You are on page 1of 149

volume 12 number 12 december 2009

E d i to r i a l
1475 Butting heads

book review
Notch signaling is essential for the maintenance of adult neural stem cells in vivo. Andreu-Agull and colleagues show that PEDF, released from endothelial cells, enhances Notch signaling in the mouse subependymal zone by inactivating a repressor of Notch target genes. On the cover are daughter cell pairs stained for epidermal growth factor receptor (red), the intracellular domain of Notch (green) and DAPI (blue). (pp 1514 and 1481)

1477 Am I Making Myself Clear? A Scientists Guide to Talking to the Public By Cornelia Dean Reviewed by Dario L Ringach

2009 Nature America, Inc. All rights reserved.

news and views


1479 Neuronal death or dismemberment mediated by Sox14 Jeannette M Osterloh & Marc R Freeman see also p 1497 1481 Pigment epithelium-derived growth factor: modulating adult neural stem cell self-renewal Andrew Chojnacki & Samuel Weiss see also p 1514 1483 Hippocampal theta rhythms follow the beat of their own drum Laura Lee Colgin & Edvard I Moser see also p 1491

b r i e f c o m m u n i c at i o n s
1485 Experience-dependent compartmentalized dendritic plasticity in rat hippocampal CA1 pyramidal neuronsv J K Makara, A Losonczy, Q Wen & J C Magee 1488 Self-modulation of neocortical pyramidal neurons by endocannabinoids S Marinelli, S Pacioni, A Cannich, G Marsicano & A Bacci 1491 Self-generated theta oscillations in the hippocampus R Goutagny, J Jackson & Sylvain Williams see also p 1483 1494 The pathways of interoceptive awareness S S Khalsa, D Rudrauf, J S Feinstein & D Tranel

Experience-dependent dendritic plasticity (p 1485)

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, NY10013-1917. Tel: (212) 726 9319, Fax: (212) 696 0978. Annual subscription rates: USA/Canada: US$225 (personal), US$3,520 (institution). Canada add 5% GST #104911595RT001; Euro-zone: 287 (personal), 2,795 (institution); Rest of world (excluding China, Japan, Korea): 185 (personal), 1,806 (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 PUBAGREEMENT #40032744. Printed by Publishers Press, Inc., Lebanon Junction, KY, USA. Copyright 2009 Nature Publishing Group. Printed in USA.

volume 12 number 12 december 2009

articles
1497 A genetic pathway composed of Sox14 and Mical governs severing of dendrites during pruning D Kirilly, Y Gu, Y Huang, Z Wu, A Bashirullah, B C Low, A L Kolodkin, H Wang & F Yu see also p 1479 1506 Nardilysin regulates axonal maturation and myelination in the central and peripheral nervous system M Ohno, Y Hiraoka, T Matsuoka, H Tomimoto, K Takao, T Miyakawa, N Oshima, H Kiyonari, T Kimura, T Kita & E Nishi 1514 Vascular niche factor PEDF modulates Notch-dependent stemness in the adult subependymal zone C Andreu-Agull, J M Morante-Redolat, A C Delgado & I Farias see also p 1481 1524 Adult generation of glutamatergic olfactory bulb interneurons M S Brill, J Ninkovic, E Winpenny, R D Hodge, I Ozen, R Yang, A Lepier, S Gascn, F Erdelyi, G Szabo, C Parras, F Guillemot, M Frotscher, B Berninger, R F Hevner, O Raineteau & M Gtz 1534 Glial precursors clear sensory neuron corpses during development via Jedi-1, an engulfment receptor H-H Wu, E Bellmunt, J L Scheib, V Venegas, C Burkert, L F Reichardt, Z Zhou, I Farias & B D Carter 1542 Leucine-rich repeat transmembrane proteins instruct discrete dendrite targeting in an olfactory map W Hong, H Zhu, C J Potter, G Barsh, M Kurusu, K Zinn & L Luo
Clearing corpses in the developing DRG (p 1534)

Adult glutamatergic interneuron generation (p 1524)

2009 Nature America, Inc. All rights reserved.

1551 Structural requirements for the activation of vomeronasal sensory neurons by MHC peptides T Leinders-Zufall, T Ishii, P Mombaerts, F Zufall & T Boehm 1559 Dynamic DNA methylation programs persistent adverse effects of early-life stress C Murgatroyd, A V Patchev, Y Wu, V Micale, Y Bockmhl, D Fischer, F Holsboer, C T Wotjak, O F X Almeida & D Spengler 1567 Amyloid- as a positive endogenous regulator of release probability at hippocampal synapses E Abramov, I Dolev, H Fogel, G D Ciccotosto, E Ruff & I Slutsky 1577 Input normalization by global feedforward inhibition expands cortical dynamic range F Pouille, A Marin-Burgin, H Adesnik, B V Atallah & M Scanziani 1586 Microcircuitry coordination of cortical motor information in self-initiation of voluntary movements Y Isomura, R Harukuni, T Takekawa, H Aizawa & T Fukai 1594 Attention improves performance primarily by reducing interneuronal correlations M R Cohen & J H R Maunsell

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
Motor cortex microcircuitry for voluntary movements (p 1586)

See back pages.

nature neuroscience

iii

e d i to r i a l

Butting heads
A recent controversy on sport-related dementia underscores the need for comprehensive epidemiology studies.

A
2009 Nature America, Inc. All rights reserved.

ccumulating evidence has suggested that professional players of American football might be at increased risk of early-onset dementia. This October, a string of heated debates on this issue culminated in a US congressional House Judiciary Committee hearing in which the National Football League (NFL) denied a possible link between concussions, chronic traumatic encephalopathy (CTE) and other dementia-related neurodegenerative disorders, despite scientific evidence to the contrary. As an antitrust-exempted franchise that generates billions of dollars annually, the NFL has a deep financial stake in this issue. However, although few would disagree with the notion that repeated head trauma is likely to have bad consequences, there is surprisingly little epidemiological data on the prevalence, both among the general public and in professional athletes, or the factors that could potentially increase the risk of CTE. Instead of denying culpability, the league should stop stonewalling and should aid comprehensive studies that help define the exact neurological risks of playing the sport and work with medical professionals to refine the sport so as to make it safer for professionals and amateurs alike. CTE is a relatively rare neurodegenerative disease with progressive dementia and extensive tau-immunoreactive neurofibrillary tangles throughout the brain. Originally identified in boxers, it is linked to multiple bouts of head injury and its clinical symptoms can be very similar to those of other dementia-linked disorders, such as Alzheimers disease. As with many other neurodegenerative diseases, definitive diagnosis can only be made on postmortem inspection. A limited number of single case studies from postmortem examination of former football players and retrospective cohort studies of boxers have bolstered the plausible association between the two phenomena, but there have been no long-term epidemiology studies to address this issue. By design, epidemiology studies cannot establish causality; they can only identify risk factors for a given disease. Nevertheless, the correlations seen in such studies could provide a wealth of information and help seed future experiments that could explain the disease etiology. Designing and executing a good epidemiology study can be extremely difficult, particularly when it comes to neurodegenerative disorders. For example, the University of Michigan conducted a study (which has yet to be peer-reviewed and published) that reports a higher incidence of dementia and cognitive deficits in former NFL players. However, because this study was based on telephone interviews that retrospectively surveyed subject-reported health history, one caveat is that this design may cause selective self-reporting and increase potential bias on the basis of the specific questions being asked. In addition, the accuracy of subjects remembering correct information or reporting pertinent details could be questionable if they suffer from memory decline in the first place. The best epidemiology studies use a prospective design rather than a retrospective one. Ideally, a group of subjects are monitored by

healthcare professionals over many years, allowing accurate diagnoses of disease progression and factoring in any diagnostic variability. An example of one such study is the recently completed Alzheimers Disease Neuroimaging Initiative (ADNI), which provided a 5-year longitudinal study of 800 elderly volunteers at various stages of cognitive decline. Even ADNI, however, comes with caveats, as the timing and duration of the study is critical. How early and how long should the subjects be monitored? For many disorders such as Parkinsons or Alzheimers, there are neuronal changes that can occur long before the subjects show symptoms, and catching subjects early enough may also be difficult. Diagnostic parameters that may cause subtle brain changes may be missed entirely if the patients are not studied early enough. Despite the potential design issues and interpretational caveats, good epidemiology studies can inspire basic scientists and open new research avenues. An epidemiology study, for example, linked pesticide exposure to an increased rate of idiopathic Parkinsons disease1, which was instrumental in creating a rodent model of Parkinsons disease2 and investigating mitochondrial dysfunction in this disorder. Ideally, however, epidemiologists and basic scientists should work in concert to exchange information and help spur each others research efforts. As Richard Mayeux, an epidemiologist working on Alzheimers research at Columbia University notes, meetings that bring together molecular, genetic and epidemiology researchers have been useful to refine study parameters and to learn about risk factors that can be applied to improve molecular or genetic models of disease. As for the current debate of whether football-related head injuries are a risk factor for CTE, the statistics-obsessed NFL has a wealth of information that could be tapped to help address this gap. The NFL keeps extensive records on each players performance and injuries ( neurological or otherwise) and has the necessary resources to fund regular and independent evaluations of the health of its players. At the recent congressional hearing, the NFL promised to make such data available. It will be essential to have this material evaluated by scientists who are not affiliated with either the league or the players advocacy group to avoid the NFLs and professional players intrinsic conflicts of interest. A prospective large-scale epidemiological study may not be possible using just professional football players, but a case-cohort comparison comparing football players with other professional athletes who do not undergo as many head collisions as football players do (such as baseball players) would be useful. Instead of trying to discredit the existing evidence, the NFL and other professional sports organizations would do well to help scientists, clinicians and the public better understand and prevent sports-associated risks in the first place. L
1. Butterfield, P.G., Valanis, B.G., Spencer, P.S., Lindeman, C.A. & Nutt, J.G. Neurology 43, 11501158 (1993). 2. Betarbet, R. et al. Nat. Neurosci. 3, 13011306 (2000).

nature neuroscience volume 12 | number 12 | december 2009

1475

book review

Scientists talking to the public: is there anyone out there?


Am I Making Myself Clear? A Scientists Guide to Talking to the Public
By Cornelia Dean Harvard University Press, 2009 288 pages, hardcover, $19.95 ISBN 0674036352

Reviewed by Dario L Ringach


In her new book, Am I Making Myself Clear? A Scientists Guide to Talking to the Public, Cornelia Dean, a journalist and former editor of the popular science section at the New York Times, extends a compelling invitation to researchers to participate more in public life, to explain their work to science journalists, to contribute to national policy debates and to do so not only when their funding is at stake. Why do scientists need to talk to the public? Because social policies need to be decided on the basis of rational grounds and facts, including issues ranging from climate change, to the goals of the space program, to the protection of endangered species and the use of embryonic stem cells. Neuroscience research using animals, for example, is being increasingly challenged by some who argue it has nothing to contribute to the development of treatments and cures for human disease. If the public and policymakers do not hear the voice of scientists, if they are not presented with the facts, it may only be a matter of time before a large segment of the public will be asking why are we doing (and why they are paying for) such work. Unfortunately, there are some obstacles between scientists and journalists. Dean correctly notes that the vast majority of researchers do not consider talking to the public as part of their obligation to society. She points out that our current academic system not only fails to promote public discourse, but it actually tends to suppress it. In this context, Donald Kennedy, past editor of Science and President Emeritus of Stanford University, is quoted explaining how researchers expressing an interest in public communication are rapidly treated to a lecture on the dangers of being Sagan-ized, a reference to the widely believed notion that the noted Cornell astrophysicist was denied a place in the National Academy as a result of his public television series Cosmos. Agreeing with this, Alan Leshner, publisher of Science and former director of the National Institute of Mental Health, is quoted as saying that public outreach efforts [must be] among the metrics used to decide promotion and tenure.

The author is in the department of Neurobiology, University of California Los Angeles, Los Angeles, California, USA. e-mail: dario@ucla.edu

Another hurdle to communication that is noted by Dean is one of perception; scientists tend to view journalists as superficial, sensationalist and focused on controversy, whereas journalists view researchers as boring, users of unintelligible jargon, caveating things to death and being unable to identify central points. But the single major issue that appears to drive scientists away from the media is the need for journalistic balance, which leads to many researchers being pitted in public against some crank that will provide the opposing view, resulting in what some view as an overweighting of dissent. The author concedes this problem has no easy solution, representing the major issue in her coverage of science and engineering. A number of chapters provide an excellent practical guide to interacting with the media. This includes tips about how to tell your story, the importance of sound bites, the effective use of your time in radio and TV, writing letters to the editor, Op-Ed pieces and books, writing your own blog, standing on the witness stand, and engaging in policy making. These practical recommendations are all excellent, but may be difficult to follow for the newcomer. The key, Dean says, is to prepare ahead of time. Know your most important points and be ready to deliver them in seconds. When it comes to discussing the actual science, make sure you know your audience, avoid jargon and practice every day by explaining your work to your family, neighbors and whoever is sitting next to you on the plane. Many scientific leaders and politicians are quoted throughout the book in memorable passages encouraging scientists to participate more in public life. In a particularly compelling excerpt from a meeting of the American Association for the Advancement of Science (AAAS), Congressman Sherwood Boehlert tells his audience that Scientists should participate actively, even avidly, in policy debates. Indeed, both as educated citizens and as professionals with relevant knowledge, scientists ought to feel obligated to contribute to policy making in their communities, in the nation and even in the wider world. Addressing another meeting of AAAS, his congressional colleague John Porter brings an entire hall into wild applause by offering the view that Scientists are by every measure the most respected people in America. They are listened to. But if the public and policymakers never hear your voices, never see... science, never understand its methods, the chance of its being high on the list of national priorities will be very low. [...] You can sit on your fingers or you can go outside your comfort zone and get into the game and make a difference for science. Neither we, nor AAAS, nor any other group can do it all for you. Science needs you. Your country needs you. America needs you... fighting for science!. I couldnt agree more. It is time for all of us to have a more active role in society and provide our input to public policy; we must go beyond the publication of scientific articles. I admit my own recent incursion into the media was not driven by any of the above considerations, but from threats by animal rights extremists. I dont recommend you take the same road. Instead, read this powerful book and you will be easily convinced that it is your obligation to devote time outside the laboratory to communicating the wonderful work you are doing, your excitement, its importance and how the public stands to benefit from it. Is there anyone out there ready to talk about science? I certainly am. L

2009 Nature America, Inc. All rights reserved.

nature neuroscience volume 12 | number 12 | december 2009

1477

news and views

Neuronal death or dismemberment mediated by Sox14


Jeannette M Osterloh & Marc R Freeman
The pruning of unneeded axons and dendrites is crucial for circuitry maturation, but poorly understood on the molecular level. During Drosophila metamorphosis, the transcription factor Sox14 acts as a context-dependent mediator of death, axonal or dendritic pruning. Its transcriptional target Mical acts specifically in dendrite pruning.
2009 Nature America, Inc. All rights reserved. Neuronal death and neurite degeneration occur in a wide variety of contexts in the developing nervous system. If excess neurons are generated or if neurons become functionally superfluous at later developmental stages, they often undergo programmed cell death. Likewise, in the final stages of neural circuit assembly, exuberant or superfluous neuronal projections are selectively eliminated while functional connections are preserved1. This latter event is quite remarkable (functionally it is equivalent to regionally restricted cell death), somehow the cell decides to completely destroy only one compartment (the pruned axon or dendrite), but the remainder of the cell remains perfectly healthy. How specific neurons are selected for death or individual neurites are selected for auto-destruction, as well as the nature of the underlying molecular pathways that mediate these events, remains poorly defined. In this issue, Kirilly et al.2 identify two molecules, Sox14 and Mical, that are essential for fine-tuning neural cell numbers and dendritic projection patterns. The Drosophila pupa is a superb system in which to study developmentally regulated cell death and neurite pruning. Much of the larval nervous system is preserved during the larva-to-adult transition, but it needs to be rewired to integrate properly with developing adult structures. Larval neurons show a diversity of behaviors during metamorphosis; some are no longer needed and undergo programmed cell death, others
The authors are at the Howard Hughes Medical Institute and Department of Neurobiology, University of Massachusetts Medical School, Worcester, Massachusetts, USA. e-mail: marc.freeman@umassmed.edu

Mical Developmental ecdysone pulse 20HE ECR USP sox14 Sox14 da I, IV Dendritic pruning Sox14 da II, III Apoptosis mical Mica Mical

Cytoskeletal breakdown

Severing and degeneration of neurite

Figure 1 Sox14 and Mical function during developmental sensory dendrite pruning in Drosophila. Ecdysone initiates a host of developmental changes in the Drosophila pupal nervous system, including cell death, axon pruning and dendritic pruning. Ecdysone binds to a nuclear receptor heterodimer (consisting of the ECR and ultraspiracle, USP), resulting in the activation of target genes, including sox14. Sox14 is required for apoptosis in type II and III da neurons and for pruning in type I and IV da neurons. In type I and IV da neurons, Sox14 activates expression of dendrite pruning genes, including mical. Mical is found throughout the cell, but, through unknown mechanisms, it only promotes severing of the dendritic compartment. 20HE, 20-hydroxyecdysone.

prune axons and initiate new axonal growth to connect with adult-specific tissues, and still others prune their dendrites. All these events are initiated by a developmentally regulated pulse of the steroid hormone 20- hydroxyecdysone (ecdysone)3,4. Ecdysone is therefore a master regulator that not only controls neuronal rewiring, but also controls remodeling of the entire body of Drosophila during metamorphosis. The larval dendritic arborization (da) neurons, which are found in the peripheral nervous system, are a particularly good system in which to assay dendritic pruning and neuronal cell death. These fall into four classes (termed IIV) on the basis of morphology. Class II and III da neurons are eliminated at

etamorphosis through apoptotic cell death, m whereas Class I and IV da neurons survive and exhibit near complete dendritic pruning, but have no morphological changes in the soma or axon1. Kirilly et al.2 started by simply live-imaging Class IV da neuron dendrites (specifically a subset termed ddaC) as pruning was initiated. ddaC dendrites that were destined to be pruned first developed blebs that were highly dynamic, moving throughout the dendritic compartment, but not into the soma or axon. Next, a break developed in the dendrite at its base next to the soma, physically separating it from the cell body. Subsequently, ddaC dendrites underwent widespread degeneration, with no sign of directionality

nature neuroscience volume 12 | number 12 | December 2009

1479

news and views


to the progression of degeneration. These gross changes in cellular morphology, which have also been observed by other groups in additional da neurons5, are similar to what is seen in fixed preparations of pruned axons in the Drosophila CNS and are similar to Wallerian degeneration in live preparations of mouse axons, arguing for a potential conservation of axon degenerative mechanisms in these quite different contexts. Kirilly et al.2 then investigated how the classes of da neurons are differentially programmed to undergo cell death or dendritic pruning. Previous studies had identified genes whose expression is responsive to ecdysone signaling during metamorphosis. On the basis of this list, the authors performed a small-scale RNA interference (RNAi) screen looking for genes whose knockdown caused severe dendrite-pruning defects. Knocking down the sox14 gene using transgenic RNAi approaches (sox14RNAi) was found to potently block dendritic pruning of ddaCs. Sox14 encodes a high-mobility group (HMG) transcription factor that belongs to the highly conserved Sox family of developmental regulators. To confirm a role for Sox14 in pruning, the authors next produced mutations in the sox14 gene, generated genetic mosaic clones such that the sox14 mutation was homozygous only in Class IV da neurons, and again found that lack of Sox14 function blocked dendritic pruning. Pruning defects observed in the sox14 mutants were rescued by overexpression of Sox14. Moreover, overexpression of Sox14 in ddaCs was sufficient to initiate premature dendritic pruning before the ecdysone pulse. These data argue that Sox14 induction is necessary and sufficient to induce dendritic pruning. Sox14 function is not limited to dendrite severing, but also extends to activation of programmed cell death and axon pruning. Programmed cell death in Class II and III ddaA/B/F neurons was suppressed by sox14RNAi treatment and in sox14 mutations. Likewise, axonal pruning of mushroom body neurons was suppressed by sox14RNAi. Thus, Sox14 can initiate all three of the neuron rewiring events observed during metamorphosis: cell death and dendrite or axon pruning. How Sox14 functions in these different cells to execute each of these programs remains unclear. For example, how could Sox14 promote dendrite pruning, but not axon pruning, in peripheral da neurons, while doing apparently the opposite in central mushroom body neurons? The authors suggest that discriminating between cell death and pruning may simply entail modulating Sox14 levels; Sox14 appears to be expressed at higher levels in Class II and III da neurons that undergo death versus Class I and IV, which exhibit dendrite pruning. Clearly, testing the predictions of this model and identifying downstream targets of Sox14 could address these questions. To identify other factors involved in pruning (including potential downstream targets of Sox14), Kirilly et al.2 performed a forward genetic screen for mutants that suppressed ddaC pruning. One mutant line exhibited a near full suppression of ddaC pruning. However, in contrast with sox14 mutants, apoptotic elimination of ddaA/B/F neurons remained normal, as did pruning of mushroom body neuron axons. These defects mapped to mical, which encodes a multi-domain cytosolic protein containing an N-terminal flavoprotein monooxygenase (FM) domain, and calponin homology and LIM domains in the central region, followed by a proline-rich domain, coiled-coil domain and PDZ-binding motif at the C terminus. Overexpression of full-length Mical rescued the mutant phenotype, whereas overexpression of a truncated Mical that was missing the N-terminal FM domain did not. Thus, FM domain function appears to be essential for Mical activity in dendrite severing. Although Mical seems to only drive pruning in dendrites, antibodies to Mical were present throughout the dendrite, soma and axon. Thus, polarized activation, rather than localization, may underlie its role in selectively destroying dendrites. Mical expression in ddaC neurons is initiated after the ecdysone pulse. Could the mical gene be a target of Sox14? Several lines of evidence suggest that mical activation is downstream of Sox14 and the receptor for ecdysone, EcR. First, overexpression of Sox14 led to an increase in Mical in ddaCs. Second, Sox14 was found to bind to the mical promoter in ecdysone-treated cell lines. Third, the accelerated pruning phenotype that resulted from early overexpression of Sox14 could be suppressed by blocking Mical function. Both mical and sox14 also appeared to be downstream of EcR, as blocking ecdysone signaling with a dominant-negative EcR blocked induction of mical and sox14 in both da neurons and mushroom body neurons. Notably, Sox14 appears to be a major downstream target of EcR for nervous system reorganization during metamorphosis and Mical seems to be an important regulator of its role in ddaC dendrite pruning. For example, Sox14 overexpression rescued pruning in EcR dominant negativetreated ddaCs, as did Mical overexpression. Moreover, in sox14 mutant ddaCs that also expressed dominant-negative EcR, overexpression of Mical could partially rescue pruning defects. Together, these data argue that Mical acts downstream of EcR, potentially directly downstream of Sox14, and is an important mediator of Sox14-dependent control of dendritic pruning (Fig. 1). As with any exciting study, a number of new questions arise from this work. For example, what other genes are regulated by Sox14? Clearly Sox14 is a major mediator of cell death and dendritic and axonal pruning. How is the specificity of the response (death or dismemberment) of the cell encoded? How is Mical, found throughout the cell, activated only in dendrites? This protein has a number of interaction domains, some of which bind actin, but the key feature that allows for dendrite severing in a localized fashion remains undefined. One possibility is that Mical somehow converges with local caspase activation, which has been described in pruned Drosophila dendrites6,7, to drive destruction, but this remains to be determined. Also, are dendrite severing and subsequent degeneration genetically separable events? For example, does Mical simply cause localized dendrite severing near the soma while other mechanisms drive degeneration? If so, how does Mical know where to sever the dendrite? If Mical is required for dendrite destruction, how does it drive dendrite disassembly? This and other recent studies in the field highlight the fact that neurite degeneration is a complex process that can be initiated by a variety of molecular mechanisms. Genetic tools have now been identified that discriminate between dendrite pruning, axon pruning and Wallerian degeneration. Is there a unifying pathway of neurite destruction into which different activating pathways feed? Are all of these processes of degradation distinct from one another? In many ways, the field is positioned similar to that of the cell death field two decades ago; we know (or strongly suspect) these events are active processes of auto-destruction, but we still need to delineate the underlying genetic pathways. In the future, clarifying precisely how Sox14 dictates the death or only dismemberment of neurons and how Mical mediates dendrite-specific localized severing and destruction should help us take important steps in this direction.

2009 Nature America, Inc. All rights reserved.

1. Luo, L. & OLeary, D.D. Annu. Rev. Neurosci. 28, 127156 (2005). 2. Kirilly, D. et al. Nat. Neurosci. 12, 14971505 (2009). 3. Kuo, C.T., Jan, L.Y. & Jan, Y.N. Proc. Natl. Acad. Sci. USA 102, 1523015235 (2005). 4. Brown, H.L., Cherbas, L., Cherbas, P. & Truman, J.W. Development 133, 275285 (2006). 5. Williams, D.W. & Truman, J.W. Development 132, 36313642 (2005). 6. Williams, D.W., Kondo, S., Krzyzanowska, A., Hiromi, Y. & Truman, J.W. Nat. Neurosci. 9, 12341236 (2006). 7. Kuo, C.T., Zhu, S., Younger, S., Jan, L.Y. & Jan, Y.N. Neuron 51, 283290 (2006).

1480

volume 12 | number 12 | December 2009 nature neuroscience

news and views

Pigment epithelium-derived growth factor: modulating adult neural stem cell self-renewal
Andrew Chojnacki & Samuel Weiss
The vascular niche-derived factor PEDF enhances Notch signaling in adult neural stem cells via an unexpected mechanism involving nuclear export of a transcriptional repressor, to promote both proliferation and multipotentiality.

The potential to use neural stem cells (NSCs) for brain repair stems from their capacity for self-renewal. According to conventional wisdom, NSCs are restricted to three distinct methods of dividing. They can divide asymmetrically to generate another NSC and a committed cell, in a process that renews the stem cell population, symmetrically to expand the population, or symmetrically to extinguish the population. These modes of division must be balanced to ensure the maintenance of the NSC population, prevent their unrestricted proliferation and limit the premature or inappropriate differentiation of progeny. Recent evidence for the close apposition of adult periventricular NSCs and blood vessels1,2 have confirmed previous 3,4 findings that factors derived from the vasculature contribute to the regulation of the adult NSC pool. Vascular cues must cross-talk and collaborate with other important stem cell regulatory pathways such as Notch signaling5,6 to determine adult NSC self-renewal and expansion. A new study by Andreu-Agull et al.7 reveals that the vasculature-derived PEDF3,4 promotes the Notch signaling dependent renewal of adult periventricular NSCs through an unconventional mechanism. Specifically, PEDF increased Notch signaling in dividing NSCs in vitro. However, it did not do this by promoting the cleavage of Notch to its transcriptionally active intracellular domain (NICD), but it instead enhanced the ability of NICD, in complex with its partner C promoterbinding factor 1 (CBF-1), to initiate transcription. The authors have previously reported that the Notch transcriptional target Hes1 is upregulated by PEDF4. Here, they found that PEDF activated a Hes1-luciferase reporter without increasing NICD concentration, but this still required the presence and activity of both NICD and CBF1.

The authors are in the Department of Cell Biology and Anatomy, Hotchkiss Brain Institute, University of Calgary, Calgary, Alberta, Canada. e-mail: akchojna@ucalgary.ca or weiss@ucalgary.ca

Andreu-Agull et al.7 next sought to identify precisely which cells in the adult periventricular area in vivo were responsive to PEDF. Because of their quiescent nature, adult NSCs retain labels incorporated during division (such as the thymidine analog BrdU) for extended periods of time, and these label-retaining cells have been commonly equated with adult NSCs8. In mice engineered to express enhanced green fluorescent protein (EGFP) under the control of four CBF1responsive elements, the authors found that proliferating label-retaining cells could be neatly classified into EGFP-high (high Notch signaling) and EGFP-low (low Notch signaling) subpopulations of approximately equal size. Administration of PEDF into the lateral ventricle markedly increased the proportion of EGFP-high label-retaining cells at the expense of the EGFP-low population. If one were to stop here and assume that all label-retaining cells are NSCs, one might conclude that PEDF promotes the symmetrical division of NSCs with high levels of Notch signaling, generating more of these cells at the expense of NSCs with low levels of Notch signaling. However, the authors subsequent experiments suggested a more interesting conclusion. Sorting adult periventricular cells for high or low EGFP expression revealed that EGFP-high cells generated larger neurospheres than EGFP-low cells. Neurospheres are clusters of undifferentiated cells that are the products of division commonly used to retrospectively identify NSCs. EGFPhighderived neurospheres maintained higher subcloning efficiency over multiple passages and contained a higher percentage of multipotent cells than EGFP-lowderived neurospheres. PEDF treatment of EGFPlow cells increased their capacity to generate multipotent primary neurospheres and these neurospheres were able to generate more secondary or tertiary neurospheres even in the absence of PEDF, suggesting that PEDF increased their self-renewal capacity. PEDF did not increase the size of primary EGFPhighderived neurospheres, but it enhanced

the proportion of progeny with high CBF1 activity and their ability to generate secondary or tertiary neurospheres. These data suggest that PEDF enhances Notch signaling by increasing CBF1 activity. How does PEDF promote CBF1 activity? The answer turned out to be complicated. Immunostaining dissociated cultures of adult periventricular cells for NICD, Andreu-Agull et al.7 found that, 24 h after plating, 45% of cell pairs (generated by a single-cell division) consisted of one cell expressing high levels of NICD and one with low levels of NICD. PEDF did not increase the proportion of NICD-high cell pairs, indicating that it does not increase self-renewal by promoting symmetric cell divisions in vitro. Instead, PEDF increased the proportion of NICD-high/low cell pairs in which both cells expressed high levels of epidermal growth factor receptor (EGFR); EGF is a mitogen for adult periventricular NSCs9. Chromatin immunoprecipitation and luciferase assays identified the Egfr promoter as a target of Notch signaling. Thus, PEDF promotes the renewal of adult NSCs expressing low levels of NICD by increasing the proportion that is responsive to mitogenic EGF signaling (Fig. 1a). The nuclear receptor co-repressor (N-CoR) suppresses Notch-induced astrogliogenesis in the developing brain10. Andreu-Agull et al.7 asked whether N-CoR was involved in PEDFinduced maintenance of CBF-1 activity and the subsequent upregulation of EGFR. Chromatin immunoprecipitation revealed that N-CoR bound to a CBF1-binding element in the Egfr promoter and was displaced from this element by PEDF treatment. Furthermore, overexpression of full-length N-CoR abrogated PEDF-induced activation of Hes-1 or EGFRluciferase reporters, whereas a r epression-dead N-CoR had no effect. The authors also observed that PEDF treatment decreased nuclear N-CoR and increased its cytoplasmic localization. The PEDF-induced nuclear export of N-CoR and subsequent increase in neurosphere formation were blocked by overexpression of p65nuclear factor B (NF-B) lacking the nuclear export signal and

2009 Nature America, Inc. All rights reserved.

nature neuroscience volume 12 | number 12 | December 2009

1481

news and views a b


PEDF Receptor? Ligand-dependent Notch cleavage/ activation NICD

Adult periventricular NSC NICD CBF1 EGFR

NCoR
+PEDF Quasi symmetric/ asymmetric division NICD CBF1 EGFR b NICD CBF1 EGFR PEDF Asymmetric division NICD CBF1 EGFR
p65-NFKB

p65-NFKB Activation

p65-NFKB dependent nuclear export

EGFR

Self-renewal Multipotentiality

NCoR CBF1

Egfr/Hes1

2009 Nature America, Inc. All rights reserved.

Neurosphere formation

Derepression

Figure 1 PEDFs effects on adult periventricular NSC self-renewal. (a) During NSC cell division, PEDF increases Notch signaling in daughter cells with low levels of NICD by promoting CBF1-dependent transcriptional activation, but not by increasing symmetric cell division. PEDF-mediated increases in CBF1 activity in the daughter cell upregulates EGFR and Hes-1 (data not shown), resulting in a precursor cell with an intermediate capacity for self-renewal as revealed by in vitro assays. (b) Mechanistically, PEDF activates an unknown receptor, eventually activating NF B. Increased NFB signaling causes the p65NF-Bdependent nucleo-cytoplasmic shuttling of N-CoR away from CBF1 binding sites, allowing NICD/CBF1-dependent transcription of Egfr and Hes1. For simplicity, not all of the required components of NFB or Notch signaling have been represented.

transactivation domain. Overexpression of a p65 subunit lacking only the transactivation domain did not reduce PEDF increases in neurosphere formation. Together, these in vitro data suggest that p65-mediated export of N-CoR from the nucleus is responsible for the increase in CBF1 activity in NSCs after PEDF treatment. Notably, Andreu-Agull et al.7 confirmed these findings in vivo. PEDF reduced nuclear expression of N-CoR, but increased EGFR expression in label-retaining cells; a C-terminal
Figure 2 An alternative interpretation of PEDFs role in regulating adult periventricular NSCs. Quiescent NSC may have lower levels of Notch signaling and, subsequently, may be less responsive to EGF signaling because of reduced EGFR expression. This may function to maintain NSC quiescence by making the cells less responsive to EGF-stimulated proliferation, thereby insulating the cells from replicationassociated DNA mutations and the astrogliogenic effects of Notch and EGF signaling. PEDF could act alone to partially activate NSCs or in conjunction with an unknown factor (factor X) to transiently increase the renewal and proliferative capacity of adult periventricular NSCs by increasing levels of Notch and consequently responsiveness to EGF signaling. Activation of the adult periventricular pool would then result in increased neurogenesis.

PEDF fragment that antagonized PEDF signaling did the reverse. Thus, PEDF activates NF-B signaling in adult periventricular NSCs, which results in p65-mediated nuclear

export of N-CoR away from CBF1 binding sites, allowing the CBF1-NICD complex to initiate transcription of Hes1 and EGFR (Fig. 1b). Ultimately, this increases the capacity

Quiescent adult periventricular NSC

NICD CBF1 EGFR

+PEDF Partially activating division NICD CBF1 EGFR Quiescent NICD CBF1 EGFR Partially activated

+PEDF + Factor X Activating division NICD CBF1 EGFR Activated NSC status In vitro characteristics Neurosphere formation Ability to be passaged Multipotentiality

1482

volume 12 | number 12 | December 2009 nature neuroscience

news and views


of PEDF-stimulated NSCs with low levels of Notch signaling to proliferate and generate multiple cell types. Andreu-Agull et al.7 have thus elucidated the pathways used by PEDF to regulate adult periventricular NSC function. The identity of the receptor that initiates PEDF signaling, however, remains unknown. Pharmacological inhibition of the only known putative PEDF receptor, adipose triglyceride lipase11, did not affect PEDF actions on adult periventricular NSCs. If expression of the PEDF receptor is restricted to adult periventricular NSCs, its use as a prospective marker would substantially aid in their study. The interpretation of PEDFs role in regulating adult NSC function, however, could depend on which population of label-retaining cells one selects as the true stem cell. Presuming that NSCs with high levels of Notch signaling and EGFR are at the top of the hierarchy might mean that PEDF functions to extend the cell- generating capacity of NSCs with low levels of Notch signaling (Fig. 1a). Alternatively, one could assume that label-retaining cells with low Notch signaling and EGFR expression are the uiescent NSCs at the top of the q hierarchy; in this case, PEDF would function to activate them (Fig. 2). Indeed, intraventricular infusion of PEDF promotes NSC division, whereas inhibition of PEDF signaling decreases it4. Furthermore, injury-mediated activation of adult periventricular NSCs increases their expression of EGFR12 and Notch signaling13. It may also be more reasonable for a quiescent non-EGFR expressing NSC to progress to an active EGFR-expressing NSC, initiated in part by PEDF, to generate transit-amplifying cells known to express EGFR12. The inability of NSCs with low levels of NICD to generate NSCs with high levels of NICD in vitro could simply be a cell culture limitation and their decreased capacity to generate neurospheres could be a reflection of their quiescent state. Tracing the fate of NSCs expressing high or low levels of NICD/ EGFR in vivo may provide some insight in the future. Because EGFR is a direct target of Notch signaling, quiescent NSCs may need shielding from excessive Notch signaling, as increasing levels of EGFR and EGF signaling have been demonstrated to inhibit neural precursor proliferation and promote astrogliogenesis14. In this regard, it is worth noting that Numb physically interacts with NICD to inhibit Notch signaling, but, paradoxically, it has been found to maintain the neural precursor pool during development15.Whether Numb and PEDF signaling interact and have opposing roles in regulating adult NSC quiescence or self-renewal would also be an interesting avenue for future investigations.
1. 2. 3. 4. Shen, Q. et al. Cell Stem Cell 3, 289300 (2008). Tavazoie, M. et al. Cell Stem Cell 3, 279288 (2008). Shen, Q. et al. Science 304, 13381340 (2004). Ramrez-Castillejo, C. et al. Nat. Neurosci. 9, 331339 (2006). 5. Nyfeler, Y. et al. EMBO J. 24, 35043515 (2005). 6. Hitoshi, S. et al. Genes Dev. 16, 846858 (2002). 7. Andreu-Agull, C., Morante-Redolat, J.M., Delgado, A.C. & Farias, I. Nat. Neurosci. 12, 15141523 (2009). 8. Chojnacki, A.K., Mak, G.K. & Weiss, S. Nat. Rev. Neurosci. 10, 153163 (2009). 9. Reynolds, B.A. & Weiss, S. Science 255, 17071710 (1992). 10. Hermanson, O., Jepsen, K. & Rosenfeld, M.G. Nature 419, 934939 (2002). 11. Notari, L. et al. J. Biol. Chem. 281, 3802238037 (2006). 12. Doetsch, F., Petreanu, L., Caille, I., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. Neuron 36, 10211034 (2002). 13. Givogri, M.I. et al. Dev. Neurosci. 28, 8191 (2006). 14. Burrows, R.C., Wancio, D., Levitt, P. & Lillien, L. Neuron 19, 251267 (1997). 15. Qu, Q. & Shi, Y. J. Cell. Physiol. 221, 59 (2009).

