect on the spiking of postsynaptic Purkinje cells, either by delaying or advancing spontaneous spikes, depending on the reversal potential. This is comparable to the effect of basket and stellate cell connections with Purkinje cells, where unitary inputs can also substantially inhibit Purkinje cell spiking19,34. Given that basket and stellate cell inputs do not appear until the second postnatal week11, this indicates that the Purkinje-Purkinje synapses may have a dominant effect on excitability in the first postnatal week, which is further ensured by the axosomatic location of their synaptic contacts. Notably, the basket/stellate cell connections with Purkinje cells undergo a marked 11-fold weakening during development (from P11 to P31)35. There does not seem to be a similar downregulation of cell mean synaptic conductance over the age range that we studied (see also ref. 7), although there appeared to be a marked reduction in the connectivity rate, with the majority of Purkinje-Purkinje synapses being pruned by the third postnatal week (Fig. 8). Early in development, the Purkinje-Purkinje cell synapse shows robust short-term depression over a wide range of presynaptic firing rates. Following the second postnatal week, however, the synaptic dynamics eventually become facilitating (also see ref. 7). A similar change in synaptic dynamics during development has also been shown

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

471

ARTICLES

© 2009 Nature America, Inc. All rights reserved.

at other synaptic connections36, including at other inhibitory synapses onto Purkinje cells35. Because Purkinje cells are spontaneously active18,19, even at young ages, these synapses are probably tonically depressed over the normal range of Purkinje cell firing in young mice, particularly during the first postnatal week of development, when the synaptic reversal potential was depolarizing and short-term depression was strongest (Supplementary Fig. 1). This implies that in young animals, paradoxically, a pause in Purkinje cell spiking may release the synapse from this tonically depressed state32,33, producing an enhanced synaptic conductance. Traveling waves of activity in cerebellar cortex Although traveling waves of activity are crucial early on for proper circuit development in several CNS regions12–14, this pattern of activity has, to the best of our knowledge, not yet been described in the developing cerebellum. The mechanisms controlling early waves in other brain regions are diverse and can vary throughout development24. However, depolarizing GABAergic transmission early in development appears to be important for the production of waves in several circuits37, which is consistent with our findings in the cerebellum. The cerebellar cortex has been thought to lack recurrent excitatory feedback, as the only intrinsic glutamatergic connections (made by granule cells and unipolar brush cells) are strictly feedforward. Our finding of functional connections between Purkinje cells, coupled with the demonstration that these connections can be excitatory early in development, provides a new pathway for recurrent local excitation. This pathway has some marked differences from excitatory feedback connections in cortical circuits, which have been proposed to serve many functional roles, from gain control, signal restoration, input selection, information storage and working memory, through to their role in pathological states such as epilepsy38,39. First, the excitation in the Purkinje cell recurrent network is ontogenetically transient, as it switches to inhibition later in development and is eventually pruned. Second, the projection is directional; Purkinje cells tend to project in only one direction along the sagittal axis and reciprocal connections are therefore rare. Consequently, this connection is unlikely to sustain positive feedback and reverberant activity, as in cortical circuits, because the network lacks the mutual reinforcement required for this to occur. Our simple model of the Purkinje cell network, constrained by our functional and anatomical data, generated traveling waves of activity in the Purkinje cell population, the direction and speed of which depend on the GABAergic reversal potential. We directly validated our model by demonstrating the existence of traveling waves of activity in juvenile Purkinje cells in sagittal slices. Traveling waves have previously been shown to be an emergent property in network models (for example, see ref. 40). However, our model is the first to show, to the best of our knowledge, waves of activity that rely on a known anatomical asymmetry in a single type of synaptic connection constrained directly by experimental data and thus differs from previous models showing that Purkinje cell collaterals can promote synchrony in the form of standing waves or oscillations in the Purkinje cell population8,41. The anatomical origin of the waves (together with the kinetics of the synaptic connection and the spontaneous firing of the Purkinje cells) helps to account for the different network dynamics in the Purkinje cell network compared with more conventional recurrent feedback-based excitatory and inhibitory network models, which exhibit synchrony, but no clear waves22,42. It will be interesting to determine whether anatomical asymmetries in axonal projections found in other brain regions can also support waves and the extent to which waves documented experimentally in other brain areas (for example, see ref. 43) are linked to possible asymmetries in axonal projections.

472

Functional implications of traveling waves What makes traveling waves different from other forms of oscillatory activity? Although theoretical and experimental evidence for oscillatory activity in the form of standing waves in the cerebellar cortex is abundant (for a review, see ref. 44), our data represent, to the best of our knowledge, the first evidence for traveling waves in the cerebellar cortex. Traveling waves are not incompatible with oscillatory activity in the form of standing waves. Indeed, traveling waves might appear to be standing waves if the particular spatial properties of the wave are not captured in the recording configuration. It is also possible that both forms of activity may coexist, as both high-frequency standing wave oscillations and lower-frequency traveling waves can arise from a population of spiking neurons. However, traveling waves are different from standing waves in that they propagate information across neuronal tissue; thus, the former contain directional information in addition to the temporal and spatial information that is contained in both types of waves. During development, traveling waves may therefore be important in the formation of functional maps and local subnetworks. Indeed, a recent study in cat visual cortex45 found that traveling waves were essential for a computational model to predict how synaptic plasticity rewires cortical maps after injury. In the same model, standing wave oscillatory activity, however, lead to incorrectly predicted remapping45. Recent work in the retina demonstrated the existence of two temporally and mechanistically distinct forms of waves: standing waves in early development and traveling waves later in development. Furthermore, these later traveling waves are required to ensure the proper formation of the ON and OFF subnetworks in the retina46. What is the functional role of these sagittal waves of activity in the developing Purkinje cell population? These waves could represent the sagittal counterpart of the activity patterns that are postulated to spread along the ‘beam’ of active parallel fibers47,48; the ordered propagation of the waves could provide a timing signal that is important for activating Purkinje cells associated with different components of a movement sequence. This in effect forms a functional compartment in the developing cerebellar cortex, whereby each lobule supports two traveling waves that move along opposite lobule banks (Supplementary Fig. 7 online). One possibility is that the compartmentalization of the cerebellar cortex in the sagittal domain, which may include zebrin-expressing Purkinje cell clusters30 and the extent of sagittal spread of climbing fiber collaterals49, is partly governed by sagittal waves, together with appropriate activity-dependent synaptic plasticity learning rules. Another possible and not mutually exclusive function of these early waves may be that they are involved in establishing the proper pattern of synaptic connectivity of Purkinje cell inputs to the DCN. Indeed, the effect of the waves of activity on the downstream neurons in the DCN will depend on the detailed pattern of synaptic connectivity: that is, how the cerebellar folium is mapped onto individual DCN neurons, which is currently poorly understood. If input from Purkinje cells along a folium is represented topographically in the DCN, these waves may be involved in the normal wiring of the DCN. Such a wiring strategy might permit directionally dependent dendritic computation50, which would be sensitive to different directions of propagation of the waves in cerebellar cortex. In many brain regions, early spontaneous waves precede sensory input and are thought to provide a critical substrate in the organization of neuronal circuitry, which is further refined later in development by sensory input. We suggest that traveling waves could thus represent a cerebellar analog of these self-organizing, presensory input circuit mechanisms. Further investigations of these waves are required to gain a better understanding of their functional roles during the development of the cerebellar circuit.

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES METHODS Experiments. All experiments were carried out in accordance with the animal care and handling guidelines approved by the UK Home Office. Acute cerebellar slices were prepared from P3–25 L7-tau-gfp mice15 or, in a few cases, Gad65-egfp mice16,17 using standard techniques19. All electrophysiological experiments were carried out at 33–35 1C. For confocal and electron microscopy, L7-tau-gfp mouse pups (P8 or P18) were transcardially perfused with fixative before we carried out immunohistochemical and immmunogold labeling. For further details, see the Supplementary Methods.

© 2009 Nature America, Inc. All rights reserved.

Network modeling. A network simulation consisting of 50 synaptically connected Purkinje cells was implemented in NEURON (http://www.neuron. yale.edu/) using a biophysical model of the Purkinje cell based on an existing model for spontaneously firing Purkinje cells23. The anatomical and physiological parameters of the model were tuned to replicate our experimental data. Further details are available in the Supplementary Methods. Data analysis and statistics. Data are reported as means ± s.e.m. unless otherwise indicated. Data analysis was performed using Igor Pro (Wavemetrics) and Matlab (MathWorks). Comparisons were made using either paired twotailed Student’s t tests or unpaired two-tailed Student’s t tests, assuming unequal variances. Further details are available in the Supplementary Methods. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank B. Clark, I. Duguid, F. Edwards, S. Ho, T. Ishikawa, M. London, E. Rancz, A. Roth and S. Smith for helpful discussions and for comments on the manuscript. We are grateful to S. du Lac, G. Szabo´ and F. Erde´lyi for providing transgenic mice, to J. Gruendemann for providing tissue for reconstructions, to B. Clark for help with perfusions, and to L. Ramakrishnan and K. Powell for expert assistance with histology and Neurolucida reconstructions. This work was funded by a European Molecular Biology Organization Long-Term Fellowship and a Royal Society Dorothy Hodgkin Fellowship to A.J.W., a Feodor Lynen Fellowship of the Alexander von Humboldt Foundation to H.C., a European Young Investigator Award and a Wellcome Trust project grant to Z.N., a Marie-Curie Intra-European fellowship and Medical Research Council Career Development Award to P.J.S., and a Wellcome Trust Senior Research Fellowship and a grant from the Gatsby Foundation to M.H. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/

1. Ivry, R. Cerebellar timing systems. Int. Rev. Neurobiol. 41, 555–573 (1997). 2. Chan-Palay, V. The recurrent collaterals of Purkinje cell axons: a correlated study of the rat’s cerebellar cortex with electron microscopy and the Golgi method. Z. Anat. Entwicklungsgesch. 134, 200–234 (1971). 3. Ramon y Cajal, S. Histologie du Systeme Nerveux de l’homme et des Vertebres (Maloine, Paris, 1911). 4. Larramendi, L.M. & Lemkey-Johnston, N. The distribution of recurrent Purkinje collateral synapses in the mouse cerebellar cortex: an electron microscopic study. J. Comp. Neurol. 138, 451–459 (1970). 5. Hamori, J. & Szentagothai, J. Identification of synapses formed in the cerebellar cortex by Purkinje axon collaterals: an electron microscope study. Exp. Brain Res. 5, 118–128 (1968). 6. De Camilli, P., Miller, P.E., Levitt, P., Walter, U. & Greengard, P. Anatomy of cerebellar Purkinje cells in the rat determined by a specific immunohistochemical marker. Neuroscience 11, 761–817 (1984). 7. Orduz, D. & Llano, I. Recurrent axon collaterals underlie facilitating synapses between cerebellar Purkinje cells. Proc. Natl. Acad. Sci. USA 104, 17831–17836 (2007). 8. Maex, R. & De Schutter, E. Oscillations in the cerebellar cortex: a prediction of their frequency bands. Prog. Brain Res. 148, 181–188 (2005). 9. Sotelo, C. Cellular and genetic regulation of the development of the cerebellar system. Prog. Neurobiol. 72, 295–339 (2004). 10. Gianola, S., Savio, T., Schwab, M.E. & Rossi, F. Cell-autonomous mechanisms and myelin-associated factors contribute to the development of Purkinje axon intracortical plexus in the rat cerebellum. J. Neurosci. 23, 4613–4624 (2003). 11. Altman, J. Postnatal development of the cerebellar cortex in the rat. II. Phases in the maturation of Purkinje cells and of the molecular layer. J. Comp. Neurol. 145, 399–463 (1972). 12. Feller, M.B. Spontaneous correlated activity in developing neural circuits. Neuron 22, 653–656 (1999). 13. Ben-Ari, Y. Developing networks play a similar melody. Trends Neurosci. 24, 353–360 (2001).

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

14. Katz, L.C. & Shatz, C.J. Synaptic activity and the construction of cortical circuits. Science 274, 1133–1138 (1996). 15. Sekirnjak, C., Vissel, B., Bollinger, J., Faulstich, M. & du Lac, S. Purkinje cell synapses target physiologically unique brainstem neurons. J. Neurosci. 23, 6392–6398 (2003). 16. Bali, B., Erde´lyi, F., Szabo´, G. & Kovacs, K.J. Visualization of stress-responsive inhibitory circuits in the GAD65-eGFP transgenic mice. Neurosci. Lett. 380, 60–65 (2005). 17. Erde´lyi, F. et al. GAD65-GFP transgenic mice expressing GFP in the GABAergic nervous system. FENS Abstr. 1, A011.3 (2003). 18. Raman, I.M. & Bean, B.P. Resurgent sodium current and action potential formation in dissociated cerebellar Purkinje neurons. J. Neurosci. 17, 4517–4526 (1997). 19. Ha¨usser, M. & Clark, B.A. Tonic synaptic inhibition modulates neuronal output pattern and spatiotemporal synaptic integration. Neuron 19, 665–678 (1997). 20. Mittmann, W. & Ha¨usser, M. Linking synaptic plasticity and spike output at excitatory and inhibitory synapses onto cerebellar Purkinje cells. J. Neurosci. 27, 5559–5570 (2007). 21. Eilers, J., Plant, T.D., Marandi, N. & Konnerth, A. GABA-mediated Ca2+ signaling in developing rat cerebellar Purkinje neurones. J. Physiol. (Lond.) 536, 429–437 (2001). 22. Vida, I., Bartos, M. & Jonas, P. Shunting inhibition improves robustness of gamma oscillations in hippocampal interneuron networks by homogenizing firing rates. Neuron 49, 107–117 (2006). 23. Khaliq, Z.M., Gouwens, N.W. & Raman, I.M. The contribution of resurgent sodium current to high-frequency firing in Purkinje neurons: an experimental and modeling study. J. Neurosci. 23, 4899–4912 (2003). 24. Firth, S.I., Wang, C.T. & Feller, M.B. Retinal waves: mechanisms and function in visual system development. Cell Calcium 37, 425–432 (2005). 25. Yanik, M.F. et al. Neurosurgery: functional regeneration after laser axotomy. Nature 432, 822 (2004). 26. Mejia-Gervacio, S. et al. Axonal speeding: shaping synaptic potentials in small neurons by the axonal membrane compartment. Neuron 53, 843–855 (2007). 27. O’Donoghue, D.L., King, J.S. & Bishop, G.A. Physiological and anatomical studies of the interactions between Purkinje cells and basket cells in the cat’s cerebellar cortex: evidence for a unitary relationship. J. Neurosci. 9, 2141–2150 (1989). 28. Brody, C.D. Correlations without synchrony. Neural Comput. 11, 1537–1551 (1999). 29. de la Rocha, J., Doiron, B., Shea-Brown, E., Josic, K. & Reyes, A. Correlation between neural spike trains increases with firing rate. Nature 448, 802–806 (2007). 30. Hawkes, R. & Leclerc, N. Purkinje cell axon collateral distributions reflect the chemical compartmentation of the rat cerebellar cortex. Brain Res. 476, 279–290 (1989). 31. Song, S., Sjo¨strom, P.J., Reigl, M., Nelson, S. & Chklovskii, D.B. Highly nonrandom features of synaptic connectivity in local cortical circuits. PLoS Biol. 3, e68 (2005). 32. Pedroarena, C.M. & Schwarz, C. Efficacy and short-term plasticity at GABAergic synapses between Purkinje and cerebellar nuclei neurons. J. Neurophysiol. 89, 704–715 (2003). 33. Telgkamp, P. & Raman, I.M. Depression of inhibitory synaptic transmission between Purkinje cells and neurons of the cerebellar nuclei. J. Neurosci. 22, 8447–8457 (2002). 34. Midtgaard, J. Stellate cell inhibition of Purkinje cells in the turtle cerebellum in vitro. J. Physiol. (Lond.) 457, 355–367 (1992). 35. Pouzat, C. & Hestrin, S. Developmental regulation of basket/stellate cell-Purkinje cell synapses in the cerebellum. J. Neurosci. 17, 9104–9112 (1997). 36. Reyes, A. & Sakmann, B. Developmental switch in the short-term modification of unitary EPSPs evoked in layer 2/3 and layer 5 pyramidal neurons of rat neocortex. J. Neurosci. 19, 3827–3835 (1999). 37. Ben-Ari, Y. Excitatory actions of gaba during development: the nature of the nurture. Nat. Rev. Neurosci. 3, 728–739 (2002). 38. Douglas, R.J. & Martin, K.A. Recurrent neuronal circuits in the neocortex. Curr. Biol. 17, R496–R500 (2007). 39. Connors, B.W. & Telfeian, A.E. Dynamic properties of cells, synapses, circuits and seizures in neocortex. Adv. Neurol. 84, 141–152 (2000). 40. Cohen, A.H. et al. Modelling of intersegmental coordination in the lamprey central pattern generator for locomotion. Trends Neurosci. 15, 434–438 (1992). 41. de Solages, C. et al. High-frequency organization and synchrony of activity in the purkinje cell layer of the cerebellum. Neuron 58, 775–788 (2008). 42. Geisler, C., Brunel, N. & Wang, X.J. Contributions of intrinsic membrane dynamics to fast network oscillations with irregular neuronal discharges. J. Neurophysiol. 94, 4344–4361 (2005). 43. Lee, S.H., Blake, R. & Heeger, D.J. Traveling waves of activity in primary visual cortex during binocular rivalry. Nat. Neurosci. 8, 22–23 (2005). 44. De Zeeuw, C.I., Hoebeek, F.E. & Schonewille, M. Causes and consequences of oscillations in the cerebellar cortex. Neuron 58, 655–658 (2008). 45. Young, J.M. et al. Cortical reorganization consistent with spike timing, but not correlation, dependent plasticity. Nat. Neurosci. 10, 887–895 (2007). 46. Kerschensteiner, D. & Wong, R.O. A precisely timed asynchronous pattern of ON and OFF retinal ganglion cell activity during propagation of retinal waves. Neuron 58, 851–858 (2008). 47. Braitenberg, V. Functional interpretation of cerebellar histology. Nature 190, 539–540 (1961). 48. Eccles, J.C., Szentagothai, J. & Ito, M. The Cerebellum as a Neuronal Machine (Springer-Verlag, Heidelberg, 1967). 49. Oberdick, J., Baader, S.L. & Schilling, K. From zebra stripes to postal zones: deciphering patterns of gene expression in the cerebellum. Trends Neurosci. 21, 383–390 (1998). 50. Rall, W. Theoretical significance of dendritic trees for neuronal input-output relations. in Neural Theory and Modeling (ed. Reiss, R.F. 73–97 (Stanford University Press, Stanford, California, USA, 1964).

473

ARTICLES

Transformation of odor representations in target areas of the olfactory bulb

© 2009 Nature America, Inc. All rights reserved.

Emre Yaksi1,3,4, Francisca von Saint Paul1,2,4, Jo¨rn Niessing1,2, Sebastian T Bundschuh1,2 & Rainer W Friedrich1,2 The brain generates coherent perceptions of objects from elementary sensory inputs. To examine how higher-order representations of smells arise from the activation of discrete combinations of glomeruli, we analyzed transformations of activity patterns between the zebrafish olfactory bulb and two of its telencephalic targets, Vv and Dp. Vv is subpallial whereas Dp is the homolog of olfactory cortex. Both areas lack an obvious topographic organization but perform complementary computations. Responses to different odors and their mixtures indicate that Vv neurons pool convergent inputs, resulting in broadened tuning curves and overlapping odor representations. Neuronal circuits in Dp, in contrast, produce a mixture of excitatory and inhibitory synaptic inputs to each neuron that controls action potential firing in an odor-dependent manner. This mechanism can extract information about combinations of molecular features from ensembles of active and inactive mitral cells, suggesting that pattern processing in Dp establishes representations of odor objects.

The brain extracts object information from a continuous stream of sensory inputs by successive transformations of stimulus representations. In the visual system, for example, activity patterns representing basic stimulus features such as position and orientation are transformed multiple times to create representations of shapes, objects and faces in higher brain regions1. In olfaction, molecular features of an odor are initially detected by a large number of odorant receptors2 and represented by the activation of discrete combinations of glomeruli in the olfactory bulb3–11. Glomerular activation patterns are then reorganized within the olfactory bulb and conveyed by mitral cells to multiple pallial and subpallial target areas12–18. It remains unclear, however, how odor representations are transformed downstream of the olfactory bulb to create higher-order representations of olfactory objects. Most, if not all, sensory systems contain topologically organized functional maps of stimulus features that are thought to organize transformations of activity patterns between brain areas by imposing spatial constraints on connectivity19. In the olfactory bulb, glomeruli responding to chemically related compounds are often concentrated within defined regions and form a fractured ‘chemotopic’ map of molecular features3–11. Primary chemical features (for example, functional groups) are mapped onto relatively large regions, whereas secondary features (for example, chain length) are mapped within these regions in a more overlapping fashion4–6,9–11. In the largest target area of the olfactory bulb, the piriform cortex, low-resolution maps indicate that odor-evoked activity is widespread20–23. However, activity may not be randomly distributed24, and a topographic organization has been reported in other telencephalic target areas25–27. Detailed

measurements of activity patterns are therefore required to examine whether pattern processing in higher olfactory brain areas is constrained by topological principles. The processing of activity patterns also depends on the integration of synaptic inputs by higher-order neurons. If activity in a small fraction of convergent inputs is sufficient to drive action potentials in a downstream neuron, higher-order neurons may pool inputs of multiple mitral cells, resulting in broadened tuning curves. Alternatively, more elaborated integration strategies may enable higher-order neurons to detect specific patterns of activity across their inputs28,29. Recent studies using odor mixtures suggest that neurons in olfactory cortex integrate mitral cell inputs in a complex fashion23,30,31, but the principles of pattern processing in higher olfactory brain areas remain poorly understood. We used the zebrafish system to analyze the transformation of activity patterns between the olfactory bulb and two telencephalic target areas. In teleosts, the major forebrain areas are homologous to those in other vertebrate classes but, with the exception of the olfactory bulb, undergo a morphogenetic process (eversion) that rearranges their relative positions during development32. As a consequence, the olfactory bulb and its major projection areas are optically accessible from the ventral side. Using high-resolution Ca2+ imaging and electrophysiological recordings, we analyzed odor-evoked activity of individual neurons and populations in the olfactory bulb, in a subpallial target area (Vv) and in a pallial area homologous to olfactory cortex (Dp). Our results indicate that Vv pools mitral cell inputs and forms overlapping odor representations,

1Max Planck Institute for Medical Research, Heidelberg, Germany. 2Friedrich Miescher Institute for Biomedical Research, Basel, Switzerland. 3Present address: Harvard Medical School, Department of Neurobiology, Boston, Massachusetts 02115, USA. 4These authors contributed equally to this work. Correspondence should be addressed to R.F. ([email protected]).

Received 17 December 2008; accepted 4 February 2009; published online 22 March 2009; doi:10.1038/nn.2288

474

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES Figure 1 Odor responses in target areas of the Telencephalon OB Vv olfactory bulb. (a) Confocal reflection image of the 20 µm Lys (repeat) Lys Val AAs zebrafish forebrain (adult, ventral view). BAs Nucs Myelinated axons appear bright. Dashed lines LOT Dp 100 show approximate locations of glomeruli responding to amino acids (AAs, red), bile acids MOT Lys (BAs, yellow) and nucleotides (Nucs, blue). AC, –25 Vv 20 % Dp anterior commissure; Dp, posterior zone of dorsal AC 2s Lys Lys (repeat) Val telencephalic area; LOT, lateral olfactory tract; MOT, medial olfactory tract; OB, olfactory bulb; 100 µm Vv, ventral nucleus of ventral telencephalic area. (b) Ca2+ signals in Vv and Dp evoked by odor 20 µm Lys stimulation (single trials). Dots in the left image indicate positions of somata. Response maps are reproducible (lysine (Lys) versus Lys repeat) and stimulus dependent (Lys versus valine (Val)). Overlap between response patterns evoked by different stimuli is high in Vv but low in Dp. Traces show the time course of Ca2+ signals in the somata, depicted by arrows.

b

∆F/F (%)

a

© 2009 Nature America, Inc. All rights reserved.

whereas Dp extracts information about combinations of molecular features by interactions between multiple synaptic pathways. RESULTS Imaging response patterns in the olfactory bulb and its targets We measured responses to odors with different primary and secondary molecular features in adult zebrafish using a brain preparation that permits optical access to the olfactory bulb and telencephalon (Fig. 1a). To identify odor-responsive telencephalic areas, we first mapped odor responses using widefield Ca2+ imaging (Supplementary Fig. 1 online). The strongest responses were observed in two direct targets of the olfactory bulb, the ventral nucleus of the ventral telencephalon (Vv) and the posterior zone of the dorsal telencephalon (Dp)12–16. Vv is a subpallial area in the medial telencephalon that may correspond to parts of the septal formation in tetrapods, but the exact homology relationships remain to be defined. Dp is a pallial area in the posteriorlateral telencephalon that is homologous to olfactory cortex32. Bile acids and amino acids, which activate segregated glomeruli in the medial and lateral olfactory bulb, respectively5,6, evoked overlapping responses in both telencephalic areas (Supplementary Fig. 1). The functional segregation of medial and lateral glomeruli in the olfactory bulb is therefore not mapped onto the medial (Vv) and lateral (Dp) target areas in the telencephalon. We analyzed odor-evoked activity patterns in the olfactory bulb, Vv and Dp at single-neuron resolution using two-photon microscopy33 after bolus loading of a synthetic Ca2+ indicator34,35 (rhod-2-AM, n ¼ 45 fish, or Oregon Green 488 BAPTA-1-AM, n ¼ 2; Fig. 1b). Within the olfactory bulb, responses of mitral cells and interneurons were distinguished using the fluorescent mitral cell marker, HuC-YC35–37. The stimulus panel comprised nine natural odors with different primary and secondary molecular features (Supplementary Fig. 2 online): a mixture of nine amino acids, a mixture of three bile acids, a mixture of two nucleotides, three individual amino acids (tyrosine, valine and lysine), two individual bile acids (TCA and TCDCA) and one individual nucleotide (ATP). These stimuli activate glomeruli in different areas of the olfactory bulb (Fig. 1a) and are therefore diagnostic for the topographic organization of glomerular response patterns5,6. Responses at different depths were measured during repeated applications of the same stimuli. We measured somatic Ca2+ signals from a total of 1,213 mitral cells (404 ± 96 mitral cells per fish, mean ± s.d., n ¼ 3 olfactory bulbs), 4,899 interneurons (1,633 ± 600 interneurons per fish, mean ± s.d., n ¼ 3 olfactory bulbs, the same olfactory bulbs as used for mitral cell imaging), 658 neurons in Vv (110 ± 41 neurons per fish, mean ± s.d., n ¼ 6 fish) and 2,002 neurons in Dp (667 ± 252 neurons per fish, mean ± s.d., n ¼ 3 fish). We first analyzed

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

the topological organization of activity patterns and then the response properties of individual neurons. Mitral cells responding to amino acids were located predominantly in the ventrolateral olfactory bulb (Fig. 2). We sometimes observed a small group of amino acid–responsive mitral cells in the medial olfactory bulb. Bile acids activated mitral cells mainly in the medial olfactory bulb, along with some mitral cells in the posterior-lateral olfactory bulb. Mitral cell responses to amino acids and bile acids were therefore clearly segregated. Nucleotides evoked fewer responses concentrated in the small posterior-lateral and medial regions that also responded to amino acids and bile acids. These response patterns are consistent with the distributions of activated glomeruli6. Interneurons responsive to amino acids and bile acids were more intermingled, particularly in the center of the olfactory bulb, although a slight bias toward the lateral and medial olfactory bulb, respectively, was observed (Fig. 2b,c). More interneurons than mitral cells responded to both stimulus classes (Fig. 2b). Interneurons responsive to nucleotides were fewer in number and often also responded to the other stimulus classes. In Vv, all odors evoked sparse, scattered activity without any obvious spatial organization (Figs. 2c and 3a). Many neurons responded to odors with different molecular features, and activity patterns evoked by different stimuli overlapped extensively. In Dp, neurons responding to amino acids, bile acids or nucleotides were scattered and intermingled, but few neurons responded to multiple stimulus classes (Figs. 2c and 3b). In some fish, responses to amino acids and bile acids were biased slightly toward the posterior and anterior part of Dp, respectively. Nevertheless, responses to both stimulus classes were distributed throughout the entire brain area. Whereas Ca2+ signals of Vv neurons were sustained, responses of Dp neurons were more transient (Fig. 1b). The reconstruction of the underlying firing rate changes by temporal deconvolution35 indicates that odor-evoked firing patterns of mitral cells and Vv neurons can last several seconds, whereas responses of Dp neurons are transient (Supplementary Fig. 3 online; see below). Although activity patterns across neuronal populations were dynamic, the gross topographic relationships between different activity patterns changed little during the time course of an odor response in any of the brain areas (Supplementary Fig. 3b). Quantitative analysis of chemotopy To quantify the topological organization of activity patterns, we first measured the mean focality by an index that becomes zero when responsive neurons are randomly distributed and approaches one when responsive neurons are tightly clustered38. The average focality of mitral

475

ARTICLES

a

AA mix

BA mix

b

Nuc mix

AA mix

MC V

IN V

D

D

Nuc mix

L

L

M

M P

A

AA mix versus BA mix

A

P BA mix versus Nuc mix

r = 0.32 Lys versus TCA

© 2009 Nature America, Inc. All rights reserved.

BA mix

TCA versus ATP

r = –0.12

AA mix versus BA mix

AA mix versus Nuc mix

r = 0.43

r = 0.38

r = –0.01 Lys versus TCA

Lys versus ATP

r = 0.32

r = –0.14

Odor 1 / odor 2 response ratio > 3:1 < 1:3

Nuc mix versus BA mix

r = 0.45 ATP versus TCA

r = 0.20

c MC

AA mix versus Nuc mix

r = 0.24 Lys versus ATP

r = 0.51

IN

Vv

r = 0.33 Dp

V D

Figure 2 Three-dimensional activity patterns across mitral cells and L interneurons (INs) in the olfactory bulb. (a) Above, patterns of Ca2+ signals A P across 379 mitral cells (MC) evoked by odors representing different primary M AA mix versus BA mix molecular features, measured by two-photon microscopy. Each plot symbol shows the position of one mitral cell. Gray dots depict neurons with Ca2+ signals o10% of the maximum in each plot. The size of spheres represents the magnitude of the Ca2+ signal between 10% and 90% of the maximum. Maxima were determined by averaging the largest 10% of responses. The distance between ticks is 50 mm on all axes. Below, overlay of response patterns evoked by mixtures or individual compounds representing different primary molecular features. Sphere size depicts the magnitude of the Ca2+ signal evoked by the more potent stimulus, as above. Color encodes the ratio of Ca2+ signals evoked by the two stimuli such that mitral cells responding with similar magnitude to both stimuli appear dark. AA, amino acid; BA, bile acid; Nuc, nucleotide; r, Pearson correlation coefficient. (b) Patterns of Ca2+ signals across 1,430 INs. Stimuli and conventions as in a. (c) Thresholded surface contours of volume activity maps generated from data in Figures 2 and 3 (AA mixture versus BA mixture), showing spatial relationships between larger regions. Patterns of somatic Ca2+ signals were binned into voxels (6.4 mm), convolved with a three-dimensional Gaussian kernel (s, 62 mm; cutoff, 95 mm) and thresholded at 4% DF/F. Volume activity evoked by amino acids and bile acids is shown in blue and yellow, respectively; overlapping volume activity appears gray.

cell activity patterns was significantly higher (0.31 ± 0.16, mean ± s.d.) than the focality in the other populations (interneurons: 0.11 ± 0.08, Vv: 0.13 ± 0.13, Dp: 0.10 ± 0.10, P o 0.001 in all cases, Wilcoxon ranksum test; Fig. 4a), confirming that odor-evoked activity patterns become more widespread downstream of mitral cells. Activity patterns are considered chemotopic when responses to defined molecular features are spatially concentrated (focal) and show little overlap with responses to other chemical features (low correlation of activity patterns). We quantified chemotopy using an index that is high when these conditions are fulfilled (Supplementary Methods online). The chemotopy index of mitral cell activity patterns (101.9 ± 23.3, mean ± s.d.) was clearly and significantly higher than in all other neuronal populations (interneurons: 11.1 ± 0.22, Vv: 7.9 ± 0.42, Dp: 24.4 ± 0.4, P o 0.001 in all cases, Wilcoxon rank-sum test; Fig. 4b). In Dp, the chemotopy index (24.4 ± 0.4) was slightly higher than for interneurons or Vv neurons (P o 0.001), which probably reflects the slight bias of responses to amino acids and bile acids toward the posterior and anterior part, respectively. We obtained similar results using other measures of chemotopy (Supplementary Methods). The topographic organization of mitral cell activity patterns is therefore not maintained in Vv and Dp. Response properties of higher-order neurons Individual neurons in different brain areas showed obvious differences in their responses to the nine different stimuli (Fig. 4c and

476

Supplementary Fig. 4 online). To quantitatively compare odorresponse properties, we first determined the fractions of neurons that responded to one, two or all three of the mixtures representing different primary molecular features (amino acid mixture, bile acid mixture or nucleotide mixture). The criterion for a response was that the odor-evoked Ca2+ signal exceeded a detection threshold defined as two times the s.d. of the Ca2+ signal in trials without odor stimulation. Most mitral cells (53%) responded to only one stimulus class, and responses to increasing numbers of stimulus classes became progressively less frequent (two classes: 30%, three classes: 17%; Fig. 4d). Most interneurons and Vv neurons, in contrast, responded to multiple stimulus classes (interneurons: one class: 26%, two classes: 40%, three classes: 34%, Vv: 36%, 25% and 39%, respectively), whereas the feature selectivity of Dp neurons was similar to that of mitral cells (48%, 31% and 21%, respectively). We further quantified the selectivity of responses to the nine odors by the lifetime sparseness39, which can vary between zero (equal responses to all stimuli) and one (selective response to only one stimulus). The mean lifetime sparseness of mitral cell responses was 0.53 ± 0.35 (mean ± s.d.) and significantly higher than the lifetime sparseness of interneuron responses (0.42 ± 0.27, P o 0.001, Wilcoxon rank-sum test; Fig. 4e). In Vv, lifetime sparseness was significantly lower than in all other brain regions (0.30 ± 0.29, P o 0.001), whereas lifetime sparseness in Dp (0.53 ± 0.31) was not significantly different from that of

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

a

BA mix

AA mix

Nuc mix

Vv V D L M P

A

AA mix versus BA mix

Nuc mix versus BA mix

r = 0.78 Lys versus TCA

r = 0.81 ATP versus TCA

© 2009 Nature America, Inc. All rights reserved.

r = 0.71

b

AA mix

AA mix versus Nuc mix

r = 0.70 Lys versus ATP

r = 0.80 r = 0.77 Odor 1 / odor 2 response ratio < 1:3 > 3:1 BA mix

Nuc mix

Nuc mix versus BA mix

AA mix versus Nuc mix

Dp V D L M

P

A

AA mix versus BA mix

r = –0.06 Lys versus TCA

r = 0.40 ATP versus TCA

r = –0.04

r = –0.07 Lys versus ATP

r = 0.33

r = 0.17

Figure 3 Three-dimensional activity patterns in Vv and Dp. (a) Ca2+ signals across 148 neurons in Vv, evoked by odors representing different primary molecular features. Conventions as in Figure 2. (b) Patterns of Ca2+ signals across 871 neurons in Dp, evoked by the same odors. AA, amino acid; BA, bile acid; Nuc, nucleotide; r, Pearson correlation coefficient.

mitral cells. Hence, Vv neurons, but not Dp neurons, are more broadly tuned than neurons in the olfactory bulb. The overlap of odor-evoked population activity was quantified by the Pearson correlation coefficient (Fig. 4f,g). Correlations between mitral cell activity patterns were low (0.28 ± 0.19, mean ± s.d.), consistent with the dissimilar molecular properties of the stimuli. Correlations between interneuron activity patterns were significantly higher (0.41 ± 0.14, P o 0.01, Wilcoxon rank-sum test), and activity patterns in Vv were more highly correlated than in all other brain areas (0.65 ± 0.11, P o 0.001), both within and across stimulus classes (Fig. 4f,g). The correlation of activity patterns in Dp, in contrast, was low (0.29 ± 0.22) and not significantly different from that of mitral cell activity patterns. Hence, activity patterns in Vv overlap substantially, whereas those in Dp are distinct. Responses of higher-order neurons to binary odor mixtures We next measured responses to individual compounds and their binary mixtures to examine the mechanisms by which neurons in Vv and Dp

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

integrate sensory inputs. The component stimuli comprised one bile acid (TDCA) and three amino acids, two of which (tyrosine and tryptophan) share the same secondary molecular feature (aromatic side chain), whereas the third (arginine) has different secondary molecular properties (basic side chain) (Supplementary Fig. 2). We first analyzed responses of mitral cells (Fig. 5a) and determined for each neuronmixture pair whether the neuron responded to the individual components, their binary mixture or both. A response was counted if the evoked Ca2+ signal exceeded the detection threshold (two times the s.d. in trials without odor). To avoid artifacts resulting from signals close to the detection threshold, neuron-mixture pairs were included in the analysis only when at least one response exceeded three times the detection threshold. All mixture-responsive mitral cells also responded to at least one of the components, and 96% of component-responsive mitral cells also responded to the mixture (n ¼ 59 mitral cell–odor pairs from four fish; Fig. 5a,b). Hence, mixture- or componentselective responses of mitral cells were rare. The mean population response to the mixture across all mitral cells was 158% of the mean component responses. To examine the relationship between mixture and component responses, we mapped the magnitude of each mixture response onto the interval between the component responses such that the smaller component response is represented by a value of zero and the larger component response is represented by one. Mixture responses frequently had values near zero or one (Fig. 5b). Hence, responses to the mixture are often similar to one of the component responses (‘component dominance’), consistent with previous electrophysiological observations40. We next quantified suppression and synergism. Synergism was defined as a response to the mixture that is greater than the larger component response, multiplied by a safety factor to minimize false positives caused by the natural response variability. The safety factor was defined as 1 + CVrep, where CVrep is the mean coefficient of variation of responses to repeated applications of the same stimulus. Suppression was defined accordingly as a response to the mixture that is smaller than the larger component response, multiplied by a safety factor defined as 1 – CVrep. Most mitral cell–mixture pairs (56%) showed no synergism or suppression (Fig. 5b), consistent with the finding that the mixture response is often dominated by the larger component response40 (Fig. 5b). Suppression occurred in 34% of cases and included responses dominated by the smaller component, whereas synergism was rare (10%). Hence, more mitral cells are active in response to the mixture but the individual mitral cell responses often resemble the component responses. In Vv (n ¼ 38 neuron-mixture pairs in five fish), the mean population response to the mixture amounted to 142% of the mean component responses. Mixture- or component-selective responses were rare (3%) and their frequency was not significantly different from that of mitral cells (P ¼ 0.6, w2 test; Fig. 5a,b). However, responses to the mixture were often larger than the largest component response (Fig. 5b) and the frequency of mixture suppression and synergism was significantly different from mitral cells (P ¼ 0.007, w2 test; Fig. 5b): mixture suppression occurred more rarely (18%), whereas synergism occurred more frequently (42%). We found that 19% of mixture responses were larger than the sum of both component responses multiplied by the safety factor, 1 + CVrep (Fig. 5b). Vv neurons therefore do not show component dominance but tend to respond more strongly to a mixture than to the components alone. In Dp (n ¼ 240 neuron-mixture pairs in five fish), the magnitude of the population response to the mixture was 119% of the mean component response. Individual Dp neurons responded significantly

477

ARTICLES

478

Lifetime sparseness

f

Pattern correlation

Chemotopy index

AA mix Lys Tyr Val BA mix TCDCA TCA Nuc mix ATP

e

Neuron #

more often than mitral cells or Vv neurons in a mixture- or componentselective fashion (20% of neuron-mixture pairs; P ¼ 0.009, w2 test; Fig. 5a,b). Responses did not show obvious component dominance. Rather, the magnitude of the mixture response often fell in between the component responses (Fig. 5b), and the frequency of mixture suppression and synergism was significantly different from that of mitral cells (P ¼ 0.0008, w2 test; Fig. 5b). Many mixture responses (44%) showed suppression, but synergism also occurred more frequently than in mitral cells (17%). We found that 11% of mixture responses were larger than the sum of both component responses multiplied by the safety factor, 1 + CVrep (Fig. 5b). Dp neurons therefore show complex mixture interactions that do not directly reflect mixture interactions in mitral cells. To examine whether the pronounced mixture interactions in Dp depend on the spatial segregation of activated mitral cells, we separately analyzed responses of Dp neurons to mixtures of two amino acids (n ¼ 133 neuron-mixture pairs) and mixtures of one amino acid and one bile acid (n ¼ 107). The frequency of mixture- or componentselective responses, the distribution of relative response magnitudes and the occurrence of suppression and synergism were similar and were significantly different from those for mitral cells (P = 0.01 and P = 0.07 for mixture selectivity/component selectivity, P = 0.04 and P = 0.01 for suppression/synergism, w2 test; Fig. 5c). Furthermore, we tested whether mixture interactions in Dp differ in response to amino acids with the same secondary molecular feature (tyrosine and tryptophan (Tyr-Trp), both aromatic, n ¼ 55 neuron-mixture pairs) and amino acids with different secondary molecular features (tyrosine and arginine (Tyr-Arg), aromatic and basic, n ¼ 78). The frequency of mixture suppression and synergism was significantly different from that of mitral cells for Tyr-Arg mixtures (P ¼ 0.03, w2 test), but not for Tyr-Trp mixtures (P ¼ 0.9, w2 test), suggesting that mixture interactions may change when stimuli become similar. However, the frequency of mixture- or component-specific responses and the distribution of

g

Neuron #

AA mix Lys Tyr Val BA mix TCDCA TCA Nuc mix ATP

AA mix Lys Tyr Val BA mix TCDCA TCA Nuc mix ATP

AA mix Lys Tyr Val BA mix TCDCA TCA Nuc mix ATP

Neuron #

© 2009 Nature America, Inc. All rights reserved.

Neuron #

c

d

Fraction of neurons (%)

b

AA mix Lys Tyr Val BA mix TCDCA TCA Nuc mix ATP

a

Focality

Figure 4 Topological organization and response *** *** n.s. *** properties of neurons in different brain areas. *** n.s. *** Dp MC IN Vv *** *** n.s. *** (a) Mean focality (± s.d.) of activity patterns in *** different brain areas, averaged over all animals 50 0.4 100 40 and odor stimuli. P-values (Wilcoxon rank-sum 0.3 30 test): mitral cell (MC) versus interneuron (IN) 20 0.2 50 10
5; MC versus Vv, 10
6; MC versus Dp, 10
5; 10 0.1 IN versus Vv, 0.17; IN versus Dp, 0.13; Vv versus 0 1 2 3 1 2 3 1 2 3 1 2 3 0 Dp, 0.93. (b) Mean chemotopy index (±s.d.) of 0 Number of odor classes that evoked a response MC IN Vv Dp MC IN Vv Dp activity patterns, averaged over all animals and odor pairs. P-values (Wilcoxon rank-sum test): n.s. n.s. 10
13 for all comparisons. (c) Ca2+ signals (color*** * MC IN *** *** coded) evoked by nine odor stimuli with different *** ** *** *** 25 25 *** primary and secondary molecular features *** 0.8 0.8 (Supplementary Fig. 2) across 25 randomly 0.6 selected neurons from each population. 0.6 0.4 (d) Percentage of neurons responding to one, two 0.4 or three odor mixtures (amino acid mixture (AA 0.2 0.2 mix), bile acid mixture (BA mix) and nucleotide 1 1 0 0 mixture (Nuc mix); MCs, n ¼ 1,213; INs, ∆F/F (%) MC IN Vv Dp MC IN Vv Dp 0 50 n ¼ 4,899; Vv, n ¼ 658; Dp, n ¼ 2,002). (e) Average lifetime sparseness (± s.d.) of Dp Vv Correlation response profiles to the nine odor stimuli. 0 1 25 25 P-values (Wilcoxon rank-sum test): MC versus IN,
20
40 10 ; MC versus Vv, 10 ; MC versus Dp, MC IN Vv Dp 0.86; IN versus Vv, 10
37; IN versus Dp, 10
39; AA mix Lys Tyr Vv versus Dp, 10
64. (f) Average Pearson Val BA mix correlation (± s.d.) of activity patterns evoked by TCDCA TCA Nuc mix the nine odor stimuli. P-values (Wilcoxon rankATP 1 1 sum test): MC versus IN, 0.004; MC versus Vv,
10
9
10 10 ; MC versus Dp, 0.81; IN versus Vv, 10 ; IN versus Dp, 0.02; Vv versus Dp, 10 . (g) Color-coded correlation matrices depicting the pairwise similarity (Pearson correlation coefficient) between activity patterns evoked by different stimuli, averaged over animals.

relative response magnitudes were similar for Tyr-Trp and Tyr-Arg mixtures and significantly different from those of mitral cells (P ¼ 0.04 and P ¼ 0.003, respectively, w2 test; Fig. 5d). Mixture interactions in Dp therefore show little, if any, dependence on odor classes, indicating that they are not strongly influenced by the positions of mitral cells in the olfactory bulb. Integration of synaptic inputs in Dp The frequent occurrence of mixture suppression suggests that odor responses of Dp neurons are controlled not only by excitatory inputs but also by inhibitory synaptic pathways. To test this hypothesis, we first injected the GABAA receptor antagonist gabazine into Dp. Gabazine strongly and reversibly enhanced odor-evoked Ca2+ signals (n ¼ 4 fish; Fig. 6a) and changed the time course of odor responses (Supplementary Fig. 5 online), demonstrating that odor responses in Dp are under GABAergic control. We further examined the mechanisms controlling odor responses by whole-cell patch clamp recordings (n ¼ 115 neuron-odor pairs in 32 neurons, usually five repetitions of each stimulus). Subthreshold odor responses were identified when the mean membrane potential within at least five consecutive 100-ms time bins during odor stimulation was significantly different from the baseline (P o 0.01, Wilcoxon rank-sum test). Suprathreshold responses were identified as significant changes in firing rate using the same procedure after convolving individual action potentials with a Gaussian kernel (s ¼ 25 ms; other values yielded similar results). In the absence of odors, most Dp neurons were silent or fired spontaneous action potentials at low rates. Upon odor stimulation, subthreshold responses occurred in 93% of the recorded neuronodor pairs (Figs. 6b,c). The membrane potential response was usually a transient depolarization (70% of neuron-odor pairs), but we also observed hyperpolarizing (10%) or multiphasic

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

2s

x ∆F/F (%) –6

10 5 0 0

Arg+Tyr

Neuron–mix pairs (%)

1

100

P = 0.6 3%

50

0

Arg+Tyr

0

10 5 0

0

10 5 0 0

80

40

40

20 14

0

15 10 5 0

M

C

Smaller comp.

100

P = 0.04 15 %

P = 0.007

60 40

42

40 20 18 0

50

0

P = 0.0008

60 40 39

44

20 17 0

Larger comp.

20 15 10 5 0 0

1

80

80

1

P = 0.9

60 40

54 33

20

13

0

nl y .o

Neuron–mix pairs (%)

15 10 5 0 0 1 Mixture response (relative to components)

P = 0.03

60 55

40 20 0

30 15

“N Su eut pp ral” re Sy ssio ne n (> rg Su ism m )

M

ix

nl y

.

0 ix

0

M

20

50

p. o

20

22 %

om

42

P = 0.003

C

0 1 Mixture response (relative to components)

38

Neuron–mix pairs (%)

0

40

100

p. &

5

P = 0.02

60

om

10

80

C

Neuron–mix pairs (%)

15

“N Su eut pp ral” re s Sy sio ne n (> rg Su ism m )

M

ix

.o nl

y

y

ix

nl

M

p. o

om

C

Neuron–mix pairs (%)

0 .

Neuron–mix pairs (%)

P = 0.007 21 %

p. &

10

0

Dp: Tyr-Arg Mix

50

om

y

ix

46

nl

M

P = 0.04

60

Dp: AA-BA Mix 100

34

20

1

0 1 Mixture response (relative to components)

Neuron–mix pairs (%)

50

15

p. o

p. & om C

19 %

80

56 40

1

15

Dp: Tyr-Trp Mix Neuron–mix pairs (%)

P = 0.01

d

Larger comp. Neuron–mix pairs (%)

100

Smaller comp. Neuron–mix pairs (%)

Neuron–mix pairs (%)

Dp: AA-AA Mix

Neuron–mix pairs (%)

50

.

* * *

P = 0.009 20 %

.o nl y

*

100

ix

Neuron–mix pairs (%)

Neuron #

* * *

1

c

80

60

Dp

10

om

Tyr

Arg

Dp

80

Vv

4

Neuron #

0

15

0

C

© 2009 Nature America, Inc. All rights reserved.

Tyr

Arg

Vv

24

50

20

“N Su eut pp ral” re s Sy sio ne n (> rg Su ism m )

Time

4%

Neuron–mix pairs (%)

1

100

Larger comp. Neuron–mix pairs (%)

Neuron #

x o x

Smaller comp.

MC

Neuron–mix pairs (%)

x

Neuron–mix pairs (%)

b

Arg+Tyr

Neuron–mix pairs (%)

Tyr

7

Neuron–mix pairs (%)

MC Arg

Neuron–mix pairs (%)

a

Supplementary Fig. 3). These results indicate that Dp neurons receive direct or indirect synaptic input from functionally diverse sets of mitral cells but respond with action potentials only to specific input patterns.

Neuron–mix pairs (%)

responses (13%) (Fig. 6b). Despite this broad subthreshold tuning, action potential firing occurred in only 16% of the neuron-odor pairs. The firing pattern was usually a short burst (Fig. 6b, cell 23), consistent with the kinetics of Ca2+ signals (Figs. 1b and 5a and

Figure 5 Mixture interactions. (a) Responses of seven mitral cells (MCs), four Vv neurons and ten Dp neurons to two individual compounds and their binary mixture. Neurons in each panel were imaged simultaneously. ‘x’ and ’o’ indicate dominance of stronger and weaker component responses, respectively. Light and dark stars depict mixture suppression and synergism, respectively. (b) Left, percentage of neuron-mixture pairs that responded to the mixture and at least one component (Comp. & Mix.), to one or both components only (Comp. only) and to the mixture only (Mix. only). P-values indicate probability that distributions are different from those observed in mitral cells (w2 test). Middle, histogram of mixture response magnitudes relative to component responses. Response magnitudes were scaled so that the smaller component response equals 0 and the larger component response equals 1. Right, frequency of neutral mixture responses, suppression and synergism. Numbers indicate percentages. Narrow bars on the right (4Sum) indicate percentage of mixture responses larger than the sum of the component responses multiplied by the safety factor, 1 + CVrep. P-values depict the probability that distributions are different from those observed in MCs (w2 test). Data from all MCs (n ¼ 59 neuron-mixture pairs), Vv neurons (n ¼ 38) and Dp neurons (n ¼ 240). (c) Same analysis for responses of Dp neurons to amino acid (AA)–AA mixtures (n ¼ 133) and AA-bile acid (BA) mixtures (n ¼ 107). (d) Same analysis for responses of Dp neurons to Tyr-Trp mixtures (n ¼ 55) and Tyr-Arg mixtures (n ¼ 78).

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

479

ARTICLES

a ∆F/F (%)

50

–12.5 40 µm

b

c

Cell 28

No response

Arg

Suprathreshold 16%

7% TDCA

77%

© 2009 Nature America, Inc. All rights reserved.

Cell 2

Lys

Subthreshold

Cell 23

dynamically controlled by the balance of excitatory and inhibitory inputs, which in turn is determined by the activity pattern across mitral cells (Fig. 6e).

d

Arg

Tyr –40 20 mV 1s

e

+17 pA

–60

–80 mV

Arg + Tyr

Olfactory bulb

Ex. Inh.

Ex. Inh.

Ex. Inh.

Vthreshold Vrest

To elucidate the mechanisms controlling action potential initiation, we measured responses to odors that evoked a subthreshold depolarization again during injection of depolarizing current that caused spontaneous action potentials. In all cases (n ¼ 19 neuronodor pairs), the subthreshold response was reversed and inhibited action potential firing (Fig. 6d). Hence, odor-evoked synaptic input drives the membrane potential toward a reversal potential near the spike threshold that is, most likely, determined by a mixture of excitatory and inhibitory conductances. The kinetics of the membrane potential responses in the absence or presence of current injection were usually not mirror symmetric (Fig. 6d), indicating that changes in different conductances follow different time courses. These data indicate that action potential firing of Dp neurons is

480

DISCUSSION We analyzed transformations of odor representations between the olfactory bulb and two telencephalic target areas, Vv and Dp. In both areas, odors evoked distributed activity, indicating that transformations of odor representations are not governed by severe topographic constraints. Vv neurons seem to pool multiple inputs, resulting in broad tuning curves and overlapping odor representations, whereas, in Dp, multisynaptic circuits detect defined olfactory bulb output patterns and mediate the association of different molecular features. Dp may therefore establish representations of higher-order olfactory objects.

0 pA Arg 2s

Dorsal posterior telencephalon

MCs

Figure 6 Inhibitory control of action potential initiation in Dp neurons. (a) Ca2+ signals evoked by odor stimulation (Arg + TDCA) in Dp before, during and 5 min after injection of gabazine (single trials, low magnification). (b) Whole-cell recordings of odor responses in Dp (overlays of multiple trials). (c) Frequency of response types in Dp. (d) Response of a Dp neuron to odor stimulation (Arg) at rest (black) and during depolarizing current injection (17 pA, gray). (e) Working model of pattern processing in Dp. Each Dp neuron (black) receives excitatory and inhibitory input from mitral cells (MCs) via different synaptic pathways (gray, inhibitory neurons). The precise architecture of these pathways remains to be analyzed; pure feed-forward connectivity is assumed here only for simplicity. Activation of an MC subset (arrows) results in a mixture of excitatory (Ex.) and inhibitory (Inh.) synaptic inputs (black and gray waveforms, respectively) and clamps the membrane potential to an odor-dependent value in each Dp neuron (below). Action potential firing requires simultaneous activity of multiple excitatory inputs and weak inhibitory input, which occurs only in a few Dp neurons in response to each odor. Owing to the coexistence of excitatory and inhibitory input pathways, responses of Dp neurons depend both on excitatory responses of some mitral cells and on inhibitory or neutral responses of others. Such a mechanism would allow Dp neurons to detect defined activity patterns across mitral cells and associate different molecular features.

Topological transformations of odor representations A major challenge in the topological analysis of odor representations is the measurement of activity patterns evoked by multiple stimuli across a large fraction of neurons. We addressed this problem by combining two-photon imaging of somatic Ca2+ signals with a small-animal model. The segregation of glomerular responses to different primary molecular features5,6 was largely preserved in mitral cells, even though mitral cell activity patterns are reorganized locally38. In Vv and Dp, however, responses to odors with different molecular features were widespread and intermingled. In Dp, we observed small differences between the distribution of amino acid– and bile acid–evoked activity, consistent with differences between odor-evoked immediate-early gene expression patterns in piriform cortex24. The distribution of activity in Dp may therefore not be random, but the primary topographic organization of glomeruli is not transmitted to Vv and Dp. These results indicate that transformations of odor representations between the olfactory bulb and higher brain areas are not severely constrained by topography. Why, then, are odor representations topographically organized at early stages of olfactory processing? One possibility is that response spectra of higher-order neurons are too complex to be mapped along a low number of dimensions. Alternatively, odor representations in target areas of the olfactory bulb may be topographically organized not according to obvious molecular features but according to other, unknown, variables. Finally, the chemotopy of glomerular maps may configure neuronal interactions in the olfactory bulb for the local processing of activity patterns. This hypothesis is consistent with the local reorganization of mitral cell activity patterns

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

© 2009 Nature America, Inc. All rights reserved.

resulting in pattern decorrelation38,41–43, a computation that may support the representation and storage of olfactory information in higher brain areas44,45. The chemotopic organization at early stages may therefore promote computations important for higher-order processing of olfactory information. Mechanisms of pattern transformations Vv and Dp use different read-out strategies and extract complementary information from olfactory bulb output patterns. The broad tuning of Vv neurons implies that responses of the same Vv neuron can be triggered by different—even uncorrelated—mitral cell activity patterns. Moreover, Vv neurons often responded more strongly to a mixture than to the individual components. The broadening of tuning curves in Vv may therefore be mediated, at least in part, by pooling of inputs from functionally diverse mitral cells. Response properties of Dp neurons were more complex and seem to be shaped by at least two mechanisms. First, a subset of Dp neurons showed mixture synergism and responded to binary mixtures but not their components, suggesting that action potentials are triggered by simultaneous activity of converging inputs. This conclusion is consistent with in vitro studies in rodents28,46 and responses of neurons in piriform cortex to optical stimulation of mitral cells29. Second, the frequent occurrence of mixture suppression and the disinhibitory effect of gabazine demonstrate that odor responses of Dp neurons are strongly influenced by inhibition. Inhibitory input is likely to come from interneurons within Dp (E.Y. and R.W.F., unpublished data), whereas excitatory input may originate from mitral cells, other Dp neurons or both. Our electrophysiological results show that combinations of excitatory and inhibitory inputs clamp the membrane potential to a defined, odor-dependent value that is often below the threshold for action potential initiation. Odor responses are therefore controlled by neuronal circuits that transform olfactory bulb output patterns into patterns of excitation and inhibition across Dp neurons, resulting in odor-specific firing patterns after thresholding. Our results give rise to a working model for pattern processing in Dp (Fig. 6e). An important feature of this model is that action potential firing of Dp neurons is controlled by excitatory and inhibitory input pathways that originate from different, but potentially overlapping, sets of mitral cells. The coexistence of excitatory and inhibitory pathways can explain the pronounced mixture interactions and permits Dp neurons to extract information not only from excitatory but also from neutral or inhibitory mitral cell responses. Dp neurons are therefore likely to detect combinatorial activity patterns across olfactory bulb output neurons and mediate the association between different molecular features. In insects, transformations of odor-encoding activity patterns between the antennal lobe and mushroom body involve a combination of convergent excitatory input, polysynaptic inhibitory input and thresholding47,48, suggesting that higher-order neuronal circuits in insects and vertebrates share basic features. Unlike in the mushroom body, however, odor-evoked activity in Dp does not seem to be ultrasparse. In the olfactory cortex of mammals, individual neurons integrate inputs originating from different mitral cells18,28,29, receive excitatory and inhibitory inputs17,46 and show mixture suppression and synergism23,30,31, consistent with properties of Dp neurons. It will therefore be interesting to compare Dp and its mammalian homolog in more detail. Functional implications of pattern transformations As Vv and Dp perform different transformations, they are likely to subserve distinct functions. Vv is connected to limbic structures and

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

has strong reciprocal connections to the hypothalamus14,16, suggesting that it may be involved in the control of general behavioral or physiological states. Activity patterns in Dp were more distinct, but the average response selectivity and pattern overlap were not significantly different from the previous stage of processing. Dp therefore transforms one combinatorial odor representation into another. Odor representations in the olfactory bulb and Dp are qualitatively different because Dp neurons associate multiple molecular features, whereas mitral cell responses are constrained by the molecular receptive ranges of individual odorant receptors. Hence, Dp seems to establish synthetic representations of complex odor objects18,31. The pronounced mixture interactions associated with this transformation may explain, in part, why the perceived quality of an odor mixture is often distinct from the components alone. In addition, transformations in Dp and homologous areas may subserve other functions such as pattern storage and pattern completion during the formation and recall of odor memories31,44,45. METHODS Animals, dye loading and odor stimulation. Experiments were done in an explant of the intact brain and nose40. We delivered odor stimuli through a constant flow directed at the nares using a computer-controlled, pneumatically actuated HPLC injection valve (Rheodyne). All animal procedures were performed in accordance with official animal care guidelines and approved by the Federal Republic of Germany and the Veterinary Department of the Canton of Basel-Stadt (Switzerland). We dissolved 50 mg of rhod-2-AM (n ¼ 45 experiments) or Oregon Green 488 BAPTA-1-AM (n ¼ 2) in 16 ml of DMSO/Pluronic F-127 (80/20; Invitrogen), diluted it 1:10 (n ¼ 37) or 1:5 (n ¼ 10) in artificial cerebrospinal fluid (aCSF) and injected the mixture into the brain through a pipette with a tip diameter of approximately 1–2 mm. Imaging and electrophysiology. We measured odor-evoked Ca2+ signals using a 20 water immersion objective (NA 0.95; Olympus or NA 1.0; Zeiss) and a custom-made multiphoton microscope49. Within the olfactory bulb, we measured responses throughout the ventral hemisphere, including medial regions that have not been examined in previous studies35,38. We performed temporal deconvolution as described using a time constant of three seconds35. Gabazine (SR 95531 hydrobromide; 100 mM in aCSF; Tocris Bioscience) was injected into Dp from a patch pipette. We carried out whole-cell patch clamp recordings in Dp using a Multiclamp 700B amplifier (Molecular Devices) and borosilicate glass pipettes (9–15 MOhms) filled with 130 mM potassium gluconate, 10 mM sodium gluconate, 10 mM sodium phosphocreatine, 4 mM NaCl, 4 mM magnesium ATP, 0.3 mM sodium GTP, 10 mM HEPES (pH 7.25, B300 mosm). Measurements were not corrected for junction potentials. Experimental methods and data analysis procedures are described in more detail in Supplementary Methods. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank P. Caroni, T. Oertner, B. Roska and members of the Friedrich laboratory, particularly M. Wiechert, for stimulating discussions and comments on the manuscript. We are grateful to W. Denk for support. This work was supported by the Novartis Research Foundation, the Max Planck Society, the Deutsche Forschungsgemeinschaft (GK 791; SFBs 488, 780; FOR 643), the Minna-James-Heineman Foundation and the Boehringer Ingelheim Fonds (to E.Y. and S.T.B.). AUTHOR CONTRIBUTIONS E.Y. performed widefield imaging experiments, measured three-dimensional activity patterns and performed analyses of three-dimensional activity patterns and single-neuron responses; F.v.S.P. measured responses to binary mixtures, performed pharmacological experiments and electrophysiological recordings, and analyzed the data; J.N. measured responses to binary mixtures; S.T.B. participated in the construction of imaging and odor stimulation equipment and

481

ARTICLES helped with experiments; R.W.F. is the principal investigator, conceived the study, constructed equipment, performed some of the binary mixture experiments, contributed to the data analysis and wrote the manuscript.

© 2009 Nature America, Inc. All rights reserved.

Published online at http://www.nature.com/neuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Van Essen, D.C. & Gallant, J.L. Neural mechanisms of form and motion processing in the primate visual system. Neuron 13, 1–10 (1994). 2. Buck, L. & Axel, R. A novel multigene family may encode odorant receptors: a molecular basis for odor recognition. Cell 65, 175–187 (1991). 3. Meister, M. & Bonhoeffer, T. Tuning and topography in an odor map on the rat olfactory bulb. J. Neurosci. 21, 1351–1360 (2001). 4. Uchida, N., Takahashi, Y.K., Tanifuji, M. & Mori, K. Odor maps in the mammalian olfactory bulb: domain organization and odorant structural features. Nat. Neurosci. 3, 1035–1043 (2000). 5. Friedrich, R.W. & Korsching, S.I. Combinatorial and chemotopic odorant coding in the zebrafish olfactory bulb visualized by optical imaging. Neuron 18, 737–752 (1997). 6. Friedrich, R.W. & Korsching, S.I. Chemotopic, combinatorial and noncombinatorial odorant representations in the olfactory bulb revealed using a voltage-sensitive axon tracer. J. Neurosci. 18, 9977–9988 (1998). 7. Xu, F. et al. Odor maps of aldehydes and esters revealed by functional MRI in the glomerular layer of the mouse olfactory bulb. Proc. Natl. Acad. Sci. USA 100, 11029–11034 (2003). 8. Xu, F., Greer, C.A. & Shepherd, G.M. Odor maps in the olfactory bulb. J. Comp. Neurol. 422, 489–495 (2000). 9. Mori, K., Takahashi, Y.K., Igarashi, K.M. & Yamaguchi, M. Maps of odorant molecular features in the mammalian olfactory bulb. Physiol. Rev. 86, 409–433 (2006). 10. Igarashi, K.M. & Mori, K. Spatial representation of hydrocarbon odorants in the ventrolateral zones of the rat olfactory bulb. J. Neurophysiol. 93, 1007–1019 (2005). 11. Takahashi, Y.K., Kurosaki, M., Hirono, S. & Mori, K. Topographic representation of odorant molecular features in the rat olfactory bulb. J. Neurophysiol. 92, 2413–2427 (2004). 12. Finger, T.E. The distribution of the olfactory tracts in the bullhead catfish, Ictalurus nebulosus. J. Comp. Neurol. 161, 125–141 (1975). 13. von Bartheld, C.S., Meyer, D.L., Fiebig, E. & Ebbesson, S.O. Central connections of the olfactory bulb in the goldfish, Carassius auratus. Cell Tissue Res. 238, 475–487 (1984). 14. Rink, E. & Wullimann, M.F. Connections of the ventral telencephalon (subpallium) in the zebrafish (Danio rerio). Brain Res. 1011, 206–220 (2004). 15. Levine, R.L. & Dethier, S. The connections between the olfactory bulb and the brain in the goldfish. J. Comp. Neurol. 237, 427–444 (1985). 16. Wong, C.J. Connections of the basal forebrain of the weakly electric fish, Eigenmannia virescens. J. Comp. Neurol. 389, 49–64 (1997). 17. Neville, K.R. & Haberly, L.B. Olfactory cortex. in The Synaptic Organization of the Brain (ed. Shepherd, G.M.) 415–454 (Oxford University Press, Oxford, 2004). 18. Wilson, D.A., Kadohisa, M. & Fletcher, M.L. Cortical contributions to olfaction: plasticity and perception. Semin. Cell Dev. Biol. 17, 462–470 (2006). 19. Kaas, J.H. Topographic maps are fundamental to sensory processing. Brain Res. Bull. 44, 107–112 (1997). 20. Illig, K.R. & Haberly, L.B. Odor-evoked activity is spatially distributed in piriform cortex. J. Comp. Neurol. 457, 361–373 (2003). 21. Rennaker, R.L., Chen, C.F., Ruyle, A.M., Sloan, A.M. & Wilson, D.A. Spatial and temporal distribution of odorant-evoked activity in the piriform cortex. J. Neurosci. 27, 1534–1542 (2007). 22. Sharp, F.R., Kauer, J.S. & Shepherd, G.M. Laminar analysis of 2-deoxyglucose uptake in olfactory bulb and olfactory cortex of rabbit and rat. J. Neurophysiol. 40, 800–813 (1977).

482

23. Zou, Z. & Buck, L.B. Combinatorial effects of odorant mixes in olfactory cortex. Science 311, 1477–1481 (2006). 24. Zou, Z., Li, F. & Buck, L.B. Odor maps in the olfactory cortex. Proc. Natl. Acad. Sci. USA 102, 7724–7729 (2005). 25. Yan, Z. et al. Precise circuitry links bilaterally symmetric olfactory maps. Neuron 58, 613–624 (2008). 26. Schoenfeld, T.A. & Macrides, F. Topographic organization of connections between the main olfactory bulb and pars externa of the anterior olfactory nucleus in the hamster. J. Comp. Neurol. 227, 121–135 (1984). 27. Nikonov, A.A., Finger, T.E. & Caprio, J. Beyond the olfactory bulb: an odotopic map in the forebrain. Proc. Natl. Acad. Sci. USA 102, 18688–18693 (2005). 28. Franks, K.M. & Isaacson, J.S. Strong single-fiber sensory inputs to olfactory cortex: implications for olfactory coding. Neuron 49, 357–363 (2006). 29. Arenkiel, B.R. et al. In vivo light-induced activation of neural circuitry in transgenic mice expressing channelrhodopsin-2. Neuron 54, 205–218 (2007). 30. Yoshida, I. & Mori, K. Odorant category profile selectivity of olfactory cortex neurons. J. Neurosci. 27, 9105–9114 (2007). 31. Barnes, D.C., Hofacer, R.D., Zaman, A.R., Rennaker, R.L. & Wilson, D.A. Olfactory perceptual stability and discrimination. Nat. Neurosci. 11, 1378–1380 (2008). 32. Wullimann, M.F. & Mueller, T. Teleostean and mammalian forebrains contrasted: evidence from genes to behavior. J. Comp. Neurol. 475, 143–162 (2004). 33. Denk, W., Strickler, J.H. & Webb, W.W. Two-photon laser scanning fluorescence microscopy. Science 248, 73–76 (1990). 34. Stosiek, C., Garaschuk, O., Holthoff, K. & Konnerth, A. In vivo two-photon calcium imaging of neuronal networks. Proc. Natl. Acad. Sci. USA 100, 7319–7324 (2003). 35. Yaksi, E. & Friedrich, R.W. Reconstruction of firing rate changes across neuronal populations by temporally deconvolved Ca2+ imaging. Nat. Methods 3, 377–383 (2006). 36. Li, J. et al. Early development of functional spatial maps in the zebrafish olfactory bulb. J. Neurosci. 25, 5784–5795 (2005). 37. Higashijima, S., Masino, M.A., Mandel, G. & Fetcho, J.R. Imaging neuronal activity during zebrafish behavior with a genetically encoded calcium indicator. J. Neurophysiol. 90, 3986–3997 (2003). 38. Yaksi, E., Judkewitz, B. & Friedrich, R.W. Topological reorganization of odor representations in the olfactory bulb. PLoS Biol. 5, e178 (2007). 39. Willmore, B. & Tolhurst, D.J. Characterizing the sparseness of neural codes. Network 12, 255–270 (2001). 40. Tabor, R., Yaksi, E., Weislogel, J.M. & Friedrich, R.W. Processing of odor mixtures in the zebrafish olfactory bulb. J. Neurosci. 24, 6611–6620 (2004). 41. Friedrich, R.W., Habermann, C.J. & Laurent, G. Multiplexing using synchrony in the zebrafish olfactory bulb. Nat. Neurosci. 7, 862–871 (2004). 42. Friedrich, R.W. & Laurent, G. Dynamic optimization of odor representations in the olfactory bulb by slow temporal patterning of mitral cell activity. Science 291, 889–894 (2001). 43. Tabor, R., Yaksi, E. & Friedrich, R.W. Multiple functions of GABA(A) and GABA(B) receptors during pattern processing in the zebrafish olfactory bulb. Eur. J. Neurosci. 28, 117–127 (2008). 44. Hasselmo, M.E., Wilson, M.A., Anderson, B.P. & Bower, J.M. Associative memory function in piriform (olfactory) cortex: computational modeling and neuropharmacology. Cold Spring Harb. Symp. Quant. Biol. 55, 599–610 (1990). 45. Marr, D. Simple memory: a theory for archicortex. Philos. Trans. R. Soc. Lond. B Biol. Sci. 262, 23–81 (1971). 46. Luna, V.M. & Schoppa, N.E. GABAergic circuits control input-spike coupling in the piriform cortex. J. Neurosci. 28, 8851–8859 (2008). 47. Perez-Orive, J. et al. Oscillations and sparsening of odor representations in the mushroom body. Science 297, 359–365 (2002). 48. Turner, G.C., Bazhenov, M. & Laurent, G. Olfactory representations by Drosophila mushroom body neurons. J. Neurophysiol. 99, 734–746 (2008). 49. Wachowiak, M., Denk, W. & Friedrich, R.W. Functional organization of sensory input to the olfactory bulb glomerulus analyzed by two-photon calcium imaging. Proc. Natl. Acad. Sci. USA 101, 9097–9102 (2004).

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

Gating multiple signals through detailed balance of excitation and inhibition in spiking networks

© 2009 Nature America, Inc. All rights reserved.

Tim P Vogels1,2 & L F Abbott1 Recent theoretical work has provided a basic understanding of signal propagation in networks of spiking neurons, but mechanisms for gating and controlling these signals have not been investigated previously. Here we introduce an idea for the gating of multiple signals in cortical networks that combines principles of signal propagation with aspects of balanced networks. Specifically, we studied networks in which incoming excitatory signals are normally cancelled by locally evoked inhibition, leaving the targeted layer unresponsive. Transmission can be gated ‘on’ by modulating excitatory and inhibitory gains to upset this detailed balance. We illustrate gating through detailed balance in large networks of integrate-and-fire neurons. We show successful gating of multiple signals and study failure modes that produce effects reminiscent of clinically observed pathologies. Provided that the individual signals are detectable, detailed balance has a large capacity for gating multiple signals.

Experimental observations1,2 as well as theoretical arguments3,4 suggest that excitation and inhibition are globally balanced in cortical circuits. In a globally balanced network, each neuron receives large but approximately equal amounts of excitation and inhibition that, on average, cancel each other. Spontaneous activity is driven by fluctuations in the total synaptic input, leading to asynchronous and irregular patterns of spiking5–8. Such networks have been used to study signal propagation and to determine conditions that support various signaling schemes9–16. However, neurons in these networks are typically only part of a single signaling pathway, and the transmitted signals cannot be gated or rerouted. Cognitive processing requires signal paths to change dynamically according to the information content of the signal and the processing demands of the receiver17. This requires precise control and gating of signalcarrying pathways. We propose a mechanism for gating based on an extension of the concept of globally balanced networks to local cortical circuits in a form that we call ‘detailed balance’. Detailed balance implies that, in addition to an overall or global balance, neurons receive equal amounts of excitation and inhibition on subsets of their synaptic inputs that correspond to specific signaling pathways. Activation of a balanced pathway produces little response in the excitatory neurons of the signal-receiving region, but responses can be gated ‘on’ by a command signal that disrupts the detailed balance. We analyze properties of the resulting gating mechanism and examine some of its failure modes. We show that the mechanism can gate the propagation of signals from multiple different sources to a single group of neurons, and we determine its capacity for gating large numbers of signals.

RESULTS We explored the idea of detailed balance in a large network of roughly 20,000 integrate-and-fire neurons with both short- and long-range connectivity (Fig. 1a,b; Methods). With appropriately adjusted parameters, this network operates in a globally balanced manner, producing irregular, asynchronous activity in the absence of any time-varying or random external input5–8. The distribution of firing rates for the network is approximately exponential with an average firing rate per neuron of 8 Hz (Fig. 1c), the distribution of average membrane potentials is approximately gaussian with a mean of
60 mV (Fig. 1d), the distribution of interspike intervals (ISIs) is broad with peaks reflecting normal firing and bursting (Fig. 1e), and the distribution of coefficients of variation for the ISIs is centered at a value slightly greater than 1 (Fig. 1f). Average excitatory and inhibitory membrane currents are of approximately equal magnitude and the net current is near zero, indicating the globally balanced state of the network. This network model is intended to provide a sparse representation of the neurons over a fairly large area, not a full description of a single local circuit such as a cortical column. Signal gating To investigate signal gating within the network, we embed a twolayered pathway with ‘sender’ and ‘receiver’ subnetworks (Fig. 1b). These should be viewed as parts of distinct cortical regions. The connections from the sender region are directed to both excitatory and inhibitory neurons in the receiver region. Such targeting to inhibitory interneurons is consistent with data on the specific targeting of long-range excitatory projections to inhibitory interneurons18. To generate a signal, we drive the neurons in the sender subnetwork with a set of external Poisson spike trains at various rates. This causes neurons

1Center for Neurobiology and Behavior, Department of Physiology and Cellular Biophysics, Columbia University College of Physicians and Surgeons, New York, New York, USA. 2Volen Center for Complex Systems, Department of Biology, Brandeis University, Waltham, Massachusetts, USA. Correspondence should be addressed to T.P.V. ([email protected]).