2009 Nature America, Inc. All rights reserved.

Hippocampal theta rhythms follow the beat of their own drum


Laura Lee Colgin & Edvard I Moser
The firing of most hippocampal neurons is modulated by the theta rhythm, but its not clear how and where the rhythm is generated. A study now shows that the required machinery for theta generation lies in local circuits of the hippocampus.
The hippocampal theta rhythm is one of the largest synchronous signals in the mammalian brain. Theta rhythms are essential for linking distributed cell ensembles during key functions of the hippocampus, such as learning and spatial navigation. The search for its origins began more than 50 years ago when lesions of the medial septum were found to disrupt hippocampal theta1. Subsequent studies found that medial septal neurons discharged synchronously with hippocampal theta, leading researchers to suggest the edial-septum diagonal-band area as the m theta pacemaker2,3. In this issue, Goutagny et al.4 provide fresh insight into this matter by showing that theta rhythms with properties similar to those of the living animal emerge in a whole-hippocampus in vitro preparation without input from external sources and without the addition of any drugs that might mimic aspects of septal stimulation. The septal pacemaker hypothesis has dominated models of theta generation since its inception. The hypothesis was challenged by reports of theta-like activity in hippocampal slices lacking input from the medial septum, but treated with agonists of acetylcholine or metabotropic glutamate receptors5,6. However, it was never determined whether these slice rhythms were analogous to theta in behaving animals. In some preparations, the theta oscillations appeared as intermittent bursts that partially resembled epileptic activity7. In others, theta was seen only when AMPA receptors were silenced6, a condition that may never occur in the living animal. The question of whether intrahippocampal circuits are capable of theta generation has therefore remained open. Goutagny et al.4 provide evidence that drug- and septal inputfree hippocampus can generate and sustain its own theta oscillation. They measured field potentials and single-neuron activity from CA1 of the isolated hippocampi, which exhibited oscillatory activity in the theta frequency range (310 Hz; Fig. 1). Depth profiles showed a near-complete phase reversal between the pyramidal cell layer and stratum radiatum, suggesting that the rhythmic pattern was generated in CA1 in a manner similar to that of the intact animal8. As in behaving

The authors are at the Kavli Institute for Systems Neuroscience and Centre for the Biology of Memory, Medical-Technical Research Centre, Norwegian University of Science and Technology, Trondheim, Norway. e-mail: edvard.moser@ntnu.no

nature neuroscience volume 12 | number 12 | December 2009

1483

news and views


nimals9, CA1 a principal cells and interneurons preferentially fired near the trough of theta oscillations in the CA1 pyramidal cell layer. The theta activity was abolished by antagonists of GABA or AMPA/kainate receptors. Theta rhythms were coherent across large parts of the longitudinal axis of CA1, but coherence between CA1 and CA3 oscillations was low. Removal of inputs from CA3 did not block the oscillations in CA1, suggesting that the two subfields have independent generators for theta activity. These results clearly suggest that theta activity is an inherent property of the hippocampal network emerging from local connectivity. External inputs, such as those from the medial septum, are not strictly required for generating or maintaining theta activity in the hippocampus. The induction of theta rhythm in a complete hippocampal preparation provides important clues about the intrahippocampal circuit mechanisms that may sustain theta-patterned activity. In contrast with standard transverse slices, the complete preparation preserves the majority of intrahippocampal connections in the longitudinal plane, suggesting that these are more involved in theta generation than had been previously thought. The fact that theta activity was maintained in CA1 after it was disconnected from CA3 suggests that the intrinsic interneurons in CA1 are critical for generating the oscillations. Several classes of CA1 interneurons are strongly modulated by hippocampal theta activity in intact animals10,11 and many of these interneurons extend their axons for hundreds of micrometers or even millimeters longitudinally. These longitudinal connections could be important for enabling theta rhythms in the intact hippocampi to become oherent along widespread regions along c the septotemporal axis of CA1. The possible involvement of such interneurons would also explain why theta activity is difficult to induce in standard hippocampal slices. Is there more than one hippocampal oscillator involved in theta generation? The study by Goutagny et al.4 suggests that there are multiple generators and that these are distributed along the septotemporal axis of the hippocampus. When theta activity was recorded at two locations in CA1, one dorsal and one ventral, and the regions between were silenced by procaine, both regions continued to oscillate, but the ventral oscillator was slower than the dorsal oscillator by about 1 Hz. When the connections between the areas were intact, however, the faster oscillator appeared to be entraining the slower one, suggesting that the oscillators are normally coupled. The existence of multiple weakly coupled oscillators with

Behaving rat 0.5 mV 1s

Whole hippocampus in vitro

0.1 mV 1s
Figure 1 The upper trace shows theta recorded from the CA1 apical dendrites of a freely moving rat (L.L.C. and E.I.M., unpublished data), whereas the lower trace shows theta waves in the wholehippocampus preparation of Goutagny et al.4 (their Fig. 1c). Note that the in vivo and in vitro patterns are almost identical in frequency, wavelength and waveform.

2009 Nature America, Inc. All rights reserved.

decreasing frequencies along the septotemporal axis suggests a possible mechanism for the systematic changes in theta phase reported along the septotemporal axis in behaving animals12 and potentially sheds light on the timing of theta-related information processing in the hippocampus. The presence of theta activity in the isolated hippocampal preparation does not, however, exclude a role for inputs from the medial septum. Although these inputs may not be necessary for the rhythm itself, they may, for example, be involved in synchronizing oscillators at different septotemporal levels. It is also important to be aware that there may be more than one mechanism for theta generation and that other mechanisms may require septal input. Theta oscillations in behaving animals appear in two forms13. One occurs during active movement and is resistant to muscarinic acetylcholine receptor antagonists. The other is associated with immobility and is abolished by muscarinic antagonists. Goutagny et al.4 found that the theta rhythms were unaffected by a muscarinic antagonist, suggesting that they were analogous to movement-related theta. Thus, the septal pacemaker hypothesis may still prevail for theta that requires muscarinic receptors. Many other interesting questions are raised by the findings of Goutagny et al.4. For example, if CA1 contains multiple theta oscillators along its longitudinal axis, are these discrete or do they form a continuous overlapping network? If they are discrete, how many are there and how are their borders defined? Are differences in oscillator frequencies related to the differences in intracellular oscillation

frequencies for pyramidal cells at different eptotemporal s locations14? Moreover, what are the components of the CA1 circuit responsible for theta generation? Which types of interneurons are involved and what are their relative contributions? Do these interneurons correspond to the classes of theta-related interneurons that have been defined in vivo11? And finally, can the whole-hippocampus preparation switch from theta activity to other hippocampal network states or are external inputs needed for such transitions? The introduction of spontaneously occurring, stable theta rhythms in an in vitro preparation by Goutagny et al.4 paves the way for such analyses in the years to come.
1. Green, J.D. & Arduini, A.A. J. Neurophysiol. 17, 533557 (1954). 2. Petsche, H., Stumpf, C. & Gogolak, G. Electroencephalogr. Clin. Neurophysiol. 14, 202211 (1962). 3. Stewart, M. & Fox, S.E. Trends Neurosci. 13, 163168 (1990). 4. Goutagny, R., Jackson, J. & Williams, S. Nat. Neurosci. 12, 14911493 (2009). 5. Konopacki, J., Bland, B.H. & Roth, S.H. Brain Res. 451, 3342 (1988). 6. Gillies, M.J. et al. J. Physiol. (Lond.) 543, 779793 (2002). 7. Williams, J.H. & Kauer, J.A. J. Neurophysiol. 78, 26312640 (1997). 8. Bland, B.H. Prog. Neurobiol. 26, 154 (1986). 9. Csicsvari, J., Hirase, H., Czurko, A., Mamiya, A. & Buzsaki, G. J. Neurosci. 19, 274287 (1999). 10. Buzski, G. Neuron 33, 325340 (2002). 11. Somogyi, P. & Klausberger, T. J. Physiol. (Lond.) 562, 926 (2005). 12. Lubenov, E.V. & Siapas, A.G. Nature 459, 534539 (2009). 13. Kramis, R., Vanderwolf, C.H. & Bland, B.H. Exp. Neurol. 49, 5885 (1975). 14. Maurer, A.P., Vanrhoads, S.R., Sutherland, G.R., Lipa, P. & McNaughton, B.L. Hippocampus 15, 841852 (2005).

1484

volume 12 | number 12 | December 2009 nature neuroscience

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

Experience-dependent compartmentalized dendritic plasticity in rat hippocampal CA1 pyramidal neurons


Judit K Makara1, Attila Losonczy1,2, Quan Wen1,2 & Jeffrey C Magee1 The excitability of individual dendritic branches is a plastic property of neurons. We found that experience in an enriched environment increased propagation of dendritic Na+ spikes in a subset of individual dendritic branches in rat hippocampal CA1 pyramidal neurons and that this effect was mainly mediated by localized downregulation of A-type K+ channel function. Thus, dendritic plasticity might be used to store recent experience in individual branches of the dendritic arbor. The arrival of highly correlated input patterns onto individual apical oblique and basal dendrites causes hippocampal CA1 pyramidal cells (CA1PCs) to generate local fast sodium spikes whose propagation strength varies among branches14. The propagation strength of dendritic spikes fundamentally determines their effect at the soma, as only strongly propagating spikes evoke precisely timed action potential output2. Previously, we reported that spike propagation strength is enhanced by associative pairing of dendritic spikes with backpropagating action potentials or muscarinic acetylcholine receptor activation (branch strength potentiation, BSP2). This raises the possibility that dendritic plasticity participates in hippocampal mnemonic functions by providing a 2009 Nature America, Inc. All rights reserved.

echanism for storing complex features encoded in highly correlated m input patterns. We hypothesized that experience in a spatially and socially rich environment57 could induce compartmentalized changes in spike propagation strength in CA1PC dendrites. To test this, we compared spike propagation in perisomatic dendrites of CA1PCs in acute slices prepared from rats that were exposed to an enriched environment for 27 d (Supplementary Methods and Supplementary Fig. 1) with that of control rats. Experiments were conducted in accordance with institutional regulations (Janelia Farm Institutional Animal Care and Use Committee). Most perisomatic dendrites form families consisting of primary parent dendrites, terminal daughter dendrites and occasionally intermediate segment(s). The strength of local spike propagation is hierarchically distributed in dendritic families2. Primary parent dendrites usually express strong spikes (rate of rise, dV/dt > 2 V s1, see ref.2 and Supplementary Methods), whereas propagation in most subordinate branches is weak (dV/dt < 2 V s1), with a subgroup of daughter dendrites in which propagation is strong enough to spread into the parent dendrite and evoke its strong spike2 (see below). We first systematically measured the propagation strength of dendritic spikes in every possible branch of perisomatic apical oblique dendritic

a
20 m

Control

D1

D2

D1 D2

P dV/dt P ~dV/dt P dV/dt D dV/dt D dV/dt T dV/dt D = dV/dt T

Enriched 20 m

d
P D1 D2

Figure 1 Hierarchical distribution of propagation strength of dendritic spikes in CA1 pyramidal cells of control and enriched rats. ( a,b) Cells loaded with Alexa488 from control (a) and enriched (b) rats. Ovals indicate input sites on a primary parent dendrite (P) and the connected terminal daughter dendrites (D1 and D2). (c,d) Vm (upper) and dV/dt (lower) traces evoked at the input sites in a and b, respectively. In both cells, input onto P evoked a strong spike with a single d V/dt component (dV/dtP). Input onto D1 evoked a strong compound spike (dV/dtT) with two components, dV/dtD (D1s own spike) and dV/dtP (the spike propagated into P). dV/dtD could be resolved (gray traces, shifted for clarity) at more distal input sites (c) or by hyperpolarization (d). Input onto D2 evoked a weak spike that did not propagate into the parent branch (d V/dtD = dV/dtT). Scale bars represent 1 mV (Vm traces) or 1 V s1 (dV/dt traces), 2 ms. (e,f) Cumulative probability of primary parent dendrites ( e, dV/dtP) and of the total spike of terminal daughter dendrites ( f, dV/dtT) in control (green) and enriched (yellow) cells. Dashed line indicates 2 V s 1. Error bars represent s.e.m.
1Howard

D1 D2 dV/dtP ~dV/dt P dV/dt D dV/dt D dV/dt T dV/dt D = dV/dt T

Cumulative probability

Cumulative probability

1.0 0.8 0.6 0.4 0.2 0 0 4 8 12 16 24 dV/dt P (V s1)

1.0 0.8 0.6 0.4 0.2 0 0 4 8 12 20 30 dV/dt T (V s1)

Hughes Medical Institute, Janelia Farm Research Campus, Ashburn, Virginia, USA. 2Present address: Department of Neuroscience, Columbia University, New York, New York, USA (A.L.), Department of Physics, Harvard University, Cambridge, Massachusetts, USA (Q.W.). Correspondence should be addressed to J.K.M. (makaraj@janelia.hhmi.org) or J.C.M. (mageej@janelia.hhmi.org). Received 28 May; accepted 18 September; published online 8 November 2009; corrected online 17 November 2009 (details online); doi:10.1038/nn.2428

nature neuroscience VOLUME 12 | NUMBER 12 | december 2009

1485

b r i e f c o m m u n i c at i o n s a
dV/dt (V s1) 30 25 20 15 10 5 0 P 30 25 dV/dt (V s1) 20 15 10 5 0 P D D

b
Cumulative probability

1.0 0.8 0.6 0.4 0.2 0 0 1 2 3 4 dV/dt D (V s1) 5

Figure 2 Coupling and dV/dtD in strong and weak dendritic families. (a) Spike dV/dt of strong parent dendrites (P, dV/dtP) and of connected daughter dendrites (D, dV/dtT) in control (upper) and enriched (lower) cells. Red lines indicate coupled P-D pairs, and blue lines indicate noncoupled P-D pairs. Dashed line indicates 2 V s1. (b) Cumulative probability of dV/dtD in all terminal daughter branches of strong parent dendrites in control (green) and enriched (yellow) cells. ( c) Data are presented as in b, but for terminal branches connected to weak parent dendrites.

c
Cumulative probability

1.0 0.8 0.6 0.4 0.2 0 0 0.5 1.0 1 dV/dt D (V s )

2009 Nature America, Inc. All rights reserved.

families by evoking local Na+ spikes with two-photon uncaging of MNI-glutamate and measuring spike dV/dt at the soma (Fig. 1ad). The spike strength of primary parent dendrites had a similar distribution in control rats as has been previously reported2. The majority (79.6%) of primary branches expressed strong spikes, 18.5% had weak spikes and no spike could be evoked in 1.9% (total n = 54). In cells from enriched rats, the proportion of strong primary parent branches was 68.9%, whereas 21.3% had weak spikes and no spikes could be evoked in 9.8% (total n = 61; Fig. 1e). Thus, the proportion of weak or nonspiking primary branches (resulting in weak dendritic families) showed a tendency to increase in enriched rats (control, 11 of 54; enriched, 19 of 61; P = 0.189), but most primary dendrites still expressed strong spikes (strong dendritic families). Although weak primary parent dendrites had similar propagation strength in the two groups (control, 1.40 0.18 V s1, n = 10; enriched, 1.39 0.11 V s1, n = 13; P = 0.95), dV/dtP of strong primary parent branches was slightly larger in enriched cells (control, 6.12 0.76 V s 1, n = 43; enriched, 8.21 0.82 V s1, n = 42; P = 0.026). Next, we analyzed the distribution of spike strength evoked in terminal daughter branches, tested at 1540 m from the branch point. The total spike strength (dV/dtT; definitions in Fig. 1ad and Supplementary Methods) was markedly different between control and enriched terminal branches (Fig. 1f), as terminal dendrites with strong spikes were much more frequent in enriched cells (24 of 84, 28.6%) than in control cells (9 of 83, 10.8%; P = 0.004). To further understand this, we separately looked at the spike strength in terminal branches connected to strong versus weak parent dendrites (Fig. 2). The effect of a terminal branch spike at the soma is in part determined by the strength of the primary branch to which it is connected. If the primary branch is strong, the strength of the terminal branch is determined by whether the local terminal
Figure 3 A-type K+ channel function is reduced by enrichment in terminal daughter dendrites of strong dendritic families. ( a,b) Left, image stacks of cells loaded with Alexa488 from control (a) and enriched (b) rats. Ovals indicate input sites on a primary parent dendrite (P) and connected terminal daughter dendrite (D). Right, Vm (upper) and dV/dt (lower) traces evoked from parent (P) and daughter (D) before (Ctr) and during (Ba 2+) bath application of 200 M Ba2+. Scale bars represent 1 mV (Vm traces) or 1 V s1 (dV/dt traces), 2 ms. (c,d) Summary of the effects of Ba2+ on dV/dtD (c, individual values before (Ctr) and during Ba2+; d, change in Ba2+ relative to Ctr) in control (green) and enriched (yellow) cells. Error bars represent s.e.m.

branch spike can overcome the impedance load of the branch point and recruit the strong spike of the parent dendrite (coupled branches) or not (noncoupled branches)2. Accordingly, in coupled terminal daughter branches connected to strong parent dendrites, the total spike (dV/dtT) has two components, the first related to the spike of the daughter branch itself (dV/dtD) and the second of similar size as the spike of the parent dendrite (~dV/dtP; see Fig. 1c,d)2. Coupling can be expressed by the spike ratio, calculated as the ratio of d V/dtT to dV/dtP. The spike ratio had a bimodal distribution in both control and enriched cells, with noncoupled branches having spike ratios on average ~0.16 and coupled branches with spike ratios of ~1 (Fig.2a and Supplementary Fig. 2). However, comparing the frequency of terminal daughter branches that were coupled to strong parent dendrites, we found that coupling probability was twofold higher in enriched cells (23 of 67 branches, 34.3%) than in control cells (14 of 86 branches, 16.3%; P = 0.0097; Fig. 2a). The higher probability of coupled daughter-parent dendrites in enriched cells may result from stronger spike propagation in the daughter segment. Indeed, dV/dtD in terminal daughter branches of strong parent dendrites was increased compared with control (control: 0.79 0.04 V s1, median of 0.73 V s1, n = 86; enriched: 1.42 0.12 V s1, median of 1.10 V s1, n = 67; P < 0.001; Fig. 2b). In fact, in some enriched terminal branches (13 of 67, 19.4%), dV/dtD was larger than 2 V s1, a value that was rarely observed in control cells. The effect on dV/dtD could not be explained by a difference in the distance of the input site along oblique dendrites (Supplementary Fig. 3), and the alterations in branch excitability were uniformly observed in a large fraction of the enriched cell population as opposed to large changes

Control 20 m

Ctr

Ba

2+

P D dV/dt P dV/dtD

dV/dtD

Enriched

D Ctr

Ba2+

P D
20 m

dV/dt P

dV/dtD

dV/dtD

c
dV/dt D (V s1)

2.0 1.5 1.0 0.5 0

dV/dtD (Ba2+) / dV/dtD (Ctr)

2.5

* ***

2.6 2.2 1.8 1.4 1.0

***

Ctr

Ba2+

Ctr

Ba2+

Control Enr

1486

VOLUME 12 | NUMBER 12 | december 2009 nature neuroscience

b r i e f c o m m u n i c at i o n s
in only a small fraction (Supplementary Fig. 4). Notably, we found a substantial correlation between dV/dtP and dV/dtD in enriched, but not in control cells (Supplementary Fig. 2). Thus, enrichment forms dendritic families that are composed of both very strong parent and daughter dendrites. Notably, these changes did not result from a general increase of dendritic excitability, as d V/dtD was not altered in daughter branches of weakly spiking parent dendrites (control, 0.56 0.05 V s1, n = 18; enriched, 0.52 0.03 V s1, n = 21; P = 0.735; Fig. 2c), and the number of strongly spiking parent dendrites and the spike strength in weak parent dendrites (see above) did not increase in enriched cells. The enhancement of terminal branch spike strength in strong dendritic families closely resembled that observed during BSP2. The underlying mechanism of BSP is a downregulation of A-type K + currents2 that are sensitive to low concentrations of Ba 2+ (refs. 2,8). Therefore, we examined the effect of 200 M Ba2+ on comparable weak terminal branches of strong dendritic families in control and enriched cells (Fig. 3). Ba2+ reliably increased dV/dtD in control cells (from 0.74 0.11 V s1 to 1.23 0.13 V s1, 176 11%, n = 10, P < 0.001; Fig. 3a,c,d), but had a much weaker effect in enriched cells (from 0.93 0.13 V s1 to 1.12 0.18 V s1, 120 6%, n = 11, P = 0.019; Fig. 3bd; main effect, P = 0.003, repeated measures ANOVA). This indicates that downregulation of A-type K+ currents is important for the effect of enrichment on terminal branch spike strength, similar to BSP in vitro. Because dendritic geometry may affect spike propagation9,10, we tested whether the morphology of the studied dendrites was changed by our short enrichment exposure. Digital reconstruction of perisomatic apical dendritic arbors of Alexa488-filled CA1PCs11 from slices of control (n = 10) and enriched (n = 10) rats did not reveal any differences in arbor radius, total dendritic length, number of branch points and Sholl analysis (Supplementary Fig. 5). In summary, exposure to an enriched environment leads to compartmentalized changes in the distribution of dendritic spike propagation in CA1PCs, indicating that the electrical properties of individual dendritic branches can be modified by in vivo experience. The increase of excitability in specific branches is mediated primarily by decreases in the activity of Ba2+-sensitive voltage-gated K + channels (probably Kv4.2, ref. 2). Thus, recent experience induces plasticity of dendritic branch excitability that shares many features with the previously reported in vitro associative plasticity referred to as BSP. Therefore, we would infer that the enrichmentinduced alterations reflect an in vivo BSP-like storage process that takes place during the learning of a complex environment, triggered by the repeated occurrence of highly correlated input patterns arriving during awake and sleep sharp waves 1214. Nevertheless, contribution of other factors, such as physical activity or altered stress levels, cannot be ruled out. Future experiments modifying hippocampal activity patterns and BSP induction/expression during experience should help to understand the role of compartmentalized regulation of dendritic excitability in hippocampal information storage.
Note: Supplementary information is available on the Nature Neuroscience website. AUTHOR CONTRIBUTIONS J.K.M. and J.C.M. designed the study. J.K.M. and A.L. conducted the electrophysiological recordings. J.K.M. analyzed most of the electrophysiological data. Q.W. performed and analyzed cell reconstructions. J.K.M. and J.C.M. wrote the paper.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
Losonczy, A. & Magee, J.C. Neuron 50, 291307 (2006). Losonczy, A., Makara, J.K. & Magee, J.C. Nature 452, 436441 (2008). Golding, N.L. & Spruston, N. Neuron 21, 11891200 (1998). Ariav, G., Polsky, A. & Schiller, J. J. Neurosci. 23, 77507758 (2003). Rampon, C. & Tsien, J.Z. Hippocampus 10, 605609 (2000). van Praag, H., Kempermann, G. & Gage, F.H. Nat. Rev. Neurosci. 1, 191198 (2000). Nithianantharajah, J. & Hannan, A.J. Nat. Rev. Neurosci. 7, 697709 (2006). Gasparini, S., Losonczy, A., Chen, X., Johnston, D. & Magee, J.C. J. Physiol. (Lond.) 580, 787800 (2007). 9. Vetter, P., Roth, A. & Husser, M. J. Neurophysiol. 85, 926937 (2001). 10. Migliore, M., Ferrante, M. & Ascoli, G.A. J. Neurophysiol. 94, 41454155 (2005). 11. Wen, Q., Stepanyants, A., Elston, G.N., Grosberg, A.Y. & Chklovskii, D.B. Proc. Natl. Acad. Sci. USA 106, 1253612541 (2009). 12. ONeill, J., Senior, T. & Csicsvri, J. Neuron 49, 143155 (2006). 13. Wilson, M.A. & McNaughton, B.L. Science 265, 676679 (1994). 14. Ndasdy, Z., Hirase, H., Czurk, A., Csicsvri, J. & Buzski, G. J. Neurosci. 19, 94979507 (1999). 1. 2. 3. 4. 5. 6. 7. 8.

2009 Nature America, Inc. All rights reserved.

nature neuroscience VOLUME 12 | NUMBER 12 | december 2009

1487

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

Self-modulation of neocortical pyramidal neurons by endocannabinoids


Silvia Marinelli1, Simone Pacioni1, Astrid Cannich2, Giovanni Marsicano2 & Alberto Bacci1 Control of pyramidal neuron excitability is vital for the functioning of the neocortex. Somatodendritic slow self-inhibition (SSI) allows inhibitory neurons to regulate their own activity, but the existence of similar mechanisms in excitatory cells has not been shown. We found that in rodents endocannabinoids mediated SSI and long-term modulation of inhibitory connections in layer 2/3 pyramidal neurons with a distinct dendritic morphology, suggesting that a glutamatergic network in cortical circuits is self-regulated. In the neocortex, modulation of pyramidal neuron intrinsic excitability underlies several cognitive functions. For example, tonic hyperpolarization of cortical pyramidal neurons has been proposed to promote contrast adaptation during sensory processing1.

2009 Nature America, Inc. All rights reserved.

Low-threshold spiking inhibitory interneurons can generate a persistent, self-induced hyperpolarization that is mediated by autocrine action of endocannabinoids2,3. Pyramidal neurons can synthesize endocannabinoids46 and they express type 1 cannabinoid receptors (CB1Rs; Supplementary Figs. 1 and 2)68. Whether excitatory pyramidal neurons can produce endocannabinoid-mediated SSI is, however, unknown. We carried out whole-cell current-clamp recordings from layer 2/3 pyramidal neurons (n = 540) in acute neocortical slices from rats and mice (Supplementary Methods). Intracellular stimulations of pyramidal neurons, eliciting ten trains of action potentials at 10 or 50 Hz (inter-train interval = 20 s), evoked SSI consisting of a longlasting (>20 min) membrane hyperpolarization (P < 0.001; Fig. 1 and Supplementary Table 1) that was associated with a reduction of membrane resistance (P < 0.001; Fig. 1) in 31% (51 of 163) of pyramidal neurons (SSI-positive pyramidal neurons). Conversely, 69% of recorded layer 2/3 pyramidal neurons (112 of 163) did not show SSI (SSI-negative pyramidal neurons; P > 0.1; Fig. 1 and Supplementary Table 1). In the presence of the CB1R antagonist AM-251, SSI-inducing stimuli failed to trigger SSI in 91% of pyramidal neurons ( n = 46, P< 0.001, 2 test, control versus AM-251treated pyramidal neurons; Fig. 1e and Supplementary Table 1). Furthermore, SSI was absent

Spike trains Spike trains Figure 1 Endocannabinoid-mediated slow self-inhibition in neocortical layer 2/3 pyramidal neurons. (a,b) Current-clamp traces of a SSI-positive (SSI+, a) and a SSI-negative (SSI, b) layer 2/3 pyramidal neuron SSI+ (gray bar indicates ten trains of spikes at 50 Hz, 20-s intertrain intervals). SSI Negative deflections represent responses to negative current injections 30 s (30 pA). Action potentials are truncated for display purposes. Top 50 s insets, firing behavior of the same neurons. Scale bars represent 25 mV 5 Spike trains and 250 ms. (c) Summary plot of subtracted membrane potential AM-251 Control 150 (Vm) versus time of SSI-positive neurons. Shaded box, application of 0 100 SSI-inducing spike trains. (di) Plots of Rm (normalized to pretrain values) 5 50 versus Vm in response to SSI-inducing stimuli in all tested conditions. n = 163 n = 46 10 0 Filled symbols indicate SSI-positive neurons, showing a hyperpolarization 20 10 0 10 20 10 0 10 0 5 10 15 20 25 Vm (mV) associated to a reduction of Rm. (j) Voltage-clamp recording of a Time (min) pyramidal neuron (60 mV) in the presence of tetrodotoxin (0.5 M), CB1/ WT Intracellular BAPTA Intracellular THL 200 6,7-dinitroquinoxaline-2,3-dione (10 M), D()-2-amino-5-phosphonovaleric acid (100 M) and gabazine (10 M). There was an outward shift of the 100 holding current in response to ten depolarizing voltage steps from 60 to n = 87 n = 67 n = 43 n = 48 0 mV (shaded area). (k) The same stimulus protocol in another pyramidal 0 10 0 10 20 10 0 10 20 10 0 10 20 10 0 10 20 neuron failed to induce a shift in the holding current. Vertical deflections Vm (mV) Vm (mV) in j and k are responses to tests for passive properties and recording 0 Pre-steps 60 mV Post-steps * 120 stability. (l) Cumulative probability plot of the current required to hold 100 the cell at 60 mV (Ihold) before (pre-steps, black line) and after 50 60 (post-steps, gray line) voltage steps to 0 mV for all tested neurons n = 76 (n = 76). Note the positive shift of the holding current (Kolmogorov-Smirnov 0 0 2.5 min 2 min 100 0 100 200 SSI SSI+ test, P < 0.03). (m) Average plot of the current necessary to hold Ihold (pA) pyramidal neurons at 60 mV before (black) and after (white) the depolarizing steps in SSI-negative and SSI-positive neurons. * P < 0.001. Error bars represent s.e.m. All experimental procedures were approved by national committees on animal health and care (Italian Ministry of Health, INSERM and French Ministry of Agriculture and Forestry).

10 mV

Percentage Rm

Percentage Rm

Vm (Vm)

Cum. prob. (%)

5 mV

50 pA

1European

Brain Research Institute, Rome, Italy. 2Endocannabinoids and Neuroadaptation Group, U862 INSERM NeuroCentre Magendie Universit Bordeaux 2, Bordeaux, France. Correspondence should be addressed to A.B. (a.bacci@ebri.it). Received 28 July; accepted 25 September; published online 15 November 2009; doi:10.1038/nn.2430

1488

VOLUME 12 | NUMBER 12 | december 2009 nature neuroscience

50 pA

Ihold (pA)

b r i e f c o m m u n i c at i o n s a
Figure 2 SSI is mediated by persistent CB1R activation and is mimicked by CB1R agonists. (a) Current-clamp recording of an SSI-positive layer 2/3 pyramidal neuron. Application of AM-251 10 min after SSI-inducing stimuli slowly terminated SSI. (b) Summary Vm time-course plots relative to AM-251 treatment 5 min (white) and 10 min (black) after SSI induction. (c) Current-clamp recording of a pyramidal neuron responding to the synthetic CB1R agonist WIN-55,212 (1 M) with hyperpolarization and reduced Rm (left). Right, summary plot (n = 4) of changes in Vm after application of the endocannabinoid 2-AG and subsequent application of AM-251 (AM). ** P < 0.01, paired t test. (d) Rm versus Vm plots of WIN-55,212treated (left) and 2-AGtreated (right) neurons. Black symbols indicate pyramidal neurons responding to CB1R agonists. Error bars represent s.e.m.

Spike trains

AM-251

b
Vm (mV)

5 min 6 Spike trains AM-251

5 mV

6 0 15 Time (min) 30 5 Vm (mV) 0 5 mV

c
2009 Nature America, Inc. All rights reserved.

WIN 55,212

2 min

**
Control 2-AG AM 2-AG WIN

10 150 Percentage Rm 100 50

d
Percentage Rm

150 100 50 0

n = 20 10 0 10 10 Vm (mV) 0

n = 19 10

in 91.8% of pyramidal neurons (n = 67) from mice lacking the gene for CB1Rs (Cnr1/, referred to as CB1/), resulting in a reduction of SSI incidence as compared with their wild-type littermates (23 of 87 pyramidal neurons, 26.44%, P < 0.01, 2 test; Fig. 1f,g), which had similar SSI occurrence and magnitude as rats (P > 0.5, 2 test; Fig.1d and Supplementary Table 1). Endocannabinoid-mediated SSI of pyramidal neurons resulted from activation of G proteincoupled inward-rectifying K+ (GIRK) channels, as determined by changes of I-V relationships induced by the synthetic CB1R agonist WIN-55,212 (1 M), and Ba2+ effects (Supplementary Fig. 3). The remaining hyperpolarization present in CB1 knockout mice had similar SSI kinetics and might have been caused by activity-dependent activation of K+ channels and/or other mechanisms. To test whether pyramidal neuron SSI is a result of an autocrine action of neo-synthesized 2-arachidonoylglycerol (2-AG) 3, we attempted to induce SSI while intracellularly perfusing pyramidal neurons with the Ca2+ chelator BAPTA (10 mM, n = 43), which prevents endocannabinoid production6,9, or tetrahydrolipstatin (THL 3 M; n = 48), an inhibitor of 2-AG synthesis10. In both conditions, pyramidal neuron SSI was prevented (P < 0.005 both treatments, 2 test; Fig. 1h,i and Supplementary Table 1). The incidence of SSI in THL-treated neurons was statistically similar to that of slices treated with AM-251 or from CB1/ mice (2 test, P > 0.1 both cases). Consistent with the idea that SSI is triggered by the same neurons that fire action potentials, SSI was similarly detected when network activity was neutralized in volt age-clamp experiments (Fig. 1jm). Local perfusion of AM-251 5 or 10 min following SSI induction reversed the membrane hyperpolarization in SSI-positive pyramidal neurons (n = 5 and n = 4, respectively; Fig. 2a,b), indicating that there was persistent CB1 activation during SSI. Late AM-251 applications failed to depolarize
nature neuroscience VOLUME 12 | NUMBER 12 | december 2009

SSI-negative pyramidal neurons (n = 22; data not shown), which were therefore not already under the influence of self-inhibition. The synthetic CB1R agonist WIN-55,212 (1 M) and the endocannabinoid 2-AG (10 M) persistently hyperpolarized pyramidal neurons to a similar extent as was observed during SSI (35% and 31%, respectively; 2 test, P > 0.8). This was associated with a similar change of Rm (P > 0.1 for WIN or 2-AG versus SSI, t test; Fig. 2c,d and Supplementary Table 1) and was terminated by AM-251 (Fig. 2c). Notably, spike trains delivered after WIN or 2-AG applications failed to induce SSI (n = 19; data not shown). Conversely, WIN hyperpolarized some SSI-negative pyramidal neurons (4 of 18 neurons, 6.9 1.9 mV; data not shown). Altogether, these data suggest that pharmacological CB1 activation occludes SSI, but action potential firing is unable to induce SSI in some CB1R-expressing pyramidal neurons. Cortical pyramidal neurons are classified into different subtypes depending on electrophysiological, morphological and topographic features11. We could not detect differences in action potential waveform and firing characteristics in SSI-positive versus SSI-negative layer 2/3 pyramidal neurons (Supplementary Table 2). However, apical, but not basal, dendrites of SSI-positive pyramidal neurons were less branched and extended more superficially than those of SSI-negative cells (n = 9 and n = 8, respectively; P < 0.05; Fig. 3ac). SSI-positive and SSI-negative pyramidal neurons did not differ in total dendritic length (P > 0.05; Fig. 3d) and were equally distributed (P > 0.05; Fig. 3e). Thus, SSI-positive and SSI-negative neurons are two morphologically different pyramidal neuron subtypes. Somatodendritic expression of CB1Rs in glutamatergic neurons has not been morphologically demonstrated6, despite the wide expression of CB1 mRNA in glutamatergic cells (77%; Supplementary Fig.1)7. Using CB1 immunocytochemistry with pepsin12, we found that very low levels of somatic CB1Rs were expressed in a small percentage of pyramidal neurons (Supplementary Fig. 2). These immunohistochemical data provide further evidence for CB1R expression in somatodendritic compartments of neocortical layer 2/3 pyramidal neurons. Does pyramidal neuron SSI interact with other5,13 endocannabinoidmediated signaling in the neocortex? We found that SSI-positive pyramidal neurons could evoke depolarization-induced suppression of inhibition more commonly than SSI-negative cells (89% versus 35% for SSI-positive versus SSI-negative pyramidal neurons, n =9 and n = 31, respectively; Supplementary Fig. 4), suggesting that either SSI-negative pyramidal neurons receive less inhibitory contacts expressing CB1Rs than SSI-positive pyramidal neurons or a large fraction of SSI-negative pyramidal neurons is unable to produce endocannabinoids on depolarization. Notably, SSI-inducing repeated depolarizations elicited a long-lasting reduction of evoked inhibitory postsynaptic currents in all SSI-positive, but in much smaller percentage of SSI-negative, pyramidal neurons (100% versus 25%, SSI-positive versus SSI-negative pyramidal neurons, n = 9 and 31, respectively; Supplementary Fig. 4). This depolarization-induced
1489

b r i e f c o m m u n i c at i o n s a b c
Number of intersections
SSI+ SSI Basal dendrites 0 0 200 400 0 100 200 Distance from soma (m)

15

Apical dendrites

20

SSI+

Dendrite length (mm)

SSI

d
3.0 SSI+ SSI

Distance (mm)

Pial surface I II/III IV V VI

0.5 1.0 1.5

1.5

50 m

0 Basal Apical

SSI+

SSI

2009 Nature America, Inc. All rights reserved.

Figure 3 SSI identifies a morphological subtype of neocortical pyramidal neurons. (a,b)Somatodendritic compartments of a biocytin-filled and digitally reconstructed SSI-positive (a) and SSI-negative (b) pyramidal neuron in neocortical layer 2/3. (c) Plot of dendrite intersections (apical, left; basal, right) with virtual circles of increasing radii versus distance from the soma of SSI-positive and SSI-negative pyramidal neurons. The shaded box indicates significantly different branching between the two groups (P< 0.05). Arrowheads indicate apical dendrite farther extension in SSI-positive pyramidal neurons. (d)Plot of total basal and apical dendritic length in SSI-positive and SSI-negative neurons. (e) Distribution across cortical layers of SSI-positive and SSI-negative pyramidal neurons. Error bars represent s.e.m.

long-term depression of inhibition (d-LTDi) was terminated by late applications of AM-251, suggesting that d-LTDi was a result of persistent activation of presynaptic CB1Rs. Persistent CB1R signaling in SSI and d-LTDi might result from either a switch into a constitutively active configuration of the receptor14 or a constant 2-AG presence. Both of these possibilities imply that intracellular molecular changes occur in the post- and/or presynaptic neurons, translating transient postsynaptic depolarizations into a permanent activation of CB1R signaling. d-LTDi shares the key feature of persistent CB1 signaling with long-term depression of autaptic glutamatergic transmission in cultured neurons15, but it differs from other forms of endocannabinoidmediated synaptic plasticity of GABAergic responses that require CB1R activation for induction, but not expression, and highfrequency presynaptic stimulation inducing heterosynaptic plasticity5. Thus, repeated postsynaptic depolarizations induce autocrine SSI and also stimulate persistent retrograde endocannabinoid signaling to dampen inhibitory synaptic inputs. Such dualistic effects might depend on spatial aspects of endocannabinoid mobilization and could be important for the modulation of neocortical networks. The long-lasting decrease of GABAergic responses impinging the same pyramidal neurons that produced SSI might counterbalance the persistent depression of pyramidal neuron excitability. Thus, SSI and d-LTDi might provide fine modulation of the network at the singlecell level. Alternatively, when the whole network undergoes cycles of intense repetitive activity, the entire SSI-positive glutamatergic population might generate more global effects, with probably important consequences for information processing of the neocortex1.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank N. Berretta (Fondazione Santa Lucia) for insightful discussion, and P. Mendez-Garcia (European Brain Research Institute) and J. Loureno (INSERM) for critically reading the manuscript. We are grateful to M. Ammassari-Teule (CNRInstitute of Neuroscience and Fondazione Santa Lucia) for the use of

Neurolucida, and to M. Watanabe (Hokkaido University School of Medicine) for the gift of CB1 antibody. This work was supported by the Giovanni Armenise-Harvard Foundation (Career Development Award to A.B.), the European Commission (a Marie Curie International Reintegration Grant to A.B.), the European Research Council (ERC) under the European Communitys 7th Framework Programme (FP7/2007-2013, ERC grant agreement number 200808, A.B.), the Italian Institute of Technology (A.B.), Avenir INSERM (in partnership with the Fondation Bettencourt-Schueller, G.M.), Agence National de la Rechereche (G.M.), Region Aquitaine (G.M.), the European Commission Coordination Action ENINET (contract number LSHM-CT-2005-19063 to A.B. and G.M.). A.B. is the 2007/2008 National Alliance for Research on Schizophrenia and Depression Henry and William Test Investigator. AUTHOR CONTRIBUTIONS S.M. and A.B. designed the experiments, S.M. conducted all of the electrophysiological experiments, and S.P. performed the immunhistochemistry and morphological analyses. S.M., S.P. and A.B. analyzed the data. A.C. and G.M. performed the in situ hybridization and provided the mutant mice. S.M., G.M. and A.B. wrote the manuscript.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
Carandini, M. & Ferster, D. Science 276, 949952 (1997). Bacci, A., Huguenard, J.R. & Prince, D.A. Nature 431, 312316 (2004). Marinelli, S. et al. J. Neurosci. 28, 1353213541 (2008). Trettel, J., Fortin, D.A. & Levine, E.S. J. Physiol. (Lond.) 556, 95107 (2004). Chevaleyre, V., Takahashi, K.A. & Castillo, P.E. Annu. Rev. Neurosci. 29, 3776 (2006). 6. Kano, M., Ohno-Shosaku, T., Hashimotodani, Y., Uchigashima, M. & Watanabe, M. Physiol. Rev. 89, 309380 (2009). 7. Hill, E.L. et al. J. Neurophysiol. 97, 25802589 (2007). 8. Fortin, D.A. & Levine, E.S. Cereb. Cortex 17, 163174 (2007). 9. Stella, N., Schweitzer, P. & Piomelli, D. Nature 388, 773778 (1997). 10. Hashimotodani, Y., Ohno-Shosaku, T., Maejima, T., Fukami, K. & Kano, M. Neuropharmacology 54, 5867 (2008). 11. Spruston, N. Nat. Rev. Neurosci. 9, 206221 (2008). 12. Watanabe, M. et al. Eur. J. Neurosci. 10, 478487 (1998). 13. Freund, T.F., Katona, I. & Piomelli, D. Physiol. Rev. 83, 10171066 (2003). 14. Leff, P. Trends Pharmacol. Sci. 16, 8997 (1995). 15. Kellogg, R., Mackie, K. & Straiker, A. J. Neurophysiol. 102, 11601171 (2009). 1. 2. 3. 4. 5.