Received 9 December 2008; accepted 15 January 2009; published online 22 March 2009; doi:10.1038/nn.2276

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

483

ARTICLES

10

5

5

0 0 10 20 30 Avg. firing rate (Hz)

–65 –55 Avg. mem. pot. (mV)

10 5 0 10 100 1,000 ISI (ms)

15 10 5 0 0

1 2 ISI CV

a

No signal

50 0 0

100

300

500

700

Time (ms)

b

Single trial Single cell average

–50 –60

100

300

500

700

e f

Mem. curr. (nA) Pop. rate (Hz) P.r. (Hz)

d

0

100

300

500

700

0

100

300

500

700

0

100

300

500

700

0.2 0.1 0.0 –0.1 –0.2

100 50 0 20

30 Cell no.

c

Receiver

484

Unbalanced signal

100

–70

Figure 2 Detailed balance in a network. No signal (left column): all neurons fire at background rates. Balanced signal (middle): sender neurons fire in a correlated manner in response to oscillatory input and project the input firing pattern to both excitatory and inhibitory receiver neurons. Inhibitory receiver neurons reproduce the input pattern, preventing their excitatory neighbors from doing the same. Unbalanced signal (right): by decreasing the responsiveness of the inhibitory receiver neurons, the signal balance in the excitatory receiver neurons shifts in favor of excitation, and the signal is revealed in their firing pattern. All firing rates and averages are calculated in 5-ms bins. (a) Population (pop.) average firing rate of the sender neurons responding to a sinusoidally varying input. (b) Voltage trace of membrane potential (mem. pot.) in a randomly selected excitatory receiver neuron. Red trace, single trial. Black trace, average subthreshold membrane potential over 100 trials. (c) Average membrane currents (curr.) of the excitatory receiver neurons. Red and blue, excitatory and inhibitory currents, respectively; black, net current (including voltage-dependent leak and constant background currents). (d) Blue trace, average firing rate of the inhibitory receiver neurons. Red histogram, average firing rate of the excitatory receiver neurons. (e) Average population firing rate (p.r.) of the entire network. (f) Spike raster for 30 randomly chosen excitatory receiver neurons.

Balanced signal

Sender

in the sender area to fire in a manner that mimics the input signal (Fig. 2a). In the balanced state, excitation from the sender network, a simple oscillatory signal in the example of Figure 2, activates both excitatory and inhibitory subpopulations in the receiver region. The resulting inhibitory activity (blue, Fig. 2d) produces a local countersignal that cancels the excitatory membrane currents (Fig. 2c) and generates only modest firing-rate fluctuations in the excitatory receiver neurons (red, Fig. 2d,f). The signal path is hence gated ‘off’ in the default (balanced) state of the signal-carrying pathway. Signal propagation within this network can be gated ‘on’ in several ways, all of which involve unbalancing the excitatory and inhibitory pathways between the sender and receiver regions. The main requirement is a mechanism that differentially modulates the net excitatory and net inhibitory pathways from the sender to the receiver region19. A possible candidate is cholinergic modulation, which satisfies the basic requirements of cell-specific targeting20,21 as well as relatively rapid response times22–24. Rather than modeling such modulation in detail, in the following examples detailed balance is disrupted by decreasing the gain or responsiveness of local inhibitory interneurons in the receiver region. This modulation, in keeping with the strong effects of attention seen for inhibitory neurons25, corresponds to changing the input–output transfer function so that the same synaptic current generates a smaller response. Although the examples we show focus

on modulation of inhibition, any combination of modulations that increases the ratio of excitatory to inhibitory transmission along the signaling pathway will perform similarly. To unbalance the signal in Figure 2, we reduced the response gain of the local inhibitory neurons in the receiver region to 15% of its control value. (We discuss more modest gain modulations below.) Gain modulation that decreases the amplitude of the firing-rate modulations of inhibitory neurons in the receiver region (Fig. 2d, blue trace) reduces the inhibition of excitatory neurons in this region, leaving the bulk of the excitatory synaptic current uncanceled (Fig. 2c). This produces robust firing in the excitatory receiver neurons that is locked to the temporal pattern of the input signal (Fig. 2b,d,f). Overall network activity is relatively unaffected by these changes (Fig. 2e) because the modulated interneurons provide only a small fraction of the total inhibition to the network. Average subthreshold membrane potentials (excluding action potentials and their subsequent refractory periods) of the excitatory receiver neurons in the balanced

Pop. rate (Hz)

10

f Percentage

e 15

Percentage

d 15

Percentage

Percentage

c

© 2009 Nature America, Inc. All rights reserved.

Figure 1 Network connectivity and properties. (a) All excitatory and 65% of the inhibitory neurons are connected randomly with a connection probability of 2% (red). The other 35% of the inhibitory neurons (blue) have local connectivity, targeting their nearest neighbors. (b) An embedded signal pathway is created by selecting a group of sender neurons (green) that target either excitatory or locally inhibitory neurons (red and blue, respectively, throughout the figures) in a signal-receiving region (red shading) of the network. (c–f) Asynchronous background activity in the network model. Distributions for network neurons of average (avg.) firing rates (c), average membrane potentials (mem. pot.) (d), ISIs plotted on a semilog scale (e) and coefficients of variation (CV) for those ISIs (f). Arrows indicate the means of the distributions.

b

Mem. pot. (mV)

a

20 10 0 Time (ms)

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

a

b

Balanced Unbalanced

1.0 Amplitude ratio

Receiver rate (Hz)

120

60

0

Receiver rate (Hz)

0.8 0.6 0.4 0.2 0.0

0

c

Balanced Unbalanced

60 120 Sender rate (Hz)

180

10

30 50 Oscillation (Hz)

70

200 150 100 50 0

Fire rate (Hz)

0

500

1,000

0

500 Time (ms)

1,000

200 150 100 50 0

f

**

*

40 ms

g

125 Hz

60 Peak response (Hz)

e

20 Hz

90 Hz 50 Hz 15 Hz

40

20

0 1

10 100 Rise time (ms)

1,000

and unbalanced states (Fig. 2b, black trace) differ by only 4 mV, but this is sufficient to produce markedly different firing patterns. Response properties To further quantify the gating mechanism, we studied responses to different types of input (Fig. 3a,b). The firing rates of excitatory receiver neurons are relatively unaffected by constant input rates in the balanced gated-off state (Fig. 3a, solid trace) but rise sharply as a

function of input rate when the pathway is gated ‘on’ (Fig. 3a, dashed trace). The rise begins to saturate at high rates because of the residual inhibition produced locally, even at low gain. Gating is also evident in the amplitudes of firing-rate fluctuations for excitatory neurons in the receiver region when the input signal is oscillatory (Fig. 3b). In addition, gating occurs when filtered white noise (with a 50-ms time constant12) is used as the input signal (Fig. 3c,d). In the gated-on state, this complex, irregular signal is transmitted with similarity values12,14 (defined in the Methods) of B90%, sufficient to propagate the signal across several layers14. In the balanced, gated-off state (Fig. 3d), the output of the excitatory receiver group is greatly decreased in amplitude, and the similarity between input and output is reduced to B25%. A close inspection of the responses in the balanced state (Fig. 3d) shows that detailed balance does not completely cancel signals when input firing rates change rapidly. Rapid changes in the signal can evoke a response in the excitatory receiver neurons before inhibition can balance excitation because of the time lag between the monosynaptic excitation and the canceling disynaptic inhibition acting on the receiver neurons. This effect would be even more marked if excitatory and inhibitory synapses were subject to different amounts or types of shortterm plasticity. Such partial gating of transients allows large signals with rapid onsets to be propagated, which may induce upstream control circuits to activate gain modulation and open the gate. To further investigate this effect, we activated the sender neurons with step-like input rates of various step sizes and rise times (Fig. 3e). Short rise times produce fairly strong responses in excitatory receiver neurons that

a

1.0

b

3c

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

Similarity

Similarity

0.6 0.4 0.2

40 60 ∆Gain (%)

80

100

0

1

2

3


syn. strength (nS)

d

1.0

1.0 0.8

Similarity

0.8 0.6 0.4

0.0

0.4

0.0 20

0.2

0.6

0.2

3d

0.0

c

Synapse loss

1.0 0.8

0.8

Figure 4 Gain properties. (a) Maximum values of the cross-correlations (termed ‘similarity’; see Methods) between the sender region and the excitatory (red) and inhibitory (blue) receiver cells for different gains. Solid lines, symmetric gain reduction; dashed lines, asymmetric gain reduction, wherein only the gain of the excitatory synapses onto the inhibitory receiver cells is changed. Circles denote the parameters used in Figure 3. (b) Similarity values between excitatory receiver activity and the signal in the balanced (gated-off) state as a function of increasing the variability (s.d. s) of the synaptic (syn.) strengths of the excitatory (green) and inhibitory (blue) pathways. The arrows mark the variability limits beyond which the tails of the strength distributions become rectified to zero. (c) Effect of reducing the number of inhibitory receiver neurons on the ability to gate signals off. Similarity values in the balanced state for decreasing numbers of inhibitory receiver cells, without and with synapse strength compensation (solid and dotted lines, respectively). (d) Operation of the gating mechanism with only 20 inhibitory receiver neurons by compensating synapse strength and shortened refractory times to allow for more rapid inhibitory firing. Similarity between the signal and the excitatory (red trace) and inhibitory (blue trace) receiver activity is plotted as a function of change in inhibitory gain.

Similarity

© 2009 Nature America, Inc. All rights reserved.

d

Figure 3 Response analysis. (a) Firing rates of the excitatory receiver neurons as a function of different constant sender firing rates, in the balanced (solid trace) and unbalanced (dashed trace) states. (b) Ratio of receiver to sender excitatory firing-rate oscillation amplitudes at different oscillation frequencies, in the balanced (solid trace) and unbalanced (dashed trace) states. (c,d) Response to a random, time-filtered signal in the unbalanced (c) and balanced (d) states. Red trace, average firing rate of the excitatory receiver neurons; black histogram, rates of the sender neurons. Deviations from the signal in c and from the average background rate in d are gray. (e) Schematic of an input step. Step size (*) and step duration (**) vary independently. (f) Average responses of the excitatory receiver neurons in the balanced state to instantaneous steps of different sizes. (g) Peak amplitude of the responses in these neurons to steps of different sizes (legend) and rise times (horizontal axis).

Synapse compensation factor 4.6 2.2 1.2 0.7 0.4 0.2 0.0

10

30 50 No. of available cells

70

0.6 0.4 0.2 0.0 20

40 60 ∆Gain (%)

80

100

485

ARTICLES

Sender rate (Hz)

a

No signal & balanced

No signal & unbalanced

Signal & unbalanced

Signal & balanced

120 60 0 100

200

300

b

400 500 Time (ms)

600

700

800

900

120 60

© 2009 Nature America, Inc. All rights reserved.

d

0 Receiver rate (Hz)

c

120 60 0 120 60 0 100

200

300

400 500 Time (ms)

600

700

800

900

Figure 5 Network pathologies. (a) Average firing rate of the sender neurons without and with an oscillatory input. (b–d) Responses of excitatory (red histogram) and inhibitory (blue trace) receiver neurons with correct tuning (b); weakened local inhibition, leading to a gating deficit (c); or a hyperactive receiver region causing a response to the gating modulation (d). Conditions shown in the different columns are no signal and no modulation, no signal but gated on, signal on and gated on, signal on but gated off. Firing rates are calculated in 5-ms bins.

depend on the step size (Fig. 3f,g), even in the balanced gated-off state, but these diminish as the rise time of the step increases, illustrating the transient nature of this transmission. The gain changes used to gate signals on have been fairly large, so we next examined different degrees and types of gain modulation in the inhibitory receiver neurons. Beginning with no gain change (DGain ¼ 0), we decreased the responsiveness and thus the firing rate of the inhibitory receiver population. This causes the firing rate of the excitatory receiver neurons and its similarity to the sender signal in the gated-on state (Fig. 4a, solid red trace) to increase. At DGain B80%, the signal similarity of the activity of the inhibitory neurons goes rapidly to zero (Fig. 4a, solid blue trace), and the similarity of the excitatory receiver activity plateaus at B90%. Alternatively, it is possible to reach this same plateau level with a gain shift of only 30% (Fig. 4a, dotted traces) by modulating the inhibitory population asymmetrically, which means modifying only the responsiveness to excitatory inputs. The gating mechanism is robust to many (but not all; see below) different perturbations of the network. To study the effect of synaptic variability, we computed the similarity of responses in the receiver region to the signal, when it is gated off, as a function of the variability in the strengths of the inhibitory (Fig. 4b, blue trace) or excitatory (Fig. 4b, green trace) sender synapses onto the excitatory receiver cells. Synaptic strength variance does not have a large effect in either case until the variance becomes high enough to force substantial numbers of synapses to zero strength (which occurs at a different point for excitatory and inhibitory synapses because of their different initial strengths), changing the mean synaptic strength. After this point, the high degree of variability in the excitatory synaptic strengths makes it difficult to shut the signal off (Fig. 4b).

486

We also tested the robustness of gating a signal off to the loss of its most critical components, the inhibitory neurons in the receiver region (Fig. 4c). The effect of decreasing the number of available interneurons (originally 73) is roughly linear (solid trace), until gating off fails completely when less than 40 cells are available. It is possible to partially rescue gating by upregulating the strengths of all remaining inhibitory synapses proportionately to the number of deleted neurons and thus deleted synapses (dotted trace). However, gating still fails when less than 25 inhibitory cells are available because such a small population of inhibitory neurons cannot fire a sufficient number of action potentials to provide balancing membrane currents, even if their synapses are strengthened to unrealistically high values. If the inhibitory receiver neurons are allowed to spike at rates as high as 600 Hz, it is possible to further rescue the balance mechanism and to successfully gate signals in as many as 600 excitatory cells with as few as 20 inhibitory neurons (Fig. 4d). Pathologies The basic requirement to achieve a state of detailed balance is local inhibition strong enough to cancel signals in the gated-off state. In addition, the gain modulation used to unbalance and gate ‘on’ a pathway must not have an excessively destabilizing effect on the global excitatory–inhibitory balance of the network. With this in mind, we examined more ways in which network gating can fail when tuning is relaxed by studying gated off and gated on states with no signal and in the presence of an oscillatory signal (Fig. 5). With proper tuning (Fig. 5b), the excitatory neurons of the receiver subnetwork respond robustly only when the signal is present and gating is on, although there is a weak transient response when the signal is present but gating is off. We considered two different detuning conditions. First, we reduced the strength of all synapses from the local inhibitory neurons by 60% (Fig. 5c). This causes baseline firing rates in the absence of a signal to rise slightly, but the effect is not large because the bulk of the inhibition is not affected. Little change is seen in the response to the gain modulation alone, but activating the input shows that the gating mechanism has been compromised. Because of the weakened inhibition, excitatory inputs to the excitatory receiver neurons cannot be fully cancelled by local inhibition, and the signal can never be fully gated off. We also detuned the detailed balance by increasing the strength of excitatory synapses within the receiver area by 60% (Fig. 5d). Excitatory synapses onto excitatory and inhibitory neurons were both modulated in this way, so a rough balance is still maintained within the receiver region. As in the case of reduced inhibition, enhanced excitation slightly elevates firing rates in the gated-off, no-signal condition. Activating gain modulation to open the gate causes a substantial elevation in the firing rate of excitatory receiver neurons, even when no signal is present. Thus, with altered excitation, the receiver neurons respond to the gating signal as if it were an input. This means that the network falsely transmits internally generated activity (the gating signal) as if it were an external signal. By contrast, in this condition signal responses in both the gated-on and gated-off states appear normal. We address the implications of these findings in the Discussion. Multiple signals One of the advantages of the gating mechanism we propose is that a group of receiver neurons can remain responsive to one set of incoming signals even while other sets are being cancelled by balancing inhibition. This gives the mechanism the capacity to gate multiple signals. As a first example, the network we have been considering is expanded so that it

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES signal S1 + S2 and the firing rate of the receiver cells (Fig. 6f,g). This is much the 150 0.8 same as in the single-signal case (Fig. 4a, 100 50 solid red trace). 0.6 50 How many signals can be canceled and then 0 0 gated ‘on’ by a population of inhibitory neu0.4 d 100 b 150 rons? To address this question, we studied a * ** 0.2 single excitatory receiver neuron, rather than 100 50 the full network that we have been considering 0 50 up to this point. As discussed above for the 0 0 two-signal case, configuring a network for 0 50 100 0 500 1,000 1,500 2,000 2,500 ∆Gain S2 (%) multiple signals takes a fair amount of Time (ms) modification and readjustment, and this e 1.0 a f 100 g 1.0 b* made it unpractical to consider a wide range ** 0.8 0.6 of different numbers of signals within a full 0.6 50 network. Instead, we set up the mechanism of 0.4 0.2 0.2 b ** a* detailed balance in a single integrate-and-fire 0 –0.2 neuron that receives 800 excitatory and 200 0 50 100 20 60 100 80 40 40 80 inhibitory inputs. The critical component in ∆Gain S2 (%) ∆Gain (%) ∆Gain S1 (%) ∆Gain S2 (%) determining the capacity of detailed balance Figure 6 Gating two signals in a network. (a,b) Average response of the excitatory receiver subnetwork to for switching multiple signals is the populatwo simultaneously delivered signals (S1 and S2). Colored areas indicate difference between the average tion of inhibitory afferents, because they are firing rate in the receiver region (red) and either S1 (purple, a) or S2 (green, b). First column: both signal less numerous than their excitatory counterpathways are balanced, the signals are off. Second and third columns: signal pathways 2 and 1, parts and must cancel the excitatory effects of respectively, are unbalanced by shifting the gain of their respective inhibitory receiver populations to the signals while at the same time allowing a 15% of their control values. (c,d) Similarity values between S1 and the excitatory receiver activity particular signal to get through when modu(c) and S2 and the same excitatory receiver activity (d) for all possible combinations of the two gain modulations. Both signals reach similarity values of 485%. (e) Similarity values for S1 and S2 for lated. We chose 200 inhibitory inputs to independent gain changes. Left of the gray line only the gain for S1 is manipulated while the gain for S2 match the number that seem to influence a remains 100%, and vice versa on the right side. Black circles indicate the gain values used for a,b for single pyramidal neuron in cortical circuits26. the regions denoted by the asterisks. (f) Similarity values as in c,d but measured for the combined signal The single-neuron model acts much like S1 + S2. (g) Similarity values between S1 + S2 and the excitatory receiver subnetwork activity as a any of the excitatory receiver neurons in the function of combined (equal) inhibitory gains, taken from the results along the diagonal of f. full network because we adjusted its input to match what a typical neuron receives when can gate two signals, rather than one. We then discuss the capacity of the network is intact. The excitatory and inhibitory neurons that detailed balance for switching a large number of signals using a model provide input to this model neuron are represented by Poisson spike trains generated from firing rates that encode various numbers of that simulates a single neuron in the networks we have been using. We introduced a second signal to the receiver group. (To accom- signals directly rather than through other model neurons. For this modate two signals, S1 and S2, we modified the network architecture reason, we refer to them as inputs or afferents rather than as neurons— slightly to allow two separate groups of B70 inhibitory neurons; see although, of course, they correspond to neurons in the full network. Methods.) We then compared the average firing rate of the excitatory Detailed balance is achieved by distributing the signals across the receiver subnetwork compared to each of these signals, S1 and S2 excitatory and inhibitory afferents and adjusting synaptic strengths (Fig. 6a,b). The colored bars indicate the difference between the so that all signals are cancelled in the default state. To gate a particular average firing rate in the receiver region (red) and each of these signals. signal on, we set the gains of all the inhibitory afferents that carry that When both pathways are balanced (‘signals off’), firing in the receiver signal to zero, essentially shutting them off. We used this extreme form region stays roughly at the background level, except for transient of modulating because we wished to compute the maximum capacity responses as described above. When one of the pathways is unbalanced of the system, not a capacity limited by restricting the amount of to allow propagation of its signal, the response of the receiver group modulation. Each signal consisted of a mean firing rate plus independent filtered follows that signal accurately, as indicated by the small divergence between the appropriate signal and response pair. Activity of the white noise fluctuations (as used in Fig. 3c,d). To begin (Fig. 7), we receiver subnetwork does not follow the signal that is in the off state, distributed M signals across the afferents so that each excitatory and as indicated by the larger difference regions for the gated-off signals. inhibitory afferent carries only one signal. The maximum number of This finding can also be confirmed using the similarity to S1 and S2, signals that can be distributed in this way is M ¼ 200, the number of respectively, for all combinations of the two levels of inhibitory gain inhibitory afferents. Performance, measured in terms of a similarity reduction (Fig. 6c,d). The regions where similarity is high for either index (see Methods), falls rapidly as a function of the number of signals signal are well separated from each other, and both signals reach being gated, and this way of distributing and gating multiple signals similarity values above 85% in the regions where they are gated ‘on’. does not allow switching of more than B10 signals (Fig. 7a,d). The Furthermore, when only one of the two signals is gated on, the other problem is not in canceling out the signals that are not being gated ‘on’ signal tends to weaken, and similarity between receiver activity and the but in being able to fully gate the chosen signal on. One problem with the scheme of having one signal per inhibitory gated-off signal can even become negative because of inhibitory overshoot (Fig. 6e). To compare the gating of two signals with the gating of afferent is that the number of afferents being gain-modulated to zero to one, it is useful to examine the similarity value between the combined gate a given signal on is small. For example, only 10 inhibitory afferents Signal 1 on

**

© 2009 Nature America, Inc. All rights reserved.

[

NUMBER 4

Similarity

∆Gain S1 (%)

Similarity

NATURE NEUROSCIENCE VOLUME 12

c 100

∆Gain S1 (%)

*

Similarity

Signal 2 on

∆Gain S1 (%)

Signals off

Fire rate (Hz)

Fire rate (Hz)

a

[

APRIL 2009

487

ARTICLES 1 signal per synapse All signals active

Rate (Hz)

~40 synapses per signal 5 signals active

10 simul. signals

10 signal paths

10 simul. signals 140 80 20

140 80 20 100

300

140 80 20

140 80 20

300

100 simul. signals

100 140 80 20

300

100 signal paths

Similarity value

f 1.0 0.8 0.6 0.4 0.2 0.0 0 100 200 No. of input signals

Figure 7 Multiple signals into a single cell. (a–c) Firing rates of the output signal (red), averaged over 200 runs, compared to the input signal (black). First row: 10 simultaneous (simul.) signal paths. Second row: 25 simultaneous signals paths. Third row: 100 simultaneous signal paths. (a) Each afferent to the model neuron carries only one signal. (b) Each afferent carries 40 signals. (c) Each afferent can carry 40 signals but only 5 signals are present at a given time. (d–f) Similarity values as a function of the number of signals being gated in a–c. (d) With one signal per afferent, gating is limited to less than about 20 signals. (e) Overlapping several signals onto each afferent improves performance slightly. (f) When only 5 signals are present at a given time, large numbers of signals can be gated when 40 signals are carried on each afferent (solid trace), but performance is still limited if each afferent carries a single signal (dashed trace).

100 300 Time (ms)

100 300 Time (ms)

e

0 100 200 No. of input signals

300

140 80 20

140 80 20

1.0 0.8 0.6 0.4 0.2 0.0

300

25 signal paths

100

100 simul. signals

140 80 20

d Similarity value

300

25 simul. signals

100

300 100 Time (ms)

© 2009 Nature America, Inc. All rights reserved.

140 80 20 100

25 simul. signals

100 Rate (Hz)

per signal c b ~40Allsynapses signals active

Similarity value

Rate (Hz)

a

1.0 0.8 0.6 0.4 0.2 0.0 0 200 400 600 No. of signal channels

because of the balancing inhibition, the mean excitatory input is canceled, but this cancellation is subject to fluctuations due to the spiking nature of the inputs. As a result, there is a fundamental limitation in the number of signals from which a single signal can be extracted, independent of the method by which this is done. To show that the limited capacity (Fig. 7a,b) is a result of this fundamental restriction on firing-rate coding and not a limitation of the detailed-balance approach, we restricted the number of signals present on the afferents to the neuron at any given time. In other words, we set up the postsynaptic neuron so that it could extract any one out of M input signals, but at any given time we restricted the number of signals present to 5 out of these M possibilities (Fig. 7c). This seems reasonable for an in vivo switching situation: out of the myriad of possible stimuli that can activate a neuron, only a few are likely to be present at any given time. The number 5 is arbitrary; the key is to restrict the number of signals at any given time to a value that does not make the signals undetectable owing to the M1/2 scaling problem discussed above. When the input signals are restricted in this way, the capacity is still limited when each afferent is only allowed to carry one signal, but when each signal is distributed across 40 afferents, the switching capacity is much larger (Fig. 7c,f). There is no decrease in similarity for the gated on signal over the entire range from 1 to 600 possible signals. This shows that the limited performance for the onesignal-per-afferent case (Fig. 7a) is a result of having too few afferents per signal. By contrast, the poor performance when 40 afferents are used per signal (Fig. 7b) does not represent a limitation of detailedbalance gating but rather a basic limitation of rate-based coding. When this latter limitation is avoided, detailed balance can switch very large numbers of signals.

carry any particular signal when M ¼ 20. This problem can be addressed by distributing the M signals across the excitatory and inhibitory afferents so that each afferent carries more than one signal (Fig. 7b). Gating works best if each signal is carried on B40 inhibitory and B160 excitatory afferents (chosen randomly for each signal from all available afferents of each type), which means that each inhibitory afferent carries, on average, 40M/200 signals, rather than 1 as before. In this case, because of the overlap in the signals, when a particular signal is gated ‘on’ by setting the gains of the inhibitory afferents carrying that signal to zero, this upsets the detailed balance for the other, ungated signals. To compensate for this, the gains of the remaining inhibitory afferents are adjusted so that the ungated signals are cancelled as nearly as possible by the remaining active inhibitory afferents. In other words, through a procedure discussed in the Methods, the gains on the remaining active afferents are increased to compensate for those missing owing to gating ‘on’ of the chosen signal. Performance with this distribution of signals is better than in the one-signal-per-afferent case, but detailed balance still cannot handle more than 30 signals (Figs. 7b,e). What is limiting the ability of the detailed balance scheme to switch large numbers of signals? The limitation is, in fact, not a deficit of the detailed balance scheme but a fundamental problem with encoding multiple signals using Table 1 Synapse modifications firing rates, which no switching scheme can avoid. This is the problem of keeping firing Inhibition Hyperexcitable Two rates positive. As discussed above, each signal Synapse groups Healthy deficit receiver signals corresponds to a mean firing rate plus positive sender-receiver 0.8 nS Dgex-ex 1.1 1.1 1.1 0.9 and negative fluctuations about this mean. Dgexglobal net-receiver 1.0 1.0 1.6 1.0 Dgex-ex Although the mean rate carries no information sender-receiver 1 1.0 1.0 1.0 1.1 Dgex-loc inh about the signal, it cannot be set to zero or half net-receiver 1 1.0 1.0 1.6 1.3 Dg ex-loc inh of the signal would be lost owing to firing-rate local 1.5 nS All 1.0 0.4 1.0 1.0 rectification. Adding together M signals results Dginh receiver 1-receiver 3.1 3.1* 3.1 3.3 Dgloc inh-ex in a total input that has a mean proportional receiver 2-receiver 4.0 Dgloc inh-ex to M and a fluctuating signal that is propornet-receiver 1 7.5 nS Dginh-loc 1.25 1.25 1.25 1.32 Dginhglobal inh tional to only M1/2 because the M signals are net-receiver 2 1.2 Dginh-loc inh independent of one another. Thus, there is a net-receiver 0.75 Dginh-ex strong tendency for the mean input to drown Overview of synaptic alterations, sorted by synapse type (rows) and purpose of modification (columns). The synaptic strengths out the signals to which we want the post- appearing in column 2 were multiplied by the factors appearing in columns 4, 5 and 6. The asterisk indicates that this value is 3.1 synaptic neuron to respond27. Of course, times the scaled value (scaled by 0.4).

488

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

© 2009 Nature America, Inc. All rights reserved.

ARTICLES DISCUSSION We have proposed an extension of the concept of global balance that offers an alternative to the more traditional model of gating by using inhibitory neurons to deactivate a signaling pathway14,28,29. This inverts the usual scheme of allowing signal propagation by default and disrupting the signal flow to gate a pathway off. Instead, in our model, the balanced gated off state is the default. In complex networks, gating presumably occurs in parallel along the many pathways responsible for transmitting different aspects of a stimulus. It may be easier for a system to keep track of what is ‘interesting’ in a broadband signal stream than to keep track of all the ‘uninteresting’ stimulus features that should be suppressed. If one feature of a stimulus warrants further processing, a control mechanism can select it by unbalancing its respective module, allowing the signal to propagate further downstream. We used gain modulation to unbalance signal-carrying pathways and gate signals on, but ordinary subtractive inhibition of inhibitory interneurons could also be used in our scheme. Other discussions of signal switching in balanced neural circuits, either by shifting inhibition or through gain modulation, can be found in refs. 2,19,30. Continuous temporally and strengthwise correlated (balanced) excitatory and inhibitory input activity has been reported in vivo through recordings from pairs of pyramidal cells in the rat somatosensory cortex during spontaneous and sensoryevoked activities31. It has also been demonstrated that inhibition can be used to adjust the gain of a downstream circuit that receives balanced input and can thus control behavioral responses in a context-dependent manner32. These findings fit well into the framework of our hypothesis. Although we found that gating is robust to several perturbations, it would be interesting to study how a homeostatic mechanism might impose and maintain a detailed balance. Spike timing–dependent plasticity has been shown to generate a global balance between excitation and inhibition33,34, but detailed balance is likely to require some further competitive as well as homeostatic mechanisms. In developing systems, this might involve tuning AMPA and immature, excitatory GABA synapses to the same degree. Failure to maintain a precise balance between excitation and inhibition due to various abnormalities of synaptic transmission is commonly hypothesized as a basis for mental disorders such as schizophrenia35–39 and autism40,41. Although it is natural to think that such an imbalance might lead to basic instabilities, such as those associated with epilepsy, it is more difficult to understand how they would lead to cognitive and behavioral disorders. Our results provide a suggestion. If we associate the local inhibitory neurons in our model network with parvalbumin-positive inhibitory neurons in cortex, the failure of gating in this model with reduced inhibition (Fig. 5c) could provide a functional basis for the hypothesis that reduced GABA production in parvalbumin-positive interneurons may contribute to gating problems in schizophrenia35. Similarly, the inability to discriminate between external and internal activity could be related to the hallucinations and delusions that have been hypothesized to arise from defective dopaminergic regulation36,37 or NMDA current anomalies38. Although these latter anomalies correspond to hypofunction of NMDA conductances, this is associated with a hyperexcitability of the affected circuits39, so we modeled the overall effect by increasing excitation (Fig. 5d). The mechanism we have proposed makes a distinctive prediction concerning inhibitory activity in a signal-receiving region. Although excitatory neurons receiving a signal should respond more vigorously in an attentive (gating on) than in a nonattentive (gating off) state, at least some local inhibitory interneurons should

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

follow a signal more reliably in the inattentive state and should decorrelate their activity from the stimulus with attention. In studying the capacity of detailed balance to switch multiple signals, we encountered a fundamental limitation of multisignal encoding using firing rates that affects any switching scheme. However, once this limitation was avoided by keeping too many of the possible signals from being present at any given time, we found a large capacity to gate multiple signals. Thus, detailed balance offers a powerful and dynamic way of controlling signal flow in complex and multiply interconnected circuitry. METHODS Neuron model. The model used for all our simulations is a leaky integrateand-fire neuron, characterized by a time constant, t ¼ 20 ms, and a resting membrane potential, Vrest ¼
60 mV. Whenever the membrane potential crosses a spiking threshold of
50 mV, an action potential fires and the membrane potential is reset to the resting potential, where it remains clamped for a 5 ms refractory period. To set the scale for currents and conductances in the model, we used a membrane resistance of 100 MO. We modeled synapses onto each neuron as conductances, so the subthreshold membrane voltage obeys t

dV ¼ ðVrest
VÞ + gex ðEex
VÞ + ginh ðEinh
VÞ + Ib dt

Reversal potentials were Eex ¼ 0 mV and Einh ¼
80 mV. The synaptic conductances gex and ginh are expressed in units of the resting membrane conductance. When the neuron receives a presynaptic action potential, the appropriate postsynaptic variable increases, gex - gex + Dgex for an excitatory spike and ginh - ginh + Dginh for an inhibitory spike. Otherwise, these parameters obey the equations tex

dgex ¼
gex dt

and tinh

dginh ¼
ginh dt

with synaptic time constants tex ¼ 5 ms and tinh ¼ 10 ms. Ib is a constant background current used to maintain network activity (see below). The integration time step for the simulations was 0.1 ms. We implemented all simulations in C. Network architecture. We studied a network of 20,164 leaky integrate-and-fire neurons, laid out on a 142  142 grid. Neurons were either excitatory or inhibitory. The ratio of inhibitory neurons was roughly 1 in 4, but the geometric organization of neurons on the grid constrained the final numbers to 15,123 excitatory cells and 5,041 inhibitory cells. Inhibitory neurons were divided into two groups of 3,361 and 1,680 neurons differing in their connectivity pattern. All excitatory neurons and 65% of the inhibitory neurons had a random connectivity of 2% to the rest of the network (red cells in Fig. 1a). The 1,680 inhibitory neurons of the second group each targeted 40% of their 500 closest neighbors, thus acting locally42 (blue cells in Fig. 1a). To avoid boundary effects, we implemented the network with the topology of a torus. We used 20,000 cells because this was the largest network that we could study within reasonable computation times. It has been shown previously that the activity in such networks becomes independent of size at about 10,000 neurons8,11,16. Other network parameters were chosen in keeping with both general properties of cortical circuits and previous work11,12,14,16,42. Signal path. In additional to the general architecture, we introduced a specific pathway from one region of the network to another, which we call sender and receiver subnetworks (Fig. 1b). The two subnetworks were chosen to be sufficiently distant from each other to exclude possible interactions through local inhibitory neurons. Synapses from a given excitatory sender neuron were allocated to contact either the excitatory or the locally inhibitory neurons of the receiver region, but not both. This division is made for the sake of tuning simplicity. The numbers of neurons and projecting synapses for the sender subnetwork were chosen to supply each of the B500 excitatory receiver and B70 inhibitory receiver neurons with 50 synapses from the sender subnetwork, a number necessary for critical spike propagation with appropriate tuning of

489

ARTICLES

© 2009 Nature America, Inc. All rights reserved.

synaptic strengths (see below). In addition, the number of neurons in both subnetworks was chosen to be as large as possible without interfering with overall network activity during signal propagation14. The final numbers were 494 excitatory–excitatory sender neurons, 234 excitatory–inhibitory sender neurons, 463 excitatory receiver neurons and 73 inhibitory receiver neurons. Tuning conditions. Except for synapses along the signaling pathway and those mentioned further below, all synapses of the same cell type had the same strength. A background current (Ib) of 0.03 nA was delivered to every neuron and the three sets of strengths were adjusted to allow asynchronous background global ¼ activity within the network. The postsynaptic conductances of Dgex local ¼ 1.5 nS and Dg global ¼ 7.5 nS correspond to 0.5 mV excitatory 0.8 nS, Dginh inh postsynaptic potentials and
0.4 mV and
1.1 mV inhibitory postsynaptic potentials, respectively, as obtained from spike-triggered averages in the active network. To propagate signals from the sender subnetwork to the excitatory receiver subnetwork, we set the synapses between those two groups to sender-receiver ¼ 0.9 nS. The synapses between sender neurons and inhibitory Dgex-ex receiver neurons were left unchanged, and the inhibitory synapses between the inhibitory and the excitatory neurons in the receiver subnetwork were receiver-receiver ¼ 4.65 nS. The extra tonic excitation that strengthened to Dgloc inh-ex the inhibitory receiver neurons receive from background activity through the projections from the sender subnetwork was compensated by strengthening net-receiver ¼ 9.4 nS. Under these global inhibition to these cells to Dginh-loc inh conditions the system is sufficiently balanced to prevent correlated inputs in the sender subnetwork from modulating the firing pattern of the excitatory receiver subnetwork. To allow propagation, the balance between the excitatory and the inhibitory signal was modified by decreasing the gain of the inhibitory receiver neurons. In integrate-and-fire neurons such a gain change is equivalent to reducing the strength of all synapses by 85% in the case of symmetric gain changes and by 30% in the asymmetric case in which only the response amplitude to excitatory inputs is altered. These values were chosen to minimize similarity values in the gated-off state (Fig. 4a). To compensate for cell loss (Fig. 4c), we calculated the decrease of overall synaptic strength and redistributed the difference equally among the remaining synapses in the pathway. The gating mechanism can function with only 20 inhibitory receiver neurons when we strengthen their synapses threefold, to 15 nS (approximately the same strength as two globally inhibitory neurons), and allow them to fire at rates of up to 600 Hz (Fig. 4d). Two input signals. To propagate and control an additional signal to the receiver region, a second inhibitory receiver subnetwork is necessary. Some of the globally acting inhibitory neurons in the receiver region can be recruited as locally inhibitory for that purpose by generating a new local architecture for them. An additional set of Poisson input spike trains is generated and connected synaptically to both the shared excitatory receiver subnetwork and the new inhibitory receiver subnetwork. As before, the synapses of the Poisson population are tuned to drive their respective target subnetworks in the absence of additional correlated signal input. To balance two active signals at the same time, some of the synapses within the network must also be retuned. See Table 1 for a complete listing of all modified synapses. Pathologies. We chose two ways to disrupt the detailed balance mechanism (Table 1). First, we introduced a deficiency in the locally inhibitory neurons, including those of the inhibitory receiver subnetwork, by decreasing their local ¼ 0.6 nS. Second, we induced a hyperexcitability of synaptic strengths to Dginh the receiver region by increasing all the excitatory synapses from the rest of net-receiver ¼ 1.28 nS and the network onto the receiver neurons to Dgex-ex net-receiver ¼ 1.28 nS. These two manipulations are independent and can Dgex-loc inh be combined without retuning. Multiple input signals to a single cell. In the later part of the paper, we modeled the gating of multiple signals in a single integrate-and-fire cell that receives 800 excitatory and 200 inhibitory synapses modeled as Poisson processes with temporally changing spiking probabilities. To avoid unrealistically high membrane currents when many inputs arrive at the cell, synaptic strengths were tuned down to gex ¼ 0.014 nS and Dginh ¼ 0.044 nS with postsynaptic potentials of 0.13 mV and
0.22 mV at Vrest, respectively. When

490

we drove each afferent with more than one signal, the overlap (effectively a summation of spiking probabilities for each synapse) demanded a rescaling of the input rates to a dynamic signal range between 0 and 150 Hz. The following procedure is used to compute the inhibitory gain factors needed to gate ‘on’ one signal among many. First, to specify which signal is connected to which inhibitory afferent, we define an N  M matrix B with Bia ¼ 1 if inhibitory afferent i receives signal a and BP ia ¼ 0 if it does not. The total inhibitory input due to signal a is then Ca ¼ iBia. We next choose a particular signal, say signal 1, to gate on. We do this by setting the gains for all the inhibitory neurons receiving signal 1—that is, all neurons with Bi1 ¼ 1—to zero. We then adjust the firing rates (gains) of the remaining inhibitory afferents to compensate for these missing afferents for all other signals (missing because their gains are zero). If n afferents receive signal 1, we define an (N – n)  (M – 1) matrix B˜, which is just B with signal 1 and all of the afferents connected to signal 1 removed. Define C˜ to be a vector with M – 1 components given by C˜a ¼ Ca+1 for a ¼ 1, 2, y, M – 1. The inhibition missing because the afferents receiving signal 1 have been turned off can be replaced if the firing rates of the afferents not receiving signal 1 are multiplied by a row vector of gain factors a such that aB˜ ¼ C˜. This equation is ‘solved’, in the sense of minimizing the square of the difference between the two sides summed over a, by setting a ¼ pinv(B˜)C˜, where pinv(B˜) is the pseudoinverse of B˜. The gains of all inhibitory afferents receiving signal 1, by contrast, are set to zero. This determines the complete set of gains used to gate signal 1 on and leave all other signals off. To gate signal 1 off, all the gains are set to 1. A similar procedure is used for any other signal that we wish to gate. Response properties of the balance mechanism. To supply a signal to the network, we generated Poisson input spike trains with a firing rate r0(t) as a source of input to the network. Each input spike generated by that group increased the excitatory synaptic conductances in neurons of the sender region by gex - gex + Dg0. The synaptic strength Dg0 was tuned so that the firing rates of the sender neurons reproduce the input signal, that is, they track the input firing rate r0(t). We used a correlation measure12,14 to determine how similar the firing rates in the receiver region were to the input. To do this, we calculated the population firing rate r(t) in 5-ms bins by counting spikes and also determined its time-averaged value r¯. The correlation is then CðtÞ ¼

hðr0 ðtÞ
r0 Þðrðt + tÞ
r Þsr0 it hðr0 ðtÞ
r0 Þðr0 ðt + tÞ
r0 Þsr it

where the brackets denote an average over time, r0(t) and r¯0 are the firing rate and average for the input, and sr and sr0 are the s.d. values of the corresponding firing rates. We used the activity of the input as a reference rather than the sender subnetwork to distinguish signal transmission from propagation of fluctuations arising in the sender. Signal propagation between subnetworks is then characterized by reporting the maximum value (over t) of C(t), which we call the similarity14. For the analysis of multisignal gating in a single integrate-and-fire cell, we used a similarity measure that was not normalized by sr to avoid overestimating the quality of the output signal in cases when the output firing rate was greatly diminished. ACKNOWLEDGMENTS The idea of detailed balance was originally suggested to us by G. Turrigiano. Research supported by the US National Science Foundation (IBN-0235463), the Swartz Foundation, the Patterson Trust Fellowship Program in Brain Circuitry and a US National Institutes of Health (NIH) Director’s Pioneer Award, part of the NIH Roadmap for Medical Research, through grant number 5-DP1OD114-02. Thanks to J. Peelle, M. Schiff, P. Jercog and R. Yuste for suggestions. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/

1. Shu, Y., Hasenstaub, A. & McCormick, D.A. Turning on and off recurrent balanced cortical activity. Nature 423, 288–293 (2003). 2. Haider, B., Duque, A., Hasenstaub, A.R. & McCormick, D.A. Neocortical network activity in vivo is generated through a dynamic balance of excitation and inhibition. J. Neurosci. 26, 4535–4545 (2006).

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

© 2009 Nature America, Inc. All rights reserved.

ARTICLES 3. Shadlen, M.N. & Newsome, W.T. Noise, neural codes and cortical organization. Curr. Opin. Neurobiol. 4, 569–579 (1994). 4. Troyer, T.W. & Miller, K.D. Physiological gain leads to high ISI variability in a simple model of a cortical regular spiking cell. Neural Comput. 9, 971–983 (1997). 5. Amit, D.J. & Brunel, N. Model of global spontaneous activity and local structured activity during delay periods in the cerebral cortex. Cereb. Cortex 7, 237–252 (1997). 6. van Vreeswijk, C. & Sompolinsky, H. Chaos in neuronal networks with balanced excitatory and inhibitory activity. Science 274, 1724–1726 (1996). 7. Brunel, N. Dynamics of networks of randomly connected excitatory and inhibitory spiking neurons. J. Physiol. (Paris) 94, 445–463 (2000). 8. Kumar, A., Schrader, S., Aertsen, A. & Rotter, S. The high-conductance state of cortical networks. Neural Comput. 20, 1–43 (2008). 9. Abeles, M. Corticonics: Neural Circuits of the Cerebral Cortex (Cambridge University Press, Cambridge, UK, 1991). 10. Aertsen, A., Diesmann, M. & Gewaltig, M.O. Propagation of synchronous spiking activity in feedforward neural networks. J. Physiol. (Paris) 90, 243–247 (1996). 11. Diesmann, M., Gewaltig, M.O. & Aertsen, A. Stable propagation of synchronous spiking in cortical neural networks. Nature 402, 529–533 (1999). 12. van Rossum, M.C., Turrigiano, G.G. & Nelson, S.B. Fast propagation of firing rates through layered networks of noisy neurons. J. Neurosci. 22, 1956–1966 (2002). 13. Vogels, T.P., Rajan, K. & Abbott, L.F. Neural networks dynamics. Annu. Rev. Neurosci. 28, 357–376 (2005). 14. Vogels, T.P. & Abbott, L.F. Signal propagation and logic gating in networks of integrateand-fire neurons. J. Neurosci. 25, 10786–10795 (2005). 15. Destexhe, A. & Contreras, D. Neuronal computations with stochastic network states. Science 314, 85–90 (2006). 16. Kumar, A., Rotter, S. & Aertsen, A. Conditions for propagating synchronous spiking and asynchronous firing rates in a cortical network model. J. Neurosci. 28, 5268–5280 (2008). 17. Posner, M.I. ed. Cognitive Neuroscience of Attention (Guilford Press, New York, 2004). 18. Germuska, M., Saha, S., Fiala, J. & Barbas, H. Synaptic distinction of laminar-specific prefrontal-temporal pathways in primates. Cereb. Cortex 16, 865–875 (2006). 19. Salinas, E. Context-dependent selection of visuomotor maps. BMC Neurosci. 5, 47–68 (2004). 20. Disney, A.A., Aoki, C. & Hawken, M.J. Gain modulation by nicotine in macaque v1. Neuron 56, 701–713 (2007). 21. Disney, A.A. & Aoki, C. Muscarinic acetylcholine receptors in macaque V1 are most frequently expressed by parvalbumin-immunoreactive neurons. J. Comp. Neurol. 507, 1748–1762 (2008). 22. Disney, A.A., Domakonda, K.V. & Aoki, C. Differential expression of muscarinic acetylcholine receptors across excitatory and inhibitory cells in visual cortical areas V1 and V2 of the macaque monkey. J. Comp. Neurol. 499, 49–63 (2006). 23. Xiang, Z., Huguenard, J.R. & Prince, D.A. Cholinergic switching within neocortical inhibitory networks. Science 281, 985–988 (1998).

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

24. Gil, Z., Connors, B.W. & Amitai, Y. Differential regulation of neocortical synapses by neuromodulators and activity. Neuron 19, 679–686 (1997). 25. Mitchell, J.F., Sundberg, K.A. & Reynolds, J.H. Differential attention-dependent response modulation across cell classes in macaque visual area V4. Neuron 55, 131–141 (2007). 26. 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). 27. Abbott, L.F. Theoretical neuroscience rising. Neuron 60, 489–495 (2008). 28. Anderson, C.H. & Van Essen, D.C. Shifter circuits: a computational strategy for dynamic aspects of visual processing. Proc. Natl. Acad. Sci. USA 84, 6297–6301 (1987). 29. Olshausen, B.A., Anderson, C.H. & Van Essen, D.C. A neurobiological model of visual attention and invariant pattern recognition based on dynamical routing of information. J. Neurosci. 13, 4700–4719 (1993). 30. Pouille, F. & Scanziani, M. Routing of spike series by dynamic circuits in the hippocampus. Nature 429, 717–723 (2004). 31. Okun, M. & Lampl, I. Instantaneous correlation of excitation and inhibition during ongoing and sensory-evoked activities. Nat. Neurosci. 11, 535–537 (2008). 32. Baca, S.M., Marin-Burgin, A., Wagenaar, D.A. & Kristan, W.B. Jr. Widespread inhibition proportional to excitation controls the gain of a leech behavioral circuit. Neuron 57, 276–289 (2008). 33. Song, S., Miller, K.D. & Abbott, L.F. Competitive Hebbian learning through spike-timing dependent synaptic plasticity. Nat. Neurosci. 3, 919–926 (2000). 34. Morrison, A., Aertsen, A. & Diesmann, M. Spike-timing-dependent plasticity in balanced random networks. Neural Comput. 19, 1437–1467 (2007). 35. Lewis, D.A., Hashimoto, T. & Volk, D.W. Cortical inhibitory neurons and schizophrenia. Nat. Rev. Neurosci. 6, 312–324 (2005). 36. Seeman, P. Dopamine receptors and the dopamine hypothesis of schizophrenia. Synapse 1, 133–152 (1987). 37. Moore, H., West, A.R. & Grace, A.A. The regulation of forebrain dopamine transmission: relevance to the pathophysiology and psychopathology of schizophrenia. Biol. Psychiatry 46, 40–55 (1999). 38. Tamminga, C.A. Schizophrenia and glutamatergic transmission. Crit. Rev. Neurobiol. 12, 21–36 (1998). 39. Jackson, M.E., Homayoun, H. & Moghaddam, B. NMDA receptor hypofunction produces concomitant firing rate potentiation and burst activity reduction in the prefrontal cortex. Proc. Natl. Acad. Sci. USA 101, 8467–8472 (2004). 40. Rubenstein, J.L. & Merzenich, M.M. Model of autism: increased ratio of excitation/ inhibition in key neural systems. Genes Brain Behav. 2, 255–267 (2003). 41. Tabuchi, K. et al. A neuroligin-3 mutation implicated in autism increases inhibitory synaptic transmission in mice. Science 318, 71–76 (2007). 42. Aviel, Y., Mehring, C., Abeles, M. & Horn, D. On embedding synfire chains in a balanced network. Neural Comput. 15, 1321–1340 (2003).

491

ARTICLES

Phase-to-rate transformations encode touch in cortical neurons of a scanning sensorimotor system

© 2009 Nature America, Inc. All rights reserved.

John C Curtis1,2,4 & David Kleinfeld2,3 Sensory perception involves the dual challenge of encoding external stimuli and managing the influence of changes in body position that alter the sensory field. To examine mechanisms used to integrate sensory signals elicited by both external stimuli and motor activity, we recorded from rats trained to rhythmically sweep their vibrissa in search of a target. We found a select population of neurons in primary somatosensory cortex that are transiently excited by the confluence of touch by a single vibrissa and the phase of vibrissa motion in the whisk cycle; different units have different preferred phases. This conditional response enables the rodent to estimate object position in a coordinate frame that is normalized to the trajectory of the motor output, as defined by phase in the whisk cycle, rather than angle of the vibrissa relative to the face. The underlying computation is consistent with gating by an inhibitory shunt.

The perception of object location relative to the body depends on tracking sensor position—eyes for seeing or fingers for touching—as much as on the activation of those sensors by features of an object. Over a half century ago, von Holst1 emphasized that one cannot hope to understand sensation without consideration of the effects ‘‘produced on the sensory-receptors by the motor impulses which initiate a muscular movement.’’ von Holst factored the signals required for sensation into three components. One is an afferent signal that originates from environmental influences—for example, light for the case of looking and pressure for the case of touching—and is denoted ex-afference. A second component is an afferent signal that results from activation of sensory receptors by self-motion and is called reafference. The motor-driven sensory input can involve the same receptors that encode external stimuli, as in the case of peripheral reafference, or a separate group of receptors, as in the case of proprioception. A final sensory component may be provided by an efference copy of the motor command; this corresponds to the intended rather than actual motor activation of sensory receptors. The ex-afferent component can interact with one or both motor signals (that is, reafference or efference copy) to produce a perceptually stable representation of the identity and location of external stimuli relative to a changing body configuration. The coexistence and possible interaction of ex-afference, reafference and efference copy signals has been demonstrated from peripheral to thalamocortical levels2. In gaze control, reafferent signals of actual eye position and efference copy of the intended position of gaze3 gate the input to vestibular nuclei as part of the vestibular ocular response4. In the visual system, neurons in cat thalamus5 and primary visual cortex6 respond to visual stimuli (the ex-afferent signal) and to the stimulation of extra-ocular muscle proprioceptors (a reafferent signal). Further,

interactions between ex-afference visual signals and a presumed reafference of eye position have been observed at multiple levels of cortical processing in primates7. Responses to combinations of ex-afference, reafference and efference copy signals lie at the heart of transformations to place sensory input in body-centered coordinates. Yet mechanistic and conceptual understandings of how ex-afference and reafference interact to generate such transformations are lacking. Rats sweep their vibrissae through space with stereotypical rhythmic motions as they locomote and search for objects in their immediate environment. Multiple features of the rat vibrissa system make it an ideal nervous system for studying the interaction of ex-afference and reafference. First, behavioral work has shown that rats can determine the position of an object relative to that of its head through the use of a single moving vibrissa8. This implies that the underlying computation of touch in a head-centered coordinate system depends on the interaction of an ex-afference signal (that is, vibrissa contact) with either a reafference or an efference copy that reports vibrissa position. A substrate for such interactions is provided by anatomical connections among sensory and motor areas, at the levels of brainstem through cortex9,10, that form nested feedback loops11,12. Efferent signals give rise to rhythmic motor activity that results in stereotypical whisking behavior13. This motion in turn generates a robust peripheral reafference that is locked to the phase of the vibrissae in the whisk cycle14 and strongly modulates the output of neurons in vibrissa primary sensory (S1) cortex15–17, with different neurons having different preferred phases15. Recent evidence suggests that reafference and ex-afference signals are communicated along parallel pathways from the brainstem to cortex18,19. Thus ex-afferent and reafferent signals associated with vibrissa-based touch are likely to remain separate until they are allowed to interact in vibrissa S1 cortex.

1Division

of Biological Sciences, 2Computational Neurobiology Graduate Program and 3Department of Physics, University of California, San Diego, California, USA. address: Systems Neurobiology Laboratory, The Salk Institute for Biological Studies, San Diego, California, USA. Correspondence should be addressed to D.K. ([email protected]).

4Present

Received 24 November 2008; accepted 23 January 2009; published online 8 March 2009; doi:10.1038/nn.2283

492

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

© 2009 Nature America, Inc. All rights reserved.

ARTICLES

a

b

c

d

Figure 1 Experimental setups and data acquisition. (a) Free ranging apparatus. The rat cranes from a perch to contact a piezoelectric touch sensor that is held in position by a pneumatic piston and imaged with a high-speed camera. Five seconds after the initial contact the sensor is withdrawn and the rat receives a liquid food reward. Spike signals from stereotrodes in the rat’s cortex and EMG electrodes in the mystacial pad, along with contact, positional and video data, are logged. (b) Body restrained apparatus. The animal is wrapped in a sack that is held in a plastic tube. All other experimental features as in a, except that food is delivered from a port close to the animal’s head. (c) Examples of primary data surrounding contact events, including vibrissa position data, the rectified touch signal and accompanying video frames, the rectified rEMG and lowpass filtered rectified rEMG, and the neuronal activity from both channels of the stereotrode. We further show the spike times from two single units after sorting of the spike data; the bottom plots show the respective waveforms and autocorrelations. (d) The left strip shows the response of unit 1 in c to touch across multiple trials, together with the touch signal. The right strip shows the response during free whisking, together with the concurrent rEMG. The averages relative to contact and to the peak of the rEMG signals are shown at the bottom of the panel.

Here, we test the hypothesis that ex-afference touch signals are modulated by reafference signals associated with sensor motion to form a representation of object location relative to the animal’s body plan. We ask the following questions: first, what is the nature of exafferent vibrissa touch signals encoded by single units in vibrissa S1 cortex? Past results consider only responses that are induced by passive rather than active vibrissa movement. Second, is touch represented in a coordinate system that is matched to the region currently scanned, defined by the phase of the vibrissa in the whisk cycle, or one that spans the full range of vibrissa position? Past results15–17 imply that the reafferent signal encodes phase, which suggests but does not establish that touch is also encoded in terms of phase. Finally, how can known cortical circuits give rise to the observed interaction of ex-afferent and reafferent signals? RESULTS Rats were trained to palpate a sensor with their vibrissae in either a freeranging (Fig. 1a) or a body-constrained (Fig. 1b) behavioral configuration. In both paradigms, whisking was accompanied by large movements of the head, so that contact of a vibrissa with the sensor spanned all possible phases of the whisk cycle (Supplementary Fig. 1 online); there was a small but significant (P o 0.01) excess of touch events at protraction over retraction. Animals that succeeded in this task underwent surgery to implant a microwire head stage20 above vibrissa S1 cortex to record broadband electrical activity. These signals were subsequently sorted into single units21, as verified by the consistency of spike waveforms across instances and autocorrelation

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

functions that decay toward zero at equal time (Fig. 1c). We established the principal vibrissa22 for each of two to four electrodes in S1 cortex and trimmed all but these vibrissae; no systematic differences in whisking or touch were observed that were related to different numbers of intact vibrissae. Further, microwires were implanted into the mystacial pad to record the differential electromyogram (rEMG) of the muscles that drive the vibrissa motion. The rectified rEMG was used to deduce the phase of the vibrissae, denoted f(t), during contact events and during periods of time when rats were coaxed to whisk freely in the air16. These epochs of freewhisking13 allowed the reafferent response to whisking to be assessed (Fig. 1d and Supplementary Fig. 2 online). Videographic imaging was used to deduce the phase and angular position, y(t), of the vibrissae surrounding contact events, and to confirm that only a single vibrissa touched the sensor (Fig. 1c). The resulting spike, rEMG data, videographic data and touch sensor signals were sufficient to calculate the spiking as a function of phase in the whisk cycle during free whisking as well as spiking as a function of both phase and absolute angle during epochs of vibrissa contact (Fig. 1d and Supplementary Fig. 2). The instantaneous phase in the whisk cycle is denoted f(t); this can be expressed as f(t) ¼ [2pfwhiskt – fwhisk]modulo 2p during rhythmic whisking, where fwhisk is the whisking frequency and fwhisk is the preferred phase. A preferred phase of fwhisk ¼ 0 corresponds to the protracted position, fwhisk ¼ ± p is the retracted position, and negative (positive) angles indicate protraction (retraction). The instantaneous angular position and phase are related by y(t) ¼ ymidpoint(t) + Dy(t)
cos[f(t)], when the midpoint, ymidpoint(t), and amplitude, Dy(t), vary only slowly on the timescale of the period of whisking, that is, 1/fwhisk.

493

ARTICLES

e

a

© 2009 Nature America, Inc. All rights reserved.

b

c

f

d

g

Figure 2 Free whisking and exemplar touch responses of all single units. (a–d) Spike rates for rapidly excited (a), slowly inhibited (b), slowly excited (c) and no (d) response. The left column is a histogram of the average spike rate centered on the peaks of rEMG activity, which corresponds to the maximal protraction. Spike rates were fit with Poisson maximum likelihood estimates of a series of complex exponentials to determine the baseline firing rate and the dominant whisking frequency, amplitude and phase of each spike response (green line). The phase for the peak response is defined as fwhisk. The middle column is a histogram of the touch response, smoothed using Bayesian adaptive regression splines (green line). The last column shows the temporal waveform and autocorrelation function for the unit. The uncertainly in all cases is represented by 95% confidence intervals. (e) The spike rate for all units whose spike rate was significantly (P o 0.05) modulated by whisking versus the rate in the absence of whisking; in neither case did the vibrissae contact an object. A fit to the data of Ratewhisking ¼ slope Ratenonwhisking yields slope ¼ 1.05 ± 0.11 (mean ± 2 s.e.m.) for rapidly excited units and slope ¼ 1.03 ± 0.04 for all other units; neither slope is significantly different from unity (line). (f) Average response to touch over all single units in a given class of both whisking and touchsensitive units. The compendium is the percent of total responses across all single units. (g) The laminar distribution of single units that responded to active touch, independent of their whisking-related response (Supplementary Fig. 6).

Single units respond during free whisking as well as touch We recorded 152 single units in the vibrissa S1 cortex of nine rats, a majority of which responded to whisking in free air. No differences were seen between free-ranging (Fig. 1a) versus body-constrained (Fig. 1b) paradigms. Cross-correlations between spiking activity and the rectified rEMG show that individual units tend to spike at specific phases of the whisk cycle (Fig. 2a–d, left column). The presence of units whose spike rate remains phasically modulated in time follows from the narrow distribution of whisking frequencies13 (fwhisk ¼ 8.7 ± 1.3 Hz; mean ± s.d.). In an extension of past work, we found that the spike rate during free whisking epochs appears unchanged from that during periods of negligible mystacial rEMG activity, such as when an animal walks without whisking (Fig. 2e). Thus whisking tends to reorganize the timing of spikes rather than add new spikes, which is reminiscent of the effect of finger taps on the response of neurons in the primary somatosensory area of monkeys23. Most single units responded to active touch (Fig. 2a–d, middle column). Three broad classes of responses emerged based on trialaveraged responses: rapid excitation (Fig. 2a), slow net inhibition (Fig. 2b) similar to that seen under nonwhisking conditions24 and slow net excitation (Fig. 2c). The temporal delineation between rapidly and slowly responding cells was sharp (Supplementary Fig. 3 online), and all classes of neurons contained both narrow and broad spike

494

waveforms25 (Supplementary Fig. 4 online). Rapidly excited units responded to touch with phasic spiking that had a latency to onset of 5 to 9 ms and ranged from 12 to 44 ms in duration. These units had the greatest maximum spike rate (Fig. 2f) and fired about 2% of their action potentials in bursts (Supplementary Fig. 5 online). They were largely confined to the granular and deep infragranular layers (Fig. 2g and Supplementary Fig. 6 online), which is consistent with data gathered from anesthetized animals in which whisking was driven by electrical stimulation of the vibrissa motor nerve18,26. Both categories of slowly responding neurons encode the behavioral task per se (as well as touch) in that their spike rates change as the rat cranes and whisks vigorously as it attempts to touch the sensor (Fig. 2c, middle row). Neurons that exhibited slow excitation upon touch dominated the supragranular and infragranular layers, whereas those with slow inhibitory responses were uniformly distributed among all layers (Supplementary Fig. 6). Phase in the whisk cycle gates the rapid touch response We observed that 20% of the single units were both rapidly excited by touch and modulated by whisking. These units, designated as RE touch/whisking neurons, form the locus of our analysis on the confluence of ex-afferent touch and reafferent whisking signals. Our goal was to test whether the touch responses were modulated by the phase of

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES Figure 3 Examples of the interaction between phase in the whisk cycle, as determined from the rEMG signal, and the response of rapidly excited touch-sensitive single units. (a) Histograms of the spike responses to free whisking in air. The spike rates were fit with sinusoids (green line). (b) Histograms of the touch response, averaged over all events, as a function of time from the measured contact. The spike rates were fit with smooth curves (green line). (c) Histograms of the touch response parsed according to the phase in the whisk cycle upon contact. The phase interval is p/4 radians. The red curves correspond to the interval with the maximum response, and the black curves correspond to a lower response; these curves are scaled versions of smoothed trialaveraged touch responses in b. (d) Plots of the peak values of the touch response from the fits to each of the eight intervals of the touch responses in c; dots designate the example fits from c. The uncertainty represents the 95% confidence interval. A smooth curve through this data defines the phase of maximal touch response, denoted ftouch. (e) Curves derived from sinusoid fits of the pre- and post-touch response and smooth curves of the touch response are shown for the data of unit 3 and plotted on a logarithmic scale. The responses to touch events for whisking phases near f ¼ –(5/8)p are minimal and are close to the trough of the modulation of spiking by whisking. The touch response increases in amplitude for contact events close to the peak of the modulation of spiking by whisking, that is, near f ¼ (3/8)p.

b

c

d

Retraction Protraction

Retraction Protraction Retraction Protraction

a

© 2009 Nature America, Inc. All rights reserved.

e

the vibrissa in the whisk cycle at the time of contact. We illustrate the analysis process in terms of the data for three example units whose preferred phases span the full range of whisking (Fig. 3a–d). First, we fit sinusoids to the rectified rEMG signal surrounding each contact event as a means to model rhythmic whisking and determine the a phase in the whisk cycle at the time of contact (Fig. 3a). Second, a smooth rate function was fit to the event-averaged touch response for each unit (Fig. 3b). Third, individual touch responses were sorted into one of eight phase intervals within the whisk cycle. The average touch response within each phase interval was

Figure 4 Summary of the phase sensitivity and spike rate modulation of rapidly excited touchsensitive single units. (a) Scatter plot of the preferred phase for free whisking versus the preferred phase for touch. Shown are mean values plus 95% confidence intervals for the estimates of either phase. The data are consistent with ftouch ¼ fwhisk (P o 0.001). (b) Touch responses as a function of phase in the whisk cycle (gray), where each curve is normalized to its peak value and centered with respect to fwhisk, the preferred phase for the unit while the rat whisked in air. The solid black curve is the population average, and the dashed curve is the minimum relative spike activity. (c) The maximum spike rates during active touch (Max in inset) versus the average spike rates while whisking in air (Mean in inset). (d) Comparison of the modulation depths (equation (1)) of the spike rates for touch versus free whisking.

NATURE NEUROSCIENCE VOLUME 12

[

b

d

c

NUMBER 4

fit as a scaled version of the previously derived smooth rate function for that neuron (Fig. 3c); we note that the shape of the touch response appeared invariant to amplitude. Finally, the peak amplitude of the

[

APRIL 2009

495

ARTICLES Figure 5 Comparison of the phase dependence versus angular position dependence of the touch response, derived from videographic analysis of vibrissa position, for rapidly excited touchsensitive single units. (a) Histograms of the touch response parsed according to the phase in the whisk cycle upon contact; phase interval is p/4 radians. The red curves correspond to the interval with the maximum response, and the black curves correspond to a lower response. (b) Plots of the peak values of the touch response from the fits to each of the eight intervals of the touch responses in a; dots designate the example fits from a. The uncertainty represents the 95% confidence interval. A smooth curve through this data defines the phase of maximal touch response. (c) Histograms of the touch response parsed according to the angular position of the vibrissa upon contact. The angle relative to the midline of the animal’s head was determined from videographic analysis of vibrissa position (Fig. 1c). The red curves correspond to the interval with the maximum response, and the black curves correspond to a lower response. (d) Plots of the peak values of the touch response from the fits to each of the eight intervals of the touch responses in c; dots designate the example fits from c. (e) Comparison of the modulation depths (equation (1)) of the spike rates for touch versus phase in the whisk cycle. We show the data with phase determined from videography (green) and phase determined from the rEMG (red; Fig. 4d). All points are significantly (P o 0.05) modulated. (f) Comparison of the modulation depths (equation (1)) of the spike rates for touch versus angular position in the whisk cycle. Only 4 of 28 points are significantly (P o 0.05) modulated. (g) The angular amplitude (orange arcs) and midpoint (blue arc) for whisking for each of the RE touch-sensitive single units. The bars denote the 95% range of variation of parameters across all trials.

a

© 2009 Nature America, Inc. All rights reserved.

e

eight fitted functions defined a tuning curve of touch response versus phase in the whisk cycle (Fig. 3d). We observed that the touch response for each unit is strongly modulated by vibrissa position, such that the response is maximal at or near the preferred phase during free whisking (compare panels in Fig. 3a with those in Fig. 3d); this is highlighted when the spike response is visualized on a logarithmic scale (Fig. 3e). There is no systematic change in the amplitude of whisking as a function of where contact occurs in the whisk cycle (Supplementary Fig. 7 online). The maximum spike rate for the response to touch is about tenfold higher, on average, than the approximately 9-Hz average spike rate during whisking. The strong modulation of the touch response is suggestive of a nonlinear interaction between reafference and ex-afference. In general, RE touch/whisking units showed responses to touch that were tuned to the phase of the whisk cycle (28 of 35 units) (all responses in Supplementary Figs. 8 and 9 online). A summary shows that the phase at the maximal touch response for each unit, denoted as ftouch (Fig. 3d), is statistically equal to the unit’s preferred phase during free whisking; that is, ftouch Efwhisk (Fig. 4a). All values of preferred phase are represented, with a distribution that is biased toward retraction (Fig. 4a, side bars). All tuning curves are relatively broad, with an average half width at half maximum of 0.32p ± 0.05p radians (mean ± s.d.) (Fig. 4b, thick line); this coincides with the p/3 radian width for cosine-shaped tuning curves. We next determined whether the spike rate during free whisking is predictive of the peak rate upon touch. In contrast to naı¨ve expectations, the maximum rate upon touch was nearly independent of the average spike rate during free whisking (P ¼ 0.05)

496

b

c

d

f

g

(Fig. 4c). A related analysis considers the modulation depth of the spike rates (Fig. 4d, inset) Modulation depth Maximum spike rate  Minimum spike rate  Mean spike rate

ð1Þ

The modulation depth of the tuning curve for the touch response was, on average, four times greater than that for the free whisking response. Neither the spike rate nor the modulation depth showed a systematic dependence on preferred phase in the whisk cycle (Supplementary Fig. 10 online). Critically, the modulation depth for touch at different phases in the whisk cycle was statistically independent (P ¼ 0.8) of the modulation of the rate during free whisking (Fig. 4d). Vibrissa angle is a poor modulator of the touch response We revisited the possibility that the spike response upon contact may be a function of angular position, which depends on the midpoint angle and amplitude of the whisk, in addition to phase in the whisk cycle (Fig. 4a,b). Position data were derived from videographic images during contact events (Fig. 1c). As a control, we compared the phase of the vibrissae in the whisk cycle at the time of contact derived from the videographic data (Fig. 5a,b and Supplementary Fig. 8) with the phases derived from the rEMG data. The two sets of data closely match (Supplementary Fig. 11 online). We next considered modulation in the spike rate upon contact as a function of position (Fig. 5c,d) and observed a relatively small and insignificant modulation. The modulation of the touch response by phase in the whisk cycle was relatively high: an average of 1.2 across the 2p radian range of phase (Fig. 5e). In contrast, the modulation of the touch response by position

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES Figure 6 Example of the phase dependence as a function of whisking parameters for a rapidly excited touch-sensitive single unit. (a) The response for unit 3 (Figs. 3 and 5) averaged over all trials, for which fwhisk ¼ 0.45 p. (b,c) The touch response for unit 3 parsed according to the angular position of the vibrissa upon contact and separated into low-amplitude (41 to 121, b) versus large-amplitude (131 to 251, c) whisks, with preferred phases fwhisk ¼ 0.46p and fwhisk ¼ 0.38p, respectively. The black curve is the combined response from a. (d,e) The touch response for unit 3 parsed according to the frequency of whisking and separated into low-frequency (5 to 9Hz, d) versus high-frequency (10 to 13 Hz, e) whisks, with preferred phases fwhisk ¼ 0.37p and fwhisk ¼ 0.40p, respectively.

a

b

We have shown that these two signals are merged in a highly nonlinear manner (Figs. 3 and 4) in vibrissa S1 cortex and that contact is coded with respect to vibrissa phase rather than angular position (Fig. 5). The coding is robust and invariant with respect to changes in whisking parameters (Fig. 6). The representation of contact in a normalized, relative coordinate system that is dynamically generated by the motor trajectory is somewhat similar to the coding of visual stimuli in dynamic, objectedcentered coordinates, as occurs in parietal and premotor cortices in primates7,27. This is in contrast to the static, retinotopic coordinates instantiated by primary visual areas. Coding in phase coordinates implies that there is a common pathway for the reafferent and exafferent signals, or that these signals follow separate pathways with similar adaptation. To the extent that separate pathways are used, as has been suggested18, we propose a neuronal circuit that computes the location of objects within a sensory field (Fig. 7a).

© 2009 Nature America, Inc. All rights reserved.

c

d

e

averaged only 0.4 across the full range of whisking angles and was statistically significant (P o 0.05) in only 14% of the cases (Fig. 5f). As an average over all RE touch/whisking units, the amplitude of whisking varied by 121 between trials, or 70% of the average value of Dy, whereas the midpoint of the region of whisking varied by 61 between trials, or 35% of the average amplitude (Fig. 5g). In summary, RE touch/ whisking units encode touch in terms of phase, which is normalized to the particular amplitude and midpoint, as opposed to angular position, on a given trial. We consider the possibility that the broad tuning curve for the touch response as a function of phase in the whisk cycle (Fig. 4b) results from a preferred phase, ftouch, that is modulated by the amplitude or the frequency of whisking. Under these contingencies, the observed broad curve could be the sum of multiple narrow curves, each with a slightly different value of ftouch. For the example of unit three (Figs. 3 and 5), we observe no difference in either the preferred phase fwhisk or the width of the tuning curve when the dataset is equally divided based on trials with large-amplitude versus small-amplitude whisks (Fig. 6a–c). A similar invariance occurs when the dataset is equally divided based on trials with lower versus higher whisking frequencies (Fig. 6a,d,e), for which time delay could in principle lead to a frequency-dependent phase. Signal-to-noise constraints allowed us to perform this analysis for only 6 of the 28 RE touch/whisking units, yet all showed statistically significant invariance (P o 0.02) with regard to preferred phase and broad tuning (Supplementary Fig. 12 online). DISCUSSION Active sensing by the rat vibrissa system involves two sensory signals: a reafferent signal of motor activity that encodes the phase of the vibrissa in the whisk cycle and an ex-afferent signal that encodes touch (Fig. 2).