1490

VOLUME 12 | NUMBER 12 | december 2009 nature neuroscience

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

Self-generated theta oscillations in the hippocampus


Romain Goutagny1,2, Jesse Jackson1,2 & Sylvain Williams1 Hippocampal theta rhythm is crucial for spatial memory and is thought to be generated by extrinsic inputs. In contrast, using a complete rat hippocampus in vitro, we found several intrinsic, atropine-resistant theta generators in CA1. These oscillators were organized along the septotemporal axis and arose independently from CA3. Our results suggest that CA1 theta rhythm can emerge from the coupling of multiple autonomous hippocampal theta oscillators. Episodic and spatial memory, the ability to recollect past events and locations, requires the sequential binding of cell assemblies over a temporal metric provided by theta-frequency oscillations1. Traditional models of hippocampal theta rhythm generation emphasize the necessity of a synchronizing extrinsic rhythm generator that is thought to arise from the medial septum. However, theoretical studies2,3 suggest that the hippocampus may possess the minimal circuitry required for theta periodicity. The lack of experimental evidence to support this concept may be a result of the fact that the hippocampal slice preparation is not adequate for the study of complete intrinsic hippocampal network properties4. To circumvent this issue, we performed extracellular field recording in the intact isolated hippocampus in vitro (from rats postnatal day 1528; Fig. 1a, Supplementary Methods and Supplementary Fig. 1). After 1050 min in the recording chamber (Fig. 1bd), CA1 area exhibited self-sustained, continuous theta activity (5.1 0.2 Hz, range = 310 Hz, mean theta waveform amplitude = 0.124 0.007 mV, range = 0.0330.331 mV, n = 62) at a similar frequency as has been seen in in vivo recordings57 and that lasted up to 120 min (Supplementary Fig. 2). The wave-shape of the theta oscillation varied according to frequency and maintained a symmetrical near-sinusoidal shape similar to in vivo theta rhythm (Supplementary Figs. 3 and 4). Using
Figure 1 Descriptive properties of theta rhythm in the isolated hippocampus. (a) A photograph of the complete hippocampal preparation. (b) Spectrogram showing the emergence of continuous CA1 theta oscillations in the isolated hippocampus in vitro. (c) Representative raw data and corresponding autocorrelations from the same experiment shown in b at the times indicated. Scale bars represent 0.1 mV and 1 s. (d) Left, raw data from 75-m increments through the distal CA1 region in the middle hippocampus. Middle, averaged filtered traces from four sequential depths demonstrating the rapid complete phase reversal over 150 m. Right, current source density of the averaged theta wave (triggered from stratum radiatum), demonstrating a single sink/source dipole. Scale bars represent 0.1 mV and 200 ms.
1Douglas

2009 Nature America, Inc. All rights reserved.

identical experimental conditions, we found that theta was not generated in transverse or horizontal hippocampal slices (500 m, n = 5; Supplementary Fig. 5). The depth profile of the recorded theta activity demonstrated a near-complete phase shift from stratum oriens to stratum radiatum (164 11, range = 116208; Supplementary Fig. 6) and current source density measurements revealed the presence of a single sink-source alternation between the pyramidal cell layer (with high multiunit activity) and stratum radiatum (Fig. 1d). This hippocampal theta was observed using normal artificial cerebrospinal fluid, but was reduced or abolished by increasing either excitatory or inhibitory tone (Supplementary Figs. 7 and 8). To assess the neurotransmitters involved in this oscillation, we bath-applied different antagonists for cholinergic, glutamatergic and GABAergic neurotransmission (Supplementary Fig. 9 and

a
CA1

b 25
Frequency (Hz)

0.04

20 15 10 5 20 30 40 50 60 70 80 90 100 Time (min) 0.59

CA3

Temporal

Septal

c
1.0 0.5 0 0.5 1.0 1

35 min

60 min

100 min

0 Time (s) Voltage

+1 1

0 Time (s) Mean wave

+1 1

0 Time (s) Current

+1

d
Str. or

Str. pyr

Str. rad

2 3 Time (s)

300 ms Sink 2,500

300 ms Source 2,500

Mental Health University Institute, McGill University, Department of Psychiatry, Montral, Qubec, Canada. 2These authors contributed equally to this work. Correspondence should be addressed to S.W. (sylvain.williams@douglas.mcgill.ca). Received 8 June; accepted 29 September; published online 1 November 2009; doi:10.1038/nn.2440

nature neuroscience VOLUME 12 | NUMBER 12 | december 2009

1491

b r i e f c o m m u n i c at i o n s
Figure 2 Topography of the spontaneous theta Coherence Theta power oscillation recorded in CA1. (a) Recording per mm 1.0 1 2 3 4 5 Temporal Temporal electrodes were moved (1-mm intervals) along 0.9 1010 10 10 10 Intact 0.00 0.8 both transverse and longitudinal axes, as shown 0.05 0.7 CA3 intact 0.10 0.6 by the location of black and gray dots. The 0.15 CA3 removed 0.5 Pyr coherence was measured in both axes and the 0.20 0.4 * 0.25 0.3 CA3 removed C A1_T position of the point is indicated by the arrow 1 0.2 0.30 Rad CA3 0.1 below the plot. The change in coherence per mm 0 dropped at a rate of 0.2 0.04 per mm in CA1_T1 CA3 Septal Septal the transverse plane, whereas it dropped in the CA1_L1 CA1_L4 longitudinal axis by only 0.01 0.05 per mm Autocorrelations CA1(T) (t3 = 3.06, P = 0.055). CA1_L1, longitudinal 1; Pre CA1_L4, longitudinal 4; CA1_T1, transverse 1. 2.5 0.9 (b) Removal of CA3 did not prevent the * Pre * 0.8 2.0 emergence of theta activity. Example raw data CA1(S) 0.7 (gray) and theta-filtered (312 Hz) data (black) 0.6 1.5 of activity recorded in two separate preparations 0.5 Post 0.4 1.0 with and without CA3. Right, theta power versus 0.3 Post depth for intact (n = 6) and CA3-removed (n = 4) 0.5 0.2 CA1 (septal) conditions. (c) The experimental procedure 0.1 CA1 (temporal) 0 0 used to identify independent intrahippocampal n=7 0 +1 1 theta oscillators. Procaine (gray ellipse with Time (s) hatched lines) was infused between two recording electrodes. Representative raw data (0.1500 Hz) recorded from the septal and temporal poles of the hippocampus before (top) and 2 min after procaine infusion (bottom) between the two hippocampal sites, unmasking the presence of two separate frequencies in the two recordings, as shown in the autocorrelations. The maximal septo-temporal theta coherence between the two recording electrodes decreased following procaine infusion as a result of the two regions oscillating at two distinct frequencies (mean absolute difference = 1.3 Hz, far right). Scale bars in b and c represent 500 ms and 0.1 mV.
CA 1_ L4

Coherence

CA

1_ L1

Septo-temporal coherence

Pr Po e st

abs frequency (Hz)

2009 Nature America, Inc. All rights reserved.

Supplementary Table 1). Given that two types of hippocampal theta can be identified in vivo on the basis of sensitivity to muscarinic receptor blockers8, we first applied atropine sulfate (10 M, 10 min, n = 9) and observed no change in either the power (F2,24 = 3.07, P = 0.17) or frequency (F2,24 = 2.27, P = 0.13) of the theta oscillations. The CA1 theta oscillations were only abolished by the application of the AMPA/kainate receptor antagonist 6,7-dinitroquinoxaline2,3-dione (DNQX, 10 M, 10 min, n = 6, F2,15 = 54.75, P < 0.001; Supplementary Fig. 9) or the GABAA receptor antagonist bicuculline (5 M, 2 min, n = 6, F2,12 = 130.3, P < 0.0001; Supplementary Fig. 9). Taken together, these results indicate that the isolated hippocampus can generate self-sustained theta oscillations without cholinergic activation or afferent inputs. We then examined the topography of this theta oscillation by measuring theta-band coherence along both the longitudinal (septotemporal) and transverse (CA3CA1) axes using the distal CA1 as a reference. Theta-band coherence remained high along the longitudinal axis, but exhibited significant decreases over the transverse axis (F3,12 = 8.04, P = 0.003), with the lowest coherence in CA3 (Fig. 2a), indicating that this intrinsic theta oscillation is preferentially organized along the longitudinal axis of CA1. These data suggest that the CA1 oscillations arise independent of CA3. To definitively test this hypothesis, we surgically removed CA3 during the dissection (Supplementary Fig. 10). The removal of CA3 did not impair the ability of the CA1 region to generate self-sustained theta rhythm (Fig. 2b). No significant change in theta frequency (n = 8, 5.0 0.4 Hz, P > 0.05) or in the phase shift was observed in the absence of CA3 (n = 4, 186 15, P > 0.05). Therefore, CA1 theta oscillations can arise independent of CA3 inputs in the isolated hippocampus in vitro. These results indicate that the CA1 area of the hippocampus contains the sufficient circuitry to spontaneously and independently generate theta oscillations in vitro. As a result, the dissociable mechanisms of spatial encoding between CA1 and CA3 (refs. 1,9) may be explained by the presence of separate intrinsic theta oscillators in each region. We then tested whether local inactivation of part of the CA1 region would uncouple theta oscillations recorded 45 mm apart
1492

in the septotemporal direction. Blockade of synaptic transmission (20% procaine hydrochloride, vol/vol, 0.10.3 L) between two electrodes positioned at the septal and temporal poles reduced the peak coherence (0.71 0.09 pre, 0.16 0.10 post procaine, t6 = 5.22, P = 0.002) of theta (Fig. 2c) without altering hippocampal theta power (Supplementary Fig. 11). This uncoupling was a result of the emergence of two distinct oscillators with different frequencies (the absolute difference in peak frequency = 0.2 0.1 Hz pre versus 1.3 0.2 Hz post, t6 = 3.63, P = 0.01; Fig. 2c). Moreover, the faster oscillator following procaine showed a slight lead in the pre-procaine condition (49 26 ms), suggesting that this oscillator was entraining the slower oscillator. Similar infusions of procaine in CA3 did not alter the properties of the CA1 rhythm (Supplementary Fig. 11). These results indicate that multiple theta oscillators with different frequencies coexist along the longitudinal axis of CA1 (Supplementary Fig. 12) and that a collective population rhythm can emerge via phase entrainment of theta oscillators of different inherent frequencies. This concept is consistent with a recent study suggesting that in vivo hippocampal theta wave propagation arises from a network of weakly coupled theta oscillators10. At the cellular level, extracellular recordings of putative principle cells and interneurons (n = 29) both demonstrated significant phase locking to the extracellular field potential (159 11, P < 0.01 using the Rayleigh test for uniformity, theta peak recorded from radiatum is 180; Supplementary Fig. 13), indicating that theta oscillations in CA1 entrain pyramidal cells and interneurons. To further investigate how these cells were synaptically entrained during theta-band network activity, we performed whole-cell recordings of pyramidal cells (n = 5) and interneurons (n = 12; Fig. 3). At resting membrane potential, the principle cells did not fire action potentials, but instead had rhythmic synaptic potentials ( n = 4 cells were phase-locked to the extracellular oscillation). Principle cells were driven at rest by prominent phasic inhibitory postsynaptic potentials (IPSPs, mean amplitude of 3.8 mV, Fig. 3a,c) and clearly showed rebound spiking following the IPSP when depolarized (mean phase of 120; Fig. 3a,c) on a similar (although slightly earlier) phase as pyramidal cells in vivo5,11,12. Two pyramidal cells
VOLUME 12 | NUMBER 12 | december 2009 nature neuroscience

Pr e Po st

b r i e f c o m m u n i c at i o n s
Figure 3 Synaptic activity during the * spontaneous theta oscillations. (a) Synaptic LFP str. Sink pyr activity recorded in a pyramidal cell during theta 50 mV 20 mV 20 mV 50 mV Source oscillations. This cell was driven by prominent 200 pA 200 pA 20 mV LFP str. 0.4 s IPSPs at rest, but the IPSPs were at equilibrium 0.4 s 20 ms rad 0.1 s * near 68 mV and reversed in polarity with further membrane hyperpolarization. 0.2 PYR 0.1 (b) Synaptic activity of an example 0 Rest IPSPs fast-firing interneuron during spontaneous 2 mV theta oscillation. This cell was driven by EPSPs at rest. (c) A potential mechanism for the 10 mV 10 mV 47 mV 70 mV 47 mV generation of hippocampal theta on the basis 55 mV 52 mV of our data. Top, two cycles of extracellular 0.16 INT 0.08 theta are shown and the corresponding sink68 mV 0 70 mV source alternations are shown as in Figure 1. IPSPs Below, data from a principle cell (PYR) and an 0 mV 80 mV 80 mV interneuron (INT) are shown, with the spiking 5 mV and synaptic activity thought to underlie the 70 mV Field Field 100 V 100 V sink source alternation. The large number EPSPs of sparsely firing principle cells fired on the 0.2 s 0.2 s 0 180 360 540 720 falling phase of theta recorded from stratum Theta phase pyramidale and corresponding to the sink in this layer. This elicited EPSPs and spiking in interneurons, which subsequently feedback onto principle cells. Thus, interneuron spiking was maintained by the sparsely firing pyramidal cell network. The interneuron spike elicited IPSPs in a large principle-cell population, thus resetting the membrane potential to a common phase and corresponding to the source in stratum pyramidale and passive sink in radiatum. In addition, synaptically connected interneurons elicited IPSPs shortly after interneuron spiking (shown in red). LFP, local field potential.
Spike prob. Spike prob.

2009 Nature America, Inc. All rights reserved.

were driven by small-amplitude excitatory postsynaptic potentials (EPSPs) at rest (mean of 1.4 mV), as well as phase-locked IPSPs when the cells are depolarized at 0 mV (mean amplitude of 2.2 mV). These results indicate that CA1 pyramidal cells are paced by interneurons during spontaneous theta oscillations and that a subset of these cells also receives excitatory inputs from other principle cells13. Interneurons exhibited prominent phasic EPSPs at rest that correlated to the theta oscillation (mean amplitude of 5.4 1.1 mV, n = 9 of 12; Fig. 3b) and fired action potentials (mean frequency of 3.2 0.7 Hz, n = 7) that were phase-locked to the extracellular theta oscillation (mean phase of 195 4; Fig. 3b,c). In all of the interneurons tested ( n = 3), it was also possible to identify an inhibitory driving component that was phase-locked to the local field potentials when the cells were held at 0 mV (mean amplitude of 4.1 0.9 mV). These results indicate that CA1 interneurons are mainly driven by local principle cells during spontaneous theta oscillation and that they also received inhibitory inputs. Therefore, similar to many other biological network oscillators14, a feedback loop between principle cells and inhibitory interneurons 15 appears to be necessary and sufficient for the emergence of this in vitro theta rhythm (Fig. 3c). Our results show that the CA1 area of the hippocampus contains the sufficient intrinsic circuitry to spontaneously generate theta oscillations, indicating that the CA1 region can undergo a previously unappreciated form of long range theta synchronization without a precise clocking input (see Supplementary Discussion).
Note: Supplementary information is available on the Nature Neuroscience website.

Acknowledgments We thank M. Danik, G. Ducharme, C.K. Young, C.T. Dickson and G. Buzsaki for their comments on the manuscript. This work was supported by the Canadian Institute of Health Research, the Natural Sciences and Engineering Research Council of Canada and the Fonds de la Recherche en Sant du Qubec. R.G. was supported by the Conrad F. Harrington post-doctoral fellowship from the McGill Faculty of Medicine. J.J. received a Canadian Graduate Scholarship from the Natural Sciences and Engineering Research Council of Canada and S.W. is a Fonds de la recherche en sant du Qubec chercheur boursier senior. AUTHOR CONTRIBUTIONS R.G., J.J. and S.W. designed the experiments, R.G. and J.J. performed the experiments and analyzed the data, and R.G., J.J. and S.W. wrote the paper.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
1. Hasselmo, M.E. Hippocampus 15, 936949 (2005). 2. Traub, R.D., Miles, R. & Wong, R.K. Science 243, 13191325 (1989). 3. White, J.A., Banks, M.I., Pearce, R.A. & Kopell, N.J. Proc. Natl. Acad. Sci. USA 97, 81288133 (2000). 4. Amaral, D.G. & Witter, M.P. Neuroscience 31, 571591 (1989). 5. Bland, B.H. Prog. Neurobiol. 26, 154 (1986). 6. Buzski, G. Neuron 33, 325340 (2002). 7. Vanderwolf, C.H. Electroencephalogr. Clin. Neurophysiol. 26, 407418 (1969). 8. Kramis, R., Vanderwolf, C.H. & Bland, B.H. Exp. Neurol. 49, 5885 (1975). 9. Leutgeb, S., Leutgeb, J.K., Treves, A., Moser, M.B. & Moser, E.I. Science 305, 12951298 (2004). 10. Lubenov, E.V. & Siapas, A.G. Nature 459, 534549 (2009). 11. Fox, S.E., Wolfson, S. & Ranck, J.B. Jr. Exp. Brain Res. 62, 495508 (1986). 12. Csicsvari, J., Hirase, H., Czurko, A., Mamiya, A. & Buzsaki, G. J. Neurosci. 19, 274287 (1999). 13. Crpel, V., Khazipov, R. & Ben-Ari, Y. J. Neurophysiol. 77, 20712082 (1997). 14. Andersen, P. & Eccles, J. Nature 196, 645647 (1962). 15. Cobb, S.R., Buhl, E.H., Halasy, K., Paulsen, O. & Somogyi, P. Nature 378, 7578 (1995).

nature neuroscience VOLUME 12 | NUMBER 12 | december 2009

1493

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

The pathways of interoceptive awareness


Sahib S Khalsa1,2,4, David Rudrauf1,4, Justin S Feinstein3 & Daniel Tranel1,3 A network of cortical brain regions, including the insula and anterior cingulate cortex (ACC), has been proposed as the critical and sole substrate for interoceptive awareness. Combining lesion and pharmacological approaches in humans, we found that the insula and ACC were not critical for awareness of heartbeat sensations. Instead, this awareness was mediated by both somatosensory afferents from the skin and a network that included the insula and ACC. Together, these pathways enable the core human experience of the cardiovascular state of the body. Recent functional neuroimaging studies have highlighted a network of brain regions that include the insula and the ACC as being important for representing the visceral state of the body, as they are activated when we feel interoceptive stimuli such as the heartbeat13 and gastrointestinal sensations4. It has been proposed that the insula is the sole and critical cortical substrate for interoceptive awareness5,6, whereas it has been suggested that the ACC has a visceromotor role2,5.

2009 Nature America, Inc. All rights reserved.

It has been further proposed that the anterior insula instantiates all subjective feelings from the body and feelings of emotion5,6. In this view, interoceptive awareness would be fully absent in the complete absence of the insula. However, the same studies suggesting that the insula and ACC are involved also show that interoceptive stimuli robustly activate somatosensory cortices14. The experience of our skin blushing during embarrassment, the feeling of our heart pounding in the chest and of blood vessels pulsating in the throat during anxiety and fear intuitively suggests that perhaps the skin and its somatosensory afferent projections critically contribute to interoceptive awareness and emotion79. In this view, the insula (and ACC) would not be necessary for interoceptive awareness and, in addition to pathways projecting to the insula, other somatosensory pathways would be involved10. We therefore considered two possible interoceptive awareness pathways7,9: one involving visceral afferents projecting to the insula and another involving skin afferents projecting to somatosensory cortex (see Fig. 1 for specific alternative hypotheses). We examined the contributions of these two pathways to interoception in a patient (Roger) with virtually complete bilateral insula and ACC damage, but who had intact bilateral primary somatosensory cortex11 (Fig. 2). Rogers neurological presentation is extremely rare and represented a unique opportunity to test predictions about these pathways in humans. We assessed moment-to-moment awareness of cardiovascular sensations in response to bolus administrations of isoproterenol,

H1 H2 H3 H4 H5 Figure 1 Schematic representing possible pathways of interoceptive awareness. Two possible pathways are envisioned: one involving visceral afferents projecting to the insula (green arrow) and another involving somatosensory skin afferents (cyan arrow). Several alternative hypotheses are: interoceptive awareness is only mediated by the insula pathway (H1, standard hypothesis), interoceptive awareness is only mediated by the somatosensory pathway (H2), interoceptive awareness is independently mediated by each pathway (H3), interoceptive awareness is dependent on the simultaneous action of both pathways (H4) and interoceptive Anesthetic No Anesthetic No Anesthetic No Anesthetic No Anesthetic No awareness could be dependent on other anesthetic anesthetic anesthetic anesthetic anesthetic pathways (H5). If interoceptive awareness is only mediated by the insula pathway, then only bilateral damage to the insula pathway should abolish interoceptive awareness. If interoceptive awareness is only mediated by the somatosensory pathway, then only disrupting the somatosensory pathway should abolish interoceptive awareness. If interoceptive awareness is independently mediated by each pathway, then disruption of either pathway should not abolish interoceptive awareness, but disruption of both should. If interoceptive awareness is dependent on the simultaneous action of both pathways, then disrupting either pathway should abolish interoceptive awareness. If interoceptive awareness is dependent on other pathways, disruption of both pathways should not abolish interoceptive awareness. Our results support the idea that interoceptive awareness is independently mediated by each pathway (dashed outline). + indicates interoceptive awareness present, and indicates interoceptive awareness absent. No insula Insula No insula Insula No insula Insula No insula Insula No insula Insula

1Department 4These

of Neurology and Neuroscience Program, 2Medical Scientist Training Program and 3Department of Psychology, University of Iowa, Iowa City, Iowa, USA. authors contributed equally to this work. Correspondence should be addressed to S.S.K. (skhalsa@mednet.ucla.edu).

Received 6 June; accepted 20 August; published online 1 November 2009; doi:10.1038/nn.2411

1494

VOLUME 12 | NUMBER 12 | december 2009 nature neuroscience

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

gh
a b c d

Figure 2 Brain damage in Roger. (ah) Top, extent of damage (black) on magnetic resonance imaging views of lateral (upper left and right), ventral (middle) and mesial (lower left and right) cerebrum. Bottom, axial ( ad) and sagittal (eh) slices, with corresponding slice locations displayed at top. Ins, insula.

e
2009 Nature America, Inc. All rights reserved.

Ins

Ins

Ins

Ins

a sympathetic (beta adrenergic) agonist similar to adrenaline, as a marker of interoceptive awareness 8 (Supplementary Methods). After each bolus, participants turned a dial to track their momentto-moment experience of the intensity of heartbeat sensations. All participants provided informed written consent as approved by the General Clinical Research Center Advisory Committee and the Institutional Review Board of the University of Iowa. As expected, Roger demonstrated dose-dependent heart-rate increases that were indistinguishable from healthy comparison participants (Fig. 3a and Supplementary Table 1). However, contrary to what would be predicted from the hypotheses that interoceptive awareness is only Comparison Comparison heart rate a 25 b 30 30 Roger Comparison dial mediated by the insula pathway or that Roger heart rate 25 25 20 Roger dial interoceptive awareness is dependent on 20 20 15 15 15 the simultaneous action of both pathways 10 10 10 (Fig. 1), Roger demonstrated dose-dependent 5 5 0 0 5 changes in interoceptive awareness that were 5 5 0 comparable to healthy comparison par2.0 mg 4.0 mg 10 10 15 5 15 ticipants, albeit somewhat delayed in time 0 20 40 60 80 100 120 140 160 180 Saline (Fig. 3b and Supplementary Fig. 1). Verbal Time (s) Time (s) Dose (mg) responses recorded during the experimental c d 30 30 session also suggested that Roger perceived 25 25 10 20 20 qualitative changes in cardiovascular sensa15 15 8 tion (Supplementary Table 2). This suggests 10 10 6 5 5 N that the insula and ACC are not necessary 4 0 0 for interoceptive awareness and that either 2 5 5 2.0 mg 4.0 mg 10 10 interoceptive awareness is only mediated by 0 15 15 the somatosensory pathway or interoceptive 0 25 50 75 100 125 150 Comparison Roger Time (s) awareness is independently mediated by Time (s) each pathway. Figure 3 Heart-rate response and on-line subjective dial ratings of interoceptive awareness To further assess these two remaining changes induced by isoproterenol. (a) Roger and 11 healthy age-matched male comparison hypotheses, we applied a topical lidocaine participants exhibited equivalent dose-dependent heart-rate increases. ( b) Time course of heart-rate anesthetic to the skin covering each par- response and dial ratings. Roger and the healthy participants appropriately demonstrated doseticipants region of maximal heartbeat dependent changes in interoceptive awareness. Bolus infusions occurred at time 0. ( c) Overlap sensation, as reported during the previous map showing the region of maximal heartbeat sensation, corresponding to the area of topical anesthetic application. (d) Time course of heart-rate response and dial ratings after anesthetic isoproterenol challenge (Fig. 3c). We then application. Roger no longer demonstrated appropriate changes in interoceptive awareness, even repeated a new challenge with the highest at the two highest doses. Comparison participants interoceptive awareness was unaffected. All doses (Supplementary Methods). Roger comparison data depict means. Error bars represent s.e.m. N indicates number of participants. See again demonstrated heart-rate increases that Supplementary Figures 1 and 2 for additional results.
Heart-rate change (bpm) Heart-rate change (bpm) Heart-rate change (bpm)

were identical to healthy comparison participants (Supplementary Fig.2). However, under anesthetic, he no longer reported any changes in cardiac sensation (Fig. 3d). Verbal responses further suggested that Roger failed to experience any qualitative changes in awareness (Supplementary Table 3). On the other hand, sensation in healthy comparison participants was unaffected by anesthetic. Quantitative sensory testing demonstrated a satisfactory anesthetic effect in all participants. Taken together, these results support the hypothesis that both neural structures innervating the skin, presumably involving primary and secondary somatosensory cortices, and the network of regions damaged in Roger, including the insula and ACC, independently mediate the ability to feel the heartbeat. These results also suggest that the insula is not the sole necessary substrate for interoceptive awareness. Our results represent, to the best of our knowledge, the first empirical demonstration of this concept7,9,12,13. Although it would have served to further test these hypotheses, it was not possible to include a patient with complete bilateral somatosensory cortex damage. This type of damage is highly implausible and we have never encountered such an individual. The precise contribution of the two pathways remains to be determined, not only for their link to interoceptive awareness, but also in regard to their involvement in the experience of emotion5,7,9,14,15. It is possible that each pathway contributes to different aspects of interoception. For example, the observed delay in Rogers ratings is compatible with a role for the insula in the on-line, instantaneous

0 50

0 0. .1 2 05 0. .5 7 1.5 0 2. 0 4. 0

10

15

20

25

30

35

40

Heart-rate change (bpm)

Heart-rate change (bpm)

0 50

10

15

20

25

30

35

40

nature neuroscience VOLUME 12 | NUMBER 12 | december 2009

1495

45

45

b r i e f c o m m u n i c at i o n s
representation of the state of the body in time6. It is important to note that, beyond the insula, Roger also has bilateral damage to the ACC, orbitofrontal cortices, basal forebrain, hippocampus, amygdala and temporal poles11. Although the insula is typically considered to be the structure relevant for visceral sensation, the ACC is thought to have a visceral motor role and is therefore not expected to be critical for awareness2,6. The remaining structures are less often implicated in interoceptive awareness. However, we cannot rule out that the damage incurred to these regions affected our results. Our results challenge classic definitions of what constitutes interoception, validate functional neuroimaging findings that impli cate both brain regions (insula and ACC) and somatosensory cortices in interoceptive awareness14, and demonstrate the set of pathways that enable the core human experience of the cardiovascular state of the body. Our findings provide empirical support for a comprehensive redefinition of interoception involving afferent information that arises from anywhere and everywhere within the body14, including through the skin via pathways that are usually considered to support exteroception. Such redefinition focuses on the source of stimulation in the body and not on the intrinsic nature of the sensory pathway5,7,9,14.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank J. Bruss for figure assistance, S. Schubert and M. Bosch for isoproterenol administration, C. Sandesara, E. St. Louis, B. Olshansky and H. Shim for medical supervision, T. Grabowski for neurological expertise and J. Martins for safety monitoring. This work was supported by the US National Center for Complementary & Alternative Medicine (F31-AT003061), the US National Institute on Drug Abuse (R01-DA022549) and the US National Center for Research Resources, General Clinical Research Center Program (M01-RR-59). AUTHOR CONTRIBUTIONS S.S.K. and D.R. designed the study with help from J.S.F. S.S.K. collected the data. S.S.K. and D.R. analyzed the data. S.S.K. and D.R. wrote the paper with the help of J.S.F. and D.T.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
1. Cameron, O.G. & Minoshima, S. Psychosom. Med. 64, 851861 (2002). 2. Critchley, H.D., Wiens, S., Rotshtein, P., Ohman, A. & Dolan, R.J. Nat. Neurosci. 7, 189195 (2004). 3. Pollatos, O., Schandry, R., Auer, D.P. & Kaufmann, C. Brain Res. 1141, 178187 (2007). 4. Van Oudenhove, L., Demyttenaere, K., Tack, J. & Aziz, Q. Best Pract. Res. Clin. Gastroenterol. 18, 663680 (2004). 5. Craig, A.D. Nat. Rev. Neurosci. 3, 655666 (2002). 6. Craig, A.D. Nat. Rev. Neurosci. 10, 5970 (2009). 7. Rudrauf, D. et al. Int. J. Psychophysiol. 72, 1323 (2009). 8. Khalsa, S.S., Rudrauf, D., Sandesara, C., Olshansky, B. & Tranel, D. Int. J. Psychophysiol. 72, 3445 (2009). 9. Khalsa, S., Rudrauf, D. & Tranel, D. Psychophysiology published online, doi:10.1111/ j.1469-8986.2009.00859.x (6 July 2009). 10. Bechara, A. & Naqvi, N. Nat. Neurosci. 7, 102103 (2004). 11. Feinstein, J.S. et al. J. Clin. Exp. Neuropsychol. published online, doi:10.1080/ 13803390903066873 (17 September 2009). 12. Dworkin, B.R. Handbook of Psychophysiology (eds Cacciopo, J.T., Tassinary, L.G. & Bernston, G.G.) 482506 (Cambridge University Press, Cambridge, 2007). 13. Jones, G.E. Advances in Psychophysiology Vol 5 (eds Jennings, J.R. & Ackels, P.K.) (Jessica Kingsley Publishers, London, 1994). 14. Cameron, O.G. Psychosom. Med. 63, 697710 (2001). 15. Damasio, A.R. The Feeling of What Happens: Body and Emotion in the Making of Consciousness (Harcourt Brace, New York, 1999).

2009 Nature America, Inc. All rights reserved.

1496

VOLUME 12 | NUMBER 12 | december 2009 nature neuroscience

a r t ic l e s

A genetic pathway composed of Sox14 and Mical governs severing of dendrites during pruning
Daniel Kirilly1,7, Ying Gu1,2,7, Yafen Huang1, Zhuhao Wu3, Arash Bashirullah4, Boon Chuan Low2, Alex L Kolodkin3, Hongyan Wang5,6 & Fengwei Yu1,2,5
Pruning that selectively eliminates neuronal processes is crucial for the refinement of neural circuits during development. In Drosophila, the class IV dendritic arborization neuron (ddaC) undergoes pruning to remove its larval dendrites during metamorphosis. We identified Sox14 as a transcription factor that was necessary and sufficient to mediate dendrite severing during pruning in response to ecdysone signaling. We found that Sox14 mediated dendrite pruning by directly regulating the expression of the target gene mical. mical encodes a large cytosolic protein with multiple domains that are known to associate with cytoskeletal components. mical mutants had marked severing defects during dendrite pruning that were similar to those of sox14 mutants. Overexpression of Mical could significantly rescue pruning defects in sox14 mutants, suggesting that Mical is a major downstream target of Sox14 during pruning. Thus, our findings indicate that a previously unknown pathway composed of Sox14 and its cytoskeletal target Mical governs dendrite severing. The selective removal of exuberant or inaccurate neuronal processes without causing neuronal death, referred to as pruning, is crucial for the refinement of neural circuits in the developing nervous system1,2. Pruning shares several features with neuronal degeneration in response to injury or disease1. In injured neurons, axons distal to the lesion site rapidly degenerate, a process known as Wallerian degeneration1,35. In neurodegenerative disorders, breakage and degeneration of neuronal processes derived from the diseased neurons often occur before neuronal death6,7. In Drosophila, larval-born neurons undergo extensive remodeling or apoptosis to form the adult nervous system during metamorphosis8. The remodeled neurons survive, but prune their neuronal processes/connections before eclosion, which later regrow to become part of the adult nervous system 8. These pruning events occur primarily in the first 24 h of metamorphosis in the CNS and peripheral nervous systems (PNS)911. In the CNS, the mushroom body neurons, olfactory projection neurons and thoracic ventral neurons remodel their larval dendrites/axons to form adult connectivities5,918. In the PNS, the majority of larval neurons, including class II and III dendritic arborization sensory neurons, are eliminated via apoptosis. However, certain class I and IV dendritic arborization neurons survive to undergo large-scale dendrite-specific pruning with minimal morphological changes in their axon termini19,20. The class IV dendritic arborization neuron ddaC undergoes a stereo typed pruning process that is initiated by the severing of proximal dendrites, followed by rapid fragmentation of severed dendrites and clearance of cellular debris via phagocytosis. During dendrite pruning, severing events consistently occur at the proximal regions
1Temasek

2009 Nature America, Inc. All rights reserved.

of dendrites. Initial signs of dendrite severing include blebbing and proximal thinning of dendrites, involving the depolymerization of microtubule and actin cytoskeletons 19,21. The steroid molting hormone 20-hydroxyecdysone (ecdysone), a master regulator that controls body-plan changes in insects, regulates both these pruning events in neurons undergoing remodeling and apoptosis of select PNS neurons during metamorphosis19,20. The ubiquitin-proteasome system and Dronc caspase act to execute the pruning process, along with Ik2 kinase and the microtubule-severing factor Katanin-60L1 (refs. 2123). How ecdysone regulates a transcriptional hierarchy to mediate dendrite pruning, however, is unknown. We found that Sox14 serves as an important regulator of dendrite severing in ddaC neurons. We found that Sox14, a critical target of ecdysone signaling, was both necessary and sufficient to induce dendrite severing during pruning. Sox14 mediates dendrite severing by promoting the expression of a direct target, mical, which was also required for dendrite severing. The Mical protein consists of multiple domains that are known to interact with actin and other cytoskeletal proteins and potentially mediates cytoskeletal alterations during dendrite pruning. Thus, our findings reveal a genetic pathway composed of the transcription factor Sox14 and its downstream target Mical that mediates dendrite severing in class IV ddaC neurons in response to ecdysone. RESULTS The class I and IV dendritic arborization neurons in the PNS survive and undergo a stereotyped pruning process in each abdominal hemisegment

Life Sciences Laboratory and the 2Department of Biological Sciences, National University of Singapore, Singapore. 3Solomon H. Snyder Department of Neuroscience, Howard Hughes Medical Institute, The Johns Hopkins School of Medicine, Baltimore, Maryland, USA. 4Division of Pharmaceutical Sciences, University of Wisconsin, Madison, Wisconsin, USA. 5Neuroscience and Behavioral Disorder Program, DukeNational University of Singapore Graduate Medical School Singapore, Singapore. 6Department of Physiology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore. 7These authors contributed equally to this work. Correspondence should be addressed to F.Y. (fengwei@tll.org.sg). Received 30 June; accepted 4 September; published online 1 November 2009; doi:10.1038/nn.2415

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1497

a r t ic l e s a b

P{Mae-UAS.6.11} LA00606 60A16

1 kb

sox1413 White pupa 512 h APF Severing 1012 h APF Fragmentation 1618 h APF Clearance >24 h APF Regrowth sox1415

1,964 bp 1,838 bp

c
White prepupae

2009 Nature America, Inc. All rights reserved.

18 h APF

Wild type

sox14 RNAi

sox1413/ sox1413

sox1413/ Df(2R)BSC136

sox1415/ MARCM

sox1413/ sox1415; O/E sox14 White prepupae 18 h APF

30 Figure 1 Sox14 is required for dendrite severing of ddaC neurons during 7 21 15 24 25 23 7 10 dendrite pruning. (a) A schematic representation of dendrite pruning events, 11 20 13 15 9 including severing, fragmentation, clearance and regrowth of ddaC dendrites in the Drosophila PNS. The box indicates the estimated area where we measured 15 dendrite severing from the soma for quantification. Soma and axon is depicted in 16 10 red, while dendrites are in black. (b) A schematic diagram of the sox14 gene and deleted regions of the sox14 mutants. The P-element insertion LA00606, which 24 5 was used to generate two excision mutants (sox14D13 and sox14D15), is located upstream of the transcriptional start site. Both sox14 alleles lack the entire first 13 0 exon. The start site of the sox14 open reading frame is shown in black, indicated by an arrow. (ch) Live confocal images of ddaC neurons, visualized by the expression of ppk-Gal4driven UASmCD8-GFP. Red arrowheads point to the ddaC soma. At the beginning of the metamorphosis, during white prepupae formation, sox14 mutant ddaC neurons (dg) had normal dendrite arbor morphology, similar to that of wild-type ddaC neurons ( c). At 18 h APF, dendrite severing of ddaC neurons was complete in wild-type pupae ( c). In contrast, sox14 RNAi (d), sox14 zygotic mutants (e), hemizygote sox14/Df (2R)BSC136 (f) or sox14 MARCM clones (g) showed dendrite pruning defects. (h) The severing defects in sox14 mutant ddaC neurons could be rescued by overexpression of Sox14. Open arrowhead marks a severed dendrite. O/E, overexpressed. ( i) Quantitative analysis of the average number of primary and secondary dendrites attached to the soma of wild-type and mutant ddaC neurons at the white prepupal stage and 18 h APF. The number of samples (n) in each group is shown above the bars. Error bars represent s.d. Dorsal is up in all images. Scale bar represents 50 m. Number of primary and secondary dendrites attached to soma

during the early phase of metamorphosis19,20. We focused our attention on the class IV ddaC neuron that is located in the dorsal cluster and can be labeled by the expression of membrane-bound mCD8green fluorescent protein (GFP) using the class IVspecific pickpocket (ppk)Gal4 driver (ppk-Gal4) (Fig. 1)24. ddaC neurons started to show signs of dendritic instability at approximately 5 h after puparium formation (5 h APF), when small blebs formed along the proximal branches of dendrites (Fig. 1a and Supplementary Movie 1). These blebs migrated dynamically along the dendrites, resulting in thinning and subsequent physical breakage of the dendrites from the soma (Supplementary Movie1), namely dendrite severing. The severing of proximal dendrites was apparent at 8 h APF and completed by 12 h APF (Fig. 1a, Supplementary Fig. 1 and Supplementary Movie 1). Over the next 6 h, severed dendrites underwent rapid fragmentation and clearance by phagocytes (Fig. 1a and Supplementary Fig. 1). At 18 h APF, no larval dendrites were observed in the vicinity of the ddaC soma, whereas the soma and the axon remained intact (Fig. 1a and Supplementary Fig. 1). Any defects in dendrite-severing events can be easily recognized by the presence of dendrites that remain attached to the soma at 18 h APF.
1498

The transcription factor Sox14 mediates dendrite severing Ecdysone is an important regulator of neuronal remodeling events in mushroom body neurons of the CNS11 and dendritic arborization neurons of the PNS19,20. Ecdysone binds to a nuclear receptor heterodimer consisting of ecdysone receptor and Ultraspiracle (EcR/Usp), resulting in the activation of target gene expression during Drosophila metamorphosis25. The B1 isoform of the ecdysone receptor (EcR-B1) is highly expressed in remodeling neurons of the CNS11,18 and the PNS20. EcR-B1 expression is activated by TGF- signaling and the cohesin complex during the larval-pupal transition2628. Inactivation of EcR/Usp functions that blocks ecdysone signaling not only inhibits neuronal pruning in both systems, but also prevents cell death in apoptotic dendritic arborization neurons11,19,20. To identify previously unknown proteins that mediate dendrite pruning downstream of EcR/Usp, we carried out an RNA interference (RNAi) screen and tested genes that, from previous microarray analyses, have the potential to respond to ecdysone signaling2931. We isolated sox14, which, when attenuated by RNAi, resulted in severe dendrite pruning defects in all ddaC neurons. sox14 RNAi
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

N Ai so so x14 x1 1 4 1 /3 so 3 s x1 D ox1 4 13 f(2 4 / 1 R )B 5 s SC D ox 13 f(2 14 6 13 / R )o r so -BR x1 1 4 1 1/ M AR s 5 so CM ox1 4 x1 c 4 1 lon 15 3 e /s s O ox /E 1 so 4 15 x1 4

ty

pe

ild

UA S

-s

ox 1

a r t ic l e s a
EcR-B1 eL3

wL3

White prepupae

5 h APF

e
Sox14

eL3

wL3

g White prepupae h

5 h APF

i
2009 Nature America, Inc. All rights reserved.

eL3

wL3

k White prepupae l

5 h APF

Mical

Figure 2 Mical localization is dependent on Sox14 and EcR/Usp in ddaC neurons. (al) Expression of EcR-B1, Sox14 and Mical at the eL3, wL3, white prepupal and 5 h APF stages. ddaC neurons are marked by dashed lines. Their location was determined by ppk-Gal4driven mCD8-GFP expression. The expression of EcR-B1 ( ad), Sox14 (eh) and Mical (il) in ddaC neurons were compared at various stages. Mical was concentrated in the soma, but it was also present in the dendrites (arrowheads) and axons (arrow) (k). For each individual antibody, immunofluorescence staining at various time points was processed in the same tube and images were taken at the same gain. ( mt) Sox14 and Mical immunostaining in various mutants. Neurons overexpressing RNAi or dominant-negative constructs were labeled by the expression of mCD8-GFP (green) driven by the Gal4109(2)80 (mo,q) or ppk-Gal4 driver (r,s). Insets show the expression of the transgenes, indicated by GFP. ddaC neurons are marked by dashed lines and other dendritic arborization neurons are shown by solid lines. Sox14 expression was regulated by ecdysone signaling (mp). Sox14 expression was abolished in all of the dendritic arborization neurons when ecdysone signaling was blocked by EcR RNAi or usp RNAi (m,n). Sox14 can also be effectively downregulated in all sox14 RNAi dendritic arborization neurons (o). However, Sox14 expression was unaffected in mical15256 mutant (p). (qt) Mical expression was dependent on the ecdysone receptor complex and Sox14. Mical expression was strongly reduced in EcRDN (q), usp RNAi (r) or sox14 RNAi (s) ddaC neurons and was absent from mical15256 mutant (t). WP, white prepupae. Scale bar represents 20 m.

m
Sox14

WP

WP

WP

WP

q
Mical

EcR RNAi

usp RNAi

sox14 RNAi

mical15256/ mical15256

WP O/E EcR DN

WP usp RNAi

WP sox14 RNAi

WP mical15256/ mical15256

did not affect the overall morphology of dendritic arbors (n = 15; Fig. 1d and Supplementary Fig. 2), as compared to the wild-type arbor (n = 23; Fig. 1c and Supplementary Fig. 2). Notably, at 18 h APF, approximately nine primary and secondary dendrites were still attached to the soma (100%, n = 16; Fig. 1d,i), whereas no dendrites were observed in the wild-type ddaC neurons (100%, n = 13; Fig. 1c,i), suggesting that the severing of proximal dendrites from ddaC neurons requires Sox14. Sox14 is a high-mobility group (HMG)box transcription factor, the sole Drosophila Sox C group protein and belongs to the evolutionarily conserved Sox family. None of the other seven Sox family genes in Drosophila appears to be involved in dendrite pruning, as knockdown of these genes by RNAi did not induce an obvious pruning defect in ddaC neurons (data not shown). To further verify the role of sox14 in ddaC dendrite pruning, we generated sox14 mutants by mobilizing a P element, P{MaeUAS.6.11}LA00606, inserted upstream of the sox14 gene (Fig. 1b). We recovered two excisions, sox14D13 and sox14D15, in which the entire first exon was removed (Fig. 1b). The homozygous sox14 mutant flies survived until the late pupal stages, allowing us to observe dendrite
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

pruning defects in the early pupal stage. The defects in dendrite severing that we observed in both of the sox14 homozygous mutants were more pronounced than those induced by sox14 RNAi at 18 h APF. In these sox14 mutants, the majority of the primary and secondary dendrites remained attached to the ddaC soma at 18 h APF (100%, n = 13; Fig. 1e,i and Supplementary Movie 2) and 24 h APF (Supplementary Fig. 1), compared with those present in the white prepupae (Fig. 1e). Transheterozygotes between sox14D13 and sox14D15, or hemizygotes between sox14D13 or sox14D15 and either of the small deficiencies removing the sox14 gene, showed indistinguishable dendrite-severing defects compared to either of the homozygotes (Fig. 1f,i). In addition, we were unable to detect Sox14 protein via immunofluorescence analysis in sox14D13 (Supplementary Fig. 3) or sox14D15 mutants (data not shown). Thus, both sox14D13 and sox14D15 are either strong hypomorphic or null alleles, and we refer to them as sox14 mutants. Using the mosaic analysis with a repressible cell marker (MARCM) technique32, we generated homozygous clones for sox14 mutants. The same sox14 ddaC clones were first examined in white prepupae and subsequently at 18 h APF for potential pruning defects (Fig. 1g). All of the sox14 ddaC clones had a strong pruning defect at 18 h APF (100%, n = 7; Fig. 1g,i), suggesting that Sox14 mediates dendrite pruning in a cell-autonomous fashion. Reintroduction of Sox14 into sox14 mutant ddaC neurons largely rescued dendrite-severing defects ( n = 24; Fig. 1h,i), further confirming that the dendrite pruning defects associated with sox14 mutants are a result of the loss of sox14 function. Wild-type dorsal dendritic arborization neurons such as ddaA, ddaB and ddaF were eliminated via apoptosis at the early pupal stage (Supplementary Fig. 4). Knockdown of sox14 by RNAi using the dendritic arborization neuron driver Gal4109(2)80 inhibited neuronal death in the majority of ddaF (86%, n = 15), ddaA and ddaB neurons at 18 h APF (Supplementary Fig. 4). Therefore, loss of sox14 function causes EcR-like phenotypes with regard to dendrite pruning and neuronal apoptosis during early metamorphosis. sox14 transcripts are ubiquitously expressed during embryogenesis and are enriched in the third instar larvae and early pupae 33,34. We generated a specific antibody to Sox14 (see Online Methods) and used it to examine Sox14 expression in dendritic arborization neurons, comparing it with EcR-B1 expression (Fig. 2). EcR-B1 expression
1499

a r t ic l e s a
Df(3R)ED5438 Df(3R)Exel6155 Df(3R)swp2
MICAL

mical

c
White prepupae

85E1 85E3 85E5 85E7 85E9 85F1 85F3 85F5 85F7 85F9 85F11

15256 (Q535*)

FM

CH

LIM

PRD Coiled coil PDZ binding

18 h APF

4,723 amino acids (long form) Antibody epitope

Wild type

mical15256/ mical15256

mical15256 Df(3R)swp2 MICAL Number of primary and secondary dendrites attached to soma

mical RNAi

mical15256 MARCM

2009 Nature America, Inc. All rights reserved.

Figure 3 Mical is essential for dendrite severing of ddaC neurons during metamorphosis. White prepupae 18 h APF (a) A diagram of the mical gene in relation to the deficiencies used. (b) Sequencing the 3 22 20 24 10 genes in the 85F1-F8 region in l(3)15256 revealed a nonsense mutation in the mical gene. 20 The nonsense mutation in the mical15256 allele leads to a truncated protein with 534 amino 15 12 18 acids. Mical is a large protein containing multiple domains. The antibody that we used targets 20 10 12 the C-terminal region of the protein (red line). FM, flavoprotein mono-oxygenase domain; 5 CH, calponin homology domain. (cg) Live confocal images of ddaC neurons in wild types 16 0 and various mical mutants, labeled by the expression of mCD8-GFP driven by ppk-Gal4. Red arrowheads indicate the ddaC soma. At the white prepupal stage, the dendrite arbors of micaldefective ddaC neurons were not different from their wild-type counterparts. Wild-type neurons completely pruned their dendrites by 18 h APF (c), whereas ddaC neurons lacking mical failed to sever dendrites and proximal dendrites remained attached to the soma ( d,e). mical promoted dendrite pruning in a cell-autonomous manner, as shown by mical RNAi (f) and MARCM analysis (g). (h) Quantitative analysis of the number of primary and secondary dendrites attached to the soma of wild-type and mutant ddaC neurons at the white prepupal and 18 h APF stages. The number of samples ( n) in each group is indicated above the bars. Error bars represent s.d. Scale bar represents 50 m.
pe Ai
6 6 25 25

ty

15

15

AL

al

m )s ica w p2 M l

15

al

ic

ic

al

ic

f(3

was low in ddaC neurons at the early third-instar larval (eL3) stage, reached its peak at the wandering third-instar larval (wL3) and white prepupal stages and declined at 5 h APF (Fig. 2ad; also seen in a previous study20). In contrast, Sox14 expression was undetectable at the eL3 and wL3 stages, but increased markedly at the white prepupal stage, followed by a slight decrease after the white prepupal stage (Fig. 2eh). The intensity of Sox14 staining in the nuclei of ddaC neurons was slightly weaker than in those of the apoptotic ddaF neurons (Fig. 2g). Thus, Sox14 is expressed in both apoptotic and remodeling dendritic arborization neurons in white prepupae and prepupae, lagging slightly behind the upregulation of EcR-B1 expression. Mical, a cytosolic protein, promotes dendrite severing To search for possible downstream targets of Sox14 involved in dendrite pruning, we screened a collection of ethyl-methylsulfonate mutagenized late-pupal lethal mutations on the third chromosome (Fig. 3)35. We isolated a mutant, l(3)15256, that showed a failure of dendrite severing in ddaC neurons by 18 h APF (Fig. 3d, Supplementary Fig. 1 and Supplementary Movie 3). l(3)15256 ddaC neurons retained an average of 7.4 primary and secondary dendrites attached to the soma at 18 h APF (100%, n = 20; Fig. 3d,h) and 3 dendrites at 24 h APF (100%, n = 13; Supplementary Fig. 1), indicating that the mutant neurons had a strong severing defect. The overall larval dendrite morphology of ddaC neurons in the wL3 larvae of l(3)15256 zygotic mutant was indistinguishable from that of wildtype ddaC neurons, as judged by the number of terminal dendrites (Supplementary Fig. 2). The l(3)15256 mutation was mapped to the cytological region 85F1-F8 (Fig. 3a) and further sequence analysis in this region identified a nonsense mutation in the mical gene that predicted the production of a truncated protein with only the first 534 amino acids (Fig. 3b). Mical is a large, cytosolic, multidomain
1500

protein containing a flavoprotein mono-oxygenase domain at the N terminus, a calponin homology domain and LIM domain at the middle region, and a proline-rich domain (PRD), coiled-coil domain and PDZ-binding motif at the C terminus (Fig. 3b)36. The Mical protein was undetectable in l(3)15256 homozygous white prepupae (Fig. 2t), as compared with wild type (Fig. 2k). Hemizygotes of l(3)15256 with a small deficiency, Df(3R)swp2MICAL, that removes the mical gene36 or four other previously identified mical mutants37 also caused equally strong dendrite-severing defects (Fig. 3e,h and Supplementary Fig. 5). Therefore, l(3)15256 is probably a null allele of mical and we refer to it as mical15256. Moreover, the dendrite-pruning defects associated with mical mutants can be fully rescued by overexpressing full-length Mical (MicalFL), but not by expressing an N-terminal portion of Mical (MicalN-ter) in ddaC neurons (Supplementary Fig. 6). These results suggest that the dendrite-pruning defects observed in mical15256 mutant neurons are attributable to a loss of mical function. Thus, mical is important in dendrite severing of ddaC neurons. mical RNAi (100%, n = 12; Fig. 3f,h) and mical15256 MARCM clones (100%, n = 12; Fig. 3g,h) also showed a severing defect in ddaC neurons, similar to mical zygotic mutants. These data establish a cell-autonomous role for mical during dendrite pruning. Moreover, similar to observations in EcRDN (EcR-B1-W650A) or sox14 mutants, mical dendritic arborization neurons ddaD and ddaE failed to prune their dendritic processes by 24 h APF (Supplementary Fig. 7). Surprisingly, unlike pruning defects observed in sox14 mutants, apoptotic neurons such as ddaF neurons were eliminated at 18 h APF in mical zygotic mutants (Supplementary Fig. 4), mical MARCM clones or mical RNAi (data not shown). Thus, Mical is specifically required for dendrite severing of class I and IV dendritic arborization neurons, but is dispensable for neuronal apoptosis in other apoptotic dendritic arborization neurons during early metamorphosis.
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

ic

al

15

25

AR

ild

25

IC

a r t ic l e s a
White prepupae

6 h APF

Wild type

O/E EcR-B1 ppk-Gal4

O/E mical

FL

O/E sox14

f
2009 Nature America, Inc. All rights reserved.

ppk-Gal4

ppk-Gal4

ppk-Gal4

O/E sox14; mical15256; mical15256 Gal4221

Mical wL3 Wild type

wL3 O/E EcR-B1

wL3 O/E mical FL Number of primary and secondary dendrites attached to soma

wL3 O/E sox14

wL3 O/E sox14 11 White prepupae 6 h APF

Figure 4 Sox14 overexpression causes precocious dendrite pruning in ddaC 9 13 9 neurons in concert with upregulation of Mical. (ae) Live confocal images of 20 6 12 mCD8-GFPlabeled ddaC neurons at the white prepupal and 6 h APF stages. Red 12 12 12 arrowheads indicate the ddaC soma. In wild type, EcR-B1 or Mical-overexpressing 15 ddaC neurons, dendrites appeared to be intact at 6 h APF ( ac), whereas Sox14overexpressing ddaC neurons (red arrowhead) pruned the majority of their 21 10 dendrites (open arrowhead, d). The dendritic arbor of 6 h APF mical mutant ddaC neurons overexpressing Sox14 was similar to that observed at the white prepupal 5 stage (e). (fj) Mical expression in dendritic arborization neurons at the wL3 stage. ddaC neurons are marked by dashed lines. Class I dendritic arborization neurons 0 ddaD/E are labeled by solid lines (j). Neurons were marked on the basis of the GFP signals shown in the insets. In wild-type wL3 larvae ( f) or on overexpression of EcR-B1 (g), ddaC neurons expressed low levels of Mical. (h) Mical overexpression could be detected by the elevated Mical signals in ddaC neurons. Overexpression of Sox14 in either ddaC neurons by ppk-Gal4 (arrow, i) or in class I neurons ddaD/ E by Gal4221 (arrows, j) induced high levels of Mical (in red) and Sox14 (in blue, arrows) expression. An arrowhead indicates a ddaE neuron in i. (k) Quantification of the number of primary and secondary dendrites attached to the soma of various genotypes at the white prepupal and 6 h APF stages. The number of samples ( n) in each group is indicated above the bars. Error bars represent s.d. Scale bars represent 50 and 20 m in a and f, respectively.
pe 1
FL

Upregulation of Mical requires EcR, usp and sox14 We next determined the expression pattern of Mical and its requirement with respect to EcR, usp and sox14 using a previously characterized antibody to Mical36. We further verified the specificity of this antibody, as it did not stain mical mutants ( Fig. 2t) or mical RNAitreated ddaC neurons (data not shown), but staining with it did lead to markedly enhanced signals when Mical was overexpressed in mical mutant ddaC neurons ( Supplementary Fig.6). Mical was expressed at low levels in the eL3 and wL3 larvae ( Fig. 2i , j). However, we observed substantially elevated Mical expression at the white prepupal and 5 h APF stages in all of the dendritic arborization neurons (Fig. 2k,l). In dendritic arborization
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

neurons, Mical was localized in both dendrites (arrowheads) and axons (arrow), and in the cytoplasm in soma (Fig. 2k). Upregulation of Mical protein temporally coincided with Sox14 expression (Fig. 2el) and lagged slightly behind the EcR-B1 upregulation (Fig. 2ad). To ascertain whether Sox14 and Mical are downstream targets of EcR-B1 during dendrite pruning, we first examined whether Sox14 and Mical expression is affected in ddaC neurons when ecdysone signaling is blocked. EcR RNAi effectively downregulated EcR-B1 expression (Supplementary Fig. 8), as compared with wild type (Supplementary Fig. 8), and caused strong ddaC pruning defects at 18 h APF (data not shown) that were similar to those in usp RNAitreated or EcRDN ddaC neurons (Supplementary Fig. 1). We
1501

/E

4 so x1 4; m m ical 1 5 ic al 15 256 2

-B

so x1

ty

ild

Ec

ic

al

/E

/E

/E

56

a r t ic l e s
Figure 5 Mical acts downstream of EcR-B1 and Sox14 during severing. (ad) Top panels, Sox14 overexpression suppressed EcRDN pruning defects. Wild-type ddaC neurons pruned their dendrites by 18 h APF (a), whereas severing of dendrites was inhibited by overexpression of EcRDN (b). The pruning defect caused by EcRDN was rescued by coexpressing either wild-type EcR-B1 (c) or Sox14 (d). Bottom panels, Mical immunostaining at the white prepupal stage in various genotypes. High-level expression of Mical in wild type (a) at the white prepupal stage could be eliminated in EcRDN ddaC neurons (b). Mical expression was restored on coexpression of wild-type EcR-B1 (c) or Sox14 (d) in the EcRDN ddaC neurons. ddaC neurons are marked by dashed lines on the basis of the GFP signals shown in the insets. WP, white prepupae. (eh) Mical functioned downstream of EcR-B1 and Sox14 during dendrite pruning. MicalFL (f), but not the nonfunctional MicalN-ter (e), ameliorated the pruning defect resulting from EcRDN overexpression. Overexpression of MicalFL (h) significantly rescued (P < 0.001) the pruning defect of sox14 mutant ddaC neurons compared with the controls overexpressing MicalN-ter (g). Red arrowheads indicate ddaC soma. (ij) Quantification of the number of primary and secondary dendrites attached to the soma (i) and the number of dendrite termini of attached dendrites (j) of wild-type and mutant ddaC neurons during metamorphosis. In i and j, the number of samples (n) in each group is indicated above each bar and error bars represent s.d. ** P < 0.001. Scale bars represent 50 m (a, top panel) and 20 m (a, bottom panel).

18 h APF

18 h APF

18 h APF

18 h APF

Mical WP Wild type

WP O/E EcR
DN

WP O/E EcR DN O/E EcR-B1

WP O/E EcRDN O/E sox14

2009 Nature America, Inc. All rights reserved.

18 h APF O/E EcR DN O/E mical N-ter 18 h APF

18 h APF O/E EcR O/E mical FL


DN

18 h APF sox1413/ sox1413; O/E mical N-ter 18 h APF 120 100 80 60 40 20 0 12 19 11

18 h APF sox1413/ sox1413; O/E mical FL

i
Number of primary and secondary dendrites attached to soma

**
20 15 10 5 0 12 19
N D

11

20 23 16

Number of dendrite termini

**
14

**
14 23

20

**

16

16

16

N O 1 / ;O Es 4 1 /E ox 3 E /E 14 so / s cR D m N x1 ox ic 4 1 14 ; O al Nte 13 3 /E r /s ; ox O/ mic al F 14 E L 13 mi ; O cal N -te /E r m ic al F L

pe

used these strains to block ecdysone signaling in all of the dendritic arborization neurons (using the Gal4109(2)80 driver) or in only ddaC neurons (using the ppk-Gal4 driver) and analyzed Sox14 and Mical expression in mutant dendritic arborization neurons at the white prepupal stage. The inhibition of ecdysone signaling by EcR RNAi (n = 13; Fig. 2m), EcRDN (data not shown) or usp RNAi (n = 11; Fig. 2n) resulted in the depletion of Sox14 proteins in all of the dendritic arborization neurons. Likewise, Mical expression failed to increase at the white prepupal stage in EcRDN (100%, n = 18; Fig. 2q) and usp RNAitreated (100%, n = 17; Fig. 2r) dendritic arborization neurons, in contrast with wild-type white prepupae (Fig. 2k). We then examined whether Sox14 is required for Mical expression in ddaC neurons. Mical expression was no longer upregulated in sox14 RNAitreated (100%, n = 14; Fig. 2s) or sox14 mutant ddaC neurons (100%, n = 20; data not shown). On the other hand, EcR-B1 expression appeared to be normal in both sox14 (n = 6; Supplementary Fig. 8) and mical mutants (n = 9; Supplementary Fig. 8), and Sox14 levels in mical mutants remained the same as those in wild type (n = 11 mutants and 25 wild types; Fig. 2g,p). Taken together, Mical and Sox14 are downstream targets of EcR/Usp and Mical upregulation requires Sox14 in ddaC neurons during early metamorphosis. The temporal regulation of EcR-B1, Sox14 and Mical protein expression in mushroom body neurons was highly similar to that observed in dendritic arborization sensory neurons (Supplementary Fig. 9). High levels of sox14 and mical transcripts could be detected in mushroom body neurons at the white prepupal stage in an
Ec R /E Ec /E O /E O O

EcR-dependent fashion (Supplementary Fig. 10). Notably, sox14 RNAi caused downregulation of mical RNA in mushroom body neurons (Supplementary Fig. 10). Sox14 was able to directly bind to the mical regulatory regions in ecdysone-treated S2 cells (Supplementary Figs. 10 and 11). Furthermore, sox14 RNAitreated mutants exhibited severe axon-pruning defects in mushroom body neurons, whereas the axon branches were pruned normally in homozygous mical mutant by 24 h APF (Supplementary Fig. 10). Sox14 overexpression causes precocious severing Given that sox14 and mical are required for dendrite pruning in ddaC neurons, we next examined whether overexpression of either one is sufficient to cause precocious pruning (Fig. 4). At 6 h APF, the majority of primary and secondary dendrites were still attached to the soma in wild-type ddaC neurons (Fig. 4a), although the overall dendritic morphology was simple, presumably as a result of the gradual retraction of high-order dendrites. We therefore chose this time point for the detection of potential precocious dendrite severing when EcR-B1, Sox14 or Mical were overexpressed in ddaC neurons. Sox14, when overexpressed in ddaC neurons, substantially accelerated the progression of dendrite pruning, including severing and debris clearance (Fig. 4d). By 6 h APF, fewer than five primary and secondary dendrites remained attached to the soma in Sox14-overexpressing ddaC neurons
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

1502

O /E type ; O Ec RD / N /E R D EE N Ec cR ; RD so O B N / O x1 ;O Es 1 4 1 /E ox / E 3 14 so / s cR D E m x1 ox N ic 4 1 14 ; O al Nte 3 13 /E r /s ox ; O mic 14 /E al F L 13 m ; O ica N /E l -ter m ic al F L


D N

ild

-B

ty

Ec

Ec

/E

/E

;O

;O

ild /E O

Ec

Ec

so x1

/E

Ec

a r t ic l e s
(Fig. 4d,k), compared with approximately 14 attached dendrites in the wild-type ddaC neurons (Fig. 4a,k). Notably, almost half of the ddaC neurons retained three or fewer attached dendrites, whereas elimination of the majority of disconnected dendritic arbors was observed (40%, n = 20; Fig. 4d). Although overexpression of Sox14 reduced the density of high-order dendrites, it did not obviously affect the number of primary and secondary dendrites in white prepupal ddaC neurons (Fig. 4a,d). Thus, Sox14 is an important transcription factor that is necessary and sufficient to promote dendrite severing in ddaC neurons. Continuous overexpression of Sox14 from the late embryonic stages to the early pupal stages using the ppk-Gal4 driver did not cause neuronal apoptosis, as ddaC neurons were still present by 24 h APF (data not shown). Does Sox14 overexpression cause upregulation of Mical expression? In wild-type wL3 larvae, Mical was expressed at low levels in dendritic arborization neurons (Fig. 4f). Notably, overexpression of Sox14 led to a marked upregulation of Mical levels in more than half of the ddaC neurons (53% n = 32; Fig. 4i), as compared with neighboring dendritic arborization neurons lacking Sox14 overexpression (Fig. 4i). Elevated levels of Mical were also observed in Sox14-overexpressing ddaD/E neurons (42%, n = 12; Fig. 4j). We then removed the mical gene in Sox14-overexpressing ddaC neurons to examine whether loss of Mical suppresses Sox14-mediated precocious dendrite pruning. Notably, removal of mical in Sox14-overexpressing ddaC neurons resulted in retention of larval dendritic arbors; approximately 15 primary and secondary dendrites were still attached to the soma at 6 h APF (100%, n = 12; Fig. 4e,k) and some persisted until 18 h APF (data not shown). Overexpressing Sox14 in mical ddaC neurons also led to a morphological defect in dendritic arbors (Fig. 4e). Thus, precocious dendrite pruning induced by Sox14 overexpression requires Mical function. Overexpression of EcR-B1 had no effect on the normal progression of severing events (Fig. 4b). It was unable to induce the expression of Sox14 (data not shown) and Mical in the wL3 larvae (100%, n = 12; Fig. 4f,g), suggesting that EcR-B1 is not a rate-limiting factor during dendrite pruning of ddaC neurons. Overexpression of Mical did not promote precocious pruning in ddaC neurons (Fig. 4c), suggesting that Mical is an important, but not exclusive, downstream target of Sox14. Therefore, Mical is necessary, but not sufficient, to initiate the severing events during pruning. Mical acts downstream of EcR-B1 and Sox14 during severing To further explore whether Sox14 and Mical mediate dendrite pruning in response to ecdysone signaling, we examined whether overexpression of Sox14 or Mical can rescue the pruning defects caused by blocking ecdysone signaling (Fig. 5). Overexpression of EcRDN caused strong dendrite-pruning defects in ddaC neurons and reduced Mical protein expression (100%, n = 11; Fig. 5a,b). We first confirmed that the dendrite-pruning defects associated with EcRDN were a result of the inhibition of EcR function. Coexpression of wild-type EcR-B1 and EcRDN in ddaC neurons fully rescued pruning defects in ddaC neurons (100%, n = 19; Fig. 5c,i,j) and restored Mical (100%, n = 21; Fig. 5c) and Sox14 expression (100%, n = 14; data not shown). We then tested whether sox14 genetically interacts with EcR-B1 during dendrite pruning. Sox14 overexpression almost fully rescued the pruning defects at 18 h APF in EcRDN-expressing ddaC neurons (94%, n = 16; Fig. 5d,i,j). Moreover, Mical expression was also fully restored when Sox14 was overexpressed in EcRDN ddaC neurons (100%, n = 15; Fig. 5d). Thus, sox14 is an important target of ecdysone signaling during pruning; Sox14, after activation by EcR-B1, initiates the expression of its downstream targets, including Mical, and promotes dendrite pruning.
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

2009 Nature America, Inc. All rights reserved.

To investigate whether overexpression of Mical could rescue the pruning defects in EcRDN and sox14 mutant ddaC neurons, we over expressed MicalFL or nonfunctional MicalN-ter protein in EcRDN or sox14 mutant ddaC neurons. Notably, overexpression of MicalFL resulted in a partial, but significant, rescue of the pruning defects in EcRDN ddaC neurons (P < 0.001), as around 13 primary and secondary dendrites were attached to the soma and 55 terminal branches remained at 18 h APF (n = 23; Fig. 5f,i,j). In contrast, when a truncated version of Mical protein lacking the C-terminal PRD, coiled-coil and PDZ-binding domains was overexpressed in EcRDN ddaC neurons, 16 primary and secondary dendrites remained attached to the soma and 82 terminal branches persisted at 18 h APF (n = 14; Fig. 5e,i,j), similar to those observed in EcRDN ddaC neurons (n = 11; Fig. 5b,i,j). Overexpression of MicalFL led to even more significant rescue of dendrite pruning defects in sox14 mutant ddaC neurons (P < 0.001), as approximately 10 primary and secondary dendrites remained intact and 45 termini remained at 18 h APF (n = 16; Fig. 5h,i,j). In contrast, overexpression of nonfunctional MicalN-ter in sox14 mutants retained around 16 primary and secondary attached dendrites and 100 terminal branches, and failed to rescue dendrite pruning (n = 20; Fig. 5g,i,j). Our data suggest a genetic hierarchy composed of EcR, sox14 and mical that governs dendrite severing during early metamorphosis (Supplementary Fig. 12). DISCUSSION Here, we identified critical functions of Sox14 and its immediate target Mical in severing dendrites of dendritic arborization neurons during early metamorphosis. The proximal regions of dendrites remained attached to the soma and severing failed to occur when Sox14 or Mical was depleted. Consistent with these observations, the expression of both Sox14 and Mical were upregulated during early metamorphosis. Notably, overexpression of Sox14 was able to induce Mical expression and rescue dendrite pruning in EcRDN ddaC neurons, overriding the requirement for EcR-B1. Mical promotes dendrite pruning through its potential association with the cytoskeleton and regulation of cyto skeletal alterations (Supplementary Fig. 12). Sox14 mediates both dendrite pruning and apoptosis of dendritic arborization sensory neurons in the PNS. Dendrite pruning involving cellular fragmentation and phagocyte-dependent clearance shares features with apoptosis19,23. sox14 was recently reported to function as a potential pro-death gene and as being involved in the destruction of the midgut and salivary glands38. We found that Sox14 could also regulate apoptosis in certain dendritic arborization neurons. Loss of sox14 function inhibited apoptosis of ddaF neurons. In apoptotic ddaF neurons, Sox14 expression was often higher than that of remodeling ddaC neurons (Fig. 2g). It is possible that Sox14 functions in a dose-dependent manner to regulate both apoptosis and remodeling of dendritic arborization sensory neurons in the PNS. Supporting this idea, the pro-death gene reaper and the caspase gene dronc, which are involved in apoptosis and remodeling, were identified as potential targets of Sox14 (ref. 38). Consistently, overexpression of Sox14 is sufficient to induce cell death in cultured l(2)mbn cells38. Continuous overexpression of Sox14 in ddaC neurons did not cause neuronal apoptosis. It is likely that the pro-death effect of Sox14 may be effectively antagonized by protective machinery activated in ddaC neurons, such as upregulation of Drosophila inhibitor of apoptosis protein 1 (ref. 22). Sox group genes are known for their roles in regulating major developmental processes by activating gene transcription39. Our results provide several lines of evidence supporting the view that Mical is a downstream target of Sox14. First, similar to Sox14, Mical also
1503

a r t ic l e s
promotes dendrite severing in ddaC neurons. Second, Mical expres sion is dependent on Sox14 function, as Mical mRNA and protein are undetectable or strongly reduced in sox14 mutants. On the other hand, overexpression of Sox14 was sufficient to induce Mical expression in dendritic arborization neurons at a much earlier time point. Third, we found several potential binding sites for Sox14 in the regulatory region of the Mical gene. We also found that Sox14 specifically associated with some of these sites in a chromatin immunoprecipitation assay. Fourth, overexpression of Mical can significantly suppress the dendrite pruning defects in sox14 mutants. Fifth, the removal of Mical can largely suppress precocious dendrite pruning induced by Sox14 overexpression in ddaC neurons. How does Mical promote dendrite severing in ddaC neurons? Mical may regulate cytoskeletal alterations or rearrangements of ddaC dendrites via at least two possible mechanisms. Mical can potentially cross-link actin filaments into bundles and networks through its calponin homology domain, a well-known actin-binding domain. Consistent with this, actin and myosin filaments, although assembled properly, are misoriented and mislocalized in mical mutant muscles37. Alternatively, Mical may also regulate actin cytoskeletal dynamics through the interaction of its PRD domain with Cas (Crk-associated substrate) proteins40, which themselves regulate actin cytoskeleton organization and focal adhesion formation in non-neuronal cells41. Notably, the sole Drosophila homolog of Cas (DCAS), similar to Mical, is also enriched in the nervous system and genetically interacts with Mical to regulate axon guidance42. In response to ecdysone signaling, Mical is markedly upregulated and localizes to the dendrites of ddaC neurons. This may lead to rearrangements of dendritic actin and myofilaments that are more prone to depolymerization and disassembly in remodeling ddaC neurons. Currently, it is not known how Mical induces the severing events in dendrites but not in axons. Given the distinct local environments that axons and dendrites of ddaC neurons reside in, it is conceivable that dendritically localized Mical is activated to induce the rearrangements of dendritic actin cytoskeletons, thus inducing dendrite-severing events. Cytoskeletal rearrangements often occur as an initial sign of apoptosis43. Mical is also upregulated in apoptotic dendritic arborization neurons in response to ecdysone, but loss of mical function has no effect in neuronal apoptosis. Given the conservation of sox14 and mical in flies and mammals, it is conceivable that regulation of mical by sox14 might also be conserved and modulated by hormone responses in mammals. In response to steroid hormones, mammalian estrogen receptors, the homologs of fly EcR, are involved in mediating neurite growth and differentiation44, as well as synapse plasticity associated with learning and memory45. Thus, it will be of wide interest to investigate whether the EcR-Sox14-Mical pathway is evolutionarily conserved and involved in neuronal pruning and synaptic plasticity in mammals (Supplementary Results and Discussion). Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank H. Aberle, R. Barrio, Y.N. Jan, W.A. Johnson, the Bloomington Stock Center, Developmental Studies Hybridoma Bank (University of Iowa) and Vienna Drosophila RNAi Center for generously providing antibodies and fly stocks. We thank members of the Yu and Wang laboratories for stimulating discussions and Z.L. Ong for technical assistance. We thank W. Chia, S. Cohen, P. Rorth, S. Roy, J. Varghese and G. Feng for helpful discussions and for reading the manuscript. This work was supported by Temasek Life Sciences Laboratory (F.Y.), DukeNational University of Singapore (MOE2008-T2-1-048 and NRF-RF2009-02 to H.W.), and grants from the US National Institutes of Health and the National Institute of Neurological Disorders and Stroke (RO1NS35165) and the Howard Hughes Medical Institute (A.L.K.). D.K. is supported by the Singapore Millennium Foundation. AUTHOR CONTRIBUTIONS D.K. and Y.G. conducted the majority of the experiments and data analysis on sox14 and mical, respectively. Y.H. contributed to the biochemical experiments. Z.W. and A.L.K. provided reagents for mical. A.B. provided the pupal lethal mutant collection. B.C.L. and F.Y. cosupervised Y.G. F.Y. and H.W. conceptualized and designed the study. F.Y. supervised the project. D.K., H.W. and F.Y. 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. Luo, L. & OLeary, D.D. Axon retraction and degeneration in development and disease. Annu. Rev. Neurosci. 28, 127156 (2005). 2. OLeary, D.D. & Koester, S.E. Development of projection neuron types, axon pathways and patterned connections of the mammalian cortex. Neuron 10, 9911006 (1993). 3. MacDonald, J.M. et al. The Drosophila cell corpse engulfment receptor Draper mediates glial clearance of severed axons. Neuron 50, 869881 (2006). 4. Avery, M.A., Sheehan, A.E., Kerr, K.S., Wang, J. & Freeman, M.R. Wld S requires Nmnat1 enzymatic activity and N16-VCP interactions to suppress Wallerian degeneration. J. Cell Biol. 184, 501513 (2009). 5. Hoopfer, E.D. et al. Wlds protection distinguishes axon degeneration following injury from naturally occurring developmental pruning. Neuron 50, 883895 (2006). 6. Tsai, J., Grutzendler, J., Duff, K. & Gan, W.B. Fibrillar amyloid deposition leads to local synaptic abnormalities and breakage of neuronal branches. Nat. Neurosci. 7, 11811183 (2004). 7. Li, H., Li, S.H., Yu, Z.X., Shelbourne, P. & Li, X.J. Huntingtin aggregateassociated axonal degeneration is an early pathological event in Huntingtons disease mice. J. Neurosci. 21, 84738481 (2001). 8. Truman, J.W. Metamorphosis of the central nervous system of Drosophila. J. Neurobiol. 21, 10721084 (1990). 9. Lee, T., Lee, A. & Luo, L. Development of the Drosophila mushroom bodies: sequential generation of three distinct types of neurons from a neuroblast. Development 126, 40654076 (1999). 10. Marin, E.C., Watts, R.J., Tanaka, N.K., Ito, K. & Luo, L. Developmentally programmed remodeling of the Drosophila olfactory circuit. Development 132, 725737 (2005). 11. Lee, T., Marticke, S., Sung, C., Robinow, S. & Luo, L. Cell-autonomous requirement of the USP/EcR-B ecdysone receptor for mushroom body neuronal remodeling in Drosophila. Neuron 28, 807818 (2000). 12. Watts, R.J., Hoopfer, E.D. & Luo, L. Axon pruning during Drosophila metamorphosis: evidence for local degeneration and requirement of the ubiquitin-proteasome system. Neuron 38, 871885 (2003). 13. Awasaki, T. & Ito, K. Engulfing action of glial cells is required for programmed axon pruning during Drosophila metamorphosis. Curr. Biol. 14, 668677 (2004). 14. Watts, R.J., Schuldiner, O., Perrino, J., Larsen, C. & Luo, L. Glia engulf degenerating axons during developmental axon pruning. Curr. Biol. 14, 678684 (2004). 15. Awasaki, T. et al. Essential role of the apoptotic cell engulfment genes draper and ced-6 in programmed axon pruning during Drosophila metamorphosis. Neuron 50, 855867 (2006). 16. Zhu, S., Chiang, A.S. & Lee, T. Development of the Drosophila mushroom bodies: elaboration, remodeling and spatial organization of dendrites in the calyx. Development 130, 26032610 (2003). 17. Brown, H.L., Cherbas, L., Cherbas, P. & Truman, J.W. Use of time-lapse imaging and dominant negative receptors to dissect the steroid receptor control of neuronal remodeling in Drosophila. Development 133, 275285 (2006). 18. Schubiger, M., Wade, A.A., Carney, G.E., Truman, J.W. & Bender, M. Drosophila EcR-B ecdysone receptor isoforms are required for larval molting and for neuron remodeling during metamorphosis. Development 125, 20532062 (1998). 19. Williams, D.W. & Truman, J.W. Cellular mechanisms of dendrite pruning in Drosophila: insights from in vivo time-lapse of remodeling dendritic arborizing sensory neurons. Development 132, 36313642 (2005). 20. Kuo, C.T., Jan, L.Y. & Jan, Y.N. Dendrite-specific remodeling of Drosophila sensory neurons requires matrix metalloproteases, ubiquitin-proteasome, and ecdysone signaling. Proc. Natl. Acad. Sci. USA 102, 1523015235 (2005). 21. Lee, H.H., Jan, L.Y. & Jan, Y.N. Drosophila IKK-related kinase Ik2 and Katanin p60-like 1 regulate dendrite pruning of sensory neuron during metamorphosis. Proc. Natl. Acad. Sci. USA 106, 63636368 (2009). 22. Kuo, C.T., Zhu, S., Younger, S., Jan, L.Y. & Jan, Y.N. Identification of E2/E3 ubiquitinating enzymes and caspase activity regulating Drosophila sensory neuron dendrite pruning. Neuron 51, 283290 (2006). 23. Williams, D.W., Kondo, S., Krzyzanowska, A., Hiromi, Y. & Truman, J.W. Local caspase activity directs engulfment of dendrites during pruning. Nat. Neurosci. 9, 12341236 (2006). 24. Ainsley, J.A. et al. Enhanced locomotion caused by loss of the Drosophila DEG/ENaC protein Pickpocket1. Curr. Biol. 13, 15571563 (2003).