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

Origin of the reafferent signal A reafference signal that encodes the phase in the whisking cycle is present at the level of primary sensory neurons in the trigeminal ganglion14,28. In animals that whisk (for example, mice, rats and gerbils), vibrissa follicles are innervated with both deep and superficial nerve endings29. Both innervations are distinct and highly structured, and they terminate in different regions of the trigeminal complex. Notably, animals that do not whisk (for example, cats, guinea pigs and rabbits) either completely lack superficial follicle receptors or the innervation is sparse and largely unstructured. This led to the hypothesis that superficial nerve innervation serves as proprioceptive reafference for vibrissa motion29. The mechanism by which compression of the follicle during movement is transformed into a phase code is unknown, but it could involve adaptation to the range of whisking30. A second issue involves the possibility that the reafference and exafference form separate thalamocortical tracks19. There are four pathways—one that involves posterior medial thalamus and three that involve subdivisions of ventral posterior medial thalamus—that originate from different populations of secondary sensory cells in the trigeminal nuclei31. Recent reports provide evidence (albeit controversial) that the posterior medial thalamic pathway encodes predominantly vibrissa motion18,32, whereas at least one of the ventral posterior medial thalamic pathways encodes predominantly touch18. Model for the interaction of contact and whisking signals A biophysical model for gating of the active touch response must account for four phenomena. First, the ex-afferent touch signals and the reafferent free whisking responses are enhanced at the same phase in the whisk cycle (Fig. 4a). Second, the spike rate amplitudes and modulation depths of the spike rates for touch and free whisking responses are independent (Fig. 4c,d). Third, the modulation of the touch response is much greater than the relatively small modulation of

497

ARTICLES

a

Figure 7 Shunting-inhibition model for the phase sensitivity of active touch. (a) Circuit diagram for the three-compartment neuron that captures the essence of shunting inhibition. The whisking signal is ‘inverted’ by a local inhibitory neuron and acts as a shunt between distal excitatory touch input and the soma. (b) Example results from a numerical simulation of the circuit equation for our model; note the extra spike for ‘touch’ near the preferred phase of fwhisk ¼ 0. (c) Example result for one model neuron. The same analysis tools were used as with the data for RE/touch units (Fig. 4). The left column is a histogram of the average spike rate centered on the peaks of EMG activity, corresponding to protraction. The next two columns are histograms of the touch response, shown as a composite and by phase in the whisk cycle. The right column shows the tuning curve. (d) The trial-averaged maximal spike rates during active touch versus the average spike rates while whisking in air; 0.05 oRGB o 0.8. (e) The maximum spike rate upon contact versus the mean rate; 0 oRGSO o20 (equation (10)). (f) The modulation depth of the spike rate as a function of phase in the whisk cycle versus the modulation with phase in the whisk cycle; 0 oRGSO o20.

b

-

© 2009 Nature America, Inc. All rights reserved.

c

d

e

f

the spike rate by free whisking (Figs. 2d and 4c,d). Fourth, the whisking reafference essentially does not change the background spike rate (Fig. 2e). Summation of the ex-afferent and reafferent inputs, followed by a spike-generating mechanism whose firing rate is a steeply increasing function of input current, is a potential mechanism for the observed phenomena. However, a substantial increase in the slope of spike rate versus input current is unprecedented for cortical neurons33. A related scheme makes use of the summation of signals near threshold33. However, both whisking and touch events ride on a substantial background rate for all of our units (Figs. 2–4). Multiplication of the exafferent touch signal with the reafferent whisking signal is a potential nonlinearity that can strongly modulate the touch response by the phase in the whisk cycle. One expectation for this scheme that is implicit from studies on the multiplication of signals by neurons34–37 is that the amplitude of the touch response should track that of the whisking response. However, in contrast to this expectation, the modulation depth for touch at different phases in the whisk cycle was independent of the modulation of the rate during free whisking (Fig. 4d). We propose that shunting inhibition of a putative touch pathway by a whisking pathway provides a likely circuit to gate the touch response by the phase in the whisk cycle. Shunting inhibition38 also provides a means for one input to modulate the synaptic gain of a second input. A minimal model consists of a neuron with three compartments, each with a leak battery with resistance R and potential EL, arranged so that (i) an active zone has a bias battery, with conductance GB and potential EB, and generates spikes; (ii) a soma receives shunting—that is, GABAA-mediated synaptic input with a battery with conductance GS and inhibitory potential ES, where ES B EL; and (iii) a dendritic compartment receives excitatory synaptic input with a battery with

498

conductance GE and excitatory potential EE (Fig. 7a). The sequential arrangement of the compartments, taken for simplicity to be one electrotonic length apart, allows the inhibitory whisking input to both modulate the background spike rate and gate an excitatory touch input (Figs. 2a and 3). Putative inhibitory neurons that are strongly modulated by whisking but not touch in close proximity to rapidly excited touch cells is consistent with our laminar analysis of different classes of single units (Fig. 2g and Supplementary Fig. 6). Insight into the mechanism of shunting inhibition can be gleaned from a linear analysis of the circuit because the spike rate tracks the membrane potential in the presence of high background activity39. For simplicity, we ignore the bias current and assume that the maximal synaptic conductances are large (Fig. 7a). Contact at the preferred phase in the whisk cycle, that is, f ¼ fwhisk so that RGS ¼ 0, leads to a membrane potential of Vm B (4EL + EE)/5 at the active zone, which exceeds the rest level Vm B EL. In contrast, when contact occurs at f ¼ fwhisk ± p, so that RGS c 1, the excitatory touch input is shunted by the inhibitory whisking input, and the membrane potential falls to Vm B (EL + ES)/2 + EE/(2RGS), which is close to the rest level. The independence of touch and whisking is seen by estimating their modulation depths (equation (1)), with Vm as a surrogate for spike rates

1
EL  ES

! 0 ð2Þ Modulation depthjWhisking C

EL
EL ES 4 which approaches zero when the shunt and leak potentials are equal, and   2 3ðEL  ES Þ + EE + j2ES j 1 Modulation depthjTouch C ! ð3Þ EL ES 5 5 j3EL + ES j which approaches a constant in the same limit. Thus the modulation depth for touch can be both independent of that for whisking and larger, which is consistent with our observations (Fig. 4d).

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

© 2009 Nature America, Inc. All rights reserved.

Numerical analysis of the model (Fig. 7a) with a Hindmarsh-Rose– type mechanism for spike generation and parameters appropriate for cortical neurons40 clearly shows that the response to touch is enhanced at the preferred phase in the whisk cycle (Fig. 7b,c). Simulations with different bias conductances show that whisking has only a marginal effect on the spike rate (Fig. 7d). Simulations with different shunt conductances show that the amplitude of the touch response is independent of the amplitude of the free whisking spike rate (Fig. 7e), and that modulation in the spike rate by changes in vibrissa position is both much greater than and independent of the modulation during free whisking (Fig. 7f). The proposed circuit can, in principle, be confirmed or refuted by recording the intracellular potential from layer 4 spiny stellate or star pyramidal cells in rats that are trained to whisk17,41. A combination of ion blockers and voltage clamping should reveal whether the whisking response is mediated by inhibitory input. Spike rates and angular resolution The average ongoing rate of spiking hovers around 9 Hz (Fig. 2e), and the average modulation of the spike rate by whisking per se is ± 2 spikes per second. This corresponds to B0.4 spikes per whisk, on top of a fluctuating background of B1 spike per whisk. We estimate that the output from B200 neurons must thus be summed to achieve a resolution of p/3 radians, as set by the tuning curve (Fig. 4b), to specify the phase in the whisk cycle on a single trial basis. Decoding schemes that make use of the absence of a response at nonpreferred phases may lower this estimate. On the other hand, the consensus view of the spike rate of neurons in vibrissa S1 cortex is evolving, with evidence from intracellular studies that the ongoing rates for many neurons may lie closer to 1Hz than 10Hz42,43. A lower average rate would increase the variability and increase the estimate. With regard to contact-induced spikes, active touch leads to an average, integrated response of 2 spikes per contact within a window of B20 ms (Fig. 3b); this further coincides with the time spent in each resolvable phase interval of p/3 radians (Fig. 4b). The additional spikes generated by active touch substantially exceed the B0.4 spikes generated by whisking alone and should be sensed with high fidelity. Resolution at a scale much finer than p/3 radians may be achieved by averaging the responses from multiple neurons. Finally, for the average whisking range of ±171 (Fig. 5g), the corresponding angular resolution is B51, which approximates the typical threshold for bilateral perceptual acuity with a vibrissa44. Relation to directional tuning Directional tuning is a common metric used to quantify neuronal response of vibrissa units in the anesthetized animal45. It measures the bias in the activity of neurons as a vibrissa is deflected in different directions. Notably, directional tuning forms a fine-scale map within a cortical column46. Directional tuning may be derived from an asymmetry in the phase preference of a neuron. With phase tuning for contact defined as T(f – ftouch) (Figs. 3d, 4b and 6), the directional tuning along the anteriorposterior axis is R0 2 p dffTðf  ftouch Þgodd Dðftouch Þ ¼ R 0 p dffTðf  ftouch Þgeven where odd and even refer to the odd and even parts of the function. The directional tuning is double-valued over the whisk cycle, so that the phase preference of a neuron cannot be uniquely determined from its directional preference. Nonetheless, to the extent that active and passive touch lead to neuronal responses with similar directional preference,

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

neurons with different preferred phases in the whisk cycle are expected to conform to the map for directional tuning. METHODS Training and behavior. We successfully trained nine female Long-Evans rats (Charles River), 270 to 300 g initial weight, to whisk against a piezoelectric sensor (DT1–028K; Measurement Specialties Inc.) in return for a liquid food reward (0.2 ml per trial; LD–100; PMI Feeds). We used two behavioral paradigms. In the ‘free ranging’ paradigm (Fig. 1a), we trained unconstrained animals to perch on the edge of a platform and crane their necks to gain access to the sensor. Each trial was initiated when the rat first contacted the sensor. We collected video images at a frame rate of B100 frames per second while the rat palpated the sensor to confirm that the longest vibrissa touched the sensor. After an approximately 3-s period of palpation, the trial was terminated by removal of the sensor and concomitant pumping of the liquid reward to a nearby well on the platform. The sensor remained retracted for 5s, then was restored to its previous position so that a new trial could begin. In the ‘body constrained’ paradigm (Fig. 1b), we placed the animals in a sack and held them in a tube within proximity of a sensor. A trial began when an animal craned and initially touched the sensor and, as above, was terminated after a 3-s period of touch events. Successful learning of either of the above behaviors took about two weeks. Once training was completed, a small chamber that contained an array of 2 to 4 stereotrodes was fit over the vibrissa area of parietal cortex and secured to the skull with screws and dental acrylic20. We individually advanced the stereotrodes through the dura into cortex with a vacuum insertion technique that prevented damage to the upper layers20. We threaded fine wires into the left and right mystacial pads to record the EMG13. The care and experimental manipulation of our animals were in strict accord with guidelines from the US National Institutes of Health and have been reviewed and approved by the Institutional Animal Care and Use Committee of the University of California, San Diego. After several days of post-operative recovery, we briefly anesthetized the animals and individually stimulated each contralateral vibrissa with a brief air puff 47 in order to determine the principal vibrissa response for each stereotrode. The designation of the principal vibrissa was based on the amplitude and latency of the stimulus-locked spikes48. Once we determined the principal vibrissae across the full complement of stereotrodes, we trimmed all other vibrissae at B1 mm from the surface of the skin. The rats were returned to their behavioral setup and invariably performed the task with the single, longest vibrissa. We then acquired spiking data with the electrode that had this vibrissa as its principal vibrissa; the electrode was lowered at the start of each recording session by 80 mm, or until single unit spikes were detected. Once this electrode had been lowered through the full depth of cortex, we trimmed the longest vibrissa and proceeded to take data from the next longest vibrissa, and so forth. Estimation of electrode depth. We exploited the stereotypic form of the radial current source density to identify the lamina of each recording. After all electrodes were lowered through the cortex and data collection was completed, the rat was anesthetized with 5% (w/v) halothane. All electrodes were fully retracted, then lowered in increments of 80 mm while air puffs were delivered at 1.3 puffs per second to passively stimulate all of the vibrissae. We recorded the local field potential at each depth as an average over 100 air puffs16. The second spatial derivative was then calculated across all of the averaged local field potential responses to produce an estimate of the one-dimensional current source density profile for each electrode. Current sinks corresponded to the afferent inputs in layers 4 and 6A; these calibration data allowed us to specify the lamella and depth of every record (Fig. 2g and Supplementary Fig. 6). Data acquisition. Continuous time series from the cortical and EMG microwires were band-pass filtered from 0.35Hz (1 pole) to 10kHz (6 poles), and the piezoelectric sensor was band-pass filtered from 4 kHz (2 poles) to 8 kHz (2 poles). We sampled all data at 32 kHz and stored blocks of approximately 3 s in duration that incorporated each epoch of touch on computer disk, together with the time-locked video images. We obtained additional spike-train records, 10 s in length, as animals were coaxed to whisk in air without contact by placing their home cage just out of reach16. We digitally high-pass filtered the

499

ARTICLES

© 2009 Nature America, Inc. All rights reserved.

broadband cortical and EMG signals at 300 Hz (4 poles). Pairs of EMG signals that spanned the mystacial pad were subtracted to form the rEMG13. We collected high-speed videography (ES310 charge coupled device camera; Kodak, Inc.) acquired at 100 frames per second during trials where rats contacted the touch sensor with their vibrissae. Synchronization between video and electrophysiological data acquisition was accomplished through a hardware trigger on the Real-Time System Integration Bus (National Instruments). We calculated the angular position of the principal vibrissa on a frame-by-frame basis as the angle between one straight line that followed the midline of the snout and another straight line that followed the first 6 to 10 mm of the vibrissa shaft (Fig. 1c). The lines were drawn manually, for each frame, with the aid of a Matlab (The Mathworks)-based graphical user interface. Data analysis. An offline non-Gaussian cluster analysis algorithm21 was used to isolate spikes from an apparent common source in each cortical signal. We characterized modulation of the spike rate of single units during epochs of rhythmic whisking in air, which occurred before and after contact trials, by cross-correlating rEMG peak times with spike times. This method allowed us to normalize the correlation in terms of spike rate. We first band-pass filtered the rectified rEMG from 3 to 22 Hz and set an appropriate threshold for the signal as a means to isolate the interval that surrounded the peak of the waveform, then calculated the center of mass in these intervals to obtain a point process that represents the peaks of the rEMG. We then shifted this time series by 20 ms to account for the measured time delay between the onset of muscle activity, as measured by the rEMG, and movement of the vibrissae13. The resultant cross-correlation corresponds to the rEMG-triggered average spiking rate (Fig. 2a–d, gray histograms in left column, and Fig. 4a). The sinusoidal nature of the cross-correlation was characterized by Poisson-distributed maximum likelihood estimates (MLE) of the mean spike counts49 for a series of complex exponential functions that spanned the frequency range of 5 to 20 Hz, plus a constant term (function glmfit in Matlab with the log-link function). The frequency of the modulation was defined as the estimate with the highest likelihood among all of the complex exponential estimates in the series. We calculated the phase and amplitude of each response from the real and imaginary parts of the estimate. Finally, we normalized the cross-correlations in terms of spike rates by multiplying the spike counts in each bin by the width of the bin (2 ms) and dividing by the number of rEMG peaks in the average (Fig. 2a–d, gray histograms in left column, and Fig. 3a). We used sampling distributions of maximum likelihood estimators to construct 95% confidence intervals of the mean spike rate and parameter estimates. We estimated touch responses from contact-triggered averages, either across all trials (Fig. 2a–d, gray histograms in middle column, and Fig. 3b) or first parsed according to the phase of whisking at the time of contact (Fig. 3b, gray histograms). Smoothed values for all estimates made use of the Poissondistributed Bayesian adaptive regression splines nonparametric smoothing algorithm50 (http://lib.stat.cmu.edu/Bkass/bars). We determined the phases of touch events within the whisk cycle by fitting a series of complex exponentials to the band-pass filtered rEMG or to splineinterpolated videographic traces, centered in a 200-ms window that surrounded the time of vibrissa touch. The procedure was as described above for spike events, except that we now used a Gaussian- rather than Poisson-distributed MLE, as the rEMG data are a continuous function rather than a point process. We quantified goodness of fit by calculating the ratio of the amplitude of the fit relative to the r.m.s. residual of the fit; we discarded contact events with ratios less than two. To assess the touch responses as a function of phase in the whisk cycle, cycles in which touch occurred were first divided into eight intervals of p/4 radians. We then binned touch responses according to the interval in which the touch events occurred to produce a set of eight histograms for each single unit (Figs. 3c and 5a,e). To ensure that our results were statistically reliable across the full range of phase intervals, we excluded sessions with less than eight touch events in any phase interval. We then modeled touch responses for each whisk cycle phase interval with a Poisson-distributed MLE as scaled versions of the overall touch response, computed as described above (Fig. 3c–e, red and black curves). We used the peak amplitudes for each of the best fits to construct the tuning curve for the single unit (Figs. 3d and 5b). We calculated confidence intervals (95%) for mean spike rates from sampling distributions of maximum

500

likelihood estimators. A similar procedure was followed to assess the touch responses as a function of angle in the whisk cycle, for which videographic data were used to determine vibrissa position relative to the midline (Fig. 1c). Model. The circuit model (Fig. 7a) consists of three compartments with equal membrane capacitances, C, and resistances, R, that are joined by a resistance of R so that the compartments are one electrotonic length apart in the absence of synaptic input. We define EL, EB, ES and EE as the reversal potentials for the leak, excitatory bias, inhibitory synaptic shunt and excitatory synaptic currents, respectively, GL as the fixed leak conductance, GB as the conductance of a slowly varying bias current, and GS and GE as the conductances for the vibrissa-driven inhibitory shunt and touch-driven excitation, respectively. For constant values of the conductances, the steady state subthreshold voltage of the active zone, denoted Vm, is given by: Vm ¼ ð8+2RGS +4RGE +RGS RGE ÞEL +RGB ð5+2RGS +3RGE +RGS RGE ÞEB +RGS ð2+RGE ÞES +RGE EE 8+5RGB +2ð2+RGB ÞRGS +ð5+3RGB ÞRGE +ð2+RGB ÞRGS RGE

ð4Þ The full dynamics are found by solving five equations, which include a thirdorder Hindmarsh-Rose system to generate spikes in a cortical cell with adaptation, and additive band-limited Gaussian noise to approximate the variability in synaptic arrival time. We have dVm 1 ¼ fða+bVm +cVm2 ÞðENa  Vm Þ+ dt tm RGR ðEK  Vm ÞU+RGA ðEK  Vm ÞW+ RGB ðEB  Vm Þ+ðVs  Vm Þ+Inoise g

ð5Þ

dU 1 ¼ fU+eVm +f +gðEKA  Vm Þ2 g dt tU

ð6Þ

dW 1 fW+hðE1  Vm ÞðE2  Vm Þg ¼ dt tW

ð7Þ

dVS 1 ¼ fðES  VS ÞRGS +ðEL  VS ÞRGL +ðVm  VS Þ+ðVE  VS Þg dt tm dVE 1 ¼ fðEE  VE ÞRGE +ðEL  VE ÞRGL +ðVS  VE Þg dt tm

ð8Þ ð9Þ

where the leak term RGL(EL –Vm) in the dynamics for Vm is subsumed in the active currents, and tm ¼ 1 ms, tU ¼ 1 ms, tW ¼ 99 ms, ENa ¼ 48 mV, EK ¼ –95 mV, EKA ¼ –38 mV, ES ¼ EL ¼ –74 mV, EB ¼ EE ¼ +10 mV, E1 ¼ –75.4 mV, E2 ¼ –69 mV, RGR ¼ 26, RGA ¼ 13, RGB ¼ 0.2 (ranges 0.05 to 0.8), RGL ¼ 1, a ¼ 17.8, b ¼ 0.476 mV–1, c ¼ 3.38  103 mV–2, e ¼ 1.3  102, f ¼ 0.8, g ¼ 3.3  104 and h ¼ 1.1  103 in our simulations. The noise current has an r.m.s. value of 2.0 mV; the shunting inhibitory conductance was of the form RGS ¼

RGSO f1+ cosð2pfwhisk t  fwhisk Þg 2

ð10Þ

where RGSO ¼ 10 (ranges 0 to 20) and fwhisk ¼ 9Hz; and the excitatory touch conductance was of the form 2 2 RGE ¼ RGEO eðttEO Þ =2tEO

ð11Þ

where RGEO ¼ 40, tEO ¼ 20 ms and tEO is a random variable with a mean of 0.25 s that marks touch events. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank S.B. Mehta for assistance with spike sorting, E.N. Brown and R.E. Kass for instruction on spike-train analysis, G.A. White for electronics support, E. Ahissar, W. Denk, M. Deschenes, M.E. Diamond, A.L. Fairhall, D.N. Hill and T.J. Sejnowski for relevant discussions, D. Matthews for reading of the manuscript, and the US National Institutes of Health (NS051177), the US National Science Foundation (IGERT) and the US/Israel Binational Foundation (2003222) for financial support.

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

© 2009 Nature America, Inc. All rights reserved.

Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. von Holst, E. Relations between the central nervous system and the peripheral organ. Br. J. Anim. Behav. 2, 89–94 (1954). 2. Cullen, K.E. Sensory signals during active versus passive movement. Curr. Opin. Neurobiol. 14, 698–706 (2004). 3. Sommer, M.A. & Wurtz, R.H. Influence of the thalamus on spatial visual processing in frontal cortex. Nature 444, 374–377 (2006). 4. Roy, J.E. & Cullen, K.E. Dissociating self-generated from passively applied head motion: neural mechanisms in the vestibular nuclei. J. Neurosci. 24, 2102–2111 (2004). 5. Lal, R. & Friedlander, M.J. Effect of passive eye position changes on retinogeniculate transmission in the cat. J. Neurophysiol. 63, 502–522 (1990). 6. Ashton, J.A., Boddy, A. & Donaldson, I.M. Directional selectivity in the responses of units in cat primary visual cortex to passive eye movement. Neuroscience 13, 653–662 (1984). 7. Andersen, R.A., Snyder, L.H., Li, C.S. & Stricanne, B. Coordinate transformations in the representation of spatial information. Curr. Opin. Neurobiol. 3, 171–176 (1993). 8. Mehta, S.B., Whitmer, D., Figueroa, R., Williams, B.A. & Kleinfeld, D. Active spatial perception in the vibrissa scanning sensorimotor system. PLoS Biol. 5, e15 (2007). 9. Deschenes, M., Timofeeva, E., Lavalle´e, P. & Dufresne, E. The vibrissal system as a model of thalamic operations. Prog. Brain Res. 149, 31–40 (2005). 10. Fox, K. Barrel Cortex 14–48. (Cambridge University Press, Cambridge, 2008). 11. Kleinfeld, D., Berg, R.W. & O’Connor, S.M. Anatomical loops and their electrical dynamics in relation to whisking by rat. Somatosens. Mot. Res. 16, 69–88 (1999). 12. Ahissar, E. & Kleinfeld, D. Closed loop neuronal computations: focus on vibrissa somatosensation in rat. Cereb. Cortex 13, 53–61 (2003). 13. Berg, R.W. & Kleinfeld, D. Rhythmic whisking by rat: retraction as well as protraction of the vibrissae is under active muscular control. J. Neurophysiol. 89, 104–117 (2003). 14. Szwed, M., Bagdasarian, K. & Ahissar, E. Coding of vibrissal active touch. Neuron 40, 621–630 (2003). 15. Fee, M.S., Mitra, P.P. & Kleinfeld, D. Central versus peripheral determinates of patterned spike activity in rat vibrissa cortex during whisking. J. Neurophysiol. 78, 1144–1149 (1997). 16. Ganguly, K. & Kleinfeld, D. Goal-directed whisking behavior increases phase-locking between vibrissa movement and electrical activity in primary sensory cortex in rat. Proc. Natl. Acad. Sci. USA. 101, 12348–12353 (2004). 17. Crochet, S. & Petersen, C.C.H. Correlating membrane potential with behaviour using whole-cell recordings from barrel cortex of awake mice. Nat. Neurosci. 9, 608–609 (2006). 18. Yu, C., Derdikman, D., Haidarliu, S. & Ahissar, E. Parallel thalamic pathways for whisking and touch signals in the rat. PLoS Biol. 4, e124 (2006). 19. Pierret, T., Lavallee, P. & Deschenes, M. Parallel streams for the relay of vibrissal information through thalamic barreloids. J. Neurosci. 20, 7455–7462 (2000). 20. Venkatachalam, S., Fee, M.S. & Kleinfeld, D. Ultra-miniature headstage with 6-channel drive and vacuum-assisted micro-wire implantation for chronic recording from neocortex. J. Neurosci. Methods 90, 37–46 (1999). 21. Fee, M.S., Mitra, P.P. & Kleinfeld, D. Automatic sorting of multiple unit neuronal signals in the presence of anisotropic and non-Gaussian variability. J. Neurosci. Methods 69, 175–188 (1996). 22. Simons, D.J. Response properties of vibrissal units in rat S1 somatosensory neocortex. J. Neurophysiol. 41, 798–820 (1978). 23. Mountcastle, V.B., Talbot, W.H., Sakata, H. & Hyvarinen, J. Cortical neuronal mechanisms in flutter-vibration studied in unanesthetized monkeys, neuronal periodicity and frequency discrimination. J. Neurophysiol. 32, 452–484 (1968). 24. Krupa, D.J., Wiest, M.C., Shuler, M.G., Laubach, M. & Nicolelis, M.A. Layer-specific somatosensory cortical activation during active tactile discrimination. Science 304, 1989–1992 (2004). 25. Bartho, P. et al. Identification of neocortical principal cells and interneurons by extracellular features. J. Neurophysiol. 92, 600–608 (2004).

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

26. Derdikman, D. et al. Layer-specific touch-dependent facilitation and depression in the somatosensory cortex during active whisking. J. Neurosci. 26, 9538–9547 (2006). 27. Olson, C.R. Brain representation of object-centered space in monkeys and humans. Annu. Rev. Neurosci. 26, 331–354 (2003). 28. Leiser, S.C. & Moxon, K.A. Responses of trigeminal ganglion neurons during natural whisking behaviors in the awake rat. Neuron 53, 117–133 (2007). 29. Arvidsson, J. & Rice, F.L. Central projections of primary sensory neurons innervating different parts of the vibrissae follicles and intervibrissal skin on the mystacial pad of the rat. J. Comp. Neurol. 309, 1–16 (1991). 30. Maravall, M., Petersen, R.S., Fairhall, A.L., Arabzadeh, E. & Diamond, M.E. Shifts in coding properties and maintenance of information transmission during adaptation in barrel cortex. PLoS Biol. 5, e19 (2007). 31. Urbain, N. & Descheˆnes, M. A new thalamic pathway of vibrissal information modulated by the motor cortex. J. Neurosci. 27, 12407–12412 (2007). 32. Masri, R., Bezdudnaya, T., Trageser, J.C. & Keller, A. Encoding of stimulus frequency and sensor motion in the posterior medial thalamic nucleus. J. Neurophysiol. 100, 681–689 (2008). 33. McCormick, D.A., Connors, B.W., Lighthall, J.W. & Prince, D.A. Comparative electrophysiology of pyramidal and sparsely spiny stellate neurons of the neocortex. J. Neurophysiol. 54, 782–806 (1985). 34. Ahrens, K.F., Levine, H., Suhl, H. & Kleinfeld, D. Spectral mixing of rhythmic neuronal signals in sensory cortex. Proc. Natl. Acad. Sci. USA. 99, 15176–15181 (2002). 35. Pena, J.L. & Konishi, M. Robustness of multiplicative processes in auditory spatial tuning. J. Neurosci. 24, 8907–8910 (2004). 36. Andersen, R.A. & Mountcastle, V.B. The influence of angle of gaze upon the excitability of the light sensitive neurons of the posterior parietal cortex. J. Neurosci. 3, 532–548 (1983). 37. Salinas, E. & Abbott, L.F. A model of multiplicative neural responses in parietal cortex. Proc. Natl. Acad. Sci. USA. 93, 11956–11961 (1996). 38. Koch, C., Poggio, T. & Torre, V. Nonlinear interactions in a dendritic tree: localization, timing, and role in information processing. Proc. Natl. Acad. Sci. USA. 80, 2799–2802 (1983). 39. Richardson, M.J., Brunel, N. & Hakim, V. From subthreshold to firing-rate resonance. J. Neurophysiol. 89, 2538–2554 (2003). 40. Rudolph, M., Pospischil, M., Timofeev, I. & Destexhe, A. Inhibition determines membrane potential dynamics and controls action potential generation in awake and sleeping cat cortex. J. Neurosci. 16, 5280–5290 (2007). 41. Lee, A.K., Manns, I.D., Sakmann, B. & Brecht, M. Whole-cell recordings in freely moving rats. Neuron 51, 399–407 (2006). 42. de Kock, C.P., Bruno, R.M., Spors, H. & Sakmann, B. Layer- and cell-type-specific suprathreshold stimulus representation in rat primary somatosensory cortex. J. Physiol. (Lond.) 581, 139–154 (2007). 43. Manns, I.D., Sakmann, B. & Brecht, M. Sub- and suprathreshold receptive field properties of pyramidal neurones in layers 5A and 5B of rat somatosensory barrel cortex. J. Physiol. (Lond.) 556, 601–622 (2004). 44. Knutsen, P.M., Pietr, M. & Ahissar, E. Haptic object localization in the vibrissal system: behavior and performance. J. Neurosci. 26, 8451–8464 (2006). 45. Bruno, R.M., Khatri, V., Land, P.W. & Simons, D.J. Thalamocortical angular tuning domains within individual barrels of rat somatosensory cortex. J. Neurosci. 19, 7603–7616 (2003). 46. Andermann, M.L. & Moore, C.I. A somatotopic map of vibrissa motion direction within a barrel column. Nat. Neurosci. 9, 543–551 (2006). 47. Kleinfeld, D., Sachdev, R.N.S., Merchant, L.M., Jarvis, M.R. & Ebner, F.F. Adaptive filtering of vibrissa input in motor cortex of rat. Neuron 34, 1021–1034 (2002). 48. Armstrong-James, M., Fox, K. & Das-Gupta, A. Flow of excitability within barrel cortex on striking a single vibrissa. J. Neurophysiol. 68, 1345–1358 (1992). 49. Brown, E.N., Frank, L.M., Tang, D., Quirk, M.C. & Wilson, M.A. A statistical paradigm for neural spike train decoding applied to position prediction from ensemble firing patterns of rat hippocampal place cells. J. Neurosci. 18, 7411–7425 (1998). 50. DiMatteo, I., Genovese, C.R. & Kass, R.E. Bayesian curve-fitting with free-knot splines. Biometrika 88, 1055–1071 (2001).

501

ARTICLES

Interval time coding by neurons in the presupplementary and supplementary motor areas

© 2009 Nature America, Inc. All rights reserved.

Akihisa Mita1, Hajime Mushiake1, Keisetsu Shima1, Yoshiya Matsuzaka1 & Jun Tanji1,2 Interval timing is an essential guiding force of behavior. Previous reports have implicated the prefrontal and parietal cortex as being involved in time perception and in temporal decision making. We found that neurons in the medial motor areas, in particular the presupplementary motor area, participate in interval timing in the range of seconds. Monkeys were trained to perform an interval-generation task that required them to determine waiting periods of three different durations. Neuronal activity contributed to the process of retrieving time instructions from visual cues, signaled the initiation of action in a time-selective manner, and developed activity to represent the passage of time. These results specify how medial motor areas take part in initiating actions on the basis of self-generated time estimates.

Interval timing is an essential basis for our sensorimotor function1–3. Most humans who are required to initiate an action after a specific time period in the range of seconds are able to generate the appropriate time internally without a clock or other external time device. This capacity for generating time intervals has also been observed in nonhuman primates and other animals3–6. Previous studies using behavioral tasks that require the anticipation of event timing or decision making in the temporal domain, as well as the perception of elapsed time or discrimination of the duration of sensory signals, have demonstrated the importance of cortico-subcortical loops in mediating temporal coding and regulation of behavior7–9. Distinctions between the involvement of the cerebellum and basal ganglia in controlling different aspects of behavior have been proposed: automatic versus cognitive aspects of behavior10 or precise-timing versus threshold-setting distinctions11. As for the involvement of cortical areas, the posterior parietal and prefrontal cortex were the focus of several reports and are viewed as being critical for the estimation of elapsed time12–14. Time-dependent neuronal activity in these areas has also been related to working memory and to the mechanisms used for decision making15,16. In spite of these reports, we deemed it worthwhile to explore the role of medial motor areas in temporal behavioral control, especially in the internal generation of interval timing. Studies in subhuman primates have indicated that cells in the supplementary motor area (SMA) and presupplementary motor area (preSMA) are involved in controlling self-initiated action17,18 and in the temporal organization of multiple actions19–22. Recent studies in humans have found that a corticostriatal network that includes the preSMA and SMA is activated during explicit attentional modulation of time estimation23 and during internal generation of motor timing23,24. Here, we sought to determine whether the cells in the medial motor areas of an individual that received visual instructions to generate

multiple interval times could retrieve temporal information out of the instructions, whether individual cells signal the initiation of impending action in an interval-selective manner to initiate an action after multiple self-generated intervals and the time courses of cellular activity during a period after the reception of instruction signals and before the initiation of action. We found that cells in medial cortical motor areas, and the preSMA in particular, are important in the process of differential interval generation. RESULTS Two Japanese monkeys were trained to perform an interval-generation task that required them to determine a hold time of three different intervals before initiating a key-release movement (Fig. 1a). Each trial began when the monkey pressed a hold key. One of three instruction lights (LEDs) was illuminated 1–2 s later for 2 s. A yellow light indicated that the minimum hold time before releasing the key was 2 s, whereas the red and green lights indicated that the minimum hold times were 4 and 8 s, respectively. Correct performance of the task required that monkeys release the key at any time exceeding the minimum hold time to obtain a reward of fruit juice after a 0.5-s delay. After carrying out a block of 25 trials with this assignment of cues for the three hold times (standard task), we reversed the assignments of the color cues for the 4- and 8-s hold times (that is, the red cue indicated 8 s and the green cue indicated 4 s). Accomplishment of the reversed task for a block of 25 trials brought back the standard task. At the time when the behavioral task was reversed, the three LEDs flashed simultaneously for 2 s during the intertrial interval, signaling changes in the color hold-time assignment. Both monkeys performed this task with a success rate of over 80% (Supplementary Table 1 online). We constructed a histogram representing the distribution of monkeys’ actual hold times (the interval from the onset of the cue to the release of the key) under each task

1Department of Physiology, Tohoku University School of Medicine, Sendai, Japan. 2Brain Research Institute, Tamagawa University, Machida, Tokyo, Japan. Correspondence should be addressed to J.T. ([email protected]).

Received 2 September 2008; accepted 12 January 2009; published online 1 March 2009; doi:10.1038/nn.2272

502

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

b

a

15%

Mean = 9.239 ms s.d. 952 ms

8-s task 1–2 s 2s

7.000 ms

Key press

20%

Minimum hold time

13.000 ms Mean = 5.073 ms s.d. 555 ms

4-s task

Cue onset

Actual hold time

3.000 ms 40%

9.000 ms Mean = 2.535 ms s.d. 296 ms

2-s task Key release

Reward

Time 1.000 ms

Cue

Minimum hold time

20%

8s Standard task

Reversed 4-s task

4s

7.000 ms Mean = 5.028 ms s.d. 550 ms

© 2009 Nature America, Inc. All rights reserved.

2s

3.000 ms

Figure 1 Task sequence and distribution of waiting periods. (a) Each trial began when the monkey pressed a hold key. After 1 or 2 s, one of three instruction lights (LEDs) was illuminated for 2 s. A yellow light indicated that the minimum waiting period before releasing the key was 2 s, whereas a red or green light indicated that the monkey had to wait at least 4 or 8 s, respectively, before releasing the key. The color waiting period assignment was reversed during the forthcoming block of trials (that is, the red cue indicated 8 s and the green cue indicated 4 s). The visual cue (that is, color of the LED) for the specified waiting task (for example, 2 s) is indicated on the time axis. (b) Frequency histogram of the actual hold time under each of the five task conditions for monkey T. We calculated the mean hold time ± s.d. of the actual hold time (n ¼ 250) and plotted a Gaussian distribution on the basis of the estimated mean ± s.d. Bin width was 1 s. Ordinate, task type; abscissa, actual hold time.

9.000 ms

Selective preparatory activity before motor initiation Reversed 8-s task We next examined neuronal activity that developed during the preparatory period preceding the initiation of the key-release move7.000 ms 13.000 ms ment. We found that 125 preSMA neurons condition (Fig. 1b). We calculated the mean hold time ± s.d. for the increased their activity during the preparatory period before the two monkeys and found that the ratio of the s.d. over the mean was initiation of the key-release movement (Table 1). In a majority of these neurons (n ¼ 112, 90%), the magnitude of activity during the approximately constant, generally showing a scalar property. We examined the activity of 200 preSMA and 119 SMA cells that preparatory period differed depending on the hold time. Among them, were found to be task related (satisfying the criteria described in the 41 neurons showed specific activity exclusively during a particular hold Methods). Here, we focused on neuronal activity reflecting retrieval of time (examples of three neurons showing specific activity during the time-interval information from instruction signals, neuronal activity hold times of 8 s, 4 s and 2 s are presented in Fig. 4). For the remaining leading to interval-selective action initiation and coding of time 71 neurons, the magnitudes of activity during the preparatory period were graded. The magnitudes were greater in 36 neurons using the 2 s, intervals by a population of neurons. 4 s and 8 s order (Fig. 5), whereas in 23 neurons, the magnitudes were greater using the 8 s, 4 s and then 2 s order. Instruction responses selective for the interval time In the SMA, 115 neurons increased their activity during the We found that 75 neurons responded to visual instructions in the preSMA (Table 1). Notably, a majority of these neurons showed preparatory period. In this area, the number of neurons having an selectivity for the instructed interval time. Among them, 26 neurons showed an exclusive relationship to one of the three hold times Table 1 Distribution of the neuronal activity in the preSMA and SMA (examples of three preSMA cells that were selectively active in response classified by selectivity to 8-s, 4-s and 2-s instructions are shown in Fig. 2). The magnitude of PreSMA SMA the instruction responses was graded in another 35 neurons. In 17 neurons, the order of the instruction responses was 2 s, 4 s and then 8 s Instruction Preparatory Instruction Preparatory (Fig. 3), whereas in 10 neurons, the order was 8 s, 4 s and then 2 s. The Response type response response response response selectivity of the neurons that responded exclusively to the 2-s instruction signal may have been attributable to either the 2-s time Time specific 26 41 0 2 interval or the color of the instruction (yellow). 2-s selective 8 12 0 1 To examine the influence of the instruction color or the hold time for 4-s selective 10* 12 0 0 neurons responding to either the 4- or 8-s signals (n ¼ 67), we 8-s selective 8* 17 0 1 performed two-way ANOVA. The majority of neurons (79%, Time graded 35 71 0 55 n ¼ 53/67) showed selectivity to the hold time only, whereas selectivity 2s44s48s 10* 23 0 21 to the instruction color was found in only two neurons (four neurons 8s44s42s 17* 36 0 26 were selective to both). The remaining eight neurons were selective to Others 8* 12 0 8 neither time nor color. Thus, the instruction responses were considerColor selective 6* 0 0 0 ably more selective for the hold time than for the instruction color. The 8* 13 4 58 color-selective neurons (n ¼ 6) were not used for further analysis. In Nonselective contrast with preSMA neurons, only a small fraction of task-related Subtotal 75 125 4 115 SMA neurons responded to the instruction signals (n ¼ 4, 3.3%); none Total 200 119 of the instruction responses of SMA neurons were selective for hold *For these neurons (responding to either 4-s or 8-s signals, n ¼ 67), two-way ANOVA was time or color, as assessed by ANOVA. employed to examine the influence of the instruction color or the hold time. Reversed task

NATURE NEUROSCIENCE VOLUME 12

4s

15%

8s

[

NUMBER 4

[

APRIL 2009

Mean = 9.326 ms s.d. 950 ms

503

b

c

8-s task

20

80

50

4-s task

ARTICLES

a

20

80

50

Figure 2 Three examples of preSMA cells showing instruction responses that were selective for one of the three interval times. (a) Raster displays and spike-density functions illustrating selective neuronal activation during the 8-s task in standard and reversed assignments. (b) Selective neuronal activation during the 4-s task in standard and reversed assignments. (c) Selective neuronal activation during the 2-s task. The displays are aligned according to cue onset.

80

50

Reversed 8-s task

Reversed 4-s task

2-s task

4-s task

8-s task

Reversed 8-s task

© 2009 Nature America, Inc. All rights reserved.

Reversed 4-s task

2-s task

activity of time-nonselective pre-MA neurons expressed similar curve fits for all waiting periods (Fig. 7c,d; the estimated parameters of exponential base and magnitude for individual categories of the neuronal population are summarized in Supplementary Table 3 online). 50 20 80 The same exponential curve fitting was carried out for SMA neurons (Supplementary Figs. 2 and 3 online). We found that the activity of 74 SMA neurons satisfied the criteria for the exponential fit. Among them, 36 neu80 20 50 rons showed selectivity to the three hold-time intervals (Supplementary Table 2). Unlike Cue onset 1s Cue onset 1s Cue onset 1s preSMA neurons, SMA neurons did not have an exclusive relationship with any one of the exclusive relationship to one of the three hold times was smaller (n ¼ 2) hold times. However, 36 neurons showed a graded type of selectivity to than in the preSMA neurons. However, 55 SMA neurons showed the hold time. Notably, for one population of these neurons, the graded selectivity to the hold times. In those neurons, the magnitudes magnitude of activity changes differed greatly depending on the hold were greater for the 2 s, 4 s and 8 s order, whereas the magnitudes were time. However, the exponential base was not dependent on the hold greater for the 8 s, 4 s and 2 s order in another 21 neurons. In 58 period (Supplementary Fig. 3). The exponential base was smaller in neurons, the preparatory activity did not differ with hold time SMA neurons than in preSMA neurons, indicating that SMA neurons initiated activity changes later in the hold period and closer to the onset (Supplementary Fig. 1 online). of movement than preSMA neurons. A similar tendency for low exponential base values was observed for 38 time-nonselective SMA neurons. Population analysis of activity during the hold-time interval In a subsequent analysis, we studied whether neuronal activity during We found that neurons in the preSMA and SMA often had long-lasting activity changes during the hold period. To explore the possibility that the hold period was related to the actual interval timed by the subjects, the activity during this period could provide the basis for coding the that is, monkeys’ actual hold times (the interval from the onset of the interval timing, we attempted to characterize the activity of each cue to the release of the key). For this purpose, we carried out a timeneuron by fitting the time course of activity changes to an exponential dependent regression analysis of model (this provided a better fit than a linear model, see Methods). We neuronal activity to examine its performed an exponential function approximation for the activity of relationship to the actual inter60 each neuron (see Methods). We found that the activity of 110 preSMA val time. We found that ‘timeneurons satisfied the criteria for the exponential fit. Among them, 96 graded neurons’ in the SMA neurons showed selectivity to the three intervals. Of these, the activity and preSMA were significantly of 36 neurons had a selective relation to only one hold time. To perform related to the temporal varia60 population analysis on each category of task-related neurons, we first tions of the actual interval normalized neuronal activity by the maximum and minimum firing time (P o 0.01; Supplementary rates to obtain values between 0 and 1. We then plotted the averaged Fig. 4 online). In contrast, neuspike density functions (Fig. 6). For 60 of the 96 neurons, the activity rons having a specific relation to 60 difference was graded, depending on the hold time (Supplementary Table 2 online). Our data indicated that the activity did not depend on Figure 3 An example of preSMA whether the hold time was instructed with different colors (normal cells exhibiting instruction versus reversed tasks). responses with graded magnitudes We classified each population of neurons on the basis of fitted depending on time intervals. 60 exponential curves (Fig. 7). We fitted the hold-period activity represent- Raster displays and spike-density ing the neuronal population with graded activity in the 8 s, 4 s and 2 s functions illustrate graded order (Fig. 7a) to the degree of exponential decay or buildup. To quan- neuronal activity, which was highest during the 8-s task, titatively characterize the graded nature of this population activity, we moderate during the 4-s task and 60 calculated the exponential base and magnitude (Fig. 7b). PreSMA cells lowest during the 2-s task (8 s, 4 s with higher activities during longer waiting periods showed more dyna- and 2 s). The displays are aligned Cue onset 1s mic changes with a greater exponential base. In contrast, the population according to cue onset. 20

504

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

c

8-s task

20

40

50

20

40

50

40

50

Movement onset

1s

Movement onset

1s

one of the three intervals (time-selective neurons) were not related to the temporal variations of the actual interval time. DISCUSSION Our exploration of the role of medial motor area neurons in interval timing yielded three major findings. First, a majority of preSMA neurons responding to the instruction signal showed selectivity to the interval time rather than to the color of the visual signal. This observation indicates that the preSMA is involved in retrieving information pertaining to the duration of time to structure the forthcoming motor behavior. Second, during the preparatory period preceding the initiation of the key-release movement, a majority of preSMA neurons (and a small part of SMA neurons) showed selective activity depending on the length of the preparatory period. This finding suggests that the preSMA is involved in signaling motor initiation in a time interval– dependent manner. Third, during the period in which the animals were generating a time interval (hold period), neurons in both the preSMA and SMA showed continuous changes in activity. In a majority of cases, the time course of activity changes could be characterized by exponential decay or buildup, suggesting that these neurons contributed to coding the development of time. The abundance of preSMA neurons that were time selective suggests that the preSMA contributes more to the development of time-selective activity than the SMA. Taken together, these results provide a basis for distinguishing the medial motor areas that promote the encoding of time-interval information and make use of time-selective information to initiate actions. We found that preSMA and SMA neurons that were categorized as ‘time graded’ had a substantial relationship to the hold intervals. Thus, a considerable fraction of behavioral-task related neurons are shown to be related to the hold intervals. On the other hand, neurons that were categorized as being ‘time specific’ did not have this relationship. We interpret the results as showing that the time-specific neurons were related to the minimum hold time and were able to feed that information to timegraded neurons. Thus, it seems reasonable to assume that neurons belonging to both categories together contribute to interval timing. Previous studies have emphasized the importance of the basal ganglia and cerebellum, with their cortico-subcortical loops, in the temporal

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

8-s task

20

4-s task

50

2-s task

40

Figure 5 An example of pre-SMA cells showing preparatory neuronal activity with graded magnitudes that depended on time intervals. Raster displays and spike-density functions illustrate neuronal activity, which was highest during the 8-s task, moderate during the 4-s task and lowest during the 2-s task (8 s, 4 s and 2 s). The displays are aligned according to movement onset.

Reversed 4-s task

Reversed 4-s task

20

1s

processing of information and in timing functions7–11,25. As for cortical involvement, neuroimaging studies have demonstrated the involvement of the prefrontal and parietal cortex in time estimation or discrimination tasks6,8,9,26–28. Moreover, damage to the prefrontal cortex results in marked deficits in time estimation29,30. Conversely, the involvement of medial motor areas has also been inferred on the basis of imaging studies24,31 and studies that examined cortical potentials23. Cortical representation of elapsed time has also been studied at the single-neuron level8,12–16,32,33. In Movement onset those studies, however, aspects of time perception or estimation, rather than time production, were the focal points. It is important to recognize that mechanisms for processing temporal aspects of behavior involve diverse aspects of cognitive processes; for example, time perception, interval estimation, temporal decision, time production and the processing of episodic memory3,11,34,35. We designed our experiments to address the neural mechanisms involved in generating specified time intervals, which required animals to produce an appropriate time interval that was close to a minimum waiting period before initiating an action. To perform the timing tasks described here, the subject was required to retrieve temporal information from visual instructions, generate a correct time interval and then initiate a key-release movement without external trigger signals. Our quantitative analysis using exponential curve fitting allowed us to distinguish dynamic neuronal activity in the preSMA from that 40 in the SMA. For the activity of preSMA neurons, the base of the exponential function varied systematically depending on the specified 40 minimum waiting period; this was not observed to the same extent in SMA neurons. Our data suggest that the dynamics of decay and buildup activity in the 40 preSMA, expressed as an adjustable

50

40

Reversed 8-s task

© 2009 Nature America, Inc. All rights reserved.

20

Figure 4 Three examples of pre-SMA cells showing interval-selective preparatory activity before motor initiation. (a) Raster displays and spike-density functions illustrating selective neuronal activation during the 8-s task. The displays are aligned according to the apparent initiation of movement. (b) Selective neuronal activation for the 4-s task. (c) Selective neuronal activation for the 2-s task. The displays are aligned according to movement onset.

40

Reversed 8-s task

a

2-s task

b

4-s task

ARTICLES

40

1s

Movement onset

505

ARTICLES

Normalized activity

c

b n=7

0 Cue onset

d

1 n = 11

0 Cue onset

1 n=8

0 Movement onset 1 n = 24

information as a categorical signal and generating an appropriate interval by parametric coding. In contrast, the contribution of the SMA to this aspect of behavioral control appears to be modest. METHODS

0 Movement onset

1s

parameter for the exponential function, may have contributed to the versatility of time generation to meet varying demands for interval timing. Theoretical studies suggest that scalar properties could be based on a pacemaker-accumulator model5,34,36, a multiple time scale model37 or a memory-trace model as a continuous clock9,38. One recently proposed theoretical model38 describes the sum of cascades of declining exponentials as a possible building block to construct interval times. We propose that a buildup process fits the interval time model equally well as an exponential decay process. Multiple adaptable decay and buildup processes in the preSMA may have an essential role in time generation39. It should be added, however, that the neurons that show the selectivity pattern of 8, 4 and then 2 s for the preparatory response might be showing a change in their activity related to the passage of time, rather than to a particular hold time. Another possible explanation for some of the preSMA activity might be the selectivity related to the time to reward, as previously reported40. The preSMA has been studied extensively and is thought to participate in broad aspects of behavioral control that include the updating of motor plans19, changing from a pre-existing motor plan to another41, selecting effector-independent targeting action42, inhibiting automatic actions43,44, regulating motion sequences19–22,45,46 and switching from automatic to controlled action47. Our findings suggest that the preSMA, with its wealth of inputs from the prefrontal48 and parietal cortex49, is also involved in the cognitive control of interval timing by decoding time

a

Decay type

Behavioral task and recording procedures. The two Japanese monkeys (Macaca fuscata) used here were cared for in accordance with the Guiding Principles for the Care and Use of Laboratory Animals of the US National Institutes of Health. The monkeys were trained to perform an intervalgeneration task, described above. We used conventional electrophysiological techniques to obtain in vivo single-cell recordings from the preSMA and SMA, which were identified by criteria including the properties of neuronal responses and the effects of intracortical microstimulation50. Using electromyography, we recorded the activity of a total of 18 muscles in the forelimb and axis. Although these muscles showed movement-related activity, they showed no consistent changes in activity during the hold time of three different durations. Cortical sulci and recording locations were identified using a magnetic resonance imaging scanner and were verified via histological examination of Nissl-stained brain sections. We also monitored eye positions and velocity using an infrared corneal reflection monitoring system and checked activity in limb and trunk muscles electromyographically. Recordings were obtained from the following muscles: extensor digitorum communis, flexor digitorum profundus, extensor carpi ulnaris and radialis, flexor carpi radialis, biceps and triceps brachii, brachioradialis, deltoideus, sternomastoideus, trapezius, supraspinatus, pectoralis major, thoracic and lumbar paravertebral, iliopsoas and quadriceps. Data analysis. We examined 200 preSMA and 119 SMA neurons, which were defined as being task related by satisfying the following criteria: being analyzed during the performance of more than two cycles of standard and reversed tasks (each included 25 trials) and exhibiting task relevance on the basis of the statistical test described below. The task phase was divided into three periods: a control period (500 ms before the onset of the instruction cue), a 500-ms early hold period (beginning 500 ms after the instruction cue) and the 500-ms late hold period (beginning 1,000 ms before the initiation of the key-release movement). If neuronal activity (discharge rate) during the early or late hold period were significantly different compared with the control period (Wilcoxon

b

Buildup type

506

1

Normalized activity

0 0

c

0 Decay time

Preceding time

Decay type

Buildup type

0 Decay

d

Buildup

1

1

Decay

Exponential base

Buildup

Magnitude 1

activity

Figure 7 Results of exponential curve fitting for population activity including time-specific and time-graded preSMA cell. (a) Plots of hold-period activity (thin lines) calculated for the population of SMA neurons classified as having decaying activity (left) and buildup activity (right) with graded magnitudes in the order of 8 s, 4 s and 2 s. Thick lines denote estimated values with the exponential function. The time 0 for decay type neurons corresponds to the time of the peak activity and the time 0 for buildup type neurons corresponds to the time of peak activity during the pre-movement period. (b) Bar graphs showing the distribution of the exponential base and magnitude resulting from the exponential fit for the population of preSMA neurons classified as having graded magnitudes. (c,d) The data for the population of preSMA neurons classified as nonselective to the hold time are shown. Error bars indicate s.e.m.

Magnitude

Exponential base

1

1

Normalized

© 2009 Nature America, Inc. All rights reserved.

Normalized activity

1

Normalized activity

Normalized activity

a

Figure 6 Time courses for averaged normalized responses of time-specific and time-graded preSMA cell populations exhibiting either decay or buildup activity. (a) Decay activity specific to the 8-s time interval. (b) Buildup activity specific to the 8-s time interval. (c) Decay activity with graded magnitudes in the order of 2 s, 4 s and 8 s. (d) Buildup activity with graded magnitudes in the order of 2 s, 4 s and 8 s. Note that neuronal activity was influenced by the interval time indicated by instruction cues, not by the physical color of the instruction cues.

4s 4 s (reversed) 2s

8s 8 s (reversed)

0 0

0 Decay time

0 Decay

Decay

Buildup

Buildup

Preceding time 2-s task

4-s task

8-s task

2s

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

© 2009 Nature America, Inc. All rights reserved.

signed-ranks test, P o 0.05), then the observed activity was considered to be task related. Note that the majority of task-related neurons in our database showed either decay or buildup activity during the hold period. Although some neurons showed a significant change in activity during both the early and late holding periods (that is, biphasic or sustained activity), these neurons were few in number. To determine whether task-related activity reflected selectivity for the specified waiting period (2-s, 4-s or 8-s holding task), we carried out oneway ANOVAs for each task-related neuron (P o 0.01), followed by a post hoc pair-wise comparison using the Bonferroni test. If neuronal activity during the early or late period differed significantly for only one of the three hold times, we classified the neuron as being a time-specific neuron. If neuronal activity did not differ significantly for the three hold times, the neuron was classified as being a time-nonselective cell. If neuronal activity varied monotonically according to the length of the specified time task (2 s, 4 s and then 8 s, or 8 s, 4 s and then 2 s), then the neuron was classified as being a time-graded cell. Quantitative analysis of the activity during the hold time. To quantitatively examine neuronal activity during the hold time (animals’ actual hold times), we performed exponential curve fitting for the data obtained in each neuron. First, spike counts in 100-ms time frames were normalized against the maximum and minimum spike counts during the waiting period. Neuronal activity was classified as being of decay or buildup type depending on whether the peak of activity occurred during the initial 1 s or the last 1 s of the hold time. Subsequently, neuronal activity was evaluated by fitting to an exponential function, Activity ¼ M
B t, where M is the magnitude, B is the exponential base and t is the decay time or buildup time preceding peak activity with a 0.1-s bin width. To evaluate goodness of fit for the exponential curve fits, we performed an F test for the activity of each neuron. If the exponential curve fit was statistically significant (P o 0.05), we classified the activity as exponentially decaying or building up. These values are summarized in Supplementary Table 3. We also performed a fitting with the use of a linear function model and found that the exponential function model provided a better fit (see Supplementary Fig. 5 online). Note: Supplementary information is available on the Nature Neuroscience website. AUTHOR CONTRIBUTIONS A.M. and K.S. conducted the experiments, H.M. and Y.M. conducted the data analyses, H.M. and J.T. supervised the project and J.T. wrote the manuscript. Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Gibbon, J. Timing and Time Perception. (New York Academy of Sciences, New York, 1984). 2. Gallistel, C.R. The Organization of Learning (Learning, Development, and Conceptual Change). (MIT Press, Cambridge, Massachusetts, 1993). 3. Buhusi, C.V. & Meck, W.H. What makes us tick? Functional and neural mechanisms of interval timing. Nat. Rev. Neurosci. 6, 755–765 (2005). 4. Meck, W.H. Attentional bias between modalities: effect on the internal clock, memory, and decision stages used in animal time discrimination. Ann. NY Acad. Sci. 423, 528–541 (1984). 5. Church, R.M., Meck, W.H. & Gibbon, J. Application of scalar timing theory to individual trials. J. Exp. Psychol. Anim. Behav. Process. 20, 135–155 (1994). 6. Hinton, S.C. & Meck, W.H. The ‘internal clocks’ of circadian and interval timing. Endeavour 21, 3–8 (1997). 7. Rao, S.M., Mayer, A.R. & Harrington, D.L. The evolution of brain activation during temporal processing. Nat. Neurosci. 4, 317–323 (2001). 8. Matell, M.S., Meck, W.H. & Nicolelis, M.A.L. Interval timing and the encoding of signal duration by ensembles of cortical and striatal neurons. Behav. Neurosci. 117, 760–773 (2003). 9. Meck, W.H., Penney, T.B. & Pouthas, V. Cortico-striatal representation of time in animals and humans. Curr. Opin. Neurobiol. 18, 145–152 (2008). 10. Lewis, P.A. & Miall, R.C. Distinct systems for automatic and cognitively controlled time measurement: evidence from neuroimaging. Curr. Opin. Neurobiol. 13, 250–255 (2003). 11. Ivry, R.B. & Spencer, R.M.C. The neural representation of time. Curr. Opin. Neurobiol. 14, 225–232 (2004). 12. Leon, M.I. & Shadlen, M.N. Representation of time by neurons in the posterior parietal cortex of the macaque. Neuron 38, 317–327 (2003). 13. Janssen, P. & Shadlen, M.N. A representation of the hazard rate of elapsed time in macaque area LIP. Nat. Neurosci. 8, 234–241 (2005). 14. Genovesio, A., Tsujimoto, S. & Wise, S.P. Neuronal activity related to elapsed time in prefrontal cortex. J. Neurophysiol. 95, 3281–3285 (2006).

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

15. Sakurai, Y. Working memory for temporal and nontemporal events in monkeys. Learn. Mem. 8, 309–316 (2001). 16. Oshio, K., Chiba, A. & Inase, M. Delay period activity of monkey prefrontal neurones during duration-discrimination task. Eur. J. Neurosci. 23, 2779–2790 (2006). 17. Okano, K. & Tanji, J. Neuronal activities in the primate motor fields of the agranular frontal cortex preceding visually triggered and self-paced movement. Exp. Brain Res. 66, 155–166 (1987). 18. Mushiake, H., Inase, M. & Tanji, J. Neuronal activity in the primate premotor, supplementary, and precentral motor cortex during visually guided and internally determined sequential movements. J. Neurophysiol. 66, 705–718 (1991). 19. Shima, K. et al. Role for cells in the presupplementary motor area in updating motor plans. Proc. Natl. Acad. Sci. USA. 93, 8694–8698 (1996). 20. Shima, K. & Tanji, J. Both supplementary and presupplementary motor areas are crucial for the temporal organization of multiple movements. J. Neurophysiol. 80, 3247–3260 (1998). 21. Nakamura, K., Sakai, K. & Hikosaka, O. Effects of local inactivation of monkey medial frontal cortex in learning of sequential procedures. J. Neurophysiol. 82, 1063–1068 (1999). 22. Tanji, J. Sequential organization of multiple movements: involvement of cortical motor areas. Annu. Rev. Neurosci. 24, 631–651 (2001). 23. Macar, F., Vidal, F. & Casini, L. The supplementary motor area in motor and sensory timing: evidence from slow brain potential changes. Exp. Brain Res. 125, 271–280 (1999). 24. Coull, J.T. et al. Functional anatomy of the attentional modulation of time estimation. Science 303, 1506–1508 (2004). 25. Matell, M.S. & Meck, W.H. Cortico-striatal circuits and interval timing: coincidence detection of oscillatory processes. Brain Res. Cogn. Brain Res. 21, 139–170 (2004). 26. Onoe, H. et al. Cortical networks recruited for time perception: a monkey positron emission tomography (PET) study. Neuroimage 13, 37–45 (2001). 27. Harrington, D.L. et al. Neural representation of interval encoding and decision making. Brain Res. Cogn. Brain Res. 21, 193–205 (2004). 28. Pouthas, V. et al. Neural network involved in time perception: an fMRI study comparing long and short interval estimation. Hum. Brain Mapp. 25, 433–441 (2005). 29. Koch, G. et al. Selective deficit of time perception in a patient with right prefrontal cortex lesion. Neurology 59, 1658–1659 (2002). 30. Jones, C.R.G. et al. The right dorsolateral prefrontal cortex is essential in time reproduction: an investigation with repetitive transcranial magnetic stimulation. Exp. Brain Res. 158, 366–372 (2004). 31. Macar, F., Coull, J. & Vidal, F. The supplementary motor area in motor and perceptual time processing: fMRI studies. Cogn. Process. 7, 89–94 (2006). 32. Lucchetti, C. & Bon, L. Time-modulated neuronal activity in the premotor cortex of macaque monkeys. Exp. Brain Res. 141, 254–260 (2001). 33. Akkal, D. et al. Time predictability modulates pre-supplementary motor area neuronal activity. Neuroreport 15, 1283–1286 (2004). 34. Gibbon, J. et al. Toward a neurobiology of temporal cognition: advances and challenges. Curr. Opin. Neurobiol. 7, 170–184 (1997). 35. Buonomano, D.V. & Karmarkar, U.R. How do we tell time? Neuroscientist 8, 42–51 (2002). 36. Gallistel, C.R. & Gibbon, J. Time, rate and conditioning. Psychol. Rev. 107, 289–344 (2000). 37. Ivry, R.B. & Richardson, T.C. Temporal control and coordination: the multiple timer model. Brain Cogn. 48, 117–132 (2002). 38. Staddon, J.E.R. Interval timing: memory, not a clock. Trends Cogn. Sci. 9, 312–314 (2005). 39. Reutimann, J. et al. Climbing neuronal activity as an event-based cortical representation of time. J. Neurosci. 24, 3295–3303 (2004). 40. Sohn, J.W. & Lee, D. Order-dependent modulation of directional signals in the supplementary and presupplementary motor areas. J. Neurosci. 27, 13655–13666 (2007). 41. Matsuzaka, Y. & Tanji, J. Changing directions of forthcoming arm movements: neuronal activity in the presupplementary and supplementary motor area of monkey cerebral cortex. J. Neurophysiol. 76, 2327–2342 (1996). 42. Fujii, N., Mushiake, H. & Tanji, J. Distribution of eye- and arm-movement-related neuronal activity in the SEF and in the SMA and Pre-SMA of monkeys. J. Neurophysiol. 87, 2158–2166 (2002). 43. Nachev, P. et al. The role of the pre-supplementary motor area in the control of action. Neuroimage 36 (Suppl 2): T155–T163 (2007). 44. Sumner, P. et al. Human medial frontal cortex mediates unconscious inhibition of voluntary action. Neuron 54, 697–711 (2007). 45. Isoda, M. & Tanji, J. Participation of the primate presupplementary motor area in sequencing multiple saccades. J. Neurophysiol. 92, 653–659 (2004). 46. Shima, K. & Tanji, J. Neuronal activity in the supplementary and presupplementary motor areas for temporal organization of multiple movements. J. Neurophysiol. 84, 2148–2160 (2000). 47. Isoda, M. & Hikosaka, O. Switching from automatic to controlled action by monkey medial frontal cortex. Nat. Neurosci. 10, 240–248 (2007). 48. Lu, M.T., Preston, J.B. & Strick, P.L. Interconnections between the prefrontal cortex and the premotor areas in the frontal lobe. J. Comp. Neurol. 341, 375–392 (1994). 49. Tanne´, J., Boussaoud, D., Boyer-Zeller, N. & Rouiller, E.M. Direct visual pathways for reaching movements in the macaque monkey. Neuroreport 7, 267–272 (1995). 50. Matsuzaka, Y., Aizawa, H. & Tanji, J. A motor area rostral to the supplementary motor area (presupplementary motor area) in the monkey: neuronal activity during a learned motor task. J. Neurophysiol. 68, 653–662 (1992).

507

ARTICLES

A neural mechanism of first impressions

© 2009 Nature America, Inc. All rights reserved.

Daniela Schiller1,2, Jonathan B Freeman2,3, Jason P Mitchell4, James S Uleman2 & Elizabeth A Phelps1,2 Evaluating social others requires processing complex information. Nevertheless, we can rapidly form an opinion of an individual during an initial encounter. Moreover, people can vary in these opinions, even though the same information is provided. We investigated the brain mechanisms that give rise to the impressions that are formed on meeting a new person. Neuroimaging revealed that responses in the amygdala and the posterior cingulate cortex (PCC) were stronger while encoding social information that was consistent, relative to inconsistent, with subsequent evaluations. In addition, these responses scaled parametrically with the strength of evaluations. These findings provide evidence for encoding differences on the basis of subsequent evaluations, suggesting that the amygdala and PCC are important for forming first impressions.

Making sense of others in a social interaction is not easy, as each person is often a source of ambiguous and complex information. Despite this, when encountering someone for the first time, we are usually quick to judge whether we like that person or not. Indeed, people make relatively accurate and persistent evaluations on the basis of rapid observations of even less than half a minute1,2. Here we investigated the brain mechanisms giving rise to the impressions that are rapidly formed on meeting a new person. When confronted with multifaceted social information, people will usually not be unanimous in their evaluations. For example, a person may be both smart and lazy. Although these traits have a social valence, the former is considered to be a good quality and the latter to be bad, people also assign their own subjective value to these traits on the basis of their personal preferences. Some evaluators might value intelligence more and care less about laziness, generating a positive impression of that person, whereas an opposite valuation would result in a negative impression overall. Thus, individuals vary in their evaluations despite the fact that the same information is provided. Our objective here was to examine the neural encoding of social information reflecting this subjective valuation (that is, the weight ascribed to each bit of social information) and its correlation with subsequent impressions. Neural regions that are specifically involved in the evaluative process of impression formation should show greater responding while encoding information with higher subjective value, which would be consistent with subsequent evaluations, as compared with information that has a lower subjective value. A similar approach has been used in memory research: neural responses to items that are later remembered are different from items that are forgotten. This difference based on subsequent memory is considered to be evidence for the involvement of a brain region in memory formation and was termed the difference in memory effect (also known as the DM effect)3–5. Analogously, in social evaluation processes, information that is used or disregarded might also be

encoded differently and might therefore be predictive of subsequent evaluations. This differential neural response to information that is relevant versus irrelevant to later evaluations will be referred to as the difference in evaluation effect (or, the DE effect). Previous neuroimaging studies typically compared social impression formation with other types of cognitive processes6,7. These studies strongly point to the involvement of the dorsomedial prefrontal cortex (dmPFC). For example, the dmPFC showed increased activation, as measured with functional magnetic resonance imaging (fMRI), when subjects were engaged in impression formation compared with memorizing6. Also, in impression formation, the dmPFC was selectively recruited when subjects formed impressions of social others compared with inanimate objects7. Beyond this, the mPFC has been extensively implicated in social cognition and the processing of social information during mentalizing, self-knowledge and person perception8,9. However, there is still no direct evidence for the involvement of this region in the evaluative process of social impressions. We were interested in providing such evidence by decomposing the process of impression formation and isolating the evaluation component (that is, an affective judgment about a social other). The dmPFC is a likely candidate to sort through social information and sum it up into an evaluation of another person. However, because impression formation involves an affective judgment, it is also possible that neural systems more broadly involved in emotion and valuation processes, rather than social processing per se, are recruited. The amygdala has been implicated in both domains. There is abundant evidence across species that this region is essential for the formation and expression of affective value (both positive and negative) to neutral objects by way of associative learning10–12. In the social domain, the amygdala has been implicated in judgments of trustworthiness13,14, in the assessment of emotion from facial expression15,16 and body movements17, and was linked to implicit responses that were

1Center for Neural Science, 2Department of Psychology, New York University, New York, New York, USA. 3Department of Psychology, Tufts University, Medford, Massachusetts, USA. 4Department of Psychology, Harvard University, Cambridge, Massachusetts, USA. Correspondence should be addressed to D.S. ([email protected]) or E.A.P. ([email protected]).

Received 30 November 2008; accepted 21 January 2009; published online 8 March 2009; doi:10.1038/nn.2278

508

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES He told the student that he just wasn’t smart enough. He said that everyone else had missed the main point. He was distracted during the presentation because his shirts were not meticulously ironed and starched. He excused the new team member’s error. He explored the local woods instead of going to the mall with his friends.

Three negative descriptives (6 s each)

Evaluation irrelevant

Evaluation relevant

He picked up his roommate’s package on his way home from work.

Interval (12 s) Three positive descriptives (6 s each)

1

2

Negative

3

4

5

6

7

8

Positive

Figure 1 An example of a person profile. One out of 20 person-profiles presented to each subject. The profile consisted of six person-descriptive sentences arranged from negative to positive (or vice versa). Each profile corresponded with a male face. A 12-s interval with the face alone separated the positive and the negative segments. Subsequently, an evaluation slide instructed subjects to form their impression on an 8-point scale (1, ‘I don’t like him’; 8, ‘I like him’). On the basis of subjects’ evaluations, the positive and negative segments were assigned to either the evaluation-relevant or evaluationirrelevant conditions.

© 2009 Nature America, Inc. All rights reserved.

Interval (12 s)

negative versus positive. No valence bias was found (two-tailed t tests comparing the mean Evaluation proportion of negative evaluations to 0.5, (12 s maximum) t18 ¼ 1.35). Examining a valence bias in individual subjects using binomial tests ITI found a negative bias in 4 (P o 0.05) of the (12 s) 19 subjects. To confirm that subjects’ evaluations were not guided by the order in which positive or negative descriptive sentences were introduced in a profile, we assessed the number of evaluations that were consistent with the valence of the sentences that were presented first (primacy effect) or last (recency effect) in indicative of race bias18,19. Thus, the amygdala might participate in the a profile. No order bias was found (one-sample two-tailed t tests comparing the mean proportion of evaluations consistent with first formation of social value assigned to other people. To test these hypotheses, we developed the difference in evaluation segments to 0.5, t18 ¼ 1.12). Examining an order bias in individual procedure (Fig. 1), allowing us to sort social information encoding subjects using binomial tests found a primacy effect in 1 (P o 0.05) of trials by subsequent evaluations. More specifically, we measured blood the 19 subjects. oxygenation level–dependent (BOLD) signals using whole brain fMRI We also confirmed that none of the faces led to more negative or during exposure to different person profiles. Each profile consisted of 6 more positive evaluations. Binomial tests conducted for negative person-descriptive sentences implying different personality traits. The evaluation proportions for each face found a negative bias for 1 face sentences varied gradually in their positive to negative valence (or vice (P o 0.05) out of 20. versa) but evoked equivalent levels of arousal. A 12-s interval with the These results confirm that positive and negative descriptive sentences face alone separated the positive and the negative segments. Subse- and sentences presented first versus last were evenly distributed quently, an evaluation slide instructed subjects to form their impression between the evaluation-relevant and evaluation-irrelevant categories. on an 8-point scale. On the basis of these evaluations, we determined This point is of particular importance when examining the underlying which of the presented descriptive sentences guided evaluations (eva- neural responses. Given these results, it is clear that the difference in luation relevant) and which did not (evaluation irrelevant). For exam- evaluation effect reflects subjective weighting of person-descriptive ple, if a subject’s evaluation was positive, we assigned the positive information driven by subjects’ own interpretation of it rather than segment of the profile to the evaluation-relevant category and the nega- by more general effects such as the primacy (order) or negativity tive segment to the evaluation-irrelevant category. We then identified (valence) of the stimuli themselves. the brain regions dissociating items from each category (that is, diffeIt should be noted that order effects have been found using similar rence in evaluation effect). Notably, we correlated subjects’ BOLD signal sequential presentations20. However, these effects were shown to be with their own individual evaluations. This allowed us to identify brain dependent on procedural variations, such as the sequence of the items, regions that were consistent across subjects in processing evaluation- number and spacing of the items and instruction21. These effects were relevant information regardless of the particular stimuli that they also shown to decrease with continued practice22, similar to considered. Immediately after the scanning session, subjects underwent our findings here, as subjects were practiced and familiar with the a memory-recognition task. task beforehand. To confirm that the difference at encoding on the basis of subsequent RESULTS evaluation and the difference at encoding on the basis of subsequent Behavioral results memory are separate phenomena, we assessed whether evaluationWe scanned 19 volunteers while they performed the impression- relevant descriptive sentences were remembered better. We found no formation task, which consisted of 20 profile evaluations. To confirm memory difference between evaluation-relevant and irrelevant senthat subjects’ impressions were guided by their idiosyncratic evalua- tences (paired two-tailed t test, t18 ¼ 0.317). Examining this bias in tions, rather than by the valence of the information that was presented, individual subjects using a w2 test found an effect in 2 (P o 0.05) of we assessed the number of profiles that the subjects evaluated as the 19 subjects. One subject remembered the evaluation-relevant

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

509

ARTICLES Face with descriptive sentences versus face alone

a

Differential % BOLD signal change evaluation relevant – evaluation irrelevant

© 2009 Nature America, Inc. All rights reserved.

b 0.3

* 0.2

*

0.1

0 dmPFC

PCC

Amygdala

–0.1

information better than the evaluation-irrelevant information, whereas the other subject showed an opposite effect. Both evaluation-relevant and evaluation-irrelevant sentences were remembered significantly above chance (0.2; one-sample t test, t18 ¼ 5.37, P o 0.01, t18 ¼ 3.91, P o 0.01, respectively). We conducted an additional study on a separate group of subjects without the fMRI scan to examine memory immediately after each profile presentation. This task was similar to the one used in the scanner, with the exception that, after the evaluation slide of each profile, subjects were presented with the face along with six descriptive sentences (five new and one old). The subjects had to identify which of the sentences was previously presented with the face. Again, we found no difference in memory for evaluation-relevant and irrelevant descriptive sentences and no valence or order biases in memory (see Supplementary Results and Supplementary Fig. 1 online). Together, these results eliminate differential memory as an alternative explanation of the findings in our procedure. Thus, evaluations were not driven by episodic memory for item information, confirming that the difference in evaluation effect is different from the difference in memory effect. Neuroimaging results In our neuroimaging analysis, we examined which regions showed the difference in evaluation effect out of regions that were broadly engaged in the impression-formation task. Functional regions of interest (ROIs) were identified contrasting faces with person-descriptive information and face-alone presentations (false discovery rate o 0.05; Fig. 2 and Table 1). To examine whether the regions revealed by this contrast (Table 1) show the difference in evaluation effect, we extracted the BOLD response from each of these ROIs and compared the mean percentage BOLD signal change during the presentation of evaluationrelevant versus evaluation-irrelevant person-descriptive sentences (two-tailed t tests, P o 0.05; Table 1). The only regions showing significantly greater BOLD responses to evaluation-relevant sentences were the amygdala and the PCC (Fig. 2b), and the thalamus (P o 0.05; Table 1). There were no regions showing the opposite effect. For comparison, we examined an extensive region of dmPFC revealed by the face with descriptive sentences versus face alone contrast (Fig. 2a). However, the analysis separating dmPFC responses into evaluation-relevant versus evaluation-irrelevant person-descriptive sentences revealed no difference in evaluation effect (Fig. 2b).

510

Figure 2 Brain regions demonstrating the difference in evaluation effect out of regions broadly engaged in the impression-formation task. (a) Functional ROIs were identified by contrasting faces with person-descriptive sentences versus face-alone presentations (false discovery rate o 0.05). The dmPFC (x ¼ –7, y ¼ 24, z ¼ 53; Brodmann area 8/9), PCC (x ¼ 0, y ¼ –51, z ¼ 23; Brodmann area 23) and left amygdala (x ¼ –23, y ¼ –8, z ¼ –16) are denoted by yellow circle on the statistical activation map. The full list of regions revealed by this contrast is detailed in Table 1. (b) To examine whether these ROIs show the DE effect, we extracted the BOLD response from each of these regions (dmPFC, 6,561-mm3 voxels; PCC, 760-mm3 voxels; amygdala, 111-mm3 voxels) and compared the mean percentage BOLD signal change during the presentation of evaluation-relevant versus evaluation-irrelevant person-descriptive sentences. The differential score was calculated by subtracting evaluation-irrelevant from evaluation-relevant responses, so positive scores correspond to stronger responses to the evaluation-relevant information. A significant differential responding (twotailed t tests) was shown by the PCC and the amygdala, but not by the dmPFC (P o 0.05). Error bars indicate s.e.