2009 Nature America, Inc. All rights reserved.

1504

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
25. Thummel, C.S. Files on steroidsDrosophila metamorphosis and the mechanisms of steroid hormone action. Trends Genet. 12, 306310 (1996). 26. Zheng, X. et al. TGF-beta signaling activates steroid hormone receptor expression during neuronal remodeling in the Drosophila brain. Cell 112, 303315 (2003). 27. Pauli, A. et al. Cell typespecific TEV protease cleavage reveals cohesin functions in Drosophila neurons. Dev. Cell 14, 239251 (2008). 28. Schuldiner, O. et al. piggyBac-based mosaic screen identifies a postmitotic function for cohesin in regulating developmental axon pruning. Dev. Cell 14, 227238 (2008). 29. Lee, C.Y. et al. Genome-wide analyses of steroid- and radiation-triggered programmed cell death in Drosophila. Curr. Biol. 13, 350357 (2003). 30. Beckstead, R.B., Lam, G. & Thummel, C.S. The genomic response to 20hydroxyecdysone at the onset of Drosophila metamorphosis. Genome Biol. 6, R99 (2005). 31. Li, T.R. & White, K.P. Tissue-specific gene expression and ecdysone-regulated genomic networks in Drosophila. Dev. Cell 5, 5972 (2003). 32. Lee, T. & Luo, L. Mosaic analysis with a repressible cell marker for studies of gene function in neuronal morphogenesis. Neuron 22, 451461 (1999). 33. Crmazy, F., Berta, P. & Girard, F. Genome-wide analysis of Sox genes in Drosophila melanogaster. Mech. Dev. 109, 371375 (2001). 34. Sparkes, A.C., Mumford, K.L., Patel, U.A., Newbury, S.F. & Crane-Robinson, C. Characterization of an SRY-like gene, DSox14, from Drosophila. Gene 272, 121129 (2001). 35. Wang, L. et al. A genetic screen identifies new regulators of steroid-triggered programmed cell death in Drosophila. Genetics 180, 269281 (2008). 36. Terman, J.R., Mao, T., Pasterkamp, R.J., Yu, H.H. & Kolodkin, A.L. MICALs, a family of conserved flavoprotein oxidoreductases, function in plexin-mediated axonal repulsion. Cell 109, 887900 (2002). 37. Beuchle, D., Schwarz, H., Langegger, M., Koch, I. & Aberle, H. Drosophila MICAL regulates myofilament organization and synaptic structure. Mech. Dev. 124, 390406 (2007). 38. Chittaranjan, S. et al. Steroid hormone control of cell death and cell survival: molecular insights using RNAi. PLoS Genet. 5, e1000379 (2009). 39. Lefebvre, V., Dumitriu, B., Penzo-Mendez, A., Han, Y. & Pallavi, B. Control of cell fate and differentiation by Sry-related high-mobility-group box (Sox) transcription factors. Int. J. Biochem. Cell Biol. 39, 21952214 (2007). 40. Suzuki, T. et al. MICAL, a novel CasL interacting molecule, associates with vimentin. J. Biol. Chem. 277, 1493314941 (2002). 41. Defilippi, P., Di Stefano, P. & Cabodi, S. p130Cas: a versatile scaffold in signaling networks. Trends Cell Biol. 16, 257263 (2006). 42. Huang, Z., Yazdani, U., Thompson-Peer, K.L., Kolodkin, A.L. & Terman, J.R. Crkassociated substrate (Cas) signaling protein functions with integrins to specify axon guidance during development. Development 134, 23372347 (2007). 43. Ndozangue-Touriguine, O., Hamelin, J. & Breard, J. Cytoskeleton and apoptosis. Biochem. Pharmacol. 76, 1118 (2008). 44. Toran-Allerand, C.D., Singh, M. & Setalo, G., Jr. Novel mechanisms of estrogen action in the brain: new players in an old story. Front. Neuroendocrinol. 20, 97121 (1999). 45. McCarthy, M.M. Estradiol and the developing brain. Physiol. Rev. 88, 91124 (2008).

2009 Nature America, Inc. All rights reserved.


nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1505

ONLINE METHODS
Fly strains. The following stocks were used in this study: MicalK583, MicalK1496, MicalI696, MicalG56 (ref. 37), ppk-Gal4 (ref. 24), Df(3R)swp2MICAL (ref. 36); UASMicalFL, UAS-MicalN-ter; Gal4221 (ref. 46) and ppk-eGFP (ref. 47). The RNAi stocks included UASsox14 RNAi, UASEcR RNAi, UASusp RNAi, UAS-mical RNAi and the seven RNAi lines specific to the other Drosophila Sox family genes; all other RNAi lines used for the initial screen were obtained from the National Institute of Genetics (Japan) and the Vienna Drosophila RNAi Center48. From the Bloomington stock center, we obtained Df(3R)BSC38, Df(2R)BSC136, Df(2R)orBR11a, Df(3R)ED5438, Df(3R)ED5454, Df(3R)Exel6155, elav-Gal4, 201Y-Gal4, UAS-mCD8GFP, UASEcR-B1C655. W650A TP1-9, Gal4109(2)80, FRT42D, FRT82B, tub-Gal80 and hsFlp flies. Generation of sox14 mutants. We crossed y1;P{y+[+m8] = Mae-UAS.6.11}LA00606 flies with a fly strain carrying the 2-3 transposase to induce excision. About 300 lines were established on the basis of the absence of the y+ marker. Nine pupal lethal lines were recovered and subjected to genomic PCR and DNA sequencing analysis. The line with a 1,964-bp deletion was named sox14D13 and the line with a 1,838-bp deletion was named sox14D15.

the immunofluorescent signals seen in wild-type tissues were absent in sox14 homozygous mutants and sox14 RNAi neurons (Supplementary Fig. 3 and data not shown) and the signals could be restored on the overexpression of UAS-sox14 transgenes in a sox14 homozygous mutant wL3 larvae (Supplementary Fig. 3). Imaging and dendrite quantification. Third-instar larvae, white prepupae and prepupae until 8 h APF were directly imaged using confocal microscopy. Images of neurons were taken as maximum projections of 1540 optical sections at ~1.5-m intervals. Quantification of all live images was carried out by counting the number of primary and secondary dendrites in a 300- 300-m region of the dorsal dendritic field of the ddaC neurons, originating from the second to fifth abdominal segments. For time-lapse imaging, time-lapse frames were assembled as maximum projections of 2030 optical sections taken at ~1.5-m intervals every 5 or 6 min. Movies were assembled in ImageJ with the Image5D plugin (J. Walter, University of Munich). Stills were aligned using the Stackreg plugin (P. Thvenaz, Ecole Polytechnique Fdrale de Lausanne). Images of fixed and live samples were obtained by laser confocal microscopy on a Leica SPE or Zeiss Meta510 and processed using Photoshop (Adobe Systems). Immunohistochemistry. Primary antibodies were used at a concentration of 1:1,000 for rabbit antibody to Mical, 1:40 for mouse antibody to EcR-B1 AD4.4 (Developmental Studies Hybridoma Bank, University of Iowa), 1:50 for mouse antibody to Usp (a gift from R. Barrio, CIC BioGUNE), 1:200 for mouse antibody to Sox14, 1:1,000 for rabbit antibody to GFP and 1:100 for Cy5-conjugated goat antibody to horseradish peroxidase (Jackson Laboratories). Cy3- or fluorescein isothiocyanate (FITC)-conjugated secondary antibodies were from Jackson Laboratories. Samples were mounted in VectaShield mounting medium. Statistical analyses. For the effect of nonfunctional and functional Mical overexpression in various mutant backgrounds, statistical significance was determined using two-tailed Students t test. We considered the results to be significant when P < 0.001.
46. Grueber, W.B., Jan, L.Y. & Jan, Y.N. Different levels of the homeodomain protein cut regulate distinct dendrite branching patterns of Drosophila multidendritic neurons. Cell 112, 805818 (2003). 47. Grueber, W.B., Ye, B., Moore, A.W., Jan, L.Y. & Jan, Y.N. Dendrites of distinct classes of Drosophila sensory neurons show different capacities for homotypic repulsion. Curr. Biol. 13, 618626 (2003). 48. Dietzl, G. et al. A genome-wide transgenic RNAi library for conditional gene inactivation in Drosophila. Nature 448, 151156 (2007).

2009 Nature America, Inc. All rights reserved.

Mapping of the mical allele. l(3)15256 was isolated from a pupal lethal collection on the basis of its strong pruning defects. The mutation was mapped by meiotic recombination mapping and deficiency mapping. l(3)15256 failed to complement with the deficiency Df(3R)BSC38. Further deficiency mapping using Df(3R)ED5438, Df(3R)ED5454, Df(3R)Exel6155 and Df(3R)swp2MICAL narrowed down the cytological location to 85F1-F8. The molecular lesion in the mical gene was confirmed by PCR from genomic DNA and sequencing. MARCM analysis. We heat shocked 36-h-old embryos twice at 38 C and aged them until they reached the late third-instar larval stage. ddaC clones were selected at the wL3 stage and imaged at the white prepupal stage according to their location and the complex stereotyped dendritic arbor morphology. The same ddaC clones were examined for dendrite-pruning defects at 18 h APF. Plasmid constructs and antibody production. Full-length sox14 cDNA was obtained by PCR from genomic DNA fragments and verified by DNA sequencing. It was subsequently cloned into pUAST and the resulting construct was used for microinjection and to generate transformants (Bestgene). The C-terminal region (amino acids 440669) of Sox14 was expressed using the GST expression vector (pGEX 4T-1, Pharmacia) and the purified protein was used to immunize mice to generate antibodies to Sox14. These antibodies are specific for Sox14, as

nature NEUROSCIENCE

doi:10.1038/nn.2415

a r t ic l e s

Nardilysin regulates axonal maturation and myelination in the central and peripheral nervous system
Mikiko Ohno1, Yoshinori Hiraoka1, Tatsuhiko Matsuoka1, Hidekazu Tomimoto2, Keizo Takao35, Tsuyoshi Miyakawa35, Naoko Oshima6, Hiroshi Kiyonari6, Takeshi Kimura1, Toru Kita1,7 & Eiichiro Nishi1
Axonal maturation and myelination are essential processes for establishing an efficient neuronal signaling network. We found that nardilysin (N-arginine dibasic convertase, also known as Nrd1 and NRDc), a metalloendopeptidase enhancer of protein ectodomain shedding, is a critical regulator of these processes. Nrd1/ mice had smaller brains and a thin cerebral cortex, in which there were less myelinated fibers with thinner myelin sheaths and smaller axon diameters. We also found hypomyelination in the peripheral nervous system (PNS) of Nrd1/ mice. Neuron-specific overexpression of NRDc induced hypermyelination, indicating that the level of neuronal NRDc regulates myelin thickness. Consistent with these findings, Nrd1/ mice had impaired motor activities and cognitive deficits. Furthermore, NRDc enhanced ectodomain shedding of neuregulin1 (NRG1), which is a master regulator of myelination in the PNS. On the basis of these data, we propose that NRDc regulates axonal maturation and myelination in the CNS and PNS, in part, through the modulation of NRG1 shedding. Myelination of axons by glial cells, such as oligodendrocytes in the CNS and Schwann cells in the PNS, is essential for rapid impulse conduction. Myelination is coordinated by the interaction between axons and glial cells. Although the diameter of an axon dictates whether myelination is initiated, myelination may further induce radial growth of axons1,2. Because axonal conduction is determined by axon caliber and myelin sheath thickness, axonal maturation (radial growth of axon) and myelination are essential processes for establishing an efficient neuronal signaling network3,4. Several lines of evidence suggest that bidirectional signaling between axons and myelin maintains neuronal functions. For example, some oligodendrocyte-specific proteins, such as proteolipid protein and 2,3-cyclic nucleotide phosphodiesterase (CNP), are required to maintain axonal integrity5,6. On the other hand, NRG1, a member of the epidermal growth factor (EGF) family, induces axonal signaling and is required for glial differentiation, proliferation and myelination79. NRG1 is synthesized as a transmembrane protein and then shed from the cell surface by proteolytic cleavage in the juxtamembrane region10. ADAM proteases and BACE1 (-secretase) have been proposed as sheddases for NRG1 (refs. 1115). Because BACE1-deficient mice have a hypomyelination phenotype1315, shedding of NRG1 is thought to be important in myelination in vivo. However, the underlying mechanism by which NRG1 shedding is regulated is poorly understood. NRDc is a zinc peptidase of the M16 family that selectively cleaves dibasic sites16,17. We identified NRDc as a specific binding partner of heparin-binding EGF-like growth factor (HB-EGF) and found
1Department

2009 Nature America, Inc. All rights reserved.

that NRDc enhances HB-EGF shedding through activation of tumor necrosis factor converting enzyme (TACE, also known as ADAM17)18,19. We also found that NRDc enhances the ectodomain shedding of multiple membrane proteins, including amyloid precursor protein (APP) and tumor necrosis factor-, through activation of several ADAM proteases20,21. These results suggest that NRDc may regulate the ectodomain shedding of a wide range of membrane proteins. To explore the physiological functions of NRDc, we generated Nrd1/ mice and found that these mice had impaired axonal maturation and hypomyelination in both the CNS and PNS. Furthermore, we found that NRDc regulates axonal maturation and myelination in the CNS and PNS, in part, through the modulation of NRG1 shedding. RESULTS Small brains in Nrd1/ mice NRDc was highly expressed in brain lysates of early postnatal mice but was expressed at a lower level in the brains of adult mice (Fig. 1a). Using immunohistochemical analysis, we found that NRDc was expressed in neurons and dendrites but not in glial cells ( Fig. 1b)20. These findings were supported by in situ hybridization, which revealed that there was little or no Nrd1 mRNA in the corpus callo sum (Supplementary Fig. 1). To examine the physiological functions of NRDc, we generated Nrd1/ mice by gene targeting (Supplementary Fig. 2). No Nrd1 mRNA or protein was detected by real-time PCR (data not shown) and western blot analysis, respectively, in the brains (Fig. 1a) and all other

of Cardiovascular Medicine, Graduate School of Medicine, Kyoto University, Sakyo-ku, Kyoto, Japan. 2Department of Neurology, Graduate School of Medicine, Mie University, Tsu, Japan. 3Division of Systems Medical Science, Institute for Comprehensive Medical Science, Fujita Health University, Toyoake, Aichi, Japan. 4Japan Science and Technology Agency (JST), Institute for Bioinformatics Research and Development and Core Research for Evolutional Science and Technology, Kawaguchi, Saitama, Japan. 5Frontier Technology Center, Graduate School of Medicine, Kyoto University, Sakyo-ku, Kyoto, Japan. 6Laboratory for Animal Resources and Genetic Engineering, Center for Developmental Biology, RIKEN Kobe, Chuo-ku, Kobe, Japan. 7Present address: Kobe City Medical Center General Hospital, Chuo-ku, Kobe, Japan. Correspondence should be addressed to E.N. (nishi@kuhp.kyoto-u.ac.jp). Received 29 July; accepted 22 September; published online 22 November 2009; doi:10.1038/nn.2438

1506

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 1 Expression of NRDc in brains and Brain MEF Brain a gross CNS phenotypes of Nrd1/ mice. P5 P14 P90 P365 +/+ / +/+ / +/ (a) Immunoblot analysis of brain extracts NRDc (left, center) or total cell extracts (right) GAPDH stained with antibodies to NRDc and GAPDH. b c d Analysis of brain extracts from wild-type mice at +/+ / +/+ / P5, P14, P90 and P365 (left) and from control NRDc Nissl Merge wild-type (+/+) and Nrd1/ mice (/) at P90 (center). Analysis of the total extracts of MEFs isolated from Nrd1+/+, Nrd1+/ or Nrd1/ mice (right). Full-length blots are presented in Supplementary Figure 9. (b) Immunohistochemistry using an antibody to NRDc (green) with cerebral cortex sections g e f from wild-type mice at P30. The section was * 250 +/+ / +/+ / double-stained with fluorescent Nissl (red). 200 Scale bars represent 250 m. (c) Brains of Nrd1/ mice and control wild-type littermates 150 at P90. (d) Enhanced limb-clasping reflex of / 100 Nrd1 compared with wild-type mice at P42. (e) Hematoxylin and eosinstained sections 50 of one brain hemisphere of Nrd1/ mice and 0 wild-type littermates at P90. An enlargement +/+ / of the lateral ventricles and cortical shrinkage / / in Nrd1 brain is shown. Scale bars represent 2 mm. (f,g) Nissl-stained sections of cerebral cortex of Nrd1 mice and wild-type littermates at P90. Scale bars represent 500 m. Counting of Nissl-positive neurons at the same bregma levels (six non-overlapping fields, n = 2 per genotype, results are mean s.e.m., * P < 0.01) revealed that there was a greater neuronal cell density in the Nrd1/ cortex.

2009 Nature America, Inc. All rights reserved.

tissues tested from Nrd1/ mice (data not shown). Immunostaining of tissue sections also showed a lack of NRDc protein in the cerebral cortex of Nrd1/ mice (Supplementary Fig. 2). The levels of NRDc protein in embryonic fibroblasts (MEFs) isolated from Nrd1+/+, Nrd1+/ or Nrd1/ mice correlated with the predicted gene dosage of Nrd1 in these mice (Fig. 1a). Pups lacking NRDc were born at the expected Mendelian ratio. However, approximately 80% died within 48 h of birth. Nrd1/ pups weighed approximately 30% less than Nrd1+/+ and Nrd1+/ littermates, indicating that NRDc is indispensable for normal prenatal growth. The Nrd1/ mice that survived remained smaller than their wild-type and heterozygous littermates throughout postnatal development and

+/+

+/+

had average brain weights that were 29% lower than those of Nrd1+/+ mice at postnatal day 90 (P90) (Nrd1+/+, 521.0 30.0 mg; Nrd1/, 371.5 16.8 mg; n = 5, P < 0.001; Fig. 1c). Although the Nrd1/ mice that survived lived until 2 years of age, they had several prominent neurological disorders. For example, Nrd1/ mice exhibited enhanced limb-clasping reflexes when suspended by the tail, whereas control mice extended their limbs (Fig. 1d). To assess the underlying neuropathology, we examined the brains of Nrd1/ mice histologically. Although the gross anatomy of the Nrd1/ mouse brain was normal at P1, P14 and P30 (Supplementary Fig. 3), we observed a thin cerebral cortex and enlarged lateral ventricles at P90 (Fig. 1e). We found that the Nrd1/ cortex had a greater neuronal cell density than in the Nrd1+/+ cortex by Nissl staining (Fig. 1f,g). Furthermore, there were no differences in the number of TUNEL-positive cells detected in the cortex of Nrd1+/+ and Nrd1/ mice (data not shown), indicating that there was no excessive loss of neurons in the Nrd1/ cortex. Hypomyelination in the CNS of Nrd1/ mice We next examined the integrity of axons and myelin using silver impregnation and luxol fast blue (LFB) staining, respectively. We detected much less silver impregnation in Nrd1/ brains, especially in the corpus callosum and cortical layers adjacent to it
Figure 2 Impaired axonal maturation and hypomyelination in the CNS of Nrd1/ mice. (ad) Silver impregnation by Bielschowskey method (a,b) and LFB staining (c,d) of Nrd1/ brain (/) and wild-type brain (+/+) at P90. We found impairment of axonal maturation and hypomyelination in Nrd1/ brains, especially in the corpus callosum. (e,f) Immunohistochemistry using antibodies to NF-H ( e), CNP and MBP (f) of corpus callosum from control wild-type mice and Nrd1/ mice at P30. Scale bars represent 2 mm (a,c), 500 m (b,d,f) and 250 m (e). (g) Immunoblot analysis of brain extracts from wild-type and Nrd1/ brains at P14 showed a reduction of MBP, CNP and NF-H expression in Nrd1/ brains. On the other hand, there were no differences in Olig2, PDGFR and GST- expression (n = 3 in each genotype). Full-length blots are presented in Supplementary Figure 9.

+/+

+/+

e
NF-H

+/+

f
CNP

+/+

+/+

Blot: MBP CNP NF-H PDGFR GST- GAPDH Nissl Olig2 MBP NRDc

+/+

+/+

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

Average cell number per field

1507

a r t ic l e s a
Nrd1
+/+

Nrd1/

P14

Figure 3 Delay in the initiation of myelination and hypomyelination in Nrd1/ mice. Electron microscopic analysis of corpus callosum. (a) Electron micrographs of transverse sections at the corpus callosum from Nrd1/ mice and wild-type littermates at P14 and P90. Scale bars represent 2 m. (b) Percentage of myelinated axons in the corpus callosum of Nrd1/and wild-type littermates at P14, P30, P90 and P365. (c,d) Quantitation of myelin sheath thickness in the corpus callosum by analyzing g ratios with electron micrographs. Scatter plots of myelin thickness, expressed as g ratios, against axon diameters at P90 in Nrd1/ (black) and wild-type (red) corpus callosum are shown (c). We determined the average myelin sheath thickness (g ratio) of myelinated fibers at P30, P90, P120 and P365 and found hypomyelination in the corpus callosum of Nrd1/ mice (d). Data represent the mean s.e.m. for 300900 myelinated fibers from each group. * P < 0.03 and ** P < 0.0001. (e) Average diameters of myelinated axons in the corpus callosum at P30 and P365. Data represent the mean s.e.m. for 300900 myelinated fibers from each group. * P = 0.0002 and ** P < 0.0001.

P90

b
Percentage of myelinated axons

80 60

+/+ /

c
1.0 0.9 g ratio 0.8 0.7 0.6

P90

Nrd1+/+ Nrd1/

40 20 0

P14 P30 P90 P365

0.5

0.5

d
g ratio

0.9

** *

+/+ /

e
Axon diameter (m)

1.0 1.5 2.0 2.5 Axon diameter (m)

3.0

1.0 0.8 0.6 0.4 0.2 0

+/+ /

** **

**

0.8

0.7

0.6

(Fig. 2a,b). Similarly, LFB staining was markedly weaker in this region (Fig. 2c,d). These results suggest that the volume of axons and myelin in Nrd1/ brain is decreased, especially in the corpus callosum. Furthermore, expression of the axonal marker neurofilament-H (NF-H; Fig. 2e) and the myelin markers CNP and myelin basic protein (MBP) (Fig. 2f) were lower in the corpus callosum in Nrd1/ brain at P30. Using western blotting, we found reduced CNP and MBP protein levels at an earlier stage (P14) in Nrd1/ brain extracts (Fig. 2g), suggesting that Nrd1/ mice have a severe impairment of axonal maturation and hypomyelination starting at the early stages of these processes. Neuronal expression of NRDc was detected in whole regions of the brain, but there were some prominent regional differences. For example, NRDc was highly expressed in cortical neurons, but neurons in striatum expressed relatively low levels of NRDc (Supplementary Fig. 4), similar to the expression pattern seen in human brain22. Consistent with these results, the myelination defect in Nrd1/ striatum was not as obvious as the defect in the cortex, hippocampus and corpus callosum (Supplementary Fig. 4). These results suggest that there is a correlation between the expression level of NRDc and the extent of the myelination defect in Nrd1/ mice.
1508

On the other hand, western blots of the same set of brain extracts revealed that the expression of platelet-derived growth factor receptor (PDGFR) and oligodendrocyte transcription factor 2 (Olig2)23, markers of oligodendrocyte precursors, and GST-, a marker of mature oligodendrocyte, were not reduced in Nrd1/ mice (Fig. 2g), suggesting that oligodendroglial differentiation is normal in Nrd1/ mice. Moreover, we found via quantitative analysis of Olig2-positive cells at embryonic day 18.5 (E18.5) and P5 that there were no differences in the number of oligodendrocyte precursors in Nrd1+/+ and Nrd1/ brains (Supplementary Fig. 5). These data suggest that NRDc affects myelination, but not differentiation of oligodendrocytes. To obtain direct evidence of impaired axonal maturation and hypomyelination in the CNS, we analyzed sections of the corpus callosum in Nrd1/ mice and Nrd1+/+ littermates between P14 and P365 by electron microscopy (Fig. 3a and Supplementary Fig. 6). The proportion of myelinated axons was markedly reduced in Nrd1/ mice at all of the stages examined (Fig. 3b). At P14, only about one seventh of the Nrd1/ axons were myelinated compared with the Nrd1+/+ axons, indicating that there was a substantial delay in the initiation of myelination in Nrd1/ mice. The proportion of myelinated axons in the Nrd1/ CNS was only 21.0% at P90 and did not approach the level seen in Nrd1+/+ mice even at P365 (Fig. 3b). We next compared myelin sheath thickness in the Nrd1/ and Nrd1+/+ corpus callosum by determining the g ratio24 (axon diameter to total fiber diameter) of myelinated fibers at P30, P90, P120 and P365 (Fig. 3c and Supplementary Fig. 6). The average g ratios were higher in Nrd1/ mice at all stages (Fig. 3d), indicating that myelin sheaths were thinner in Nrd1/ mice compared with those of Nrd1+/+ mice. The average diameters of myelinated axons were also smaller in Nrd1/ mice than wild-type littermates (Fig. 3e). Although axons were hypomyelinated, the ultrastructure of myelin sheaths of Nrd1/ mice appeared to be normal and indistinguishable from those of Nrd1+/+ mice (data not shown). Together, these data indicate that Nrd1/ mice have fewer myelinated fibers in the corpus callosum, and the fibers have thinner myelin sheaths and smaller axon diameters. As NRDc is highly expressed in neurons, but either not expressed or expressed at very low levels in glial cells (Fig. 1b and Supplementary Fig. 4)20, we propose that the loss of NRDc in neurons may cause an impairment of axonal maturation, resulting in the perturbation of axonal signaling required for oligodendrocytes to properly myelinate axons and hypomyelination. Hypomyelination in the PNS of Nrd1/ mice NRDc was also expressed in spinal neurons and dorsal root ganglion neurons (Supplementary Fig. 1). To investigate the role of NRDc in
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

2009 Nature America, Inc. All rights reserved.

20

65

P3

P9

P1

P3

P3

P3

65

a r t ic l e s a b
0.9 0.8 g ratio 0.7 0.6 0.5 0.4 0.3 0 0.5 P30 0.70 1.0 1.5 2.0 Axon diameter (m) P365 Axon diameter (m) 0.70 g ratio 0.65 0.60 0.55 +/+ / 2.5

Nrd1+/+

Nrd1/

P30 sciatic nerve

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0

P365 sciatic nerve

Nrd1 Nrd1/

+/+

c d
Nrd1+/+ Nrd1/ g ratio 0.65 0.60 0.55

0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Axon diameter (m) P30 P365 Axon diameter (m) 1.3 1.2 1.1 1.0 0.9 0.8 0.7 0.6 0.5

**

+/+

1.3 1.2 1.1 1.0 0.9 0.8 0.7 0.6 0.5

**

+/+

+/+

e
2009 Nature America, Inc. All rights reserved.

Number of axons / remak bundle 45 *** *** 40 35 30 25 20 15 10 5 0 P30 P365

Ratio of segregated axon 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Nrd1+/+ Nrd1/

P30

P365

Figure 4 Impaired axonal maturation and hypomyelination in the PNS of mice. (a) Electron micrographs of sciatic nerves of Nrd1+/+ and Nrd1/ mice at P365 showed obvious hypomyelination in the Nrd1/ nerves. Scale bars represent 2 m. (b) Scatter plots of g ratios and axon diameters at P30 and P365 of Nrd1/ and wild-type sciatic nerves. (c) G ratios and axon diameters at P30 and P365 of Nrd1+/+ and Nrd1/ sciatic nerves. Data represent the mean s.e.m. for 750 myelinated fibers from each group ( n = 4 for each genotype). * P < 0.03 and ** P < 0.01. (d) Electron micrographs of Remak bundles in sciatic nerves showed altered axonal segregation in Nrd1/ bundles. Arrows indicate Schwann cell cytoplasmic processes between different axons. Note that many axons of Nrd1/ bundles are directly apposed without the cytoplasmic processes (arrowhead). Scale bars represent 500 nm. (e) The average numbers of axons in a Remak bundle at P30 and P365 were significantly reduced in Nrd1/ sciatic nerves. Data represent the mean s.e.m. for 150300 Remak bundle from each group ( n = 2 for each genotype). *** P < 0.0001. (f) Ratios of segregated axons in Nrd1/ sciatic nerves at P30 and P365 were reduced (150300 Remak bundle from each group, n = 2 for each genotype).

Nrd1/

axonal maturation and myelination in the PNS, we analyzed the sciatic nerves of Nrd1/ mice by electron microscopy and found that the sciatic nerves of Nrd1/ mice were hypomyelinated (Fig. 4a). Quantitative analysis by measuring the g ratio of myelinated fibers indicated that the myelin sheath thickness was thinner in Nrd1/ mice at both P30 and P365. The diameter of myelinated axons was also smaller in Nrd1/ nerves compared with Nrd1+/+ nerves (Fig. 4b,c). We also examined the morphology of Remak bundles (Fig. 4df). Remak bundles in Nrd1/ nerves contained approximately twofold more axons than Nrd1+/+ nerves at P30 and P365 (Fig. 4d,e). In addition, there were many unsegregated or poorly segregated axons that lacked intervening Schwan cell processes in Nrd1/ bundles (Fig. 4d,f). These findings indicate that NRDc is critical for axonal maturation and myelination of both the CNS and PNS. Notably, the observed hypomyelination and morphological changes of Remak bundles in the PNS of Nrd1/ mice were very similar to those described in Nrg1+/ and Bace1/ mice7,8,13,14,25. The level of neuronal NRDc regulates myelin thickness To determine the dose-dependent effects of NRDc expression on the CNS phenotypes, we examined brains from Nrd1+/ mice, in which NRDc expression was obviously reduced (Fig. 5a). The gross ana tomy of Nrd1+/ brains showed no cortical shrinkage, but we did find enlarged lateral ventricles (Fig. 5b). Although we detected no obvious differences between Nrd1+/+ and Nrd1+/ brains by LFB staining
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

(Fig. 5b,c), we did find decreased expression of the myelin and axon markers MBP and NF-H in Nrd1+/ corpus callosum by immuno staining (Fig. 5d,e). These findings were confirmed by western blot using whole brain lysates (Fig. 5f). Furthermore, we analyzed Nrd1+/ corpus callosum by electron microscopy and found that the myelin sheath thickness of Nrd1+/ mice was significantly thinner than that of Nrd1+/+ mice (average g ratio: Nrd1+/+, 0.725 0.020 s.e.m.; Nrd1+/, 0.761 0.001 s.e.m.; n > 1000 fibers from each group, P < 0.0001; Fig. 5g,h). The axon diameters of Nrd1+/ mice were also smaller compared with those of Nrd1+/+ mice (Nrd1+/+, 0.650 0.018 m; Nrd1+/, 0.582 0.006 m; P < 0.0001). Thus, Nrd1+/ mice have a CNS phenotype that is intermediate of those of Nrd1+/+ and Nrd1/ mice, indicating that the level of NRDc expression in neurons affects axonal maturation and myelination. Our loss-of-function approach using Nrd1/ mice revealed that NRDc is essential for axonal maturation and myelination in the CNS. To further define the role of NRDc in these processes, we used a gain-of-function approach with transgenic mice that overexpress mouse NRDc under the control of the Camk2a promoter (NRDc-Tg mice)26,27. Western blot analysis of whole brain extract of NRDc-Tg mice at P90 revealed that there was an increase in NRDc expression in the transgenic mice (Fig. 6a). The Camk2a promoter was specifically activated in neurons of cerebral cortex and hippo campus26,27, where the increased expression of NRDc was confirmed by immunostaining (Fig. 6b,c). To see the regional effect of NRDc
1509

a r t ic l e s
Figure 5 Intermediate CNS phenotype in Nrd1+/ mice. (a) Immunohistochemistry of Nrd1+/ cortex (+/) with antibody to NRDc showed the obvious reduction of NRDc expression. (b,c) LFB staining of Nrd1+/ brains showed enlarged lateral ventricles, especially in the frontal section (b), compared with wild-type brains, but no obvious cortical shrinkage. Scale bars represent 1 mm. (d,e) Immunohistochemistry of Nrd1+/+ and Nrd1+/ corpus callosum at P90 using antibodies to MBP (d) and NF-H (e) showed a decrease in the expression of these axonal and myelin markers in Nrd1+/ brain. The sections were double-stained with fluorescent Nissl. Scale bars represent 500 m. (f) Immunoblot analysis of whole-brain extracts using antibodies to NRDc, NF-H and MBP. Expression of these proteins was reduced in Nrd1+/ brains compared with wild-type littermates. Full-length blots are presented in Supplementary Figure 9. (g) Electron micrographs of Nrd1+/+ and Nrd1+/ corpus callosum at P120. Scale bars represent 2 m. (h) Scatter plots of g ratio against axon diameters of Nrd1+/+ and Nrd1+/ corpus callosum at P120.

a NRDc
+/+

b
+/+ +/

c
+/+ +/

+/

d
MBP

+/+

+/

e
NF-H

+/+

+/

f
+/+ +/ +/ Blot: NRDc NF-H MBP

+/+ Nissl

+/ Nissl

+/+

+/

GAPDH

2009 Nature America, Inc. All rights reserved.

Nrd1+/+

Nrd1+/

h
1.1 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 Nrd1+/+ Nrd1+/

overexpression, we divided mouse brains into two regions: cerebral cortex plus hippo campus (frontal region) and brain stem plus cerebellum (posterior region). Using western blot, we found increased expression of NRDc protein only in the frontal region (Fig. 6d). The expression of the myelin markers MBP and CNP was also increased in the frontal region, but not in the posterior region (Fig. 6d). We detected higher expression of MBP in corpus callosum by immunostaining (Fig. 6e). Notably, analysis of myelination in the NRDc-Tg corpus callosum by electron microscopy revealed that the average g ratio was significantly lower than in wild type, indicating that the myelin sheaths were thicker in NRDc-Tg mice than in wild types (Fig. 6fh). On the other hand, there was no significant difference in the diameter of myelinated axons in the corpus callosum between the NRDc-Tg and wildtype mice (average diameter: wild type, 0.80 0.39 m; NRDc-Tg, 0.81 0.39 m, P = 0.654). These results provide further evidence that the level of neuronal NRDc regulates myelin sheath thickness. The Camk2a promoter is activated around P5, which is after the differentiation of neuronal cells and before subcortical myelination 27. Thus, NRDc probably affects myelination by regulating axonal signals transmitted from differentiated neurons. Impaired motor activity and cognitive deficits in Nrd1/ mice To examine the effect of NRDc on neurological functions, we analyzed Nrd1/ mice with a battery of behavioral tests28. We examined neuromuscular strength with grip strength and wire-hanging tests. Although Nrd1/ mice had reduced grip strength (Fig. 7a), we found no difference between Nrd1+/+ and Nrd1/ mice in the wire-hanging test (Supplementary Fig. 7), indicating that Nrd1/ mice possess sufficient grip strength to hold their body weight. We next assessed motor coordination and balance with the beam test and rotarod test29. In the beam test, Nrd1/ mice moved much slower and slipped more frequently than Nrd1+/+ mice (Fig. 7b,c). Nrd1/ mice also had a shorter latency to fall in the rotarod test (Fig. 7d). These results indicate that there is a severe impairment of motor coordination and balance in the Nrd1/ mice.
1510

g ratio

0.5 1.0 1.5 2.0 Axon diameter (m)

2.5

In contrast with motor activity, sensory pathways were apparently conserved in Nrd1/ mice, as their reactions were similar to those of Nrd1+/+ mice in the hot-plate test (Supplementary Fig. 7). We examined cognitive functions of Nrd1/ mice with the T-maze test, in which working memory and reference memory are evaluated by a forcedalternation task and a left rightdiscrimination task, respectively30. In both tasks, Nrd1/ mice took much longer to complete a session, probably as a result of impaired motor activity (Supplementary Fig. 7 and data not shown). Although the percentage of correct choices in the forced-alternation task was lower in Nrd1/ mice, there were no differences in the left rightdiscrimination task between Nrd1+/+ and Nrd1/ mice (Fig. 7e,f). Together, these results suggest that reference memory is preserved, whereas working memory is impaired in Nrd1/ mice. Although these behavioral alterations are consistent with the axonal and myelination defects, spinal or cerebellar defects might also influence the overall performance of Nrd1/ mice. NRDc regulates NRG1 shedding through BACE1 and TACE NRG1, one of the master regulators of myelination, is synthesized as a transmembrane protein and then proteolytically shed from the cell surface8,10. Recent reports of hypomyelination in Bace1/ mice have implicated BACE1 in NRG1 shedding1315. NRG1 is also shed from the cell surface by TACE11, although there is no information about nervous system phenotypes of TACE-deficient mice as a result of their early lethality. These findings, along with the potentiating effect of NRDc on protein ectodomain shedding1921, prompted us to examine whether NRDc affects axonal maturation and myelination via regulation of NRG1 shedding by BACE1 or TACE. To determine whether NRDc potentiates BACE1/TACE activity, we carried out transfection experiments in COS7 cells. NRG1 type I was tagged with hemagglutinin (HA) at the
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 6 Transgenic overexpression of neuronal NRDc induces hypermyelination in the CNS. (a) Immunoblot analysis of total brain extracts from wild-type littermates and NRDc-Tg mice at P90 with antibody to NRDc. (b,c) We stained sections from NRDc-Tg brains with an antibody to NRDc and found neuronal overexpression of NRDc in the cortex (b) and hippocampus (c) compared with wild-type brains. Scale bars represent 250 m. (d) Brains of wild-type and NRDc-Tg mice were divided into two regions at the border of cerebral cortex and cerebellum. We analyzed total brain extracts from frontal region and posterior region by immunoblot using antibodies to NRDc, CNP, MBP and NF-H. Note that the increased expression of NRDc, CNP and MBP were only detected in the frontal region. Full-length blots are presented in Supplementary Figure 9. (e) Expression of MBP in the corpus callosum was increased in NRDc-Tg compared with wild type mice. Scale bars represent 500 m. (f) Electron micrographs of corpus callosum from NRDc-Tg and wild type mice at P30. Scale bars represent 500 nm. (g,h) Myelin sheaths were thicker in NRDc-Tg mice than in wild types. Scatter plots of g ratios against axon diameters at P30 in NRDc-Tg and wild-type corpus callosum (g) and the average g ratio (h) are shown. Data represent the mean s.e.m. for no fewer than 1,000 myelinated fibers from each group (n = 4 for each genotype).

a
ty pe N R D cTg W ild

bWild type
NRDc

NRDc-Tg

cWild type

NRDc-Tg

Blot: NRDc

Frontal NRDc-Tg

Posterior Wild type NRDc-Tg Blot: NRDc CNP MBP NF-H GAPDH MBP

Wild type

Wild type

NRDc-Tg

fWild type

NRDc-Tg

g 1.0
0.9 g ratio 0.8 0.7 0.6 0.5 0 0.5 Wild type NRDc-Tg

h
P < 0.0001 0.80 g ratio 0.76 0.74 0.70

2009 Nature America, Inc. All rights reserved.

ild

N terminus so that the full length and N-terminal fragment (NTF) of NRG1 type I could be detected with an antibody to HA. Coexpression of NRDc and BACE1 increased NTF levels in total cell lysates, but not in the culture medium (Fig. 8a,b). In contrast, coexpression of NRDc with TACE clearly increased NTF levels in the culture medium (Fig. 8a,b). These results suggest that NRDc potentiates TACE-mediated NRG1 cleavage at the cell surface, whereas NRDc enhances BACE1 cleavage of NRG1 in intracellular compartments. Because the enhancement of BACE1-mediated cleavage of NRG1 was not associated with an increase in NTF levels in the culture medium, NRDc is thought to potentiate BACE1 in endocytic pathways and direct the cleaved NRG1 to degradation31. Similar to the direct interaction of NRDc and TACE19, co-precipitation demonstrated that NRDc forms as trimolecular complex with BACE1 and NRG1, suggesting that these proteins physically and functionally interact (Fig. 8c).

a Grip strength b
(N) 1.2 1.0 0.8 0.6 0.4 0.2 0 P < 0.0001 (cm 30 20 10 0

Beam test moving speed s1) P = 0.0071 +/+ /

c
6 4 2

Beam test number of slips P = 0.034 +/+ /

d
(s) 250 200 150 100 50 0

(slips) 8

Rotarod latency to fall P = 0.0102 +/+ /

Given these findings, we examined the expression pattern of endo genous NRG1 protein in fibroblasts derived from Nrd1/ and Nrd1+/+ mice. Western blot analysis using an antibody to the C terminus of NRG1 (NRG-C), which recognizes both full-length NRG1 and the C-terminal fragment of NRG1 (CTF), revealed an increase in full-length NRG1 and a decrease in CTF in Nrd1/ cells compared with Nrd1+/+ cells (Fig. 8d). In addition, although an antibody to the intracellular N terminus of NRG1 type III detected the cleaved NTF of NRG1 type III in Nrd1+/+ cells, it barely detected the cleaved fragment in Nrd1/ cells (Fig. 8d). These results indicate that NRDc positively regulates the proteolytic cleavage of both type I and type III NRG1. Next, we examined the expression of NRG1 in brain extracts from Nrd1+/+ and Nrd1/ mice at P14. Western blotting with antibodies to the C terminus of NRG1 type I and the N terminus of NRG1 type III revealed a similar pattern as found in MEFs (Fig. 8e), although the difference between the two genotypes was less evident in the brain extract. These results are consistent with the fact that the proteolytic cleavage of NRG1 type I and type III in brain is regulated by NRDc. We then analyzed the protein expression of BACE1. The molecular
Figure 7 Impaired motor activity and cognitive deficits in Nrd1/ mice. Mice at the age of 90180 d were analyzed with a battery of behavioral tests (age-matched littermates; Nrd1+/+, n = 12; Nrd1/, n = 6). (a) Grip-strength test, measured in newtons. Nrd1/ mice had reduced forelimb grip strength. (b,c) The ability of mice to traverse a narrow beam to reach a dark box was analyzed in a beam test. Nrd1/ mice moved slower (b) and slipped more frequently (c) than Nrd1+/+ mice. (d) Latency to fall from a rotating drum was measured in a rotarod test. Nrd1/ mice had a shorter latency to fall. (e) The percentages of correct choices in the T maze forced-alteration task are shown. showing Nrd1/ mice had impaired working memory. (f) The percentages of correct choices in the T maze left rightdiscrimination task are shown. There was no difference between Nrd1+/+ and Nrd1/ mice. The genotype effect was analyzed by a two-way repeated ANOVA in all tests. Data represent the mean s.e.m.