To confirm these findings, we conducted an additional analysis using a contrast directly comparing BOLD responses during the presentation of evaluation-relevant versus evaluation-irrelevant persondescriptive sentences. As expected, this contrast revealed only the PCC (P o 0.001 corrected; Fig. 3a), the amygdala (Fig. 3b) and the thalamus (see Supplementary Results). For the latter two regions, we used a more liberal threshold (P o 0.05 and P o 0.005 uncorrected, Table 1 Talairach coordinates of regions extracted from the face with person-descriptive information 4face alone contrast (false discovery rate o 0.05) Coordinates Volume Region

Side

x

y

z

(mm3)

BA

PCC*

M

0

–51

23

23

760

Amygdala* Thalamus*

L L

–23 –7

–8 –13

–16 12

– –

111 333

Thalamus* Caudate

R R

9 9

–13 5

10 10

– –

291 170

Caudate Hippocampus

L R

–9 27

3 –11

12 –13

– –

181 125

dmPFC dmPFC

L L

–7 –9

52 26

37 55

9 8

2,930 2,997

dmPFC Superior frontal gyrus

R L

11 –4

32 –1

52 61

8 6

1,584 2,128

Ventromedial prefrontal cortex Inferior frontal gyrus

M L

0 –43

48 24

–8 –1

10 47

2,931 2,426

Inferior frontal gyrus Middle frontal gyrus

R L

44 –46

26 10

–6 27

47 8

2,917 4,478

Middle frontal gyrus Superior temporal gyrus

R L

40 –46

5 14

30 –14

6 38

1,184 4,291

Superior temporal gyrus Superior temporal gyrus (posterior)

R L

52 –50

12 –57

–13 16

38 21/22

3,833 2,615

Middle temporal gyrus (anterior) Middle temporal gyrus (anterior)

R L

53 –52

–2 –2

–18 –18

21 21

4,712 5,330

Middle temporal gyrus (posterior)

L

–54

–29

1

21

4,293

Middle temporal gyrus (posterior) Precuneus

R L

52 –45

–37 –7

4 47

21 4

2,537 5,404

Fusiform gyrus Lingual gyrus

L R

–38 21

–48 –58

–15 –1

20/37 37/19

6,371 2,414

Lingual gyrus Cerebellum

L M

–17 2

–58 –64

–1 –25

19 –

2,585 4,631

BA, Brodmann area; L, left; M, middle; R, right. *Regions showing the difference in evaluation effect (two-tailed t tests, P o 0.05).

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

b

P < 0.001 (corrected)

d

c 0.2

0.2

0.1

0.1

0

0

–0.1

–0.1

Percentage BOLD signal change

Relevant

Irrelevant

Figure 3 Brain regions revealed by the evaluation-relevant versus evaluationirrelevant contrast. (a) PCC BOLD responses (x ¼ 0, y ¼ –51, z ¼ 24; P o 0.001 corrected). (b) Amygdala BOLD responses (x ¼ –23, y ¼ –8, z ¼ –16; P o 0.05 uncorrected). (c,d) Mean percentage BOLD signal change extracted from all PCC (2,187 mm3) and amygdala (108 mm3) voxels. The mean BOLD response for each trial type is shown: evaluation relevant (black bars) and evaluation irrelevant (orange bars). Error bars indicate s.e.

respectively; minimal cluster size 4 100 mm3). No other regions of activation were observed, even at the most liberal threshold (P o 0.05 uncorrected). We determined the mean and variance of the PCC and amygdala BOLD responses (Fig. 3c,d). Finally, no regions were revealed by the opposite contrast (evaluation irrelevant versus evaluation relevant), even at a more liberal threshold (P o 0.05 uncorrected). The categorical comparison between the evaluation-relevant and evaluation-irrelevant person-descriptive information is predicated on the idea that the overall evaluation involves either the positive or negative information alone. However, it might also reflect the weighted sum of both positive and negative information. To capture this, we conducted a parametric analysis, correlating the level of subjects’ evaluations (ranging from 1 to 8) with PCC and amygdala mean BOLD responses (Fig. 4). We found that the mean BOLD response was stronger during the encoding of negative information for subjects giving negative evaluations (Fig. 4a,c) and was stronger during the encoding of positive information for subjects giving positive evaluations (Fig. 4b,d). These data suggest that PCC and amygdala responses reflect the integration of positive and negative information into an overall impression. Consistent with our behavioral findings, in which no valence bias was observed, these BOLD responses also do not map

DISCUSSION We provide new evidence for the manner in which social information is encoded in the brain to form impressions of others. When required to rapidly judge others, we appear to be efficient evaluators. We sift through available social information, weighting what matters to us. Sorting information by level of importance is a matter of subjective preference. Our results show where this sorting occurs in the brain, suggesting a neural mechanism by which impressions are formed. Subjects regarded different segments of person-descriptive information as being relevant or irrelevant for their subsequent evaluations. The idiosyncratic basis for this differential relevance might be factors such as personal experience, social values and pet peeves among others, which could make particular items more salient in the eyes of the beholder. First impressions, therefore, are tightly connected with the enduring biases subjects bring along. Such biases shape how subjects weight different types of information and which information is selected for additional processing. Even though these factors may vary widely between subjects, across subjects, the same brain regions, the PCC, the amygdala and the thalamus, dissociated these two types of information. The extent to which these regions were recruited during encoding of person-descriptive information correlated with how subjects valued it, as was evident in their subsequent evaluation scores. It is important to note that the differential encoding on the basis of subsequent evaluation effect that we report here is dissociable from the previously reported effect of differential encoding on the basis of subsequent memory. Behaviorally, evaluation-relevant items were not remembered better than irrelevant ones. Consistent with this, encoding of evaluation-relevant information did not selectively engage memory related areas such as the hippocampus or the dorsolateral prefrontal

a

Negative information 0.6

b

r = –0.95

0.4 0.2 0 –0.2 –0.4 –0.6

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

1 2 3 4 ‘I don’t like him’

5

6

7 8 ‘I like him’

0.4 0.2 0 –0.2 –0.4 –0.6

Evaluation scale

c

0.4

Negative information r = –0.93

0.3 0.2 0.1 0 –0.1 –0.2

r = 0.84

1 2 3 4 ‘I don’t like him’

1 2 3 4 ‘I don’t like him’

5

6 7 8 ‘I like him’

Evaluation scale

5

6 7 8 ‘I like him’

Evaluation scale

d BOLD signal (z scores)

Figure 4 BOLD signal in PCC and amygdala correlates with strength of evaluation. (a–d) Correlations between mean normalized PCC (a,b) and amygdala (c,d) BOLD signal (extracted from the evaluation relevant versus evaluation irrelevant contrast) and level of evaluation ranging from 1 (‘I don’t like him’) to 8 (‘I like him’) are presented. For both regions, a more negative evaluation led to a stronger mean BOLD response during the encoding of negative person-descriptive information (a,c) and a more positive evaluation led to a stronger mean BOLD response during the encoding of positive person-descriptive information (b,d). This was supported by a significant correlation between the mean BOLD response and level of evaluation. Because lower scores on the evaluation scale represent more negative evaluations, the correlation was negative (PCC, r ¼ –0.95, a; amygdala, r ¼ –0.93, c; P o 0.01) during negative information trials and positive during positive information trials (PCC, r ¼ 0.84, b; amygdala, r ¼ 0.85, d; P o 0.01).

Positive information 0.6

–0.8

–0.8

BOLD signal (z scores)

© 2009 Nature America, Inc. All rights reserved.

P < 0.05

BOLD signal (z scores)

a

onto the valence of the information per se. Instead, these regions encode social information that is subjectively meaningful and more heavily weighted in later evaluations. Thus, both positive and negative persondescriptive information receives more processing as long as it is relevant for subsequent evaluations (see Supplementary Fig. 2 and Supplementary Tables 1 and 2 online).

BOLD signal (z scores)

Evaluation relevant versus evaluation irrelevant

0.4

Positive information r = 0.85

0.3 0.2 0.1 0 –0.1 –0.2

1 2 3 4 ‘I don’t like him’

5

6 7 8 ‘I like him’

Evaluation scale

511

© 2009 Nature America, Inc. All rights reserved.

ARTICLES cortex. It appears that the relevance of particular information for subsequent evaluations does not confer a mnemonic advantage (as would be shown by a DM effect). One might expect information that is valued more to be better remembered. However, these results are not surprising given previous research showing that individuals with amnesia develop impressions of social others23 and show attitude change24 without having recollection of the information on which these affective judgments were based. Normal subjects have long been known to show such dissociation in rapid on-line impression formation25, as well as in judgments made a few months after persondescriptive information was introduced26. Taken together, these data suggest that episodic memory is probably not a mediating factor in the use of information for subsequent evaluations. Understanding the neural substrates of social cognition has been one of the core motivations driving the burgeoning field of social neuroscience. A number of studies have highlighted the dmPFC in the processing of social information6–9. Our results provide further evidence that the dmPFC is recruited to process person-descriptive information during impression formation. However, BOLD responses in this region do not dissociate evaluation-relevant from evaluation-irrelevant information, suggesting that the dmPFC is not essential for the evaluative component of impression formation. In fact, social evaluation recruits brain regions that are not socially specialized27 but are more generally involved in valuation and emotional processes. Valuation and emotional processes, as a substantial amount of research has shown, are characteristic of the amygdala. In particular, the amygdala is considered to be a crucial region in learning about motivationally important stimuli10–12. It is also implicated in social inferences that are based on facial and bodily expressions15–17, in inferences of trustworthiness13,14 and in the capacity to infer social attributes28. Moreover, the involvement of amygdala in social inferences might be independent of awareness or explicit memory. For example, increased amygdala responses were correlated with implicit, but not explicit, measures of the race bias19, as well as with presentation of faces previously presented in an emotional, but not neutral, context, regardless of whether subjects could explicitly retrieve this information26. Here we provide evidence linking the two domains of affective learning and social processing by showing that the amygdala is engaged in the formation of subjective value assigned to another person in a social encounter. Although the amygdala is typically implicated in the processing of negative affect and negative stimuli have been shown to modulate it more than positive stimuli29, we found that the amygdala processed both positive and negative evaluation-relevant information, suggesting that amygdala activity is driven by factors other than mere valence, such as the motivational importance or salience of the stimuli. This result is consistent with recent findings30,31 showing enhanced amygdala responses for both positive and negative stimuli as a function of motivational importance. Evidence related to the PCC has been more diverse. There have been reports in the social domain, such as involvement in theory of mind32 and self-referential outward-focused thought33, in memory related processes such as autobiographical memory of family and friends34, and in emotional modulation of memory35 and attention36. More recently, the PCC has been linked with economic decision making, the assignment of subjective value to rewards37 under risk and uncertainty38, and credit assignment in a social exchange39. A common denominator of these studies might be that all involved either a social or an outward-directed valuation component. Our task also encompasses these features, extending

512

the role of the PCC to value assignment to social information guiding our first impressions of others. The amygdala and the PCC are both interconnected with the thalamus as part of a larger circuitry that is implicated in emotion, arousal and learning40. Beyond the known role of the amygdala and the PCC in social-information processing and value representation, our results suggest a neural mechanism underlying the online formation of first impressions. When encoding everyday social information during a social encounter, these regions sort information on the basis of its personal and subjective importance and summarize it into an ultimate score, a first impression. Other regions, such as the ventromedial PFC, the striatum and the insula, have also been implicated in valuation processes41–45. However, these regions did not emerge in our difference in evaluation effect analysis. This might suggest a possible dissociation in the valuation network between regions engaged in the formation of value and its subsequent representation and updating. The latter regions would not be engaged during encoding and therefore would not show a difference in evaluation effect but would instead have an effect once the evaluation is formed. The amygdala and the PCC probably participate in both value formation and its representation. The difference in evaluation procedure may provide a useful tool for disentangling the different components of the valuation system and their specific contributions to social versus nonsocial evaluations. In sum, the complexity of social evaluation in forming a first impression is evident in the recruitment of multiple brain systems that are involved in social-information processing, emotion and valuation. Although it has been suggested that some neural systems are specialized for social-information processing, it seems that when it comes to evaluating others, these systems are not enough. Additional regions specialized in affective processing are recruited to evaluate other people in our initial encounters with them, providing a neural signature of first impressions. METHODS Subjects. We recruited 19 right-handed normal volunteers (12 males) between 18 and 31 years of age (mean ¼ 22.68, s.d. ¼ 4.57) for the fMRI evaluation task. The experiment was approved by the New York University Committee on Activities Involving Human Subjects. All subjects gave informed consent and were paid for their participation. Stimuli. We constructed 20 person profiles using 120 person-descriptive sentences implying different personality traits (for example, considerate: ‘‘He promised not to smoke in his apartment since his roommate was trying to quit.’’)6. These sentences were pretested (n ¼ 30) for valence (1 ¼ very negative and 8 ¼ very positive) and arousal (1 ¼ not arousing and 8 ¼ very arousing). Each profile consisted of six unique sentences, arranged sequentially from either negative to positive or positive to negative. The valence transition was gradual according to the mean valence ratings from the pretest in the following way (for negative to positive profiles): the first was very negative (1.00 o mean o 2.74), the second and third were moderately negative (2.75 o M o 4.49), the fourth and fifth were moderately positive (4.50 o M o 6.24), and the sixth was very positive (6.25 o M o 8.00). This order was reversed for positive-to-negative profiles. Two-tailed t tests were conducted between mean arousal ratings of positive and negative sentences in each profile to ensure that there were no significant differences (P 4 0.1 for all profiles). We paired each profile with a picture of a monochrome male face of neutral expression, all of which were taken with identical lighting source and camera angle (Extended Yale Database B). Procedure. Participants were told that they would see information about different people and would be asked to give their impressions of them. We presented 20 person profiles in one of four orders, counterbalanced across subjects. These four sets were designed to control for order of profile presentation, order of sentences in each profile and assignment of faces to

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

© 2009 Nature America, Inc. All rights reserved.

profiles. None of the sentences or faces repeated in each order. At the onset of each profile, the first, second and third sentences were presented with the face for 6 s each (18 s total). Following these was a 12-s interstimulus interval displaying only the face. Next, the fourth, fifth and sixth sentences were presented with the face for 6 s each (18 s total). Another face-only interstimulus interval ensued for 12 s. Subsequently, a feedback slide with the face still being apparent asked subjects to answer the question, ‘What is your impression of him?’ on a Likerttype scale between 1 (‘I don’t like him’) and 8 (‘I like him’). After subjects made a button response, they were requested to confirm their choice. Following confirmation, a ‘Thank you’ slide was displayed for the remainder of the 12 s from the onset of the feedback slide (that is, feedback slide and confirmation slide and ‘Thank you’ slide ¼ 12 s). Finally, an intertrial interval (ITI) with fixation point was displayed for 12 s, after which the next profile was presented. After the fMRI session, subjects participated in a memory task that consisted of 40 trials in one of two orders. In each trial, five faces, which were previously encountered during the fMRI task, were coupled with one sentence presented with one of these five faces during the fMRI task. Subjects had to select which of the five faces had been described with this sentence. These were sampled from all 20 profiles, with equal representation from the positive, negative, first and last segments. Evaluation analysis. To verify that subjects’ evaluations were guided by their individual subjective preferences rather than by any other effect, we probed the data for valence bias (more positive than negative evaluations or vice versa), order bias (primacy or recency effects; that is, more evaluations with the same valance as the segment presented first (first, second and third sentences) compared with last (forth, fifth and sixth sentences) or vice versa). We also examined a face bias (that is, whether particular faces led to more negative or positive evaluations). Subjects’ responses on the 8-point scale were coded as being negative (1 to 4) or positive (5 to 8). To test whether subjects were biased by valence in their evaluations, we compared the proportion of negative evaluations to the proportion that would be measured if there were zero bias (that is, 0.5) using a one-sample two-tailed t test for the group analysis and binomial tests for individual subject analyses. Similarly, to test whether subjects were biased by order, we compared the proportion of evaluations with the same valence as the segment presented first to the proportion that would be measured if there were zero bias (that is, 0.5) using a one-sample two-tailed t test for the group analysis and binomial tests for individual subject analyses. To test whether a particular face led to more negative or more positive evaluations, binomial tests comparing the proportion of negative evaluations to the proportion that would be measured if there were zero bias (that is, 0.5) were conducted for each face. An alpha level of 0.05 was set for all statistical comparisons. Memory analysis. To verify that the different types of information did not affect memory performance, we examined the memory data for valence bias (negative sentences were remembered better than positive or vice versa), order bias (sentences presented in the first segment (first, second or third presentations) were remembered better than those presented in the last segment (fourth, fifth or sixth presentations) in profile (primacy effect) or vice versa (recency effect)) and evaluation-relevance bias (evaluation-relevant sentences (having the same valence as the subsequent evaluation) were remembered better than evaluation-irrelevant sentences (with valences different from the subsequent evaluation)). To test for these memory biases, we used a recognition accuracy measure that was defined as correct recognition responses divided by the total number of sentences in category. For group analyses, we compared mean recognition accuracy in the two corresponding categories of each bias test using paired twotailed t tests. For individual subject analyses, numerical counts corresponding with the two categories were compared using w2 tests. Finally, overall recognition success was tested using a one-sample t test comparing the proportion of correct recognition responses to chance level. An alpha level of 0.05 was set for all statistical comparisons. fMRI acquisition. A 3T Siemens Allegra head-only scanner and Siemens standard head coil were used for data acquisition. Anatomical images were

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

acquired using a T1-weighted protocol (256
256 matrix, 176 1-mm sagittal slices). Functional images were acquired using a single-shot gradient echo EPI sequence (repetition time, 2.0 s; echo time, 25 ms; field of view, 192 cm, flip angle ¼ 751, bandwidth ¼ 4,340 Hz px–1 and echo spacing ¼ 0.29 ms). We obtained 39 contiguous oblique-axial slices (3
3
3-mm voxels) parallel to the anterior commissure–posterior commissure line. Analysis of the imaging data was conducted using BrainVoyager QX software package (Brain Innovation). Functional imaging data preprocessing included motion correction, slice scan time correction (using sinc interpolation), spatial smoothing using a three-dimensional Gaussian filter (4-mm full width at half maximum), and voxel-wise linear detrending and high-pass filtering of frequencies (above three cycles per time course). The structural and functional data of each participant were transformed to standard Talairach stereotaxic space46. fMRI analysis. A random-effects general linear model analysis was conducted on the fMRI signal during the evaluation task with the following predictors: evaluation-relevant person-descriptive information, evaluation-irrelevant person-descriptive information, face-alone after first information segment, face-alone after second information segment and evaluative response. The predictors were convolved with a standard canonical hemodynamic response function. Contrast analyses. The primary contrast of interest was faces with person descriptive sentences versus face alone presentations. On the basis of this contrast, we defined the ROIs on which to examine the difference in evaluation effect (see below). Regions on the statistical map showing a significant response (clusters of at least 100-mm3 contiguous voxels whose false discovery rate was o 0.05) are detailed in Table 1. To validate the primary analysis, the secondary contrast of interest was evaluation-relevant versus evaluation-irrelevant person-descriptive information, which directly explored the difference in evaluation effect (that is, regions showing neural differentiation of encoding on the basis of subsequent evaluations). We expected to reveal similar regions in these two independent analyses. Calculation of significant values in the activation map revealed by this contrast (Fig. 3a) was based on the individual voxel significance (P o 0.001 corrected) and on the minimum cluster size of 890-mm3 voxels. The probability of a false positive was determined from the frequency count of cluster sizes in the entire brain using a Monte Carlo simulation47. Because this analysis revealed only the region of the PCC, we used a gradually more liberal threshold to reveal additional regions (minimal cluster size of 100-mm3 voxels, P o 0.005 uncorrected for thalamus; Supplementary Fig. 1; P o 0.05 uncorrected for amygdala; Fig. 3b). We extracted the mean percentage BOLD signal change at peak activation from the PCC (Fig. 3c) and the amygdala (Fig. 3d) for evaluation-relevant versus evaluation-irrelevant person-descriptive sentences. Difference in evaluation effect analysis. A comparison between responses to evaluation-relevant and evaluation-irrelevant person-descriptive information was conducted on BOLD responses in ROIs revealed by the face with descriptive information versus face contrast (false discovery rate o 0.05; Table 1). For each ROI, we used paired two-tailed t tests comparing the mean percentage BOLD signal change from baseline (12-s ITI where a fixation point was presented between profile presentations) of the evaluation-relevant versus the evaluationirrelevant trials. Each trial consisted of the mean of the nine BOLD measurements (repetition time = 2 s), during which the descriptive information was presented, and which were averaged across 20 trials in each condition and across 19 subjects. The alpha level for statistical comparisons was set at 0.05. Correlation analysis. Correlations were computed between mean normalized BOLD signals and levels of evaluation ranging from 1 to 8. fMRI responses were extracted from the evaluation-relevant versus evaluation-irrelevant contrast for PCC (x ¼ 0, y ¼ –51, z ¼ 24; P o 0.001 corrected, cluster size of 2,187-mm3 voxels) and amygdala (x ¼ –23, y ¼ –8, z ¼ –16; P o 0.05 uncorrected, cluster size of 108-mm3 voxels). The fMRI signal was averaged across 19 subjects at each level of evaluation. The correlations were calculated separately for fMRI responses during the presentation of negative and positive information. Note: Supplementary information is available on the Nature Neuroscience website.

513

ARTICLES ACKNOWLEDGMENTS We thank I. Levy and D. Amodio for fruitful discussions and comments, C. Raio for assistance with data collection, and K. Sanzenbach and the Center for Brain Imaging at New York University for technical assistance. This study was funded by a Seaver Foundation grant to the Center for Brain Imaging, a James S. McDonnell Foundation grant to E.A.P. and a Fulbright award to D.S. AUTHOR CONTRIBUTIONS D.S. designed the experiments, collected and analyzed data, interpreted the data, and wrote the first draft of the manuscript. J.B.F. contributed to data collection, analysis and interpretation, and the final version of the manuscript. J.P.M., J.S.U. and E.A.P. contributed to experimental design, data interpretation and the final version of the manuscript.

© 2009 Nature America, Inc. All rights reserved.

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

1. Ambady, N. & Rosenthal, R. Half a minute: predicting teacher evaluations from thin slices of behavior and physical attractiveness. J. Pers. Soc. Psychol. 64, 431–441 (1993). 2. Uleman, J.S., Blader, S. & Todorov, A. Implicit impressions. in The New Unconscious (eds. Hassin, R. Uleman, J.S. & Bargh, J.A.) 362–392 (Oxford University Press, New York, 2005). 3. Paller, K.A., Kutas, M. & Mayes, A.R. Neural correlates of encoding in an incidental learning paradigm. Electroencephalogr. Clin. Neurophysiol. 67, 360–371 (1987). 4. Brewer, J.B. et al. Making memories: brain activity that predicts how well visual experience will be remembered. Science 281, 1185–1187 (1998). 5. Wagner, A.D. et al. Building memories: remembering and forgetting of verbal experiences as predicted by brain activity. Science 281, 1188–1191 (1998). 6. Mitchell, J.P., Macrae, C.N. & Banaji, M.R. Encoding-specific effects of social cognition on the neural correlates of subsequent memory. J. Neurosci. 24, 4912–4917 (2004). 7. Mitchell, J.P., Heatherton, T.F. & Macrae, C.N. Distinct neural systems subserve person and object knowledge. Proc. Natl. Acad. Sci. USA 99, 15238–15243 (2002). 8. Amodio, D.M. & Frith, C.D. Meeting of minds: the medial frontal cortex and social cognition. Nat. Rev. Neurosci. 7, 268–277 (2006). 9. Mitchell, J.P., Banaji, M.R. & Macrae, C.N. The link between social cognition and selfreferential thought in the medial prefrontal cortex. J. Cogn. Neurosci. 17, 1306–1315 (2005). 10. Phelps, E.A. & LeDoux, J.E. Contributions of the amygdala to emotion processing: from animal models to human behavior. Neuron 48, 175–187 (2005). 11. Everitt, B.J. & Robbins, T.W. Amygdala-ventral striatal interactions and reward-related processes. in The Amygdala: Neurobiological Aspects of Emotion, Memory and Mental Dysfunction (ed. Aggleton J.P.) 401–430 (New York, Wiley-Liss, 1992). 12. LeDoux, J.E. Emotion circuits in the brain. Annu. Rev. Neurosci. 23, 155–184 (2000). 13. Engell, A.D., Haxby, J.V. & Todorov, A. Implicit trustworthiness decisions: automatic coding of face properties in the human amygdala. J. Cogn. Neurosci. 19, 1508–1519 (2007). 14. Winston, J.S., Strange, B.A., O’Doherty, J. & Dolan, R.J. Automatic and intentional brain responses during evaluation of trustworthiness of faces. Nat. Neurosci. 5, 277–283 (2002). 15. Adolphs, R., Tranel, D., Damasio, H. & Damasio, A. Impaired recognition of emotion in facial expressions following bilateral damage to the human amygdala. Nature 372, 669–672 (1994). 16. Adolphs, R. Recognizing emotion from facial expressions: psychological and neurological mechanisms. Behav. Cogn. Neurosci. Rev. 1, 21–62 (2002). 17. Hadjikhani, N. & de Gelder, B. Seeing fearful body expressions activates the fusiform cortex and amygdala. Curr. Biol. 13, 2201–2205 (2003). 18. Cunningham, W.A. et al. Separable neural components in the processing of black and white faces. Psychol. Sci. 15, 806–813 (2004).

514

19. Phelps, E.A. et al. Performance on indirect measures of race evaluation predicts amygdala activation. J. Cogn. Neurosci. 12, 729–738 (2000). 20. Asch, S.E. Forming impression of personality. J. Abnorm. Soc. Psychol. 41, 258–290 (1946). 21. Jones, E.E. & Goethals, G.R. Order effects in impression formation. in Attribution: Perceiving the Causes of Behavior. (eds. Jones, E.E., Kanouse, D.E., Kelley, H.H., Nisbett, R.E., Valins, S., & Weiner, B.) 27–46 (New Jersey, Lawrence Erlbaum Associates, 1987). 22. Anderson, N.H. & Barrios, A.A. Primacy effects in personality impression formation. J. Abnorm. Soc. Psychol. 63, 346–350 (1961). 23. Johnson, M.K., Kim, J.K. & Risse, G. Do alcoholic Korsakoffs syndrome patients acquire affective reactions? J. Exp. Psychol. Learn. Mem. Cogn. 11, 22–36 (1985). 24. Lieberman, M.D., Ochsner, K.N., Gilbert, D.T. & Schacter, D.L. Do amnesics exhibit cognitive dissonance reduction? The role of explicit memory and attention in attitude change. Psychol. Sci. 12, 135–140 (2001). 25. Hastie, R. & Park, B. The relationship between memory and judgment depends on whether the judgment task is memory-based or on-line. Psychol. Rev. 93, 258–268 (1986). 26. Somerville, L.H., Wig, G.S., Whalen, P.J. & Kelley, W.M. Dissociable medial temporal lobe contributions to social memory. J. Cogn. Neurosci. 18, 1253–1265 (2006). 27. Frith, C.D. The social brain? Philos. Trans. R. Soc. Lond. B Biol. Sci. 362, 671–678 (2007). 28. Heberlein, A.S. & Adolphs, R. Impaired spontaneous anthropomorphizing despite intact perception and social knowledge. Proc. Natl. Acad. Sci. USA 101, 7487–7491 (2004). 29. Kim, H. et al. Contextual modulation of amygdala responsivity to surprised faces. J. Cogn. Neurosci. 16, 1730–1745 (2004). 30. Cunningham, W.A., Van Bavel, J.J. & Johnsen, I.R. Affective flexibility: evaluative processing goals shape amygdala activity. Psychol. Sci. 19, 152–160 (2008). 31. Todorov, A., Baron, S.G. & Oosterhof, N.N. Evaluating face trustworthiness: a modelbased approach. Soc. Cogn. Affect. Neurosci. 3, 119–127 (2008). 32. Fletcher, P.C. et al. Other minds in the brain: a functional imaging study of ‘‘theory of mind’’ in story comprehension. Cognition 57, 109–128 (1995). 33. Johnson, M.K. et al. Dissociating medial frontal and posterior cingulate activity during self-reflection. Soc. Cogn. Affect. Neurosci. 1, 56–64 (2006). 34. Maddock, R.J., Garrett, A.S. & Buonocore, M.H. Remembering familiar people: the posterior cingulate cortex and autobiographical memory retrieval. Neuroscience 104, 667–676 (2001). 35. Maddock, R.J., Garrett, A.S. & Buonocore, M.H. Posterior cingulate cortex activation by emotional words: fMRI evidence from a valence decision task. Hum. Brain Mapp. 18, 30–41 (2003). 36. Pessoa, L. & Padmala, S. Quantitative prediction of perceptual decisions during nearthreshold fear detection. Proc. Natl. Acad. Sci. USA 102, 5612–5617 (2005). 37. Kable, J.W. & Glimcher, P.W. The neural correlates of subjective value during intertemporal choice. Nat. Neurosci. 10, 1625–1633 (2007). 38. McCoy, A.N. & Platt, M.L. Risk-sensitive neurons in the macaque cingulate cortex. Nat. Neurosci. 8, 1220–1227 (2005). 39. Tomlin, D. et al. Agent-specific responses in the cingulate cortex during economic exchanges. Science 312, 1047–1050 (2006). 40. Taber, K.H., Wen, C., Khan, A. & Hurley, R.A. The limbic thalamus. J. Neuropsychiatry Clin. Neurosci. 16, 127–132 (2004). 41. O’Doherty, J.P. Reward representations and reward-related learning in the human brain: insights from neuroimaging. Curr. Opin. Neurobiol. 14, 769–776 (2004). 42. Schiller, D., Levy, I., Niv, Y., LeDoux, J.E. & Phelps, E.A. From fear to safety and back— reversal of fear in the human brain. J. Neurosci. 28, 11517–11525 (2008). 43. Delgado, M.R., Li, J., Schiller, D. & Phelps, E.A. The role of striatum in aversive learning and aversive prediction errors. Phil. Trans. R. Soc. Lond. B 363, 3787–3800 (2008). 44. Seymour, B. & McClure, S.M. Anchors, scales and the relative coding of value in the brain. Curr. Opin. Neurobiol. 18, 173–178 (2008). 45. Knutson, B., Rick, S., Wimmer, G.E., Prelec, D. & Loewnstein, G. Neural predictors of purchases. Neuron 53, 147–156 (2007). 46. Talairach, J. & Tournoux, P. Co-planar Stereotaxic Atlas of the Human Brain: an Approach to Medical Cerebral Imaging (Thieme, New York, 1988). 47. Forman, S.D. et al. Improved assessment of significant activation in functional magnetic resonance imaging (fMRI): use of a cluster-size threshold. Magn. Reson. Med. 33, 636–647 (1995).

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

Hierarchical cognitive control deficits following damage to the human frontal lobe

© 2009 Nature America, Inc. All rights reserved.

David Badre1,2, Joshua Hoffman3, Jeffrey W Cooney3 & Mark D’Esposito3,4 Cognitive control permits us to make decisions about abstract actions, such as whether to e-mail versus call a friend, and to select the concrete motor programs required to produce those actions, based on our goals and knowledge. The frontal lobes are necessary for cognitive control at all levels of abstraction. Recent neuroimaging data have motivated the hypothesis that the frontal lobes are organized hierarchically, such that control is supported in progressively caudal regions as decisions are made at more concrete levels of action. We found that frontal damage impaired action decisions at a level of abstraction that was dependent on lesion location (rostral lesions affected more abstract tasks, whereas caudal lesions affected more concrete tasks), in addition to impairing tasks requiring more, but not less, abstract action control. Moreover, two adjacent regions were distinguished on the basis of the level of control, consistent with previous functional magnetic resonance imaging results. These results provide direct evidence for a rostro-caudal hierarchical organization of the frontal lobes.

The function of the prefrontal cortex (PFC) is closely associated with cognitive control or the ability of humans and other primates to internally guide behavior in accordance with goals, plans and broader contextual knowledge1–18. Consider the simple example of entering a colleague’s office and finding a place to sit down. On a daily basis, in one’s own office, the chair behind the desk is the appropriate seat. In another’s office, however, we easily select the chair in front of the desk as being the socially appropriate choice. Overcoming a habitual tendency in order to coordinate behavior with an abstract social rule is an example of cognitive control. From one perspective, cognitive control mechanisms operate through a process of biased competition, whereby maintenance of a distributed neural representation of the task context (colleague’s office) configures processing throughout the action system to bias selection of an appropriate behavior (sit in the chair in front of the desk) over a competing one (sit in the chair behind the desk)8,19–21. The frontal lobes are thought to be centrally involved in coding such contextual representations to provide internal control over action14,22,23. However, the functional organization of the frontal lobe remains unknown. Although it is widely believed that separate frontal regions support distinct forms of control, there is little evidence to date of double dissociations in lateral frontal cortex and no evidence in human patients4. Thus, a fundamental goal in cognitive neuroscience is to characterize the functional organization of frontal cortex that supports the control of action. Control of action can involve abstract goals, such as deciding whether to e-mail or call a friend, as well as the concrete motor programs required to carry out these abstract goals, such as selecting the appropriate sequence of keystrokes to type an e-mail

greeting17,24–26. Thus, computational models of cognitive control and of complex action have often included hierarchical architectures that represent such actions at different levels of abstraction24–27. Consistent with the concept of hierarchically arrayed levels of control, neuroimaging studies have repeatedly demonstrated differences in functional activation along the rostro-caudal axis of lateral frontal cortex, ranging from dorsal premotor cortex (PMd; BBrodmann area 6/8) to lateral frontal polar cortex (Brodmann area 10), such that more anterior regions were associated with progressively more abstract action control further removed from the selection of a concrete motor response28–31. Across these previous studies, abstraction has been defined in different, although not necessarily mutually exclusive, ways32. Some have suggested that posterior regions are more sensitive to domain distinctions, such as spatial versus object, whereas more anterior regions are not18,33. Others propose that progressively anterior regions coordinate action over longer time scales and so can maintain action representations and mediate action contingencies over longer temporal gaps7,34. Still others have proposed that progressively anterior regions maintain more complex rules that choose a class of more specific, lower-level rules; the lowest being the rule that specifies a motor responses28. For example, the choice to write an e-mail is abstract relative to choices about what words to put in the e-mail itself. Regardless of the specific definition of abstraction, the data consistently demonstrate that more rostral regions of frontal cortex are associated with progressively abstract control demands and representations. Some theorists further interpret these data as reflecting a hierarchical organization of lateral frontal cortex, whereby control processes or representations at a given locus in the frontal lobes are influenced by more abstract control processing in ‘higher’, more anterior regions, but

1Department

of Cognitive and Linguistic Sciences, and 2Department of Psychology, Brown University, Providence, Rhode Island, USA. 3Helen Wills Neuroscience Institute and 4Department of Psychology, University of California, Berkeley, Berkeley, California, USA. Correspondence should be addressed to D.B. ([email protected]).

Received 2 December 2008; accepted 21 January 2009; published online 1 March 2009; doi:10.1038/nn.2277

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

515

ARTICLES

a

Response task Low response

Mid response

High response

+ 1

+ +

2

1

12

Low feature

Mid feature

High feature

Low dimension

Mid dimension

High dimension

1 2 34

1 Feature task Target = Target =

Positive

Target =

Negative Positive

Dimension and context tasks Dimension = shape

Shape Low context

Nonmatch

Orientation Shape Orientation Size Texture

Mid context

High context Block 3

Dimension = shape

Nonmatch

© 2009 Nature America, Inc. All rights reserved.

Shape

Dimension = orient

Match

Blocks 1 and 2

Blocks 5 and 6 Size Size Block 4

Shape Shape

Size

Size

Shape Shape

Size

Shape Shape

Size Orient Orient Texture Texture

Feature Dimension Response competition competition competition

b Response task

Feature task

Dimension task

Context task

Context competition

Low

1 response

1 target

1 dimension

100% mapping

Mid

2 responses

1 target

1 dimension

100% mapping

High

4 responses

1 target

1 dimension

100% mapping

Low

2 responses

1 target

1 dimension

100% mapping

Mid

2 responses

2 targets

1 dimension

100% mapping

High

2 responses

4 targets

1 dimension

100% mapping

Low

2 responses

2 targets

1 dimension

100% mapping

Mid

2 responses

2 targets

2 dimensions

100% mapping

High

2 responses

2 targets

4 dimensions

100% mapping

Low

2 responses

2 targets

2 dimensions

100% mapping

Mid

2 responses

2 targets

2 dimensions

50% mapping

High

2 responses

2 targets

2 dimensions

25% mapping

First order control

Second order control

Third order control

Fourth order control

not in ‘lower’, more posterior regions. Such a hierarchical influence could reflect the passing/summing of control signals from anterior to posterior in the frontal lobe31, or the reduction of uncertainty at lower levels by action pathways chosen at higher levels28 or by activating/ coordinating task sets among lower-order processors35–37. Anatomical evidence suggests that there is an asymmetry in the corticocortical connections in frontal cortex that could support such a processing hierarchy38,39. Indirect evidence from effective connectivity analysis of functional magnetic resonance imaging (fMRI) data also supports an asymmetric anterior to posterior flow of influence31,35–37. However, the neuroimaging data cannot be conclusive on this point. Indeed, some perspectives can account for a rostro-caudal functional gradient without a requirement that the processing architecture be hierarchical11,29,40,41. Thus, a fundamental issue to resolve is whether the observed rostro-caudal gradient reflects a hierarchical or nonhierarchical organization of function32. An anterior-to-posterior flow of control processing in the frontal lobes predicts that performance on tasks involving higher-order control should be impaired by disruptions to lower-order processors, even

516

Figure 1 Trial events and task analysis of the four response-selection tasks. (a) On each trial of the response task, participants chose a response key on the basis of the color of a presented square. Competition conditions were low (one response), mid (two alternative responses) and high (four alternative responses). On each trial of the feature task, the participant looked for a particular target feature (for example, a mottled texture) based on the color of the square. They made a positive response if the target feature was presented and a negative response otherwise. Competition conditions included one target feature (low), two alternative target features (mid) or four alternative target features (high). Logically, this manipulation increases the number of sets of response mappings from one to four. Thus, the number of targets may be thought of as the number of response sets. On each trial of the dimension and context tasks, the participant decided whether two objects matched along a particular dimension (for example, shape) that was cued by the color of the square. Dimension competition conditions were one dimension (low), two alternative dimensions (mid) or four alternative dimensions (high). During the context experiment, there were always two alternative dimensions, but competition was introduced by decreasing the frequency with which a given color mapped to a given dimension (low, 100%; mid, 50%; high, 25% mapping frequency). Thus, by definition, from the first order through the fourth order of the hierarchy, competition was defined by the number of responses, targets, dimensions and mappings, respectively. (b) A task analysis depicts the nested hierarchical relationship in control demands (columns) among the four tasks (rows). Color-coding highlights conditions for which competition at the response (blue), feature (yellow), dimension (green) or context (red) levels was present. Thus, this table indicates how control demands at different levels accumulate as each level of contingency is added in each task. Also, note that the low-competition condition of each task is equivalent in control demands to the mid condition of the task one level subordinate. Finally, the red outline highlights the conditions permitting a crossover interaction.

when the higher-order processors are intact. However, the reverse prediction should not hold. Performance should be unaffected for tasks involving only intact lower-order processors when higher-order processors are impaired. This hypothesized asymmetric pattern of deficit cannot be directly tested with neurophysiological methods, such as fMRI, electroencephalography or single-unit recording. Rather, it requires a lesion method that leads to isolated disruption of specific processors along the proposed hierarchical gradient. To test the asymmetry hypothesis, we asked 12 individuals with focal frontal lobe lesions and 24 age-matched controls to perform a set of four response-selection tasks (Fig. 1a and Methods) that required increasing levels of hierarchically ordered control to select a correct keypress response. In other words, from the response to feature to dimension to context task (Fig. 1b), the appropriate representations to be selected increased in abstraction. To manipulate control in each of the four response-selection tasks, there were low-, mid- and highcompetition conditions. These required either no selection (low) or selection from two (mid) or four (high) candidate representations at a given level of abstraction, respectively. For example, in the response task

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES First order control

a 5,000

Second order Third order Fourth order control control control

* * *

Response time (ms)

4,000

*

* 3,000

*

2,000

Patients Controls

* **

*

*

1,000

*

n.s. 0

b

Low Mid High Low Mid High Low Mid High Low Mid High Response Feature Dimension Context

1.0

***

0.9

‡

Proportion deficit

0.7 0.6 0.5 0.4 0.3 0.2 0.1 )

)

te xt

p(

C

on

)

on

re

p (D) p (D|L) p (D|H)

p(

D

im

en

si

tu Fe a

p(

R

es

po

ns

e)

0

p(

© 2009 Nature America, Inc. All rights reserved.