+/+ /

e (%)
100 80 60 40 20 0 1

2 Trials +/+ /

f (%)
100 80 60 40

2 Trials

1 2 3 4 5 6 Trials +/+ /

Reversal

P = 0.0042 (genotype) 3 5 Sessions 7

20 P = 0.5708 (genotype) 0 1 3 5 7

Left right discrimination P = 0.5243 (genotype trial)

Reversal learning P = 0.6435 (genotype) P = 0.1966 (genotype trial)

9 11 13 15 17 19 Sessions

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1.0 1.5 2.0 2.5 Axon diameter (m)

pe ty

c-

Tg

3.0

1511

a r t ic l e s
Figure 8 NRDc regulates NRG1 shedding through BACE1 and TACE. (a) FLAG-tagged TCL (HA) NRDc + + + + + NRDc * * BACE1 + + BACE1 + + * 3.5 NRDc, V5-tagged BACE1, TACE and HA-tagged 3.0 + NRG1 + + + Blot: TACE + + Blot: NRG1 type1 were coexpressed in COS7 cells. 2.5 (kDa) 150 NRDc 2.0 BACE1 Immunoblot analyses with the indicated (V5) 1.5 75 antibodies were performed. Note that the 1.0 BACE1 IP/NRDc 0.5 NRG-C cleaved NTF of NRG1 in the total cell lysates (V5) (FLAG) 0 150 (TCL) was increased by the addition of NRDc to TACE NRDc BACE1 (lanes 3 and 4), whereas coexpression 100 TCL (FLAG) of NRDc with TACE clearly increased the NTF 100 FL BACE1 in the culture medium (CM) (lanes 5 and 6). CM (HA) 75 (V5) NRG1 * 3.5 (b) Quantification of the cleaved NTF in the (HA) TCL 3.0 2.5 TCL and culture medium by densitometry. 50 NTF NRG-C 2.0 The ratio was arbitrarily set at 1 in control 1.5 (FL) 1.0 vectortransfected cells. Data represent the 50 NRG1 1 2 3 4 CM 0.5 (HA) mean s.e.m. in five independent experiments. 0 1 2 3 4 5 6 * P < 0.05. (c) Immunoprecipitation of the co-transfected cell lysate with an antibody to FLAG reveled a complex of NRDc, BACE1 and NRG1. Only the full length (FL) +/+ / Type III Blot: Type I of NRG1 is shown here. (d) Expression of (kDa) +/+ / NRDc (kDa) +/+ / (kDa) endogenous NRG1 type I and NRG1 type III in 150 150 100 FL FL MEFs isolated from Nrd1+/+ or Nrd1/ mice. 100 75 75 Note that the full-length of NRG1 type I was 100 NRG-C increased and the cleaved CTF was decreased 75 in Nrd1/ cells (lane 2) and that the cleaved 50 50 CTF CTF NTF 37 NTF of NRG1 type III was decreased in 1 2 Type III 75 3 4 / Nrd1 cells (lane 4). (e) Immunoblot NTF Blot: Blot: analysis of brain extracts from control BACE1 NRG-C Type III wild-type (+/+) and Nrd1/ (/) mice at GAPDH P14 (n = 3 in each genotype) with indicated 1 2 3 4 5 6 antibodies. Note that the full length of NRG1 (~140 kDa) was increased, whereas the NTF of NRG1 type III (75 kDa) was decreased in Nrd1/ brains. The mature form of BACE1 in Nrd1+/+ brain extracts is indicated by an arrow. The ratio of the mature form to immature form was higher in Nrd1+/+ than in Nrd1/ brains.

2009 Nature America, Inc. All rights reserved.

size of mature BACE1 in Nrd1/ brain was a little smaller than that in the Nrd1+/+ brain (Fig. 8e). In addition, the ratio of the mature form to immature form of BACE1 was clearly higher in Nrd1+/+ brains than in Nrd1/ brains (Nrd1+/+, 2.21 0.62; Nrd1/, 0.75 0.08; n = 3, P = 0.04), indicating that the protein maturation of BACE1 was impaired in Nrd1/ brains. These results suggest that NRDc regulates BACE1-mediated NRG1 shedding by affecting BACE1 maturation and BACE1 sheddase activity. On the other hand, the mRNA levels of Nrg1 and Bace1 were not different between Nrd1+/+ and Nrd1/ brains (Supplementary Fig. 8), indicating that NRDc post-translationally modulates the expression and activity of NRG1 and BACE1. DISCUSSION Our findings provide, to the best of our knowledge, the first in vivo evidence that NRDc is a critical regulator of axonal maturation and myelination in the CNS and PNS. In the CNS, axon diameter and myelin thickness correlated with levels of NRDc expression (homozygous < heterozygous < wild-type mice). Our results suggest that NRDc affects myelination by regulating axonal maturation, as myelin thickness is proportional to the diameter of axons1,2. Neuronspecific overexpression of NRDc, however, did not affect the axon caliber, but instead induced hypermyelination, providing additional evidence that NRDc is a critical regulator of myelination. Notably, as we did not see any differences in the number of oligodendrocyte precursor cells and the expression levels of specific oligodendrocyte precursor markers between Nrd1+/+ and Nrd1/ mice, NRDc appears to specifically regulate myelination and not oligodendrocyte differentiation or proliferation. In addition, as expression of NRDc is confined to neurons, the effect of NRDc on oligodendrocyte is
1512

noncell autonomous. Instead, our data suggest that NRDc affects axonal signaling between neurons and oligodendrocytes. NRG1 is the best-characterized neuronal factor that induces axonal signaling required for the entire program of glial differentiation, proliferation and myelination8,10,32. NRG1 is synthesized as a transmembrane protein that is proteolytically cleaved in the juxtamembrane region. ADAM proteases (TACE/ADAM17 (ref. 11) and ADAM19 (refs. 12,33)) and BACE1 (refs. 13,14) have been proposed to cleave NRG1. Furthermore, Bace1/ mice exhibit hypomyelination1315 and Adam19/ mice have delayed remyelination after injury33, suggesting that proteolytic cleavage of NRG1 is important for myelination. Our data indicate that NRDc is a critical regulator of NRG1 cleavage in vivo. First, the results of our gain-of-function experiments indicate that NRDc enhances BACE1- and TACE-mediated NRG1 cleavage. Second, loss of function in cells resulted in an increase of the fulllength NRG1 and a decrease of cleaved NRG1, indicating that NRG1 cleavage is reduced in the absence of NRDc. Third, similar to the results from Nrd1/ cells, we found a decrease of cleaved NRG1 in brain lysates of Nrd1/ mice. Nrd1/ mice also showed hypomyelination in both the CNS and PNS, where NRG1 is a master regulator of myelination. Finally, we found that NRDc formed a complex with BACE1 and NRG1, suggesting that these molecules interact functionally. These results suggest that regulation of NRG1 proteolysis by NRDc is a critical post-translational modification for myelination. Although several studies have suggested that NRG1 is involved in oligodendrocyte differentiation and CNS myelination25,32,34, the role of NRG1 in CNS myelination is still controversial as a result of the embryonic lethality of Nrg1/ mice. However, a recent study found normal myelination in a series of conditional null
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

tr N ol R D N c R D BA c C + E BA C N E R D TA c + CE TA C E

C on

tro N l R N D R D BA c c + CE BA C N E R D TA c + CE TA C E

C on

a r t ic l e s
mutants of Nrg1 that lack the gene at different stages during CNS development, suggesting that NRG1 may function differently in the PNS and CNS35. Because our data indicate that NRDc regulates myelination in both the CNS and PNS, NRDc may affect different proteins in the CNS. On the other hand, although both Nrd1/ and Bace1/ mice had hypomyelination in the CNS and PNS, no abnormalities in axon diameter and initiation of myelination have been reported in BACE1/ mice1315. Taken together, the effect of NRDc in the CNS cannot be attributed merely to NRG1 or BACE1. We have previously shown that NRDc potentiates several ADAM proteases1921 and here we found that NRDc also enhanced the capacity of BACE1 to process NRG1. The enhancing effect of NRDc on ADAM proteases is not substrate-specific, as NRDc can potentiate TACE-mediated shedding of HB-EGF, TNF-, APP and NRG1 (refs. 1921). This might also be the case for BACE1, as we found that NRDc affects the maturation of BACE1 (Fig. 8). Axonal maturation and myelination are essential in nerve conduction and aberrant in various neuropathologies. For example, the primary pathologic event of multiple sclerosis is demyelination and remyelination correlates with recovery from clinical symptoms 36. Given the dose-dependent effect of NRDc on myelin thickness, NRDc could be a potential pharmacological target of this common neurological disorder of young adults. NRDc could also be important for the regeneration of injured axons. In fact, the critical regulatory functions of NRDc in axon-oligodendrocyte signaling, axon maturation and myelination suggest that NRDc may impinge on a broad range of neurological disorders. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We are grateful to N. Nishimoto and H. Nakabayashi for technical assistance, and K. Matsumoto, E. Kimura, A. Kinoshita and A. Sehara for materials. We thank P.W. Park, T. Nishio, H. Fujiwara, T. Kaneko, F. Fujiyama and H. Kawasaki for critical reading of the manuscript. This study was supported by research grants (19041035, 20390255, 20659061, 20200068 and IBR-shien) from the Ministry of Education, Culture, Sports, Science and Technology of Japan. It was also supported by the Takeda Science Foundation, the Mochida Memorial Foundation for Medical and Pharmaceutical Research, the Suzuken Memorial Foundation, the Japan Health Foundation and the Daiichi Sankyo Sponsored Research Program. AUTHOR CONTRIBUTIONS M.O. and E.N. planned the experiments and wrote the manuscript. M.O., Y.H., T.M. and E.N. performed the experiments. H.T. carried out the histological procedures. K.T. and T.M. performed behavioral analysis. N.O. and H.K. generated the Nrd1/ mice. T. Kimura and T. Kita supervised the work.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
1. Nave, K.A. & Trapp, B.D. Axon-glial signaling and the glial support of axon function. Annu. Rev. Neurosci. 31, 535561 (2008). 2. Simons, M. & Trotter, J. Wrapping it up: the cell biology of myelination. Curr. Opin. Neurobiol. 17, 533540 (2007). 3. Hartline, D.K. & Colman, D.R. Rapid conduction and the evolution of giant axons and myelinated fibers. Curr. Biol. 17, R29R35 (2007). 4. McTigue, D.M. & Tripathi, R.B. The life, death, and replacement of oligodendrocytes in the adult CNS. J. Neurochem. 107, 119 (2008). 5. Griffiths, I. et al. Axonal swellings and degeneration in mice lacking the major proteolipid of myelin. Science 280, 16101613 (1998). 6. Lappe-Siefke, C. et al. Disruption of Cnp1 uncouples oligodendroglial functions in axonal support and myelination. Nat. Genet. 33, 366374 (2003). 7. Michailov, G.V. et al. Axonal neuregulin-1 regulates myelin sheath thickness. Science 304, 700703 (2004). 8. Nave, K.A. & Salzer, J.L. Axonal regulation of myelination by neuregulin 1. Curr. Opin. Neurobiol. 16, 492500 (2006). 9. Mei, L. & Xiong, W.C. Neuregulin 1 in neural development, synaptic plasticity and schizophrenia. Nat. Rev. Neurosci. 9, 437452 (2008). 10. Falls, D.L. Neuregulins: functions, forms and signaling strategies. Exp. Cell Res. 284, 1430 (2003). 11. Montero, J.C., Yuste, L., Diaz-Rodriguez, E., Esparis-Ogando, A. & Pandiella, A. Differential shedding of transmembrane neuregulin isoforms by the tumor necrosis factor alphaconverting enzyme. Mol. Cell. Neurosci. 16, 631648 (2000). 12. Shirakabe, K., Wakatsuki, S., Kurisaki, T. & Fujisawa-Sehara, A. Roles of Meltrin beta/ADAM19 in the processing of neuregulin. J. Biol. Chem. 276, 93529358 (2001). 13. Willem, M. et al. Control of peripheral nerve myelination by the beta-secretase BACE1. Science 314, 664666 (2006). 14. Hu, X. et al. Bace1 modulates myelination in the central and peripheral nervous system. Nat. Neurosci. 9, 15201525 (2006). 15. Hu, X. et al. Genetic deletion of BACE1 in mice affects remyelination of sciatic nerves. FASEB J. 22, 29702980 (2008). 16. Pierotti, A.R. et al. N-arginine dibasic convertase, a metalloendopeptidase as a prototype of a class of processing enzymes. Proc. Natl. Acad. Sci. USA 91, 60786082 (1994). 17. Chesneau, V. et al. N-arginine dibasic convertase (NRD convertase): a newcomer to the family of processing endopeptidases. An overview. Biochimie 76, 234240 (1994). 18. Nishi, E., Prat, A., Hospital, V., Elenius, K. & Klagsbrun, M. N-arginine dibasic convertase is a specific receptor for heparin-binding EGF-like growth factor that mediates cell migration. EMBO J. 20, 33423350 (2001). 19. Nishi, E., Hiraoka, Y., Yoshida, K., Okawa, K. & Kita, T. Nardilysin enhances ectodomain shedding of heparin-binding epidermal growth factorlike growth factor through activation of tumor necrosis factor alphaconverting enzyme. J. Biol. Chem. 281, 3116431172 (2006). 20. Hiraoka, Y. et al. Enhancement of alpha-secretase cleavage of amyloid precursor protein by a metalloendopeptidase nardilysin. J. Neurochem. 102, 15951605 (2007). 21. Hiraoka, Y. et al. Ectodomain shedding of TNF-alpha is enhanced by nardilysin via activation of ADAM proteases. Biochem. Biophys. Res. Commun. 370, 154158 (2008). 22. Bernstein, H.G. et al. Histochemical evidence for wide expression of the metalloendopeptidase nardilysin in human brain neurons. Neuroscience 146, 15131523 (2007). 23. Lu, Q.R. et al. Common developmental requirement for Olig function indicates a motor neuron/oligodendrocyte connection. Cell 109, 7586 (2002). 24. Fields, R.D. & Ellisman, M.H. Axons regenerated through silicone tube splices. II. Functional morphology. Exp. Neurol. 92, 6174 (1986). 25. Taveggia, C. et al. Neuregulin-1 type III determines the ensheathment fate of axons. Neuron 47, 681694 (2005). 26. Mayford, M., Wang, J., Kandel, E.R. & ODell, T.J. CaMKII regulates the frequencyresponse function of hippocampal synapses for the production of both LTD and LTP. Cell 81, 891904 (1995). 27. Dragatsis, I. & Zeitlin, S. CaMKIIalpha-Cre transgene expression and recombination patterns in the mouse brain. Genesis 26, 133135 (2000). 28. Yamasaki, N. et al. Alpha-CaMKII deficiency causes immature dentate gyrus, a novel candidate endophenotype of psychiatric disorders. Mol Brain 1, 6 (2008). 29. Carter, R.J. et al. Characterization of progressive motor deficits in mice transgenic for the human Huntingtons disease mutation. J. Neurosci. 19, 32483257 (1999). 30. Takao, K. et al. Impaired long-term memory retention and working memory in sdy mutant mice with a deletion in Dtnbp1, a susceptibility gene for schizophrenia. Mol Brain 1, 11 (2008). 31. Thinakaran, G. & Koo, E.H. Amyloid precursor protein trafficking, processing and function. J. Biol. Chem. 283, 2961529619 (2008). 32. Calaora, V. et al. Neuregulin signaling regulates neural precursor growth and the generation of oligodendrocytes in vitro. J. Neurosci. 21, 47404751 (2001). 33. Wakatsuki, S., Yumoto, N., Komatsu, K., Araki, T. & Sehara-Fujisawa, A. Roles of meltrin-beta/ADAM19 in progression of Schwann cell differentiation and myelination during sciatic nerve regeneration. J. Biol. Chem. 284, 29572966 (2009). 34. Taveggia, C. et al. Type III neuregulin-1 promotes oligodendrocyte myelination. Glia 56, 284293 (2008). 35. Brinkmann, B.G. et al. Neuregulin-1/ErbB signaling serves distinct functions in myelination of the peripheral and central nervous system. Neuron 59, 581595 (2008). 36. Trapp, B.D. & Nave, K.A. Multiple sclerosis: an immune or neurodegenerative disorder? Annu. Rev. Neurosci. 31, 247269 (2008).

2009 Nature America, Inc. All rights reserved.

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1513

ONLINE METHODS
Generation of mutant mice. Nrd1/ mice (accession number CDB0466K, http:// www.cdb.riken.jp/arg/mutant%20mice%20list.html) were generated by gene targeting in TT2 (ref. 37) embryonic stem (ES) cells as described at http://www.cdb. riken.jp/arg/protocol.html. The targeting vector was designed to replace exon 1 (Supplementary Fig. 2) with a DT-A/lox71/LacZ-pA/frt/PGK-Neo/frt/loxP/pA cassette (http://www.cdb.riken.jp/arg/cassette.html) and successful homologous recombination in TT2-ES cells was confirmed by PCR (forward primer #1, 5-CCT CAG CTC AGC CAC TAG CCC TTG-3; reverse primer #2, 5-TCC AAC CCG GCC TCG CTC ACA CCG A-3; numbers 1 and 2, respectively, in Supplementary Fig. 2) and Southern blotting in two clones, which yielded highly chimeric mice. We intercrossed the heterozygous offspring (F1) to obtain homozygous mutants (F2), which were genotyped with tail DNA by PCR using a mixture of three primers (forward primer #3, 5-GGA TGT CCA GCG CTA CGA CT-3; forward primer #4, 5-ATC GAT GAT ATC AGA TCC CC-3; reverse primer #5, 5-CCT CCC CTT TAC CAT CCA TC-3; numbers 3, 4 and 5, respectively, in Supplementary Fig. 2). Primers 3 and 5 amplify the wild-type allele (674 base pairs) and primers 4 and 5 amplify the mutated allele (380 base pairs) (Supplementary Fig. 2). Two strains of Nrd1/ mice, derived from two different clones of ES cells, showed identical phenotypes. All analysis of homozygous mutant mice was carried out using F2 generation, which has a 75% C57BL/6 and 25% CBA background. For the analysis of heterozygous mutant mice, F2 mice were backcrossed with C57BL/6 more than six times to exclude the possible effect of genetic heterogeneity on the phenotype of the heterozygous mice. For the generation of neuron-specific NRDc-overexpressing transgenic mice, PCR-amplified cDNA encoding mouse Nrd1 was cloned downstream of the Camk2a promotor26. The construct was excised at a unique PacI site for DNA microinjection of fertilized mouse oocytes. Routine genotyping of tail DNA was conducted by PCR (forward primer, 5-CCA TAG CAA CTG CCT CTT TG-3; reverse primer, 5-GAT CGC TGG GAG ACT TGA TG-3). Mice were housed in environmentally controlled rooms of the Institute of Laboratory Animals, Kyoto University, under the Institutes guidelines for animal and recombinant DNA experiments. Cells, plasmids and transfections. MEFs were isolated from either Nrd1+/+, Nrd1+/ or Nrd1/ embryos at E14.5 and maintained in Dulbeccos modified Eagle medium (DMEM) containing 5% fetal bovine serum (vol/vol), 100 U ml1 of penicillin, 100 mg ml1 of streptomycin and 2 mM L-glutamine. MEFs were passed according to the 3T3 protocol and immortalized38. COS7 cells were cultured in DMEM containing 10% fetal bovine serum, 100 U ml1 of penicillin, 100 mg ml1 of streptomycin and 2 mM L-glutamine. The expression plasmid for human NRDc (pcDNA3-NRD1-FLAG) was described previously19. Expression plasmids for V5-tagged human BACE1 and HA-tagged human NRG1 type I were generous gifts from A. Kinoshita and A. Sehara, respectively. The transfection of plasmids into COS7 cells was carried out using FuGENE6 (Roche) according to the manufacturers instructions. Immunoblot analysis and immunoprecipitation. The preparation of total cell extract, immunoblotting and immunoprecipitation were carried out as described previously19. For the preparation of brain extract, whole brain was homogenized by sonication in lysis buffer containing 10 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1% NP-40 (vol/vol) and protease inhibitor cocktail (Complete Mini, Roche Diagnostics), followed by ultra-centrifugation at 100,000 g for 20 min at 4 C. For the immunoblot analysis and immunoprecipitation, we used mouse monoclonal antibodies to NRDc (#23)19, glyceraldehyde-3 phosphate dehydrogenase (GAPDH, Research Diagnostics), the C terminus of neuregulin (NRG-C, Santa Cruz) for detection of the full-length and the cleaved C terminus of neuregulin type I and III, NRG1 type III (Chemicon, Millipore) for detection of the full-length and the cleaved N terminus of neuregulin type III, BACE1 (EE-17, Sigma), PDGFR (Santa Cruz), Olig2 (IBL, Japan), GST- (Stressgen), MBP (MBL), CNP (Chemicon), NF-H (Sigma), HA (HA11, Covance), FLAG M2 (Sigma) and anti-V5 (Invitrogen) as primary antibodies. Histological procedures and immunohistochemistry. Mice were anesthetized and killed by transcardial perfusion with 4% paraformaldehyde (PFA, wt/vol) in phosphate-buffered saline (PBS). Tissues were post-fixed in 4% PFA for 16 h, embedded in paraffin in the traditional manner and sectioned at a 6-m thickness

on a microtome (Leica, RM2125RT). Standard histological methods, including hematoxylin and eosin staining, Klver-Barrera staining and silver impregnation by the modified Bielschowsky method, were performed with deparaffinized sections. For immunohistochemical staining, we used rabbit polyclonal antibody to NRDc, which was raised in our laboratory39, at 8 g ml1, mouse antibody to NF-H (Sigma, 1:200); mouse antibody to CNP (Chemicon, 1:200); rat antibody to MBP (1:500) and rabbit antibody to Olig2 (IBL, 1:200). After incubation with primary antibodies, the sections were washed in PBS and incubated with secondary antibody, Alexa Fluor 488 or 594conjugated goat antibody to mouse or rabbit IgG (Invitrogen, 1:200), for 1 h, followed by counterstaining with Nissl, Neuro Trace (Invitrogen). The stained sections were examined under a fluorescence microscope and images were processed using ImageJ v1.30 (US National Institutes of Health). Electron microscopy and morphometry. Mice were anesthetized and killed via transcardial perfusion with 4% PFA and 2% glutaraldehyde (vol/vol) in 0.1 M PBS. A block of approximately 1 1 2 mm3 was removed from the body of the corpus callosum at the level of the dorsal hippocampus and from the sciatic nerve, incubated for 2 h at 4 C in the same fixative, and contrasted with 1% osmic acid (vol/vol) in PBS. Tissues were dehydrated in an ethanol gradient from 50100% and embedded in Epon. Semi-thin sections (0.9 m) were stained with toluidine blue for survey by light microscopy. Ultra-thin sections (80 nm) were cut and stained with 2% uranyl acetate (vol/vol, Watsons modified method) and Reynolds lead citrate, and analyzed with a Hitachi H-7650 transmission electron microscope. Non-overlapping digitalized images of fiber cross-sections were obtained and analyzed using Image Pro Plus 3.0 software (Media Cybernetics). The g ratio was determined by dividing the diameter of an axon (without myelin) by that of the total fiber diameter (axon + myelin sheath) from 3001,000 fibers in the corpus callosum (at least six animals per genotype). Behavioral analysis. Age-matched littermates of Nrd1+/+ (n = 12) and Nrd1/ (n = 6) mice (P90180) were analyzed with a comprehensive behavioral test battery including the grip strength test, wire-hanging test, rotarod test, beam test, hot-plate test and T-maze tests. Mice were housed in groups of three to four in clear plastic cages maintained in a room on a 12-h light/dark schedule with food and water provided ad libitum, unless specified otherwise. All experiments were performed in the light phase of the circadian cycle between 10:00 a.m. and 4:00 p.m. All behavioral testing procedures were approved by the Animal Care and Use Committee of Kyoto University. Grip strength and wire-hanging tests. Neuromuscular strength was examined by the grip strength and wire-hanging tests. A grip-strength meter (OHara & Co) was used to assess forelimb grip strength. Mice were lifted and held by their tail so that their forepaws could grasp a wire grid. The mice were then gently pulled backward by the tail with their posture parallel to the surface of the table until they released the grid. The peak force applied by mouse forelimbs was recorded in newtons. Each mouse was tested three times and the greatest value measured was used for the statistical analysis. In the wire-hanging test, mice were placed on a wire mesh that was then inverted and waved gently, so that the subject gripped the wire. Latency to fall was recorded, with a 60-s cut-off time. Rotarod test, beam test and hot-plate test. Motor coordination and balance were assessed with a rotarod test and a beam test. We performed the rotarod test using an accelerating rotarod (UGO Basile Accelerating Rotarod) by placing a mouse on a rotating drum (3 cm in diameter) and measuring the time that each mouse was able to maintain its balance on the rod as latency time to fall (s). The speed of the rotarod was accelerated from 4 to 40 rpm over a 5-min period. A beam test was performed to measure the ability of mice to traverse a narrow beam to reach a dark box29. The beam, with a rough painted surface, consisted of iron (100 cm long, 2.8 cm in diameter) placed horizontally 50 cm above the bench surface. One session of five trials was performed using the beam. Mice were allowed up to 60 s to traverse each beam. The number of sideslips was recorded for each trial by the Image OF program. The hot-plate test was used to evaluate nociception or sensitivity to a painful stimulus. Mice were placed on a hot plate at 55.0 0.3 C (Columbus Instruments) and the latency to the first hind-paw response was recorded. The hind-paw response was either a foot shake or a paw lick.

2009 Nature America, Inc. All rights reserved.

nature NEUROSCIENCE

doi:10.1038/nn.2438

T-maze forced-alternation task. The forced-alternation task was conducted using an automatic T maze that we devised30 (OHara & Co). It was constructed of white plastics runways with 25-cm-high walls. The maze was partitioned off into six areas by sliding doors that could be opened downward. The details of the protocol were described previously30. After the adaptation session, mice were subjected to a forced-alternation protocol for 7 d (one session consisting of ten trials per day, cut-off time was 50 min). Data acquisition, control of sliding doors and data analysis were performed using Image TM software (see below). T-maze left rightdiscrimination task. The left-right discrimination task was conducted using an automatic T maze and deprivation before the trials as mentioned above and previously30. On the day after the adaptation session, mice were subjected to a left rightdiscrimination task for 20 d (one session consisting of ten trials, two sessions per day, cut-off time of 50 min). The mouse was able to freely choose either the right or left arm of the T maze (A1 and A2). The correct arm was assigned to each mouse randomly. If it chose the correct arm, the mouse received a reward at the end of the arm. Choosing the incorrect arm resulted in no reward and confinement to the arm for 10 s. After the mouse consumed the pellet or the mouse stayed more than 10 s without consuming the pellet, the door that separated the arm (A1 or A2) and connecting passage-way (P1 or P2) was opened and the mouse could return to the starting compartment (S1) via the connecting passage-way. On the 11th day, the correct arm was changed for reversal learning. A variety of fixed extra-maze clues surrounded the apparatus. Image analysis. The applications used for the behavioral studies were based on the public domain NIH Image program (developed at the US National Institutes of Health and available on the Internet at http://rsb.info.nih.gov/nih-image/)

and ImageJ program (http://rsb.info.nih.gov/ij/), which were modified for each test by T.M. (available through OHara & Co). RNA analysis. Total brain RNA was isolated from snap-frozen brains (at least three mice per genotype, at P5) using Trizol (GIBCO-BRL) according to the manufacturers protocol. For real-time PCR, first strand cDNA was synthesized from total RNA using random primers and SuperScriptIII reverse transcriptase (Invitrogen). Quantitative real-time PCR was carried out using the ABI Prism 7700 Sequence Detection System and SYBR Green Master Mix following the manufacturers directions (Applied Biosystems). The results were standardized for comparison by measuring levels of Actb mRNA in each sample. Values are expressed as mean s.e.m. The mouse-specific primers that we used for real-time PCR were Nrg1 type I (5 primer, CTG TGT CTG CCT GGA A; 3 primer, CAG CCG TTG GAT CCA GA), Nrg1 type III (5 primer, GCT CTA CCA GAA GAG GGT A; 3 primer, GAT TCA CCA GTT GCA CA), Bace1 (5 primer, ACC ATC CTT CCT CAG CAA; 3 primer, GGG AAT GTT GTA GCC ACA) and Actb (5 primer, CTG ACT GAC TAC CTC ATG AAG ATC CT; 3 primer, CTT AAT GTC ACG CAC GAT TTCC). Statistical analysis. Data are presented as mean s.e.m. Statistical analyses (Students two-tailed t test, one-way ANOVA, repeated-measures ANOVA) were performed using the Statview software package (SAS institute).
37. Yagi, T. et al. A novel ES cell line, TT2, with high germline-differentiating potency. Anal. Biochem. 214, 7076 (1993). 38. Todaro, G.J. & Green, H. Quantitative studies of the growth of mouse embryo cells in culture and their development into established lines. J. Cell Biol. 17, 299313 (1963). 39. Hospital, V. et al. The metalloendopeptidase nardilysin (NRDc) is potently inhibited by heparin-binding epidermal growth factorlike growth factor (HB-EGF). Biochem. J. 367, 229238 (2002).

2009 Nature America, Inc. All rights reserved.

doi:10.1038/nn.2438

nature NEUROSCIENCE

a r t ic l e s

Vascular niche factor PEDF modulates Notch-dependent stemness in the adult subependymal zone
Celia Andreu-Agull1,2, Jos Manuel Morante-Redolat1, Ana C Delgado1 & Isabel Farias1
2009 Nature America, Inc. All rights reserved. We sought to address the fundamental question of how stem cell microenvironments can regulate self-renewal. We found that Notch was active in astroglia-like neural stem cells (NSCs), but not in transit-amplifying progenitors of the murine subependymal zone, and that the level of Notch transcriptional activity correlated with self-renewal and multipotency. Moreover, dividing NSCs appeared to balance renewal with commitment via controlled segregation of Notch activity, leading to biased expression of known (Hes1) and previously unknown (Egfr) Notch target genes in daughter cells. Pigment epitheliumderived factor (PEDF) enhanced Notch-dependent transcription in cells with low Notch signaling, thereby subverting the output of an asymmetrical division to the production of two highly self-renewing cells. Mechanistically, PEDF induced a non-canonical activation of the NF- B pathway, leading to the dismissal of the transcriptional co-repressor N-CoR from specific Notch-responsive promoters. Our data provide a basis for stemness regulation in vascular niches and indicate that Notch and PEDF cooperate to regulate self-renewal. Asymmetric division preserves adult stem cell pools while ensuring tissue renewal. However, individual stem cells appear to be capable of undergoing symmetric divisions to generate two stem (expansionary self-renewing divisions) or two committed (differentiative divisions) cells, and the balance between these division modes probably regulates the dynamics of stem-cell reservoirs1. Some intrinsic determinants are known to regulate division mode, but stem cells can respond to excessive cellular demand after injury, suggesting that their self-renewing behavior can be modulated by external signals1,2. In the specialized stem-cell microenvironments, vascular elements appear to be important for the regulation of stem cell self-renewal versus commitment both under normal and pathological conditions, but the signaling pathways involved are still under investigation3. In the subependymal zone (SEZ) of the adult mammalian brain, persistent and relatively quiescent astroglia-like neural stem cells (NSCs) support life-long production of neuroblasts and oligodendrocytes that are destined to go to the olfactory bulb and the corpus callosum, respectively, through the production of fast-dividing transitamplifying progenitors (TAPs)3,4. Clusters of neurogenic progenitors intimately apposed to brain capillaries suggest that vascular elements are an essential feature of adult NSC microenvironments57 and several endothelium-derived factors are known to regulate proliferation and/or survival of neural progenitors3. Indirect coculture experiments have shown that capillaries are also a source of NSC self-renewalpromoting molecules8 and we have recently reported that PEDF (also known as Serpin-F1) is, at least in part, responsible for these effects9. The secreted glycoprotein PEDF can promote tumor cell differentiation and neuronal survival and is a highly potent anti-angiogenic factor10, in addition to acting as an endogenous SEZ niche factor that promotes NSC expansionary divisions9. Although a molecule with phospholipase activity can seemingly bind PEDF in some cells11, the identity of the receptor(s) that could mediate the various PEDF effects, including those relating to NSC self-renewal, remains unknown. Notch signaling is important for the maintenance of stem cells in various niches12,13 and, notably, NSCs treated with PEDF have upregulated expression of Notch-downstream effectors Hes1 and 5 of the HES family of basic helix-loop-helix transcription factors8,9. Activation of Notch (14 in mammals) receptors by membranebound ligands (Jagged1/2 or Delta 14) is followed by regulated cleavage by a presenilin-secretase complex. The generated intracellular domain of Notch (NICD) then moves to the nucleus, where it binds the repressor C promoterbinding factor 1 (CBF1, also known as RBP-J and CSL) to initiate transcription of genes of the Hes and Herp families13,14. HES transcription factors negatively regulate the expression of pro-neurogenic genes and Notch signaling therefore enhances self-renewal by antagonizing differentiation, although it can also regulate proliferation and survival1518. Studies in transgenic Notch activityreporter (TNR) mice have shown that high CBF1 activity distinguishes stem cells from committed progenitors2,19,20 and lossof-function experiments have demonstrated that downregulation of Notch activity is required for the transition from radial glia/astroglialike cells to committed progenitors20,21. Putative differences in Notch activity levels in SEZ cells are unknown, but Notch activity is required to sustain NSC maintenance in this niche1517. Here, we sought to investigate potential interactions of PEDF with the Notch pathway. We found that subependymal cells that had reporter activation in TNR mice were contained in the population

1Departamento de Biologa Celular and CIBER en Enfermedades Neurodegenerativas, Universidad de Valencia, Burjassot, Spain. 2Present address: Sloan-Kettering Institute, Department of Developmental Biology, New York, New York, USA. Correspondence should be addressed to I.F. (isabel.farinas@uv.es).

Received 20 July; accepted 17 September; published online 8 November 2009; doi:10.1038/nn.2437

1514

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 1 PEDF interaction with Notch signaling regulates Notchdependent transcription and neurosphere formation. ( a) Top, Hes1 promoter activity in the absence or presence of PEDF for 24 h in cells transiently expressing luciferase reporter constructs that either contained (Hes1-luc) or lacked (HesAB-luc) CBF1-binding sites (n = 4). Bottom, activity of a specific artificial CBF1-dependent reporter that either contained consensus CBF1 recognition sequences ( 4xwtCBF1-luc) or a mutated form (4xmutCBF1-luc) in the absence or presence of PEDF for 24 h (n = 3). The effects of PEDF on CBF1-mediated transcription after genetic knockdown of Notch signaling by overexpression of a dominant negative form of CBF1 (Su(H)DBM) are shown. (b) NICD detection assessed by immunoblotting in untreated or PEDF-treated neurospheres (left) or after treatment with the -secretase inhibitor (-SI) L-658,458 (right). The table shows a quantification of the blots relative to endogenous -actin levels (n = 3). Full-length blots are presented in Supplementary Figures 7 and 11. (c) The effects of PEDF on Hes1 promoter activation after treatment with -SI in cells nucleofected with an empty vector or with a constitutively active NotchIC fragment ( n = 4) are shown. (d) Number of neurospheres generated from cells that were infected with pMXIE or pMXIE-NotchICcarrying retroviruses in the presence of PEDF, -SI or both (n = 4). (e) Fold-change in the number of secondary neurospheres formed by cultures treated with 20 ng ml 1 PEDF, 2 g ml1 Jagged1 (Jag1) or both (PEDF + Jag1), relative to those formed in the untreated condition (n = 4). Data are expressed as mean s.e.m. * P < 0.05 and ** P < 0.01.
Notch-dependent transcription

U nt re a

D M SO
NICD -actin

D F

te d

Hes1-luc

b
*
Untreated PEDF

SI
NICD -actin DMSO PEDF -SI + PEDF

HesAB-luc

0 1 2 3 4 5 6 7 Luciferase activity (au) 4xwtCBF1-luc

NICD to -actin ratio Untreated 1.51 0.40 PEDF 1.39 0.21

**
4xmutCBF1-luc 4xwtCBF1-luc 0 20 40 60 Luciferase activity (au) Untreated PEDF

c
Empty vector NotchlC 0

PE
Hes1-luc

Su(H)DBM

* *
2 4 6 8 10 12 14 Luciferase activity (au)

d
Fold-increase relative to control GFP 4 3 2 1 0

Neurosphere formation DMSO PEDF -SI -SI + PEDF

Secondary neurospheres 4 3 2 1 0

Fold-increase relative to untreated

2009 Nature America, Inc. All rights reserved.

* * *

n.s.

** * *

of astroglia/radial glia-like NSCs and that the level of Notch activity correlated with neurosphere formation potential and multipotency. Moreover, we found that different levels of Notch activation could be inherited by the two daughter cells of a neurosphere-forming cell during its first mitotic division and that PEDF intensified Notchdependent transcription in cells with low levels of Notch activation. Our biochemical data indicate that PEDF induced a non-canonical activation of the NFB pathway leading to the dismissal of nuclear receptor co-repressor (N-CoR) from CBF1 sites in specific promoters, resulting in cells that were more undifferentiated and prone to mitogenic stimulation. RESULTS PEDF enhances Notch activity in adult NSCs Because our previous observations indicate that PEDF increases NSC self-renewal and upregulates Hes1 and Hes5 expressions9, we investigated possible interactions with Notch. PEDF activated a Hes1luciferase reporter, but not a reporter lacking CBF1 sites (HesAB-luc)22, in neurosphere-forming cells; similar results were obtained with CBF1 reporters (Fig. 1a). Expression of a dominant-negative form of CBF1 (Su(H)DBM) or pharmacological inhibition of the -secretase activity (Fig. 1b) essentially abolished the effects of PEDF on Cbf1- (also known as Rbpj) and Hes1-luc reporters, respectively, but -secretase inhibition could not block the effects of PEDF on Hes1 transcription in cells expressing a constitutively active Notch intracellular fragment (NotchIC17) (Fig. 1a,c), suggesting that PEDFs effects require an active Notch. The effects of PEDF on neurosphere formation were also dependent on Notch signaling, as overexpression of Su(H)DBM (Supplementary Fig. 1) or -secretase inhibition (Fig. 1d) prevented PEDF from increasing neurosphere numbers. However, PEDF did not promote Notch cleavage. NICD has been detected in neurospheres in the absence of added Notch ligands23 and PEDF did not increase its level (Fig. 1b). Our results indicate that PEDF is not a Notch ligand, but can enhance Notchs transcriptional effects. To test whether PEDF and Notch cooperated to regulate selfrenewal, we seeded single cells in the presence of the Notch ligand Jagged1 and/or PEDF, dissociated the neurospheres after 4 d and plated the isolated cells in mitogens alone. Pre-stimulation with
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

p N MX ot IE ch IC

1 ed gg PE D F +

XI

pM

PE

Retroval infection

Pre-treatment

Jagged1 or PEDF alone increased self-renewing divisions, as the number of secondary neurospheres formed was higher than in the untreated condition. However, a more robust increase was obtained when the cells were treated with a combination of both molecules (Fig. 1e). Therefore, PEDF increased neurosphere formation and Hes1 transcription in concert with Notch, but appeared to be incapable of improving neurosphere formation when Notch signaling was very high (Fig. 1d). We next analyzed Notch activity in the SEZ of TNR mice in which enhanced green fluorescent protein (EGFP) expression is under the control of four CBF1-responsive elements 2,19,20. Immunohistochemistry revealed that 1520% cells had detectable levels of NICD and different levels of EGFP intensity (EGFPhi and EGFPlo cells; Fig. 2a). A few EGFP-positive cells were ependymocytes, identified by position, morphology and S100 immunostaining; all of them had high CBF1 activity ( Fig. 2b). However, 95% of the EGFP-positive cells were subependymal; virtually all these were positive for the astrocyte-specific l-glutamate/l-aspartate transporter (GLAST)24, but not for the astrocyte terminal differentiation marker S100, and the vast majority were also Sox2 and Ki67 positive (Fig. 2b,c). Consistently, subependymal EGFP-positive cells were rarely Mash1 or doublecortin positive (Fig. 2c and Supplementary Fig. 2). Not all of the GLAST and Sox2 double-positive cells exhibited CBF1 activity, as 22.2 5.8% were EGFPhi, 66.9 6.0% were EGFPlo and 39.2 5.9% were EGFP negative. In contrast, all of the BrdU label-retaining cells (BrdU-LRCs) were EGFP positive (Fig. 2d). These data indicate that Notch activity is highly restricted to GLAST, Sox2 and Ki67 triple-positive BrdU-LRCs and adult NSCs exhibit different levels of Notch activity. To test whether PEDF could activate Notch-dependent transcription in vivo, we infused PEDF for 24 h into the lateral ventricle of TNR mice injected with BrdU 3 weeks prior. The proportion of EGFPpositive, GLAST-negative cells did not increase after the infusion (4.0 0.9% in PEDF versus 5.4 0.5% in saline infusions, n = 3).
1515

Ja

Ja

g1

a r t ic l e s a
Merged NICD EGFP

b
S100/EGFP * *

St

LV

SEZ

EGFPhi Percentage of SEZ cells EGFPhi 3.4 0.5 EGFPlo 15.4 0.1 EGFPhi 38.4 3.5% 95.6 1.4%

4.4 1.3%

EGFP

Mash1

Merged

2009 Nature America, Inc. All rights reserved.

EGFPlo

EGFPhi

EGFP

Ki67

Merged

EGFP

BrdU-LRC

GLAST

Merged

Non-ependymal GLAST Sox2 Ki67 S100 Mash1 Saline

EGFPhi 94.4 88.5 93.6 0.0 1.2 5.1 7.0 10.5 0.3 PEDF *

EGFPlo 91.4 87.8 89.2 0.0 2.8 9.7 5.7 7.7 0.7 60 50 40 30 20 10 0

GLAST, Sox2 and Ki67 positive BrdU-LRCs EGFPhi EGFPlo EGFP 51.4 3.5 49.1 3.5 0.0

e
EGFP/GLAST/Sox2

Percentage of GLAST+/Sox2+

h
** ** Saline PEDF Saline

EGFP/GLAST/BrdU-LRC

* * *

g
Percentage of BrdU-LRCs

EGFPhi EGFPlo EGFP * Saline PEDF ** EGFPhi EGFPlo PEDF

100 80 60 40 20 0

However, we found more GLAST and Sox2 double-positive cells with high levels of EGFP, at the expense of EGFPlo, but not EGFP-negative cells, in PEDF- versus saline-infused mice (Fig. 2e,f). We also detected more EGFPhi cells and fewer EGFPlo BrdU-LRCs (Fig. 2g,h). These results are consistent with our molecular data showing that PEDF enhanced low levels of Notch transcriptional activity. PEDF and Notch together promote stemness and multipotency To investigate the neurosphere-forming potential of EGFP-positive cells, we seeded EGFPhi and EGFPlo fluorescence-activated cell sorting (FACS)-sorted cells at low density (Fig. 3ac). More than 75% of the EGFPhi cells, but only around 20% of the EGFPlo cells, formed a neurosphere after 4 d (Fig. 3b,c). These differences were not a result of distinct survival (Supplementary Fig. 3), but of differential mitogenic activation; over 85% of the EGFPhi cells had initiated a neurosphere, whereas more than 50% of the EGFPlo cells remained single and viable 48 h after plating (Supplementary Fig. 3). EGFP-negative cells survived poorly, did not divide during the 48-h period and formed almost no neurospheres (Supplementary Fig. 3). Therefore, CBF1 activity correlates with NSC activation. EGFP hi cells formed larger neurospheres and maintained high subcloning efficiency, whereas EGFP lo cells formed smaller neurospheres and their capacity to self-renew was lower and rapidly decreased in subsequent passages ( Fig. 3b , d and Supplementary Fig. 4 ). To investigate whether the level of Notch
1516

activity also correlated with multipotency, we differentiated single neurospheres ori ginating from EGFP hi or EGFP lo cells and scored clones containing different cell lineages after immunostaining for O4-positive oligodendrocytes, GFAP-positive astrocytes and III-tubulin positive neurons ( Fig. 3e ). We analyzed multipotent versus astrocyte-only clones and found that EGFP hi cells were substantially more multipotent than EGFP lo cells ( Fig. 3f ). We obtained the same results when cells were seeded clonally ( Supplementary Fig. 5 and Supplementary Fig. 6 ). Thus, the level of Notch activation positively correlates with neurosphere potential, duration of self-renewal and multipotency. To further analyze the interactions between Notch activity and PEDF, we cultured EGFPhi and EGFPlo cells with PEDF (primary) and subcultured them for two passages in the absence of PEDF (secondary and tertiary passages). PEDF increased mitogenic activation of EGFP lo cells without altering their survival (Supplementary Fig. 3), resulting in higher numbers of spheres in low-density and clonal cultures (Fig. 3c and Supplementary Fig.6). Furthermore, PEDF-treated EGFP lo cellderived neurospheres generated higher numbers of secondary and tertiary neurospheres on withdrawal of PEDF ( Fig. 3d ). Likewise, treatment with PEDF increased the frequency of tripotent EGFPlo cellderived neurospheres at the expense of astrocyte-only clones ( Fig. 3e , f and Supplementary Fig. 6 ). These results indicate that PEDF can increase self-renewal potential and plasticity in cells with low levels of CBF1 activity.
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

EGFP/GLAST/Sox2

Non ependymal cells S100 GLAST+ EGFPlo 61.7 4.5

Ependymal cells S100+ GLAST

Figure 2 Notch signaling is restricted to GLAST and Sox2 double-positive cells and can be increased by PEDF infusion. (a) Immunofluorescence for EGFP and NICD in the SEZ of TNR mice with EGFPhi and EGFPlo cells, with both types containing NICD. The percentage of each cell type is shown in the table (n = 5). LV, lateral ventricle; St, striatum. (b) EGFP immunostaining revealed that there was high CBF-1 activity in S100-positive ependymal cells. Right, immunofluorescence for GLAST, Sox2, and EGFP in TNR mice. Pie chart represents the distribution of ependymal and non-ependymal cells that were EGFP positive (n = 5). (c) EGFPhi and EGFPlo cells were Ki67 positive, but not Mash1 positive. The table contains the quantification for different markers in the subependymal EGFPhi and EGFPlo cell populations (percentage of cells that were marker positive, n = 5). Non-ependymal EGFP-positive cells in the SEZ of TNR mice were GLAST, Sox2 and Ki67 positive. (d) All of the cells retaining BrdU that was injected 3 weeks before the mice were killed (BrdU-LRCs) had CBF-1 activity (EGFP positive). The percentage of each EGFP-positive type is reflected in the table (n = 5). (eg) PEDF infusion for 24 h increased the percentage of GLAST and Sox2 doublepositive cells (e,f) or GLAST-positive BrdU-LRCs (g,h) with high CBF-1 activity at the expense of those with low CBF1 activation in TNR mice. In all cases, white arrowheads indicate EGFhi cells, empty arrowheads indicate EGFPlo cells and asterisks indicate EGF-negative cells. Data are expressed as mean s.e.m. * P < 0.05 and ** P < 0.01. Scale bars represent 20 m (a,b,e) and 10 m (c,d,g).

a r t ic l e s
Figure 3 PEDF regulates self-renewal in concert with Notch transcriptional activity. (a) Representative FACS plot showing collection gates for viable EGFPhi and EGFPlo cells from TNR mice neurospheres. (b) Images of EGFPhi and EGFPlo sorted cells (fluorescence and phase contrast) and of derived neurospheres. (c) Quantification of the clonogenic capacity (as percentage of neurospheres formed relative to seeded cells at limiting dilution) of FACSsorted EGFPhi and EGFPlo cells in the presence or absence of PEDF (n = 8). (d) Quantification of secondary and tertiary neurospheres in the absence of PEDF produced by cells from PEDFgrown primary neurospheres. (e) Representative images showing immunofluorescence for astrocytes (A, GFAP positive), neurons (N, neuronal III-tubulin positive) and oligodendrocytes (O, O4 positive) in isolated neurospheres differentiated for 7 d. (f) Distribution of unipotent (A), bipotent (A/N or A/O) and tripotent (ANO) clones among neurospheres derived from EGFPhi and EGFPlo cells (grown in the presence or in the absence of PEDF) differentiated for 7 d (n = 3 experiments from independent mice, at least 50 clones analyzed in each case). (g) PEDF treatment increased the number of self-renewing cells. FACS-sorted EGFPhi cells treated with PEDF generated neurospheres that contained more EGFPhi cells (FACS plots on the left) with high self-renewal capacity (right) (n = 3). Data are expressed as mean s.e.m. * P < 0.05, ** P < 0.01 and *** P < 0.001. Scale bars represent 100 m (b) and 50 m (f).

a
256 192 128 Cell granularity

EGFPlo

EGFPhi

b
EGFPhi

Single cells

Primary neurospheres

0 100

Percentage of neurospheres formed

c
Percentage of neurospheres formed 100 80 60 40 20 0

101 102 EGFP intensity

103

EGFPlo

64

100 80 60 40 20 0 Secondary Tertiary Secondary Tertiary EGFPlo

* * ** ***

***

Untreated PEDF

EGFPhi

EGFPlo

EGFPhi Percentage of clones

2009 Nature America, Inc. All rights reserved.

Tuj1/GFAP/O4 ANO

100 80 60 40 20 0 Untreated PEDF Untreated PEDF

**

***

ANO AN/AO A

AO

Cell granularity

g
EGFPhi-derived neurospheres

EGFPhi EGFPhi Untreated 64

EGFPlo Percentage of neurospheres formed 0 20 40 60

0 PEDF treated 64 From untreated From PEDF

In contrast, PEDF increased neither the survival nor the potential of EGFPhi cells to initially divide, produce primary spheres or differentiate to multiple lineages at either low or clonal cell density (Fig. 3cf and Supplementary Fig. 6). PEDF-treated EGFPhi cellderived neurospheres did yield more secondary clones (Fig. 3d), suggesting that PEDF enhances Notch activity to maintain an immature phenotype and self-renewal potential in the daughters of EGFPhi cells. Consistently, we observed that neurospheres formed by EGFP hi cells in the presence of PEDF contained more cells with high CBF1reporter activity and self-renewal than untreated neurospheres (Fig. 3g). These results suggest that PEDF maintains/induces CBF1 activity in dividing NSCs, leading to increased self-renewal and multidifferentiation potential. PEDF and Notch modulate NSC division outcomes Because PEDF increases self-renewing divisions in concert with Notch, we decided to study Notch activity during the first division of a NSC in vitro using a cell pair assay. Single cells were seeded for 24 h and immunostained with antibodies to NICD (specificity controls in Supplementary Fig. 7). In approximately 45% of the cell pairs, one daughter cell had high levels of NICD and the other had very low levels of NICD (NICDhi/lo), whereas equivalent high levels of NICD were observed in the rest of the cell pairs (NICDhi/hi) (Fig. 4a,b). Thus, Notch activity can be symmetrically or asymmetrically distributed. The proportion of NICDhi/lo and NICDhi/hi cell pairs did not change in the presence of PEDF (Fig. 4b), which is consistent with the observation that PEDF does not activate Notch upstream of CBF1.
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

Cell granularity

0 100

102 101 EGFP intensity

The epidermal growth factor receptor (EGFR) can also be ymmetrically or asymmetrically distributed in the progeny of dividing s fetal neural progenitors25. In adult cultures, 5055% of the cell pairs contained daughters with similar levels of the receptor (EGFRhi/hi), whereas one daughter cell expressed high levels and the other one expressed very low levels of immunofluorescence (EGFR hi/lo) in the remainder of the cell pairs (Fig. 4c,d). Notably, high NICD expression always correlated with high EGFR levels. This result, together with our observation that cells with high Notch activity are more prone to divide in EGF (Supplementary Fig. 3), suggests that Notch might regulate EGFR expression. We therefore analyzed whether responses to EGF could be under the cooperative control of Notch and PEDF. Overexpression of NotchIC in our cells substantially activated transcription of an Egfr-luc reporter (containing the 5 UTR and 980 bp of the proximal promoter; Fig. 4e), whereas overexpression of Hes1 did not (relative luciferase activity s.e.m., 2.30 0.52 in Egfr-luc and 2.82 0.87 in Egfr-luc and Hes1transduced cells, n = 3). Despite previous evidence that NICD does not activate the Egfr-luc reporter in a fibroblast cell line26, in silico analysis identified two putative CBF1-binding sites in the Egfr proximal promoter (Fig. 4f). We therefore performed chromatin immunoprecipitation (ChIP) experiments to examine the physical interaction between NICD and CBF1 proteins and the putative CBF1-containing
1517

a r t ic l e s a
Untreated NICDhi/hi DAPI Merged

b
NICDhi/hi 80 Percentage of cell pairs 60 40 20 0 Untreated PEDF EGFRhi/hi EGFRhi/lo 100 Percentage of cell pairs Untreated PEDF NICDhi/lo NICD negative

Merged

c
Untreated

EGFRhi/lo

NICDhi/lo

DAPI/Merged

d
80 60 40 20 0 NICDhi/hi NICDhi/lo

EGFRhi/hi

NICDhi/hi

DAPI/Merged

NICDhi/hi NICDhi/lo

EGFRhi/hi

NICDhi/lo

DAPI/Merged

f
580 1 kb 307 254 109 +1 +219 242 Egfr ATG

2009 Nature America, Inc. All rights reserved.

e
Empty vector NotchIC 0

Egfr-luc

Untreated PEDF

g
Input IgG CBF1 Input Serum NICD Egfr (580 to 254) Egfr (109 to 242) Hes1 (176 to +13)

* *
8 10 2 4 6 Luciferase activity (au) 12

Figure 4 PEDF and Notch regulate EGFdependent mitogenic response in adult NSCs. (a) Representative cell pairs resulting from a divided NSC immunostained for endogenous NICD showing NICDhi/hi (symmetrical) and NICDhi/lo (asymmetrical) distributions. Arrowhead indicates low levels of NICD. (b) PEDF treatment did not modify NICD distribution in cell pairs (n = 4, no less than 50 cell pairs each). (c) Immunofluorescence for EGFR and NICD in cell pairs showing symmetrical (hi/hi) or asymmetrical (hi/lo) distributions. (d) PEDF increased the number of EGFRhi/hi cell pairs at the expense of EGFRhi/lo pairs (n = 4). In untreated cultures, high NICD expression correlated with high EGFR expression. In PEDF-treated cultures, daughter cells with low NICD levels could have high levels of EGFR. (e) Luciferase assays for Egfr promoter activity in NotchIC-overexpressing cells with and without PEDF (n = 4). (f) Scheme depicting two putative sites (GTG GGA G at position 307 and GTG GGA C at position +219) that are homologous to the consensus CBF1 sequence (GTG GGA A) found by in silico analysis in the proximal promoter (1 kb) and the 5 UTR of the mouse Egfr gene. Arrows represent positions of primers used for ChIP analysis. (g) Representative PCR from ChIP analyses with antibodies to NICD and CBF1 and nonrelevant rabbit IgGs or control antiserum in the Egfr promoter. Hes1 was used as a positive control (binding site between nucleotides 176 and +13) (n = 4). Full-length gels are presented in Supplementary Figure 11. Data are expressed as mean s.e.m. * P < 0.05. Scale bars represent 5 m.

PEDF

regions on the Egfr proximal promoter, using the Hes1 promoter as a positive control22. Although antibodies to NICD and CBF1 failed to bind position +219, both were able to immunoprecipitate the Egfr promoter fragment containing the CBF1 site on position 307 (Fig. 4g). Consistent with a direct regulation of EGFR transcription by CBF1, we observed increased proportions of cells with undetectable expression of EGFR in Su(H)DBM-overexpressing cultures and increased EGFR symmetric distribution in the incipient daughters of dividing NotchIC-overexpressing NSCs (Supplementary Fig. 8). As in the case of Hes1 transcription, treatment with PEDF alone or in combination with NotchIC-overexpression promoted the activation of the Egfr-luc reporter and increased the proportion of NICDhi/lo cell pairs in which the cell with very low levels of NICD had detectable levels of EGFR (Fig. 4ce). These data indicate that EGFR is a target of Notch and that PEDF can upregulate the expression of Notch-dependent stemness genes (for undifferentiation and mitogenic activation) when Notch levels are not maximal. PEDF induces a gene-specific dismissal of repressor N-CoR Notch activity relies on a complex interplay between transcriptional coactivators and co-repressors that ultimately sets the threshold of the response14. N-CoR was initially identified as a co-repressor for unliganded hormone nuclear receptors, but it can partner with a variety of transcription factors including CBF1 (refs. 2729). Analyses in mutant mice have revealed a requirement for N-CoR in the inhibition of Notch-induced astrogliogenesis in the developing brain27, but N-CoR function and regulation in adult NSCs is largely unknown. N-CoR was present in the nuclei of most dissociated neurosphere cells, but 18.2 5.2% of the cells had detectable levels of cytoplasmic
1518

N-CoR and this percentage rose in PEDF-treated cultures (36.0 3.6%, n = 4, P < 0.01; Fig. 5a). Immunoblot detection in cellular fractionates also revealed that PEDF treatment increased the cytoplasmic levels of N-CoR (Fig. 5b). Therefore, N-CoR moves from a nuclear to a cytoplasmic location in response to PEDF. Given that PEDF enhances Notch induction of Hes1 and Egfr expression, we next examined whether CBF1-binding sites in the promoters of these genes were occupied by N-CoR and whether PEDF induced the dismissal of N-CoR from these sites. In nonstimulated cells, antibodies to N-CoR immunoprecipitated genomic DNA fragments containing known CBF1-binding sites in the Hes1 promoter22 and the putative binding site at position 307 in the Egfr promoter, but not the site at +219 or control distal regions on both promoters. (Fig. 4f and Supplementary Fig. 9), suggesting that N-CoR directly represses transcription of these genes. In PEDF-treated cultures, there was a decrease in the association of N-CoR with both promoters (Fig. 5c). The Gfap proximal promoter contains one putative repressor domain harboring a conserved CBF1 consensus-binding site that can bind CBF1 and N-CoR, and CNTF-induced dismissal of N-CoR from this site promotes astrocytic differentiation of fetal progenitors27. In nonstimulated adult NSCs, N-CoR also bound the Gfap promoter, but PEDF did not cause its displacement (Fig. 5c). Consistently, PEDF treatment did not increase the levels of Gfap mRNA (Gfap to Actb ratio s.e.m.: control, 3.2 0.8; PEDF, 2.9 0.5; n = 3). ChIP analysis on the promoter of the immune-specific Ptcra gene30 revealed that N-CoR binding to CBF1 sites was not modified by PEDF treatment (Supplementary Fig. 9). These data indicate that PEDF has a role in the modulation of Notch dependent genes that differs from that of other N-CoR regulators.
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 5 PEDF regulates Notch-dependent transcription and self-renewal by inducing the dismissal of N-CoR. (a) N-CoR immunoreactivity and nuclear DAPI staining in untreated cells or after 24 h of PEDF treatment. Arrowheads indicate cytoplasmic N-CoR labeling (n = 4). (b) Western blot from nuclear and cytoplasmic extracts of cells treated with PEDF for 24 h. PEDF treatment increased the amount of N-CoR in the cytoplasm at expense of nuclear N-CoR. Numbers indicate the ratio between the treated and untreated conditions. Specificity controls for N-CoR and phospho-histone H3 antibodies are presented in full-length immunoblots in Supplementary Figure 11. (c) ChIP assays in untreated and PEDF-treated cells using nonrelevant rabbit IgGs or antibodies to N-CoR and primers flanking the CBF1-binding sites in Egfr, Hes1 and Gfap promoters (n = 4). Uncropped gel images are presented in Supplementary Figure 11. (d,e) Overexpression of wild-type N-CoR, but not of its repression-dead form (HANcor1-C), completely abolished PEDF-induced Hes1 and EGFR transactivation in the presence of endogenous levels of NICD and in NotchICoverexpressing cells (n = 4). (f) Overexpression of N-CoR, but not of its CBF1-binding mutant form, abrogated the effects of PEDF on selfrenewal (n = 4). (g) Colocalization of endogenous N-CoR and EGFR in cell pairs formed in the absence or presence of PEDF for 24 h. Symmetric distribution of EGFR after PEDF treatment largely correlated with N-CoR cytoplasmic distribution (n = 4). Data are expressed as mean s.e.m. * P < 0.05 and ** P < 0.01. Scale bars represent 10 m (a) and 5 m (g).

a
Untreated

N-CoR/DAPI/Merged

Nucleus +

Cytoplasm +

PEDF N-CoR Phospho histone 3 -actin

PEDF

1 Untreated PEDF

0.62 Hes1-luc

1.67
pMXIE pMXIE + N-CoR pMXIE-NotchIC pMXIE-NotchIC + N-CoR pMXIE-NotchIC + HA-N-CoR C

c
In p

d
te d U nt re a
Egfr (580 to 254) Hes1 (176 to +13)

N -C oR In pu t Ig G

N -C oR

ut

* ** **
PE

Ig G

e
re at ed nt

Egfr-luc

pMXIE pMXIE + N-CoR pMXIE-NotchIC pMXIE-NotchIC + N-CoR pMXIE-NotchIC + HA-N-CoR C

D F

Gfap (265 to 147)

**
0 2 4 6 8 10 12 14 Luciferase activity (au) N-CoR/EGFR/DAPI

** **

g
Untreated

2009 Nature America, Inc. All rights reserved.

N-CoR

EGFRhi/lo

Merged

PE

**
0 2 4 6 8 10 Luciferase activity (au) 12 Empty vector N-CoR HAN-CoR C

N-CoR

EGFRhi/hi

Merged

Neurosphere number

120 80 40 0

N-CoR PEDF

EGFRhi/hi

Merged

Untreated

PEDF

Consistent with an N-CoRdependent regulation of Hes1 and EGFR transcription, overexpression of full-length Ncor1, but not of its repression-dead form (a truncated HANcor1-C lacking CBF1 sites)27, completely blocked the positive effects elicited by NotchIC overexpression or PEDF treatment on Hes1 and Egfr reporters (Fig. 5d,e). Overexpression of Ncor1, but not HANcor1 C, also blocked the increased clonogenic capacity induced by NotchIC (Supplementary Fig. 9) or PEDF (Fig. 5f). These results indicate that PEDF-induced N-CoR redistribution enhances Notch transcriptional activity at the Hes1 and Egfr genes, leading to self-renewal. Consistent with this model, 78.3 3.3% of the EGFRhi/hi cell pairs in untreated cultures, but only 39.5 9.6% of the EGFRhi/hi cell pairs in PEDFtreated cultures, expressed nuclear N-CoR (Fig. 5g). PEDF-induced dismissal of N-CoR is NF-B/p65 dependent Although little is known about the intracellular pathways activated by PEDF, its pro-survival effects are mediated by NF-B31,32. Because the NF-B subunit p65 is expressed by GFAP-positive SEZ cells33, we investigated whether this pathway was involved in signaling PEDF effects on NSCs. In nonstimulated cells, silent NF-B complexes reside in the cytoplasm in combination with inhibitory proteins of the IB family, but, following degradation of IB proteins induced by stimuluscoupled phosphorylation, p65 translocates to the nucleus, where it activates transcription34. We were able to abolish the PEDF-induced activation of Hes1 and Egfr promoters in NotchIC-transduced NSCs with the addition of the NF-B pharmacological inhibitor BAY117082 (Fig. 6a). PEDFs effects on neurosphere formation were also abrogated in the presence of BAY11-7082 (data not shown) and in cells overexpressing the constitutive repressor of NF- B,
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

SR-IB(3236)(ref. 35) (Fig. 6b). Therefore, activation of NF-B is required for PEDFs effects in Notch-competent NSCs. More specifically, the actions of PEDF on neurosphere formation were abrogated in cells transduced with a p65NF-B mutant that lacked the nuclear export signal and the transactivation domain (p65NES)36. Ectopic expression of a p65NF-B mutant that only lacked the transactivation domain (p65TA)36, however, did not block the increase in neurosphere number (Fig. 6b). Consistent with the suggestion that PEDF actions require p65 nuclear export, PEDFs effects on neurosphere formation were blocked with the addition of the pharmacological inhibitor of nuclear export leptomycin B (LMB; fold-increase in neurosphere number relative to EGF + fibroblast growth factor 2 s.e.m.: PEDF, 1.48 0.03; PEDF+LMB, 0.95 0.03; n = 3, P < 0.01). Thus, PEDFs actions on self-renewal require nucleocytoplasmic shuttling of p65, but not p65-dependent transcription of NFB-target genes. Using simultaneous immunofluorescent detection of endogenous p65 and N-CoR in PEDF-treated NSCs, we found a temporal correlation in their redistribution to the cytoplasm (Fig. 6c). Moreover, overexpression of p65NES resulted in a persistent N-CoR nuclear signal even in the presence of PEDF ( Fig. 6d), suggesting that p65 export is functionally required for PEDF-induced N-CoR redistribution. Immunoprecipitation experiments with antibodies to N-CoR followed by immunoblot for p65 in cytoplasmic and nuclear fractions revealed a physical interaction between p65 and N-CoR in both subcellular fractions in control conditions; treatment with PEDF, however, decreased the amount of p65 that could be recovered from the nucleus by N-CoR immunoprecipitation and increased the levels of cytoplasmic N-CoR/p65 complexes (Fig. 6e). Thus, p65 forms a
1519

a r t ic l e s a
NotchIC Hes1-luc

b
Fold-increase relative to untreated cells

Neurosphere formation in PEDF 1.8 1.6 1.4 1.2 1.0


pt y v SR ec -l tor B3 236 p6 5 N ES p6 5 TA

n.s.

*
Egfr-luc

**

NotchIC 0

*
2 4 6 8 10 12 14 Luciferase activity (au)

Untreated PEDF BAY 11-7082 BAY + PEDF

c
Untreated
N-CoR

d
N-CoR

p65end

DAPI/Merged

p65NES

DAPI/Merged

2009 Nature America, Inc. All rights reserved.

at

in

in

at

re

re

nt

IP: N-CoR IB: p65 -tubulin

PE

nt

10

Ig

Cytosolic fraction

Nuclear fraction

f
NotchIC + p65NES

Hes1-luc Untreated PEDF Egfr-luc

EGFR

hi/hi

PE

10

Ig

GFP

Merged

NotchIC + p65NES 0 2 4 6 8 10 12 14

EGFR

hi/lo

GFP

Merged

Luciferase activity (au)

g
Percentage of cell pairs in PEDF

EGFR 100 80 60 40 20 0 GFP p65NES p65TA EGFR


hi/hi hi/lo

hi/lo

p65NES

Merged

Figure 6 PEDF promotes dismissal of N-CoR by inducing NFB-p65. (a) Application of 0.5 M BAY11-7082 blocked the effects of PEDF on Egfr- and Hes1-luciferase reporter activities in NotchIC-expressing cells (n = 3). (b) Overexpression of SR-IB3236 and p65NES, but not p65TA, abolished PEDF-induced increases in neurosphere formation (n = 5). (c) Immunofluorescence for endogenous p65 and N-CoR in untreated and PEDF-treated cells 24 h after plating. Arrowheads indicate p65 and N-CoR cytoplasmic colocalization (n = 3). (d) Immunofluorescence with p65 and N-CoR antibodies to detect transiently expressed p65NES and endogenous N-CoR. Overexpression of p65NES blocked N-CoR nuclear export (n = 4). (e) Immunoprecipitation (IP) of N-CoR from cytosolic and nuclear fractions of untreated or PEDF-treated cells, performed with nonrelevant rabbit IgG or with antibodies to N-CoR, followed by immnunoblotting (IB) with p65 antibodies (n = 3). Full-length blots are presented in Supplementary Figure 11. (f) p65NES overexpression abrogated Egfr and Hes1 promoter activity in NotchICoverexpressing cells treated with PEDF (n = 4). (g) Quantification of EGFRhi/hi and EGFRhi/lo cell-pair distributions in cultures of GFP-, p65TA- and p65NES-transduced cells after PEDF treatment for 24 h (n = 5, no less than 50 cell pairs were scored for each condition). (h) Representative cell pairs as immunostained for EGFR and NICD in PEDF-treated cells that were previously transduced with GFP or p65NES. Data are expressed as mean s.e.m. * P < 0.05, ** P < 0.01 and *** P < 0.001. n.s., not significant, P 0.05. Scale bars represent 10 m.

PEDF

ed

Em pu

***

pu

EGFR

ed

EGFR

hi/hi

p65TA

complex with N-CoR that moves from the nucleus to the cytoplasm and this shuttling becomes enhanced in response to PEDF. Consistent with the possibility that PEDF enhances Notch activity through NF-B signaling and subcellular p65 redistribution, induction of Hes1 and Egfr transcription by PEDF were abrogated in NSCs overexpressing p65NES (Fig. 6f). Moreover, overexpression of p65NES, but not p65TA, reduced the proportion of PEDF-induced EGFRhi/hi cell pairs at the expense of EGFRhi/lo cell pairs (Fig. 6g,h). NSC division in PEDF therefore results in two daughter cells that have high levels of EGFR, both of which are capable of neurosphere formation, and this effect is dependent on p65-mediated N-CoR nuclear export. Endogenous PEDF regulates N-CoR in vivo Immunocytochemistry in the SEZ revealed cells with detectable levels of nuclear or cytoplasmic N-CoR (Fig. 7a). In TNR mice infused with PEDF for 24 h, overall CBF1-reporter levels were increased and N-CoR distribution appeared to be more cytoplasmic than in salineinfused mice (Fig. 7b). PEDF induced a 60% reduction in EGFPpositive BrdU-LRCs with nuclear N-CoR, concomitantly increasing
1520

cells with cytoplasmic N-CoR (Fig. 7c,d). Consistent with our molecular data showing that Notch- and PEDF-induced redistribution of N-CoR regulate the expression of EGFR, 92.3 3.1% of GLAST and Sox2 doublepositive BrdU-LRCs that were EGFPhi in PEDF-infused mice had high levels of this receptor. This resulted in higher proportions of EGFP and EGFR double-positive cells among GLAST and Sox2 double-positive cells after PEDF infusion; notably, we could observe cell pairs of EGFPhi, GLAST-positive cells with high levels of EGFR in vivo, which was suggestive of potential NSC symmetrical divisions (Fig. 7eg). In a previous study, we infused PEDF or a C-terminal fragment capable of antagonizing PEDF effects for 7 d into the lateral ventricle of mice injected with BrdU 3 weeks previously. PEDF increased the number of BrdU-LRCs and primary neurosphere recovery, whereas the C-terminal PEDF fragment yielded BrdU-LRC numbers that were similar to those of saline-infused mice, but yielded fewer primary neurospheres9. Immunostaining with antibodies to N-CoR and BrdU in these mice revealed that the number of BrdU-LRCs with nuclear N-CoR was increased in PEDF-infused brains and decreased in C-terminal PEDFinfused brains compared with vehicle-infused ones (Fig. 7h,i). We also found increased proportions of BrdU-LRCs with high levels of EGFR in PEDF-infused than in saline- or C-terminal PEDF infused brains (Fig. 7j,k). Virtually all of the BrdU-LRCs detected in the different infusion conditions were Ki67 positive (Supplementary Fig. 10),
Merged

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 7 Endogenous PEDF regulates N-CoR subcellular distribution and activation of NSCs. (a) N-CoR in SEZ cells can be nuclear (arrowhead) or cytoplasmic (asterisk). Right, higher magnification of the region outlined by the dashed line. (b) Confocal immunofluorescence for N-CoR, EGFP and GLAST in infused TNR mice showing cytoplasmic (asterisk) or nuclear (arrowheads) distribution of N-CoR. (c) Immunofluorescence for N-CoR, EGFP and BrdU in infused TNR mice. (d) Quantification of EGFP-positive BrdU-LRCs in the SEZ of infused TNR mice (n = 3, over 50 BrdU-positives cells were analyzed per sample). (e) Coincidence of EGFP and EGFR levels in the GLAST-positive population. EGFP hi cells had higher levels of EGFR than EGFPlo cells. (f) Immunofluorescence for EGFP, EGFR and GLAST in infused TNR mice. Arrowheads indicate EGFPhi, EGFR-positive cells. (g) PEDF infusion increased the percentage of EGFPhi, GLAST-positive cells expressing EGFR. (h) Confocal images and xz (bottom) and yz (left) orthogonal projections of BrdU-LRCs (green) after saline, PEDF or C-terminal PEDF infusions, showing absence of N-CoR immunoreactivity (arrows) or staining for N-CoR (red) in the cytoplasmic (asterisks) or nuclear (arrowheads) compartments. (i) PEDF infusion decreased and C-terminal PEDF infusion increased the proportion of cells with nuclear N-CoR (n = 3). (j) Confocal images of BrdU and EGFR immunostaining. (k) PEDF, but not C-terminal PEDF, infusion increased the proportion of BrdU-LRCs expressing EGFR (n = 3). Data are expressed as mean s.e.m. * P < 0.05 and ** P < 0.01. Scale bars represent 20 m (a,b,f) and 10 m (c,e,h,j and inset in a).

a
*
N-CoR/DAPI

b
*
N-CoR/EGFP/GLAST

Saline

PEDF

* * * * *

* * * * *

* * * * *

Percentage of EGFP+BrdU-LRCs

c
Saline

N-CoR/EGFP/BrdU-LRC

*
PEDF

* *

*
Merged

60 50 40 30 20 10 0

N-CoR distribution Saline PEDF

**

**

Nuclear Cytoplasmic

EGFPhi

EGFP

EGFR

GLAST

Saline

PEDF

g
20 16 12 8 4 0 Percentage of GLAST+/Sox2+

2009 Nature America, Inc. All rights reserved.

Saline PEDF

EGFPlo

EGFP/GLAST/EGFR

**
+ FR G /E
hi lo

*
N-CoR/BrdU-LRC PEDF

FP

i
Percentage of BrdU-LRCs

*
C-ter PEDF

80 60 40 20 0

* **

Saline PEDF C-ter PEDF

EG

EGFR/BrdU-LRC

k
Percentage of BrdU-LRCs 60 40 20 0

EG

FP
Saline PEDF C-ter PEDF

/E

** *
EGFR+

Nuclear Cytoplasmic N-CoR distribution

indicating that cycling label-retaining, but not terminally differentiated, cells were scored in our experiments. Thus, N-CoR subcellular redistribution induced by PEDF correlates with higher proportions of BrdU-LRCs with detectable levels of EGFR. These results indicate that endogenous PEDF acts physiologically to cause the dismissal of specific N-CoR repressor complexes, resulting in activation of Notchregulated stem cellrelated genes. DISCUSSION We found that the balance between symmetric and asymmetric divisions in NSCs can be modulated by endogenous PEDFs effects on Notch transcriptional activity. Mechanistically, PEDF promotes selfrenewing divisions and maintenance of a multipotent state in NSCs by inducing the dismissal of transcriptional co-repressor N-CoR from its binding to CBF1 in specific promoters through a non-canonical activation of the NF-B pathway. Because PEDF has a physiological role in the adult SEZ9, we propose that PEDF is a natural modulator of N-CoR in endogenous niches contributing to stem-cell maintenance by interaction with the Notch pathway. Notch is active in radial glia-like cells of neurogenic regions and downregulation of this pathway is required for lineage progression20,21. Consistently, we found that CBF1 activity in subependymal cells was detected together with B cell marker combinations (some ependymal cells were also labeled, consistent with a recent report37), and Notch activity levels positively correlated with in vitro stem potential. Our results indicate that EGFPlo cells also have stem-cell
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

attributes, but their potential in vitro was reduced, consistent with a lineage progression from EGFPhi cells to EGFPlo cells. The effects of PEDF on the modulation of Notch and self-renewal indicate that some cells with low Notch expression can respond to cues that will retard their differentiation. However, although treatment of EGFPlo cells with PEDF resulted in a more extended self-renewing capacity, the treatment could not induce a self-renewal as extensive as that seen in EGFPhi cells. Moreover, many EGFPlo cells became refractory (probably those with very low Notch expression); for example, if grown neurospheres are treated with PEDF, they do not produce more secondary neurospheres9, suggesting that PEDF signaling needs to be coincident with downregulation of Notch signaling during cell division. This is consistent with the finding that once CBF1 signaling has become attenuated in fetal neural progenitors, forced activation of CBF1 is ineffective at reverting them to NSCs20. Despite indirect evidences indicating that PEDF can regulate -secretase/presenilin-1 activity in endothelial cells38, our data support a model in which PEDF cooperates with Notch only at the transcriptional level. In the CNS, only the Hes1, Hes5 and Blbp (encoding the brain lipid-binding protein, also known as Fabp7) genes have been experimentally linked to CBF1 activation13. We found that the Egfr promoter is a target of CBF1, providing a molecular basis for the effects of Notch activation on proliferation. Because NSCs with high levels of EGFR generate more neurospheres in vitro and activated GFAP-positive B cells in the SEZ express EGFR39,40, our results suggest that Notch probably determines self-renewing potential, at least in
1521

Saline

FR

a r t ic l e s
part, by modulating mitogenic responsiveness. On the other hand, the Notch pathway has a dominant function in inhibiting differentiation of NSCs, as Hes1/5 repress execution of neurogenic programs13,41. Recent evidence indicates that Hes1 prevents permanent cell-cycle exit, allowing for reversibility of the quiescent state42. Therefore, coordinated expression of EGFR and Hes1 may determine stemness by promoting mitogenic response in a highly regulated quiescent population and nondifferentiation in a concerted way. It has been reported that adipose triglyceride lipase (desnutrin or iPLA2) is a putative receptor for PEDF in retinal epithelial cells 11 and PEDF can indeed regulate triglyceride metabolism in hepatocytes through adipose triglyceride lipase43. However, the compound (R)-bromoenol lactone, which irreversibly blocks 90% of the enzyme lipase activity43, was ineffective at blocking PEDFs effects on neurosphere formation (C.A.-A. and I.F., unpublished data). Although PEDF mediates its effects on NSCs through an unknown receptor, our data indicate that PEDFs actions require activation of the NF-B pathway. The Notch and NF-B pathways have been shown to interact in various ways44,45. In particular, p65 induces nucleo-cytoplasmic translocation of N-CoR and Hes1 transcription in a cell line45. GFAPpositive cells in the SEZ express components of the Notch and NF-B signaling pathways33,46 and N-CoR, and we found that the actions of PEDF do not require p65-dependent transcription of target genes, but instead require p65 subcellular redistribution. Regulation of the NF-B pathway is complex, involving not only post-translational modifications in NF-B complexes or target gene histones, but also the continuous nucleo-cytoplasmic shuttling of NF-B34. We found that N-CoR bound to p65, although our data do not exclude the possibility that other molecules, such as IB, could be implicated in the physical re-distribution of N-CoR from the nucleus to the cytoplasm44. Maintenance of a stem-cell state depends on the combined action of different gene products and it may therefore involve the activity of large regulatory complexes, including transcriptional co-repressors/ activators and chromatin remodeling factors47. Our results suggest that modulation of N-CoR in response to various signals operating in endogenous niches could orchestrate the coordinated expression of molecules involved in the regulation of stem-cell proliferation and multipotency. Other pathways in distinct cellular settings regulate N-CoR differently. Dismissal from the Gfap promoter is induced in fetal NSCs in response to CNTF and N-CoR deletion results in cortical progenitors with impaired self-renewal that spontaneously differentiate into astroglial-like cells27. Notably, Notch activation during NSC differentiation appears to be pro-gliogenic41. Thus, our results indicate that Notchs opposing actions on self-renewal and astrogliogenesis could be determined by selective modulation of N-CoR repressive activity. Many transcriptional modulators are regulated by post-translational modification and MEK kinase and Akt kinasedependent phosphorylation are involved in mediating the nuclear export of N-CoR in response to CNTF27,48. Further studies will be needed to determine how N-CoR specificity is mechanistically regulated in response to PEDF or other factors. Analysis in cell pair assays and in cell populations that differ in Notch activity levels indicate that Notch asymmetry/symmetry at stem-cell division is functionally important for the outcome of the replication process 2. It is unclear how asymmetric distribution of activated Notch is produced in adult NSCs although it may depend on the regulated partitioning of effectors such as Numb, which inhibits Notch signaling and whose segregation is dictated by cell-polarity determinants 1. Our data indicate that symmetry/asymmetry can also be modulated by extrinsic cues and microenvironments, and, more specifically, neurovascular niches
1522

can modify stem-cell fate decisions that probably determine their persistence and response to nonphysiological demand. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We are grateful to A. Bigas, J.L. de la Pompa, N. Gaiano, W.C. Greene, O. Hermanson, S. Hitoshi, A. Israel, K. Jepsen, M. Karin, R. Kopan, G. Leclercq, M.G. Rosenfeld, S. Sun, D. van der Kooy, Y. Zhan and V. Meni for kindly providing constructs and reagents. We also thank E. Porlan and H. Mira for technical advice, M.P. Rubio for excellent technical assistance, and E. Porlan and S.R. Ferrn for critical reading of the manuscript and valuable discussions. We gratefully acknowledge the help of M.A. Marqus-Torrejn with the infusion experiments and of A. Martnez and D. Gil with cytometry. We are also grateful to E. Lai for comments on the manuscript. This work was supported by grants from Ministerio de Ciencia e Innovacin (SAF Program), Ministerio de Sanidad (Red Tercel, Centro de Investigacin Biomdica en Red sobre Enfermedades Neurodegenerativas, CIBERNED), Generalitat Valenciana (Programa Prometeo) and Fundacin la Caixa. C.A.-A. was a recipient of a Formacin del Profesorado Universitario predoctoral fellowship. J.M.M.-R. and A.C.D. were funded by CIBERNED. AUTHOR CONTRIBUTIONS All of the authors designed and discussed the experiments. C.A.-A. conducted most of the cell culture and biochemistry experiments on the relationship between PEDF and Notch, Notch/EGFR asymmetry and the role of N-CoR and p65, as well as the infusion experiments and in vivo analyses. C.A.-A. and J.M.M.-R. performed the FACS experiments and analyses with TNR cells, ChIP studies and luciferase assays. A.C.D. contributed to immunohistochemistry analyses and multipotency assays. I.F. supervised the project and wrote the manuscript. COMPETING INTERESTS STATEMENT The authors declare competing financial interests: details accompany the full-text HTML version of the paper at www.nature.com/natureneuroscience/.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.