0.8

(first order of abstraction), participants selected a response on the basis of a learned mapping with a color cue presented on each trial. Competition increased as participants went from having no choice (one response, low) to having two (mid) or four (high) responses to choose from. As the hierarchical rank of the tasks increased from response to feature to dimension to context (Fig. 1b), the mid- and high-competition conditions of each task required selection of a more abstract representation than the task ranked below it, and so demanded higher-order control. For example, rather than only requiring selection of a response on the basis of a learned mapping (the response task), the mid- and high-competition conditions of the feature task (secondorder control) required selection of a set of response mappings over other competitor sets. This logic was carried up four levels of abstraction across the four tasks. In the low-selection condition of each task, competition was set equivalent to that of the mid-competition condition of the task ranked immediately below it (this logic is spelled out explicitly in Fig. 1b). For example, the low-competition condition of the feature task (Fig. 1b) required selection from among two responses, but from only one response set (defined on the basis of the target). Thus, the control demands for this condition were identical to the mid-competition condition of the response task, which also required selection from two responses, but from only one response set (Fig. 1b). In contrast, the mid- and high-competition conditions of the feature task required the selection of a response set from two or four candidate response sets, respectively. Thus, the low-competition condition provides an estimate of the contribution of lower-order control demands. Moreover, comparison of the mid- and high- with the low-competition condition provides within-task control for superficial differences between the tasks themselves. Using fMRI, we previously demonstrated that the hierarchical level of control in these tasks determines the locus of activation along

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

Figure 2 Overall performance across the four tasks. (a) Reaction time for patients (red) and controls (gray) is plotted across the four tasks that increase, from left to right, in degree (low, mid and high) and order of competition (response, feature, dimension and context). Color-coding indicates conditions across tasks that include equivalent levels of conflict at the response (blue), feature (yellow), dimension (green) or context (red) levels but no higher-order competition. Notably, the difference between patients and controls grew as higher-order control was required (* P o 0.05, ** P ¼ 0.06, error bars represent s.e.m.). (b) The proportion of patients showing deficits in each task grew quantitatively, but not reliably, from response to context (left). Notably, however, the probability of a deficit at any level, p(D), was reliably greater when conditioned on a deficit at any lower level, p(D|L), relative to when it was conditioned on a deficit at any higher level, p(D|H). (*** indicates Bayes factor of 6.7, z indicates Bayes factor of 2, error bars represent s.e.m.)

the rostro-caudal axis of lateral frontal cortex, with response, feature, dimension and context control being associated with PMd (BBrodmann area 6, Montreal Neurological Institute (MNI) standard: x ¼ –30, y ¼ –10, z ¼ 68), anterior PMd/posterior PFC (pre-PMd, BBrodmann area 8, MNI: x ¼
38, y ¼ 10, z ¼ 34), inferior frontal sulcus (IFS, BBrodmann area 45, MNI: x ¼
50, y ¼ 26, z ¼ 24; Brodmann area 9/46, MNI: x ¼
52, y ¼ 28, z ¼ 38) and frontal polar cortex (BBrodmann area 10, MNI: x ¼
36, y ¼ 50, z ¼ 6), respectively28. In contrast, the parametric increase in competition (low, mid and high) was associated with a corresponding increase in activation only at the frontal locus supporting that level of control28. Depending on the site of damage to the frontal lobe, individuals with lesions should be impaired for the mid- and high-competition conditions at the level of abstraction at which disruption of control has occurred and all conditions of tasks at more abstract levels, despite having intact control processors at these levels. In contrast, individuals with lesions should perform normally on the low-competition condition of the impaired level and all conditions of tasks at lower levels. In this study, we tested this hypothesis most directly for two hierarchical levels. Specifically, we tested that a lesion to pre-PMd should impair performance on the mid- and high-competition conditions of the feature task and all conditions of the dimension and context tasks because these all require a second order of control. However, this lesion should not impair performance on any condition of the response task or the low condition of the feature task because these only require a first order of control. In contrast, a lesion to IFS should impair performance on the mid- and high-competition conditions of the dimension task and all conditions of the context task because these all require a third order of control. However, such a lesion should not impair performance on any condition of the response or feature task or the low condition of dimension task because these only require first and second orders of control. Such a pattern of results would be direct evidence for hierarchy in the frontal lobe. RESULTS In general, individuals with frontal lesions demonstrated increasingly poor performance as demands on control increased in abstraction across the four experiments (F3,27 ¼ 10.6, P o 0.0001; Fig. 2a). Post hoc contrasts demonstrated that this interaction was partially derived from reliable increases in reaction time for the conflict conditions (mid and high) of response to feature to dimension to context tasks (F 4 6.4, P o 0.05). Differences in error rates between patients and controls also followed an increasing pattern (F3,27 ¼ 5.2, P o 0.05; see Supplementary Fig. 1 online). However, because sources of error were more variable, our analysis focused on correct trial reaction time. The increasing difference in reaction time between patients and controls across tasks could reflect control deficits in two ways:

517

ARTICLES Supplementary Table 1 online). This asymmetry provides initial support for hierarchical dependencies among deficits at the different fMRI (feature effect) levels and the aggregation account of the group data. Lesion overlap Response Feature Dimension Context Next, we considered whether these hierFeature deficit 1 2 3 archical deficit dependencies were related to the locus of damage along the rostro-caudal axis of the frontal lobes. An observerindependent method assigned patients to lesion overlap groups on the basis of their Response Feature Dimension Context Dimension deficit behavioral performance across the four tasks. Vectors were created that corresponded to the idealized behavior of a patient with a selective deficit at a particular hierarchical level, such 1 4 7 as response, feature, dimension or context Response Feature Dimension Context (Fig. 3a). These vectors served as regressors Context deficit fMRI (dimension effect) in a multiple regression on each patient’s reaction time differences from age-matched Lesion overlap controls across all conditions of all experiDimension deficit ments. The assumption of this multiple regression approach is that if a patient has Response Feature Dimension Context damage encompassing more than one level of Figure 3 Observer-independent overlap analysis. (a) Regressors were generated representing the pattern control, then their behavioral profile will be of data across tasks and conditions for an idealized deficit at each level of control on the basis of the consistent with the linear sum of these two asymmetrical hierarchical assumptions. Bars indicate difference from controls in arbitrary units. Colordeficit profiles. When the resulting partial coding highlights the conditions for which deficits should emerge for patients with impairments in correlation coefficient associated with a response (blue), feature (yellow), dimension (green) and context (red) control. (b) Results from the particular regressor was positive and signifilesions overlap analysis revealed a distinction in the peak of overlap (red) among dimension patients cant, favoring inclusion rather than exclusion around the IFS/dorsolateral prefrontal cortex and the peak of overlap (red) among feature patients in anterior dorsal premotor cortex. Color bar indicates the number of patients contributing to each colored (P o 0.1), a patient was assigned to that lesion region. Insets show correspondence between sites of lesion overlap from this study and the activation overlap group. associated with the parametric effect of dimension (top) and feature (bottom) conflict from ref. 28. The model assigned all but two of the lesion Arrows on slices are in the same position for precise comparison. patients to the feature or dimension groups and one patient was assigned to both. The higher-order control demands could increasingly challenge all patients, resulting lesion overlap maps clearly delineated adjacent, but separate, regardless of the site of their lesion, and so their performance could foci of maximal overlap along the rostro-caudal axis of the PFC for the become differentially impaired as the task complexity increases, or feature versus dimension groups (Fig. 3b). The more caudal and dorsal deficits in higher-level tasks will be more likely across patients, regard- focus of lesion overlap in the feature group, approximately pre-PMd, less of the site of their lesion, than deficits at lower level tasks because of corresponds closely to the site of activation associated with the the asymmetric dependencies predicted by hierarchy, and so the larger parametric effect of feature conflict from fMRI of healthy participants28 deficits would reflect this aggregate likelihood. In the latter case, then (Fig. 3b). The more rostral and ventral focus of lesion overlap in the the presence of an impairment at any level should increase the dimension group, straddling the IFS, corresponds almost precisely to likelihood of an impairment at all higher levels, but should not increase the site of activation associated with the parametric effect of dimension the odds of an impairment at a lower level. This can be expressed conflict from fMRI28 (Fig. 3b). The high degree of correspondence as the change over the prior probability of a deficit at any level of between the fMRI and patient lesion overlap results provides strong the hierarchy, p(D), when the probability of a deficit is conditioned convergent support for the participation of these regions in cognitive on a deficit at any lower level, p(D|L), versus a deficit at any higher control at different levels of abstraction. level, p(D|H). Consideration of the behavioral profiles of the feature and dimenA deficit was assigned for a task if a patient’s average performance on sion groups indicated that there was a crossover interaction that mid/high conflict conditions was at least 2 s.d. worse than that of age- distinguishes the behavioral profiles of these two groups of patients matched controls. The probability of a deficit on any task, p(D), was (Fig. 4a,b). The feature group was intact relative to controls through all 62% across the patients. Although there was a quantitative increase in lower-level control conditions and the low-competition condition of the frequency of deficits as tasks required higher levels of control the feature task (F1,2 ¼ 4.8, P o 0.05). A deficit was evident for the mid (Fig. 2b), these deficit frequencies were not reliably different across and high conditions of the feature task and all conditions of the tasks (F ¼ 1.2). Notably, however, the probability of a deficit at any level dimension and context tasks (Fig. 4a). In contrast, the dimension given a deficit at a lower level, p(D|L), was 91% across patients, which group was intact through the low-conflict condition of the dimension was significantly different from p(D) (Bayes factor (posterior odds/ task, and deficient for the mid and high conditions of the dimension prior odds) ¼ 6.7). In contrast, the probability of a deficit at any level task (F1,6 ¼ 49.9, P o 0.0001) and all conditions of the context task given a deficit at a higher level, p(D|H), was only 76%, which was a (Fig. 4a). Excluding the one patient assigned to both groups, these weak change over the prior probability (Bayes factor ¼ 2.0). Notably, distinct profiles produced a crossover interaction between the feature these results were not dependent on the 2 s.d. criterion for a deficit (see and dimension groups (F1,5 ¼ 15.3, P o 0.05; Fig. 4b,c). Response deficit

Low Mid High

Low Mid High Low Mid High Mid High

Low

Low Mid High

Low Mid High

Low Mid High Mid High

Low

Low Mid High

Mid High

Low Mid High Low Mid High

b

Low Mid High Low Mid High Low

518

Mid High Low

© 2009 Nature America, Inc. All rights reserved.

Predicted behavioral difference from controls

Low Mid High

a

Feature deficit

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

ARTICLES

Feature group

Dimension

Context

Response

Dimension

Mid

High

Low

Mid High

Low

Mid

Feature

Context

Dimension-only group

2,500

Response

Feature

Dimension

Context

Response

Feature

Dimension

Figure 4 Performance of dimension and feature patient groups. (a) The differences in reaction time between patients and controls in the feature (left) and dimension (right) groups are plotted across competition conditions and tasks. Colored shading highlights occurrences of a reliable stepwise increase in a feature (yellow) or dimension (green) control deficit (* P o 0.05). (b) The differences in reaction time between patients and controls in the feature (left) and dimension (right) groups, excluding the one patient that was categorized as having both feature and dimension deficits, are plotted across competition conditions and tasks. (c) The differences from controls in the reaction time change between conflict (mid/high) and nonconflict (low) conditions of the feature (left) and dimension (right) tasks are plotted for the feature-only (blue) and dimension-only (red) overlap groups. The crossover interaction supports a double dissociation between these groups. All error bars represent s.e.m.

High

Mid

Low

High

Mid

Low

High

Mid

0

Low

500

High

Mid

Low

High

Mid

Low

High

Mid

Low

High

Mid

500

1,000

High

1,000

1,500

Mid

1,500

2,000

Low

Reaction time difference from controls (ms)

2,500

2,000

0

High

0

Low

500

Mid

Mid

1,000

High

Low

Mid

High

Low

Mid

High

Feature

1,500

Feature-only group

Low

Reaction time difference from controls (ms)

Low

Mid

High

500

2,000

High

1,000

*

Low

1,500

Response

Context

The crossover interaction in behavioral performance by the patients on the feature and dimension tasks demonstrates that, although 1,200 the tasks themselves are not independent, the 800 control processors involved at each level of the Feature group hierarchy are independent, consistent with 400 Dimension group their spatial segregation. Specifically, feature0 deficit patients are impaired on the dimension task not because they have difficulty deciding – 400 Feature competition Dimension competition which dimension is relevant to their match (mid/high–low) (mid/high–low) decision (a third-order choice), but rather as a result of the subsequent determination of a It is notable that the mid-competition condition of the dimension response on the basis of the match relationship between the items task for the feature group showed a smaller difference from controls (second-order choice). We know this because when we subtracted an than the low- or high-competition conditions as a result of chance estimate of the patients’ ability to make this response selection on the variation (F1,2 ¼ 4.4, P ¼ 0.19). However, the crossover interaction is basis of a match decision (the low-conflict dimension condition), there not simply an artifact of this aspect of the experimental design. When was no difference from controls on this task. This subtraction only only data from the high- versus low-competition conditions across the works if the dimension processor can resolve conflict at the dimension feature and dimension tasks were included in the analysis, the crossover level independent of the state (damaged or healthy) of the lower-level pattern was still evident and showed a strong trend (F1,5 ¼ 5.6, feature processor. As such, the data are consistent with a central property P ¼ 0.06). Hence, the interaction does not appear to be restricted to of a hierarchy, namely that controllers at higher levels operate indepenthe mid-competition condition. Likewise, the crossover interaction dently from the status of lower-level processors. Thus, the reason that does not arise from a floor effect that obscures the differences between the feature-deficit patients failed in the dimension task was because of a the low-, mid- and high-competition conditions for the feature group. feature-level deficit. However, the reason that the dimension-deficit Error rates across these conditions for the feature group were 16% patients failed at the same dimension task was because of a dimensionand reaction times across these conditions were 3,268 ms (range: level deficit; they had difficulty deciding which dimension was relevant to their match decision (a third-order choice). Conversely, to the extent 2,060–4,477 ms), well below the response deadline (15 s). that higher-order control is not required by a task, lower-order processors may operate independently of higher-order control, which DISCUSSION These results demonstrate that performance deficits across frontal was evident in the intact performance of dimension patients on the patients grow progressively worse as contingencies are added to an feature task. Thus, the dimension and feature processors are indepenaction decision and cognitive control operates at higher orders of dent, although the feature and dimension tasks themselves are not. More broadly, the crossover interaction also provides rare direct abstraction. However, rather than deriving from a uniform pattern of progressive deficit in each patient, this pattern is the result of an lesion evidence for the widely assumed heterogeneity of function in the asymmetric effect of a control deficit at a given level on higher-level frontal lobes. For example, the crossover interaction may be consistent control tasks. Specifically, the site of damage resulting from stroke with distinct functions that have previously been associated with along the rostro-caudal axis of the frontal lobes results in a deficit at a these regions across separate experiments related to conditional predictable level of abstraction and in tasks requiring higher levels of selection of stimulus information in pre-PMd versus selection of control, but leaves performance on tasks requiring only lower levels of more abstract categorical information and high-level monitoring in IFS/mid-dorsolateral PFC42–47. control intact.

c

Reaction time difference from controls (ms)

© 2009 Nature America, Inc. All rights reserved.

Reaction time difference from controls (ms)

*

2,000

0

b

Dimension group 2,500

2,500

Low

Reaction time difference from controls (ms)

a

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

519

© 2009 Nature America, Inc. All rights reserved.

ARTICLES It is important to clarify that general difficulty, as in difficulty arising from any cause, be it increased abstraction or another factor that makes response times longer, cannot account for the results. Difficulty was manipulated in two ways in this study, both of which result in reaction time increases, but only one of which is related to the locus of damage along the rostro-caudal axis. First, there was the level of abstraction across the experiments, as additional contingencies were added to the action decision. Second, there was the degree of competition at a given level of contingency, which increased parametrically over three levels. Both of these factors produced increases in reaction time. However, the degree of abstraction was associated with the probability effects and the regional differences in lesion overlap. This is similar to the results from our previous fMRI experiment using this same task structure28 that showed that the locus of activation along the rostro-caudal axis was related to the degree of abstraction and not to the level of competition. Instead, the level of competition selectively increased the amplitude of the fMRI response in a given region, but did not determine which region was activated along the rostro-caudal axis. Therefore, given that we manipulated two types of difficulty and located these effects related to only one, a single construct of difficulty cannot fully account for these or our previous results. It should also be clarified that the lesion overlap approach taken here differs from other published approaches that define patient groups on the basis of their lesion overlap and then assess any behavioral differences48,49. Here, we defined our patient groups on the basis of a behavioral deficit profile and then looked at the regions of overlap among patients with a common profile. Although it is sometimes difficult to predict the behavioral profile of a particular patient on the basis of the location of their lesion, our results demonstrate that if a patient has a particular behavioral profile, then there is some consistency regarding the rostro-caudal locus of their lesion and how that patient will perform on other tasks requiring higher or lower levels of cognitive control. Finally, our design does not address whether lower-order control processors are differentially affected by impairments in higher-order control when between-level interactions are required to complete a task. For example, higher-order decisions could modulate the degree of competition present at lower levels, as in constraining the number of lower-level choices by choosing a higher-level path. Such a test will be required to demonstrate that higher levels modulate lower levels, an important prediction in a strong processing hierarchy. Moreover, there may be feedback influences of lower-level on higher-level control. Our findings suggest that hierarchy may be a fruitful framework in which to understand frontal lobe architecture and systems-level processing and motivate further study. METHODS Patients and controls. We recruited 11 patients (average 56.6 years, range 45– 73; 4 female) from the Northern California Veterans Administration Health Care System. An additional patient was recruited, but was unable to perform any of the tasks. Damage in all of the patients was caused by cerebral infarction of the middle cerebral artery. Testing took place at least 6 months post-stroke. The extent of damage was assessed from structural MRI or computed tomography scans. Estimates were reconstructed on normalized templates by an expert anatomist who was blind to patient performance and then digitized to assess overlap. For visualization, digital masks were overlaid on a highresolution MNI canonical image using MRIcro (http://www.sph.sc.edu/comd/ rorden/mricro.html). We enrolled 43 control participants (26 female) following screening for any history of neurological or psychiatric disorder. The entire cohort of controls ranged in age from 21 to 69 years. For each patient, controls were selected from the cohort whose age was within 5 years of the lesion patient. From this

520

selection procedure, a subset of 24 controls (12 female, ages 41–69) was included in the analysis. Patients and controls were thus matched for age (average difference was
0.14 years, t10 ¼ 0.34) and years of education (average difference was 2.1 years, t10 ¼ 2.1). Patients and controls had normal or corrected-to-normal vision, and normal color vision, as verified by the Ishihara test for color deficiency. Informed consent was obtained in accordance with procedures approved by the Committees for Protection of Human Subjects at the University of California, Berkeley and the Northern California Veterans Administration Healthcare System. Participants were paid for their participation in the study. Behavioral tasks. Patients and controls were tested on a battery of four response-selection tasks that were designed to test progressively higher degrees of hierarchically ordered control (Fig. 1a). These tasks were adapted from our previous fMRI experiment28. In the response task, participants viewed a series of colored boxes that were presented one at a time and selected a response on a keypad on the basis of the box color. Competition increased with the number of alternative responses on a given block of trials increasing from one (low) to two (mid) to four (high). In the feature task, the series of colored boxes each contained a single object that varied from trial to trial along one perceptual dimension (either texture or orientation between subjects). The participants were required to decide whether a particular target feature along that dimension (that is, a rough texture) was presented on each trial. The participant made a positive response on the keypad if the target feature was present and a negative response to any other feature. The target feature that cued a positive response for a given trial was itself cued by the color of the box. Competition increased with the number of alternative target features for a given block of trials increasing from one (low) to two (mid) to four (high). In the dimension task, the series of colored boxes each contained two objects that each varied along four dimensions (texture, shape, size and orientation) from trial to trial. The participants were required to decide whether the objects matched along only one of those dimensions on each trial. The relevant dimension was cued by the colored box. Competition increased with the number of alternative dimensions for a given block increasing from one (low) to two (mid) to four (high). The context task was identical in terms of the task instructions to the dimension experiment, except that two dimensions were always relevant across all blocked conditions. Moreover, in the context task, a given color cue could map to different dimensions on different blocks (in the dimension task, a given color always mapped to one dimension). Thus, in the context task, it is necessary to use information about the current temporal frame (the current block or the most recent instructions) to select the appropriate mapping for a given color cue. Thus, competition was manipulated by varying the frequency across blocks that a given color cue (the context) mapped to a specific dimension. Certain color-to-dimension mappings were relevant for 100% of the blocks in which that cue was encountered, other color-to-dimension mappings were relevant for 50% of blocks in which that color was encountered and other color-to-dimension mappings were relevant on only 25% of blocks in which that color is encountered. In the latter two cases, determining which color-to-dimension mapping is currently relevant required the selection of a particular color-to-dimension mapping on the basis of the instructions of the current block. In this way, as the frequency of a given color-to-dimension mapping decreases, uncertainty or competition with other mappings increases and so selection of the currently relevant mapping requires more control. Design and experimental parameters. The four tasks were tested across 2–4 sessions for each participant. To control for mapping frequencies, the context task was always performed first. The remaining three tasks were counterbalanced for order across participants with the constraint that at least one task (either feature or response) come before the performance of the dimension task. The response, dimension and context tasks included 192 trials and the feature task included 186 trials. The response, feature and dimension tasks consisted of six blocks, two of each competition condition (low, mid and high), counterbalanced for order across participants. In the context task, there were 12 short blocks that permitted manipulation of mapping frequencies from low to

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE

© 2009 Nature America, Inc. All rights reserved.

ARTICLES mid to high across blocks. The order of blocks, cycled twice, was two blocks of low-competition conditions, followed by two blocks of high-competition conditions, and finally by two blocks of mid-competition conditions. This fixed order was provided so that participants could take advantage of order as an additional cue for selecting the appropriate color-to-dimension mapping. The low, high and mid order was used to decouple fatigue or practice effects from the parametric manipulation of competition. Individual trials in all experiments were self-paced up to a limit of 15 s. However, all participants were encouraged to respond as quickly and as accurately as possible on every trial. The specific color mappings, responses and objects used in the tasks were counterbalanced across subjects and two color sets were used to minimize confusion between tasks. Where applicable in each experiment, color cue, response, feature and dimension switches were controlled for frequency across blocks of each condition. All combinations of colors and features in the feature experiment and colors and shapes in the dimension and context experiments were controlled across competition and switching conditions. Prior to performing each task, patients and controls were shown all the color mappings that they would encounter for that task, one block at a time. The mappings were covered and the participants were quizzed verbally. They then performed two practice blocks with the mapping set that they had just memorized. In the first practice block, the relevant mappings were available at the top of the screen, if a reminder was needed. The second practice block was identical to the experimental setting. Data analysis. Median reaction time was obtained for each participant from correct trials. In cases in which a control subject performed greater than 2 s.d. above the mean reaction time of the entire control cohort (n ¼ 43) or if their error rates were at chance for a given condition, that participant was excluded from the group average for that particular task. We conducted a deficit probability analysis to determine the change over the prior probability of a deficit at any level, p(D), given a deficit at a lower, p(D|L), and at a higher, p(D|H), level. First, the patient and control data were linearly corrected for simple motor speed by subtracting the reaction time for the response task low-competition condition from all other reaction times. This provides a measure of simple reaction time and estimates speed in the absence of cognitive control. The average reaction times of the mid and high conditions of each task were calculated for the patients and then standardized to a Z score on the basis of the matched control distribution. For standardized scores greater than 2 s.d., the corresponding task was coded as being deficient. For example, if the average reaction time for the mid and high conditions of the dimension task was 2 s.d. above the mean of controls for a particular patient, then this patient was listed as having a deficit at the dimension level. These deficit counts were then used to calculate the following probabilities: the probability of a deficit at each level, p(Response), p(Feature), p(Dimension) and p(Context), the base rate probability of a deficit at any level, p(D), the conditional probability of a deficit at any level given a deficit at a lower level, p(D|L), and the conditional probability of a deficit at any level given a deficit at a higher level, p(D|H). The Bayes factor is the ratio of the posterior odds to the prior odds. The convention is that a Bayes factor less than 3 is considered to be negligible, a factor between 3 and 10 is substantial or implies supportive evidence, and a factor above 10 is considered to be strong evidence50. This analysis was also conducted for deficit criteria ranging from 1–2.5 s.d. (see Supplementary Table 1). We used an observer-independent overlap method to assign patients to lesion overlap groups on the basis of their behavioral performance across the tasks. Regressors were created that reflected the predicted deficits for patients in each of the four groups, response, feature, dimension and context (Fig. 3a). These predictions derived from the hierarchy hypothesis and made three assumptions: conditions that include competition only at levels below the level of deficit will be intact, performance at the level of deficit will be worse depending on the degree of competition at that level (thus, performance will get parametrically worse with parametric increases in competition) and performance on conditions that include competition at higher levels will also be impaired. These prediction vectors were then included in a multiple regression. Patients who had a reliable and positive partial correlation coefficient were

NATURE NEUROSCIENCE VOLUME 12

[

NUMBER 4

[

APRIL 2009

included in a particular overlap group. A lenient threshold was used for inclusion (P o 0.1) to include as many patients as possible in the overlap maps. Overlap masks were created using MRIcro on the basis of the normalized lesion masks generated for each patient. The mask of one patient with a rightsided lesion was mirror flipped to permit its comparison with the left sided lesions of the other patients in the group. Behavioral averages from the group assignments, after excluding the one subject included in both groups, were used to compute the behavioral crossover interaction. Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We are grateful to R.T. Knight and D. Scabini for their help with patient recruitment and lesion characterization. We also would like to thank W. Heindel and A. Kayser for their input on revisions of this manuscript. This work was supported by the US National Institutes of Health (grants MH63901 and NS40813), the Veterans Administration Research Service and a National Research Service Award (F32 NS053337). Published online at http://www.nature.com/natureneuroscience/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Badre, D. & Wagner, A.D. Selection, integration and conflict monitoring; assessing the nature and generality of prefrontal cognitive control mechanisms. Neuron 41, 473–487 (2004). 2. D’Esposito, M. et al. The neural basis of the central executive system of working memory. Nature 378, 279–281 (1995). 3. D’Esposito, M., Postle, B.R. & Rypma, B. Prefrontal cortical contributions to working memory: evidence from event-related fMRI studies. Exp. Brain Res. 133, 3–11 (2000). 4. Duncan, J. An adaptive coding model of neural function in prefrontal cortex. Nat. Rev. Neurosci. 2, 820–829 (2001). 5. Fuster, J.M. The Prefrontal Cortex: Anatomy, Physiology and Neuropsychology of the Frontal Lobe (Lippincott-Raven Publishers, Philadelphia, Pennsylvania, 1997). 6. Goldman-Rakic, P.S. The prefrontal landscape: implications of functional architecture for understanding human mentation and the central executive. Phil. Trans. R. Soc. Lond. B 351, 1445–1453 (1996). 7. Koechlin, E. & Summerfield, C. An information theoretical approach to prefrontal executive function. Trends Cogn. Sci. 11, 229–235 (2007). 8. Miller, E.K. & Cohen, J.D. An integrative theory of prefrontal cortex function. Annu. Rev. Neurosci. 24, 167–202 (2001). 9. Passingham, R.E. The Frontal Lobes and Voluntary Action. (Oxford University Press, Oxford, 1993). 10. Passingham, R.E. & Rowe, J.B. Dorsal prefrontal cortex: maintenance in memory or attentional selection? in Principles of Frontal Lobe Function (eds. Stuss, D.T. & Knight, RT.) 221–232 (Oxford University Press, Oxford, 2002). 11. Petrides, M. Lateral prefrontal cortex: architectonic and functional organization. Phil. Trans. R. Soc. Lond. B. 360, 781–795 (2005). 12. Shallice, T. & Burgess, P.W. Deficits in strategy application following frontal lobe damage in man. Brain 114, 727–741 (1991). 13. Smith, E.E. & Jonides, J. Storage and executive processes in the frontal lobes. Science 283, 1657–1661 (1999). 14. Stuss, D.T. & Benson, D.F. The frontal lobes and control of cognition and memory. in The Frontal Lobes Revisited (ed. Perecman, E.) 141–158 (The IRBN Press, New York, 1987). 15. Bunge, S.A. How we use rules to select actions: a review of evidence from cognitive neuroscience. Cogn. Affect. Behav. Neurosci. 4, 564–579 (2004). 16. Carter, C.S. et al. Anterior cingulate cortex, error detection and the on-line monitoring of performance. Science 280, 747–749 (1998). 17. O’Reilly, R.C. & Frank, M.J. Making working memory work: a computational model of learning in the prefrontal cortex and basal ganglia. Neural Comput. 18, 283–328 (2006). 18. Courtney, S.M., Roth, J.K. & Sala, J.B. A hierarchical biased-competition model of domain-dependent working memory maintenance and executive control. in The Cognitive Neuroscience of Working Memory (eds. Osaka, N., Logie, R. & D’Esposito, M.) 369–383 (Oxford University Press, Oxford, 2007). 19. Cohen, J.D., Dunbar, K. & McClelland, J.L. On the control of automatic processes: a parallel distributed processing account of the Stroop effect. Psychol. Rev. 97, 332–361 (1990). 20. Cohen, J.D. & Servan-Schreiber, D. Context, cortex and dopamine: a connectionist approach to behavior and biology in schizophrenia. Psychol. Rev. 99, 45–77 (1992). 21. Desimone, R. & Duncan, J. Neural mechanisms of selective visual attention. Annu. Rev. Neurosci. 18, 193–222 (1995). 22. MacDonald, A.W. III, Cohen, J.D., Stenger, V.A. & Carter, C.S. Dissociating the role of the dorsolateral prefrontal and anterior cingulate cortex in cognitive control. Science 288, 1835–1838 (2000). 23. Wallis, J.D., Anderson, K.C. & Miller, E.K. Single neurons in prefrontal cortex encode abstract rules. Nature 411, 953–956 (2001). 24. Botvinick, M.M. Multilevel structure in behaviour and in the brain: a model of Fuster’s hierarchy. Philos. Trans. R. Soc. Lond. B Biol. Sci. 362, 1615–1626 (2007).

521

© 2009 Nature America, Inc. All rights reserved.

ARTICLES 25. Botvinick, M.M. Hierarchical models of behavior and prefrontal function. Trends Cogn. Sci. 12, 201–208 (2008). 26. O’Reilly, R.C., Noelle, D.C., Braver, T.S. & Cohen, J.D. Prefrontal cortex and dynamic categorization tasks: representational organization and neuromodulatory control. Cereb. Cortex 12, 246–257 (2002). 27. Hazy, T.E., Frank, M.J. & O’Reilly, R.C. Towards an executive without a homunculus: computational models of the prefrontal cortex/basal ganglia system. Philos. Trans. R. Soc. Lond. B Biol. Sci. 362, 1601–1613 (2007). 28. Badre, D. & D’Esposito, M. Functional magnetic resonance imaging evidence for a hierarchical organization of the prefrontal cortex. J. Cogn. Neurosci. 19, 2082–2099 (2007). 29. Christoff, K. & Keramatian, K. Abstraction of mental representations: theoretical considerations and neuroscientific evidence. in Perspectives on Rule-Guided Behavior (eds. Bunge, S.A. & Wallis, J.D.) (Oxford University Press, New York, 2007). 30. Koechlin, E. & Jubault, T. Broca’s area and the hierarchical organization of human behavior. Neuron 50, 963–974 (2006). 31. Koechlin, E., Ody, C. & Kouneiher, F. The architecture of cognitive control in the human prefrontal cortex. Science 302, 1181–1185 (2003). 32. Badre, D. Cognitive control, hierarchy and the rostro-caudal organization of the frontal lobes. Trends Cogn. Sci. 12, 193–200 (2008). 33. Buckner, R.L. Functional-anatomic correlates of control processes in memory. J. Neurosci. 23, 3999–4004 (2003). 34. Fuster, J.M. The prefrontal cortex—an update: time is of the essence. Neuron 30, 319–333 (2001). 35. Rowe, J.B. et al. Is the prefrontal cortex necessary for establishing cognitive sets? J. Neurosci. 27, 13303–13310 (2007). 36. Sakai, K. & Passingham, R.E. Prefrontal interactions reflect future task operations. Nat. Neurosci. 6, 75–81 (2003). 37. Sakai, K. & Passingham, R.E. Prefrontal set activity predicts rule-specific neural processing during subsequent cognitive performance. J. Neurosci. 26, 1211–1218 (2006).

522

38. Barbas, H. & Pandya, D.N. Architecture and frontal cortical connections of the premotor cortex (area 6) in the rhesus monkey. J. Comp. Neurol. 256, 211–228 (1987). 39. Petrides, M. & Pandya, D.N. Efferent association pathways from the rostral prefrontal cortex in the macaque monkey. J. Neurosci. 27, 11573–11586 (2007). 40. Christoff, K. & Gabrieli, J.D.E. The frontopolar cortex and human cognition: evidence for a rostrocaudal hierarchal organization within the human prefrontal cortex. Psychobiology 28, 168–186 (2000). 41. Christoff, K., Ream, J.M., Geddes, L.P.T. & Gabrieli, J.D.E. Evaluating self-generated information: anterior prefrontal contributions to human cognition. Behav. Neurosci. 117, 1161–1168 (2003). 42. Halsband, U. & Passingham, R.E. Premotor cortex and the conditions for movement in monkeys (Macaca fascicularis). Behav. Brain Res. 18, 269–277 (1985). 43. Hampshire, A., Duncan, J. & Owen, A.M. Selective tuning of the blood oxygenation leveldependent response during simple target detection dissociates human frontoparietal subregions. J. Neurosci. 27, 6219–6223 (2007). 44. Hampshire, A., Thompson, R., Duncan, J. & Owen, A.M. The target selective neural response–similarity, ambiguity, and learning effects. PLoS ONE 3, e2520 (2008). 45. Heekeren, H.R., Marrett, S., Bandettini, P.A. & Ungerleider, L.G. A general mechanism for perceptual decision-making in the human brain. Nature 431, 859–862 (2004). 46. Petrides, M., Alivisatos, B., Evans, A.C. & Meyer, E. Dissociation of human mid-dorsolateral from posterior dorsolateral frontal cortex in memory processing. Proc. Natl. Acad. Sci. USA 90, 873–877 (1993). 47. Rushworth, M.F. et al. Attentional selection and action selection in the ventral and orbital prefrontal cortex. J. Neurosci. 25, 11628–11636 (2005). 48. Dreher, J.C., Koechlin, E., Tierney, M. & Grafman, J. Damage to the fronto-polar cortex is associated with impaired multitasking. PLoS ONE 3, e3227 (2008). 49. Thompson-Schill, S.L. et al. Verb generation in patients with focal frontal lesions: a neuropsychological test of neuroimaging findings. Proc. Natl. Acad. Sci. USA 95, 15855–15860 (1998). 50. Jeffreys, H. Theory of Probability (Clarendon Press, Oxford, 1961).

VOLUME 12

[

NUMBER 4

[

APRIL 2009 NATURE NEUROSCIENCE