2009 Nature America, Inc. All rights reserved.

1. Morrison, S.J. & Kimble, J. Asymmetric and symmetric stem-cell divisions in development and cancer. Nature 441, 10681074 (2006). 2. Wu, M. et al. Imaging hematopoietic precursor division in real time. Cell Stem Cell 1, 541554 (2007). 3. Riquelme, P.A., Drapeau, E. & Doetsch, F. Brain micro-ecologies: neural stem cell niches in the adult mammalian brain. Phil. Trans. R. Soc. Lond. B 363, 123137 (2008). 4. Zhao, C., Deng, W. & Gage, F.H. Mechanisms and functional implications of adult neurogenesis. Cell 132, 645660 (2008). 5. Palmer, T.D., Willhoite, A.R. & Gage, F.H. Vascular niche for adult hippocampal neurogenesis. J. Comp. Neurol. 425, 479494 (2000). 6. Shen, Q. et al. Adult SVZ stem cells lie in a vascular niche: a quantitative analysis of niche cell-cell interactions. Cell Stem Cell 3, 289300 (2008). 7. Tavazoie, M. et al. A specialized vascular niche for adult neural stem cells. Cell Stem Cell 3, 279288 (2008). 8. Shen, Q. et al. Endothelial cells stimulate self-renewal and expand neurogenesis of neural stem cells. Science 304, 13381340 (2004). 9. Ramrez-Castillejo, C. et al. Pigment epithelium-derived factor is a niche signal for neural stem cell renewal. Nat. Neurosci. 9, 331339 (2006). 10. Tombran-Tink, J. & Barnstable, C.J. PEDF: a multifaceted neurotrophic factor. Nat. Rev. Neurosci. 4, 628636 (2003). 11. Notari, L. et al. Identification of a lipase-linked cell membrane receptor for pigment epitheliumderived factor. J. Biol. Chem. 281, 3802238037 (2006). 12. Chiba, S. Notch signaling in stem cell systems. Stem Cells 24, 24372447 (2006). 13. Louvi, A. & Artavanis-Tsakonas, S. Notch signaling in vertebrate neural development. Nat. Rev. Neurosci. 7, 93102 (2006). 14. Bray, S.J. Notch signaling: a simple pathway becomes complex. Nat. Rev. Mol. Cell Bio. 7, 678689 (2006). 15. Alexson, T.O., Hitoshi, S., Coles, B.L., Bernstein, A. & van der Kooy, D. Notch signaling is required to maintain all neural stem cell populationsirrespective of spatial or temporal niche. Dev. Neurosci. 28, 3448 (2006). 16. Androutsellis-Theotokis, A. et al. Notch signalling regulates stem cell numbers in vitro and in vivo. Nature 442, 823826 (2006).

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
17. Hitoshi, S. et al. Notch pathway molecules are essential for the maintenance, but not the generation, of mammalian neural stem cells. Genes Dev. 16, 846858 (2002). 18. Ohtsuka, T., Sakamoto, M., Guillemot, F. & Kageyama, R. Roles of the basic helixloop-helix genes Hes1 and Hes5 in expansion of neural stem cells of the developing brain. J. Biol. Chem. 276, 3046730474 (2001). 19. Duncan, A.W. et al. Integration of Notch and Wnt signaling in hematopoietic stem cell maintenance. Nat. Immunol. 6, 314322 (2005). 20. Mizutani, K., Yoon, K., Dang, L., Tokunaga, A. & Gaiano, N. Differential Notch signaling distinguishes neural stem cells from intermediate progenitors. Nature 449, 351355 (2007). 21. Breunig, J.J., Silbereis, J., Vaccarino, F.M., Sestan, N. & Rakic, P. Notch regulates cell fate and dendrite morphology of newborn neurons in the postnatal dentate gyrus. Proc. Natl. Acad. Sci. USA 104, 2055820563 (2007). 22. Jarriault, S. et al. Signaling downstream of activated mammalian Notch. Nature 377, 355358 (1995). 23. Campos, L.S., Decker, L., Taylor, V. & Skarnes, W. Notch, epidermal growth factor receptor, and beta1-integrin pathways are coordinated in neural stem cells. J. Biol. Chem. 281, 53005309 (2006). 24. Mori, T. et al. Inducible gene deletion in astroglia and radial gliaa valuable tool for functional and lineage analysis. Glia 54, 2134 (2006). 25. Sun, Y., Goderie, S.K. & Temple, S. Asymmetric distribution of EGFR receptor during mitosis generates diverse CNS progenitor cells. Neuron 45, 873886 (2005). 26. Zhang, Y.W. et al. Presenilin/gamma-secretasedependent processing of betaamyloid precursor protein regulates EGF receptor expression. Proc. Natl. Acad. Sci. USA 104, 1061310618 (2007). 27. Hermanson, O., Jepsen, K. & Rosenfeld, M.G. N-CoR controls differentiation of neural stem cells into astrocytes. Nature 419, 934939 (2002). 28. Kao, H.Y. et al. A histone deacetylase co-repressor complex regulates the Notch signal transduction pathway. Genes Dev. 12, 22692277 (1998). 29. Rosenfeld, M.G., Lunyak, V.V. & Glass, C.K. Sensors and signals: a coactivator/ co-repressor/epigenetic code for integrating signal-dependent programs of transcriptional response. Genes Dev. 20, 14051428 (2006). 30. Reizis, B. & Leder, P. Direct induction of T lymphocytespecific gene expression by the mammalian Notch signaling pathway. Genes Dev. 16, 295300 (2002). 31. Yabe, T., Sanagi, T., Schwartz, J.P. & Yamada, H. Pigment epitheliumderived factor induces pro-inflammatory genes in neonatal astrocytes through activation of NF-kappaB and CREB. Glia 50, 223234 (2005). 32. Yabe, T., Wilson, D. & Schwartz, J.P. NFkappaB activation is required for the neuroprotective effects of pigment epitheliumderived factor (PEDF) on cerebellar granule neurons. J. Biol. Chem. 276, 4331343319 (2001). 33. Denis-Donini, S., Caprini, A., Frassoni, C. & Grilli, M. Members of the NF-kappaB family expressed in zones of active neurogenesis in the postnatal and adult mouse brain. Brain Res. Dev. Brain Res. 154, 8189 (2005). 34. Chen, L.F. & Greene, W.C. Shaping the nuclear action of NF-kappaB. Nat. Rev. Mol. Cell Bio. 5, 392401 (2004). 35. DiDonato, J. et al. Mapping of the inducible IkappaB phosphorylation sites that signal its ubiquitination and degradation. Mol. Cell. Biol. 16, 12951304 (1996). 36. Harhaj, E.W. & Sun, S.C. Regulation of RelA subcellular localization by a putative nuclear export signal and p50. Mol. Cell. Biol. 19, 70887095 (1999). 37. Carln, M. et al. Forebrain ependymal cells are Notch-dependent and generate neuroblasts and astrocytes after stroke. Nat. Neurosci. 12, 259267 (2009). 38. Cai, J., Jiang, W.G., Grant, M.B. & Boulton, M. Pigment epitheliumderived factor inhibits angiogenesis via regulated intracellular proteolysis of vascular endothelial growth factor receptor 1. J. Biol. Chem. 281, 36043613 (2006). 39. 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, 10211034 (2002). 40. Ciccolini, F., Mandl, C., Holzl-Wenig, G., Kehlenbach, A. & Hellwig, A. Prospective isolation of late development multipotent precursors whose migration is promoted by EGFR. Dev. Biol. 284, 112125 (2005). 41. Yoon, K. & Gaiano, N. Notch signaling in the mammalian central nervous system: insights from mouse mutants. Nat. Neurosci. 8, 709715 (2005). 42. Sang, L., Coller, H.A. & Roberts, J.M. Control of the reversibility of cellular quiescence by the transcriptional repressor HES1. Science 321, 10951100 (2008). 43. Chung, C. et al. Anti-angiogenic pigment epithelium-derived factor regulates hepatocyte triglyceride content through adipose triglyceride lipase (ATGL). J. Hepatol. 48, 471478 (2008). 44. Ang, H.L. & Tergaonkar, V. Notch and NFkappaB Signaling pathways: do they collaborate in normal vertebrate brain development and function? Bioessays 29, 10391047 (2007). 45. Espinosa, L., Santos, S., Ingles-Esteve, J., Munoz-Canoves, P. & Bigas, A. p65-NFkappaB synergizes with Notch to activate transcription by triggering cytoplasmic translocation of the nuclear receptor corepressor N-CoR. J. Cell Sci. 115, 12951303 (2002). 46. Givogri, M.I. et al. Notch signaling in astrocytes and neuroblasts of the adult subventricular zone in health and after cortical injury. Dev. Neurosci. 28, 8191 (2006). 47. Hsieh, J. & Gage, F.H. Chromatin remodeling in neural development and plasticity. Curr. Opin. Cell Biol. 17, 664671 (2005). 48. Baek, S.H. et al. Exchange of N-CoR co-repressor and Tip60 coactivator complexes links gene expression by NF-kappaB and beta-amyloid precursor protein. Cell 110, 5567 (2002).

2009 Nature America, Inc. All rights reserved.

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1523

ONLINE METHODS
Mice, immunohistochemistry and in vivo infusions. Experiments were performed in TNR mice obtained from the Jackson Laboratories or in CD1 mice (Charles River Labs). Housing of mice and all experiments were carried out according to European Union 86/609/EEC and Spanish RD-1201/2005 guidelines, following protocols approved by the ethics committee of the Universidad de Valencia. Both 7-day infusions into the lateral ventricle of adult mice and BrdU administration regimes have been previously detailed9,49. For 24-h infusions, 2-month-old TNR mice received a total of 1.5 g of PEDF (Biovendor). For primary antibodies, we used rabbit antibodies to N-CoR (1:1, Abcam), GFAP (1:600, Dako), BrdU (1:1500, Megabase Research), EGFP (1:200, Molecular Probes), GLAST (1:1,000, Tocris), S100 (1:200, Dako), activated (cleaved) Notch1 (1:100, Abcam) and Ki67 (1:100, Abcam), mouse antibodies to GFAP (1:500, Sigma), BrdU (1:300, Dako) and Mash-1 (1:100, BD Biosciences), goat antibodies to Sox2 (1:40, R&D Systems) and GFP (1:200, Rockland), guinea pig antibody to GLAST (1:1,000, Chemicon), and sheep antibody to EGFR (1:100, Upstate). For N-CoR immunodetection specifically, 30-m-thick vibratome and 7-m-thick paraffin sections were boiled in 10 mM sodium citrate buffer (pH 6.0) for 30 min for antigen retrieval, rinsed twice in 0.1 M phosphate buffer, blocked in 0.1 M phosphate buffer with 10% horse serum (vol/vol, Gibco) and 0.3% Triton X-100 (vol/vol) for 1 h, and incubated with primary antibodies overnight, followed by incubation with biotinylated horse antibody to rabbit (1:200, Vector) and Cy3-conjugated streptavidin (1:200, Jackson Immunochemicals). EGFP expression in TNR mice sections was measured using Leyca Microsystems Confocal Software. The levels of immunofluorescence in each EGFP-positive cell was measured as mean pixel density above background (60 mean pixels; EGFPhi cells, 165; EGFPlo cells, 30165; EGFP-negative cells, 30). Neural stem cell culture, in vitro assays and immunofluorescence. Methods for NSC isolation and culture and neurosphere formation analysis have been previously described in detail49. We used PEDF (50100 ng ml1) from either Bioproducts MD (human recombinant produced in E. coli) or from Biovendor (human recombinant produced in HEK cells) with identical results. Jagged1 (2 g ml1) and the inhibitors L-685.458 (1 M) and BAY-11-7082 (0.5 M) were all from Calbiochem. PEDF and Jagged 1 were added to the medium at the moment of seeding. Inhibitors were added to the medium 1 h before the addition of PEDF. Neurospheres obtained from the SEZ of 23-month-old TNR hemizygous mice (Jackson Laboratories) were mechanically dissociated to a single-cell suspension and separated into EGFPhi, EGFPlo and EGFP-negative cells by FACS using a MoFlo sorter (Dako). Autofluorescence levels were determined by comparing EGFP intensity with the background fluorescence of a wild-type culture; cells in this range were retrieved as being EGFP negative. The brightest 510% of the EGFP-positive cells were collected as EGFPhi and the dimmest 1020% of the cells were recovered as EGFPlo. Cells were seeded at 2.5 cells per l (low density) or at 1 cell per well in p96 plates (clonal). Clones generated from these FACS-sorted cells were considered to be primary neurospheres in terms of PEDF treatment. Cell apoptosis and viability and the numbers of single viable cells and incipient neurospheres were determined at 24, 48 and 72 h after plating, as previously described9. For cell-pair analyses, single cells were seeded and fixed after 24 h with 4% paraformaldehyde (wt/vol) in 0.1M phosphate buffer (pH 7.4). Cell pairs were visually inspected to ensure that they were the results of cell division and at least 5055 cell pairs were counted in each condition per experiment. Differentiation was analyzed by seeding individual passage 23 neurospheres of similar sizes, collected one at a time using a pipette, in Matrigel-coated 96-well plates for 7 d in vitro (2 d in neurosphere medium without EGF and another 5 d in 2% fetal bovine serum, vol/vol) before fixation in 4% paraformaldehyde in phosphate buffer and immunocytochemical staining. No less than 50 clones were analyzed for each condition and experiment. Methods for immunofluorescence in proliferating and differentiating neurosphere cultures and in isolated cells have been previously described49. For primary antibodies, we used rabbit antibodies to NICD (1:150, Abcam), p65 (1:150, Santa Cruz) and GFAP (1:300, Dako), mouse antibodies to O4 (1:2, Developmental Studies Hybridoma Bank) and III-tubulin (1:300, Covance), goat antibody to N-CoR (1:50, Santa Cruz), and sheep antibody to EGFR (1:100, Upstate). Cell transduction and luciferase assays. We electroporated 24 g of DNA of each of the following constructs using a Nucleofector (II) (Amaxa Biosystems):

HANcor1-C (amino acids 1,5012,300), IB3236, EGFP, p65TA, p65NES, Ncor1, Su(H)DBM, NotchIC-IRES-GFP, Hes1-luc, HesAB-luc, 4xwtCBF1-luc, 4xmutCBF1-luc, 2xB-luc and Egfr-luc. In reporter assays with firefly luciferasebased constructs, 50 ng of a Renilla luciferase construct was used as an internal control. After 2436 h, transduced cells (passage 46) were dissociated, plated in the presence or absence of PEDF (alone or after a 1-h pre-treatment with inhibitor) and cultured for 24 h before being harvested for analysis. Efficiency was around 80% in all cases. The retrovirus preparation for NotchIC overexpression and NSC infection were performed as described17; the neurospheres were mechanically dissociated and transduced cells were isolated by FACS using a MoFlo sorter (Dako). Dissociated transduced cells were plated at 5 cells per l, treated as described above and cultured for 24 h before harvesting. Luciferase activity was measured in cell lysates using the Dual Luciferase Assay System (Promega) and promoter activity was defined as the ratio between the firefly and Renilla luciferase activities. Co-immunoprecipitation and immunoblot. Isolated cells from passage 46 neurospheres were treated with PEDF for 24 h before collecting the cells. For western blot, neurospheres were rinsed in phosphate-buffered saline (PBS), incubated in lysis buffer (20 mM PBS, 150 mM NaCl, 5 mM EDTA, 1% Triton X-100, 1 mM sodium orthovanadate, 1 mM NaF, 1 mM PMSF and 1 Complete (Roche)) for 15 min and centrifuged at 20,000 g for 15 min at 4 C. For preparation of nuclear and cytoplasmic fractions, cells were lysed in cytoplasmic extraction buffer containing 10 mM Tris-HCl (pH 7.6), 1.5 mM MgCl2, 10 mM KCl, 0.5 mM EDTA, 1 mM DTT, 1 mM sodium orthovanadate, 1 mM NaF, 1 mM PMSF and 1 Complete mix (Roche) for 20 min on ice and passed 40 times through a G25 needle. The mix was centrifuged at 1,000 g for 15 min and the supernatant containing the cytoplasmic extract was kept. The pellet containing the nuclear cell fraction was resuspended in lysis buffer consisting of 50 mM Tris (pH 8.0), 10 mM EDTA (pH 8.0), 1% SDS (wt/vol), 1 mM PMSF and 1 Complete mix (Roche), incubated for 10 min on ice, and sonicated three times for 30 s (Bioruptor). We used 500750 g of protein extract and saved one-tenth of the total volume for input. For co-immunoprecipitation, supernatant was diluted by adding 16.7 mM Tris (pH 8.0), 1.2 mM EDTA (pH 8.0), 1.1% Triton X-100, 90 mM NaCl, 0.01% SDS, 1 mM PMSF and 1 Complete (Roche). Immunoprecipitation was performed overnight at 4 C with 3 g of antibody to N-CoR (Affinity Bioreagents) or 3 g of rabbit IgGs (Sigma). After immunoprecipitation, 60 l of protein A/G sepharose beads (50% slurry) were added and the incubation continued for another 2 h. Beads were centrifuged and washed and the protein was eluted and subjected to immunoblotting with rabbit antibody to p65 (1:300, Santa Cruz). For western blot, we used rabbit antibodies to N-CoR (1:500) and histone 3 (1:500). Samples were subjected to electrophoresis on SDSpolyacrilamide gel followed by transfer to a nitrocellulose membrane. Membranes were blocked 1 h with 5% nonfat dry milk (wt/vol) and incubated with primary antibodies followed by incubation with horseradish peroxidase conjugated appropriate secondary antibodies. Western blots were developed using ECL reagents (Amersham) according to supplier recommendations. ChIP. After PEDF treatment for 24 h, cells (passage 48 neurospheres) were crosslinked with 1% formaldehyde (vol/vol) at 2025 C for 10 min. Glycine was added to a final concentration of 125 mM to stop cross-linking. Cells were rinsed twice with cold PBS and incubated for 10 min at 4 C in low-salt washing buffer (10 mM Tris (pH 8.0), 1 mM EDTA (pH 8.0), 0.5 mM EGTA (pH 8.0), 0.25% Triton X-100, 1 mM PMSF and 1 Complete (Roche)). After centrifugation at 1,300g, the pellet was resuspended in high-salt washing buffer (10 mM Tris (pH 8.0), 1 mM EDTA (pH 8.0), 0.5 mM EGTA (pH 8.0), 0.2 M NaCl, 1 mM PMSF and 1 Complete (Roche)) and incubated for 10 min at 4 C. Nuclei were resuspended in 1% SDS lysis buffer and chromatin was sonicated to obtain DNA fragments around 500 base pairs. One-tenth of the total volume was saved as a total input DNA control. Lysates were precleared by incubation with 50 l of Protein A sepharose (50% slurry preblocked with salmon sperm DNA, yeast tRNA and BSA) for 1 h. Immunoprecipitation was performed overnight at 4 C with 5 g of rabbit antibody to N-CoR (Santa Cruz), 6 g of rabbit antibody to CBF1 (Santa Cruz), 3 g of rabbit antibody to NICD (Abcam), 56 of g nonrelated rabbit antibody (Dako) and 3 g of nonrelated rabbit serum. After immunoprecipitation, 60 l of preblocked Protein Asepharose (50% slurry) were added and the

2009 Nature America, Inc. All rights reserved.

nature NEUROSCIENCE

doi:10.1038/nn.2437

incubation continued for another 4 h. Precipitates were thoroughly washed and extracted twice in 1% SDS and 0.1 M NaHC03 in TE buffer (10 mM Tris-HCl (pH 8.0), 0.1 mM EDTA). Elutes were incubated at 65 C overnight to reverse cross-linking, followed by 1 h incubation at 50 C with 10 M EDTA, 40 M Tris-HCl (pH 6.8) and 20 g of proteinase K (Roche). DNA fragments were recovered by Qiaquick PCR purification kit (Qiagen). For primers, we used Hes1 176 (5-GCC GCC AGA CCT TGT GCC TA-3), Hes +113 (5-CCA GAT CCT GTG TGA TCC GCA-3), Egfr 580 (5-TTC TTT CAG AGA CAT GGA GGG TTC-3), Egfr 254 (5-GAA GTC CAG CCA ATC TAT GCC AG-3), Egfr 109 (5-TGC CTG CTT TCG ATC CTC-3), Egfr +224 (5-AAT CCG AGA CAG ACG GAG-3), Gfap 265 (5-GAC TAA GCT GTT CCC TCG GC-3), Gfap 147 5-TGA GGT CAC TGT ACC CAG AG-3), Gfap 2,388 (5-TGA GCA ACT ACT AGA TCC TTG G-3), Gfap 2,026 5-TCT GCC TCT GGT GAC TTT TC-3), Ptcra 389 (5-GTA GAG CGA AGG AAC TAG GC-3) and Ptcra 150 (5-CAC CCT CTC ATA ACC TTC-3). RNA isolation and reverse transcriptionPCR analysis. Total RNA was isolated using RNeasy MicroKit (Qiagen) and 1 g of total RNA was used to synthesize

cDNA using random primers and reverse transcriptase (SuperScript II RT, Gibco, Invitrogen). For PCR analysis, 1 l of cDNA was used as the template in a reaction volume of 20 l containing 0.25 mM dNTPS (Eppendorf), 0.25 M primers (Isogen) and 1 U of Taq DNA polymerase (Promega). Transcript levels were quantified by densitometric analysis of reverse transcriptionPCR bands in ethidium bromidestained electrophoresis gels and normalized to -actin levels. For primers, we used -actin (forward, 5-CCG GGA CCT GAC AGA CTA CCT-3; reverse, 5-GCC ATC TCC TGC TCG AAG TCT A-3) and Gfap (forward, 5-CTC TAG TAC TGC TGC ATG-3; reverse, 5-TGT GAG CGT ACT TCT ATG-3). Statistical analyses. Analyses of significant differences between means were performed using two-tailed Students t tests. The arcsen transformation for normalization was applied to relative values (fold-change and percentage). In each case, n indicates the number of independent cultures or mice used. In all cases, P < 0.05 denoted statistical significance.
49. Ferrn, S.R. et al. A combined ex/in vivo assay to detect effects of exogenously added factors in neural stem cells. Nat. Protoc. 2, 849859 (2007).

2009 Nature America, Inc. All rights reserved.

doi:10.1038/nn.2437

nature NEUROSCIENCE

a r t ic l e s

Adult generation of glutamatergic olfactory bulb interneurons


Monika S Brill13,10, Jovica Ninkovic2,3,10, Eleanor Winpenny4,10, Rebecca D Hodge5,10, Ilknur Ozen4, Roderick Yang5, Alexandra Lepier1, Sergio Gascn1,2, Ferenc Erdelyi6, Gabor Szabo6, Carlos Parras7,9, Francois Guillemot7, Michael Frotscher8, Benedikt Berninger1,2, Robert F Hevner5, Olivier Raineteau4,9 & Magdalena Gtz13
2009 Nature America, Inc. All rights reserved. The adult mouse subependymal zone (SEZ) harbors neural stem cells that are thought to exclusively generate GABAergic interneurons of the olfactory bulb. We examined the adult generation of glutamatergic juxtaglomerular neurons, which had dendritic arborizations that projected into adjacent glomeruli, identifying them as short-axon cells. Fate mapping revealed that these originate from Neurog2- and Tbr2-expressing progenitors located in the dorsal region of the SEZ. Examination of the progenitors of these glutamatergic interneurons allowed us to determine the sequential expression of transcription factors in these cells that are thought to be hallmarks of glutamatergic neurogenesis in the developing cerebral cortex and adult hippocampus. Indeed, the molecular specification of these SEZ progenitors allowed for their recruitment into the cerebral cortex after a lesion was induced. Taken together, our data indicate that SEZ progenitors not only produce a population of adult-born glutamatergic juxtaglomerular neurons, but may also provide a previously unknown source of progenitors for endogenous repair.

The presence of neural stem cells (NSCs) in the adult mammalian brain has prompted hope for replacement of degenerating neurons. This potential avenue for repair has been further supported by the observation that neuroblasts originating from these NSCs are recruited to sites of injury1,2. Adult neurogenesis occurs largely in two regions of the forebrain that generate distinct types of neurons: the subgranular zone generating glutamatergic projection neurons of the dentate gyrus and the SEZ generating GABAergic interneurons of the olfactory bulb3. The potential of these regions to contribute to endogenous repair may depend on the range of neuronal subtypes that they can produce. Although earlier transplantation studies have proposed a broad potential of NSCs in regard to the generation of different neuronal subtypes4, this view has recently been challenged by the demonstration that there are distinct NSCs with restricted potential regarding subtype specification5. On the other hand, it has been suggested that apoptotic cell death in the cerebral cortex elicits the migration of neuroblasts from the SEZ with the ability to replace glutamatergic projection neurons6,7. Given that SEZ cells have so far been thought to generate only GABAergic neurons, this recruitment either suggests a high degree of plasticity of adult SEZ cells or, alternatively, there might be a yet undefined source of glutamatergic neurons in this region.
1Department

NSCs in the adult SEZ generate GABAergic interneurons integrating into the granule cell layer (GCL) or glomerular layer of the olfactory bulb8. A large part of these interneurons arise from the lateral SEZ, which derives from the ventral telencephalon, producing mostly telencephalic GABAergic neurons during development8,9. Transcription factors that are crucial for the generation of GABAergic interneurons during development, such as Dlx10 and Sp8 (ref. 11), continue to determine the postnatal and adult generation of olfactory bulb interneurons10,11. Transcription factors that are present in the dorsal telencephalon during development, such as Pax6 or Neurog2 (ref. 12), are largely restricted in their expression to the dorsal region of the adult10,13 or postnatal SEZ14. Fate-mapping experiments have shown that the adult dorsal SEZ originates from the Emx1-positive area of the embryonic telencephalon and participates in the generation of specific subtypes of GABAergic olfactory neurons, such as the dopaminergic periglomerular neurons15,16. However, during embryonic development, dorsal progenitors generate predominantly glutamatergic neurons in a Pax6-dependent manner, including those of the olfactory bulb17,18. Thus, the entire population of olfactory bulb projection neurons is thought to derive from a Pax6-expressing territory19. In the developing cerebral cortex,

of Physiological Genomics, Institute of Physiology, Ludwig-Maximilians University Munich, Munich, Germany. 2Institute for Stem Cell Research, Helmholtz Zentrum Mnchen, German Research Center for Environmental Health, Neuherberg, Germany. 3Munich Center for Integrated Protein Science CiPSM, Munich, Germany. 4Cambridge Centre for Brain Repair, Robinson Way, Cambridge, UK. 5Departments of Neurological Surgery and Pathology, University of Washington, Seattle Childrens Hospital Research Institute, Seattle, Washington, USA. 6Laboratory of Molecular Biology and Genetics, Institute of Experimental Medicine, Budapest, Hungary. 7Division of Molecular Neurobiology, Medical Research CouncilNational Institute for Medical Research, London, UK. 8Insitute of Anatomy and Cell Biology, University of Freiburg, Freiburg, Germany. 9Present address: Brain Research Institute, University of Zrich/ETH, Zrich, Switzerland (O.R.), INSERM U711, Biologie des Interactions Neurones/Glie, Hpital de la Piti-Salptrire, Btiment de la Pharmacie, Paris, France (C.P.). 10These authors contributed equally to this work. Correspondence should be addressed to M.G. (magdalena.goetz@helmholtz-muenchen.de), O.R. (raineteau@hifo.uzh.ch) or R.F.H. (rhevner@u.washington.edu). Received 15 June; accepted 8 September; published online 1 November 2009; corrected after print 11 December 2009; doi:10.1038/nn.2416

1524

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 1 Neurog2+/gfp, Tbr1 and Tbr2 are expressed in a defined dorsal region of the SEZ. (ac) Fluorescent micrographs depicting Neurog2+/gfp-, Tbr2- and Tbr1-positive cells as indicated in the dorsal SEZ and RMS. (d) Three-dimensional reconstruction of the localization of Neurog2+/gfp(green) and Tbr2-positive (red) cells in the adult forebrain indicates that these progenitors are located in the dorsal SEZ, but not in the lateral wall of the lateral ventricle or third ventricle. Coronal views of three different levels are indicated by the black lines. The histogram below depicts the distribution of Neurog2+/gfp- and Tbr2-positive cells along the rostrocaudal axis. Tbr2-positive cells were more widespread than Neurog2+/gfppositive cells and the expression of both markers dropped rapidly after the cells entered the RMS. (e,f) Neurog2+/gfp cells are immunoreactive for Tbr2 (e) and Tbr1 (f). Nuclei are visualized with DAPI (e,f). Ctx, cortex; LV, lateral ventricle; Str, striatum. Scale bars represent 20 m.

GFP

Dcx Ctx

Tbr2 Ctx

Tbr1 Ctx

RM

S
LV
RMS

RM S

LV

LV

Str

d
LV Third ventricle

GFP

Tbr2

2009 Nature America, Inc. All rights reserved.

Pax6 regulates Neurog1 and Neurog2 expression and these transcription factors are important to specify cortical neurons toward a glutamatergic fate17,18,20. In addition, the transcription factors Tbr1 and Tbr2 (also known as Eomes) are expressed in early postmitotic glutamatergic neurons and their intermediate progenitors21. Given the glutamatergic progeny of the dorsal telencephalon during development and its contribution to the adult SEZ, we searched for and found progenitors generating glutamatergic neurons in the adult dorsal SEZ. RESULTS Neurog2 and Tbr2 in a subset of dorsal SEZ progenitors We detected Neurog2- and Tbr2-immunoreactive cells in a dorsal subregion of the SEZ and the rostral migratory stream (RMS) (Fig. 1ad and Supplementary Fig. 1), whereas Pax6- and Mash1 (also known as Ascl1)-expressing progenitors were distributed in a more widespread manner (Supplementary Fig. 2). Consistent with their similar localization, most Neurog2-expressing cells (green fluorescent protein (GFP)-expressing cells in mice in which GFP had been inserted into the Neurog2 locus, referred to Neurog2+/gfp; Supplementary Fig. 1) were either Tbr2 or Tbr1 immunopositive (Tbr2, 96%, 394 Neurog2+/gfp cells; Tbr1, 70%, 151 Neurog2+/gfp cells; Fig. 1e,f). To further examine their identity, we also stained for doublecortin (Dcx), a protein that is exclusively present in neuroblasts and young postmitotic neurons. The majority of Neurog2+/gfp-positive (80 4%, 212 Neurog2+/gfp cells), Tbr2-positive (70 8%, 255 cells) and Tbr1-positive (97 2%, 265 cells) cells were also Dcx positive (Fig. 2ac). To distinguish postmitotic neuroblasts from proliferating neuronal progenitors, we stained for Ki67, a protein that is present in dividing cells, or provided the DNA-base analog BrdU, which is incorporated in proliferating cells. Tbr1-positive cells lacked Ki67 immunoreactivity and did not incorporate BrdU within a few hours of injection (1 1%, 201 cells; Fig.2), suggesting that virtually all of the Tbr1-positive cells were postmitotic neurons. Conversely, 37% of the GFP-positive cells (156cells) colocalized with Ki67 and 22% were labeled with BrdU (Fig. 2d,g), indicating that one-third of these cells comprise an actively dividing progenitor population. Likewise, a small subset of Tbr2-positive cells incorporated BrdU (Fig. 2e,g). During development, Tbr2 is mostly contained in progenitors undergoing a last round of cell division, whereas Tbr1 is present in postmitotic neurons21,22. To examine whether Tbr1-positive cells arise from proliferating progenitors, we injected BrdU intraperitoneally and waited for 3 d. A quarter of the Tbr1-positive cells labeled with BrdU ( Fig. 2g), consistent with Tbr1 being expressed when cells become postmitotic. To ensure that Tbr1-positive neuroblasts originated from Tbr2-positive progenitors, we used Tbr2BAC-gfp mice23 (Supplementary Fig. 3), which allowed us to carry out short-term fate mapping, as GFP persists for longer than
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009
Number of cells per 50-m coronal section
300 200 100 0

Dorsal Lateral Rostral

GFP

Tbr1

GFP positive

Tbr2 positive

LV

Tbr2 (ref. 24). Indeed, almost all of the Tbr1-immunoreactive cells in the RMS were GFP positive (Supplementary Fig. 3), consistent with the idea that they originate from Tbr2-expressing progenitors. Notably, none of these transcription factors (Neurog2, Tbr2 or Tbr1) were ever colocalized with GFAP, a protein that is characteristically expressed in astroglia, including NSCs (data not shown). However, we determined that the Tbr-expressing cells were derived from astrocytelike NSCs by genetic fate mapping with GLAST::creERT2 mediating recombination in GLAST-positive NSCs25 (Supplementary Fig. 4). Consistent with the recombination frequency observed on tamoxifeninduced recombination25,26, more than 60% of the Tbr2-positive cells were GFP positive, that is, derived from astrocyte-like NSCs (64% of the Tbr2-positive cells were Tbr2 and GFP double positive, 104 cells). Thus, an adult, NSC-derived subpopulation of actively dividing progenitors and young neuroblasts in the SEZ and RMS express a series of transcription factors that are normally indicative of a glutamatergic neuronal lineage. To determine whether Tbr1- or Tbr2-positive neuroblasts may nevertheless have GABAergic traits, we analyzed Dcx-positive neuroblasts in the RMS. Although most neuroblasts already express GAD, synthesize GABA27 and are GFP positive in Gad67 (also known as Gad1)::gfp28 and Gad65 (also known as Gad2)-gfp29 mice in which GFP protein expression is confined to cells expressing Gad67 or Gad65 mRNA (Fig. 3ac), Tbr1 and Tbr2 double-positive cells define a neuroblast subset that lacks both Gad67- (Fig. 3dg) and Gad65-driven GFP (Fig. 3h). Consistently, Tbr-positive cells did not contain Dlx (Fig. 3i). Moreover, also in the olfactory bulb, Tbr2-positive cells did not colocalize with any GABAergic interneuron markers, such as tyrosine hydroxylase, calbindin or calretinin (Supplementary Fig.5).
1525

a r t ic l e s a
Neurog2 +/gfp WM

Tbr2 WM

Tbr1 dSEZ

Figure 2 Neurog2, Tbr1 and Tbr2 expression defines a subset of neuroblasts. (ac) Fluorescent micrographs depicting Neurog2+/gfp, Tbr2 and Tbr1 staining in a subpopulation of Dcx-positive neuroblasts in the RMS and dorsal SEZ. (df) Neurog2+/gfp- and Tbr2-positive cells were labeled by BrdU after a short BrdU pulse, but Tbr1-positive cells ( f) were not BrdU positive. Boxed areas (gray) are shown at higher magnification at the bottom of the panels. (g) Histogram depicting the proportion of Neurog2+/gfp-, (black) Tbr2- (gray) and Tbr1-expressing (white) cells in the RMS, labeled with BrdU as indicated. Tbr2 and Neurog2+/gfp were detected in fast-proliferating cells, whereas Tbr1-positive cells were only labeled 3 d after BrdU application (n = 3 mice; number of cells analyzed: short pulse = 116 Neurog2+/gfp cells, 144 Tbr1-positive cells and 285 Tbr2-positive cells; 3 d = 190 Tbr1-positive and 166 Tbr2-positive cells). Error bars represent s.e.m. dSEZ, dorsal subependymal zone; WM, white matter. Scale bars represent 20 m.

Dcx

Dcx

Percentage of marker-positive cells labeled with BrdU

35 30 25 20 15 10 5 0

BrdU short pulse BrdU pulse + 3 d

Thus,Tbr1-positive neuroblasts, Tbr2-positive neuroblasts in the RMS and the neurons in the olfactory bulb did not express any traits of GABAergic neurons, suggesting that the Tbr-expressing neuroblasts may be predisposed to differentiate into a different, possibly glutamatergic, phenotype. Arrival of Tbr2-positive neuroblasts in the olfactory bulb Although Tbr2 and Tbr1 are clearly contained in migrating neuro blasts in the RMS, hardly any of the Tbr2- or Tbr1-immunoreactive cells present in the glomerular layer of the olfactory bulb were BrdU labeled at postnatal or adult stages (Supplementary Fig. 6), whereas they could be readily labeled by injection of BrdU at embryonic stages (Supplementary Fig. 6). Thus, Tbr1- or Tbr2-positive cells in the olfactory bulb are generated during embryonic development. This implies that the Tbr2-positive dividing progenitors present in the adult RMS either do not arrive or survive in the adult olfactory bulb, or, alternatively, that the expression of these transcription factors is transient and lost on arrival. To distinguish between these possibilities and to elucidate the fate of the Tbr2-positive progenitors in the RMS, we followed their progeny using Tbr2BAC-gfp mice for short-term fate mapping. GFP and Dcx double-positive cells were present in the RMS at the entrance to the olfactory bulb (Fig. 4a,b) and were distributed toward the glomerular layer (Fig. 4c). Some GFP-positive cells in the glomerular layer incorporated BrdU following 3 weeks of administration via their drinking water (Fig. 4di), suggesting that the progeny of the adult dividing Tbr2-positive progenitors indeed arrived in the olfactory bulb, where these cells then downregulated Tbr2 (Fig. 4jm) and Tbr1 protein (data not shown). Consistently, Tbr2-driven GFP vanished with a delay, as reflected by the lower GFP levels in the few juxtaglomerular cells generated and labeled by BrdU 3 weeks earlier that were negative for Tbr2 protein (Fig. 4jm).
1526

Adult Neurog2+ progenitors generate glutamatergic neurons To follow the neuronal differentiation of progenitors that have expressed Neurog2 and/or Tbr2 into later stages by a permanent marker, we used Cre-mediated recombination to activate a persistent reporter gene. As virtually all Neurog2-positive progenitors expressed either Tbr1 or Tbr2 (Fig. 1e,f), we used a mouse line expressing Cre under control of the Neurog2 enhancer element 1 (ref. 30) (E1Neurog2-cre) crossed with the Z/EG25 reporter line (referred to as E1Neurog2-cre; Z/EG ). As expected, Neurog2-creinduced GFP was detected in cells stained for Tbr1 in the RMS (Supplementary Fig. 7) and GFP-positive cells were not detectable in the ventral and lateral SEZ, but were present in the dentate gyrus (Supplementary Fig. 7). To assess whether adult-generated cells of the Neurog2 lineage can be found in the olfactory bulb, we stained newly generated neurons for Dcx. GFP and Dcx double-positive cells were indeed found in the glomerular layer of the olfactory bulb (Fig. 4n). To confirm that these cells are adult generated, we analyzed GFP-positive cells for incorporation of BrdU (3 weeks in drinking water followed by a 3-week BrdUfree period) and found many BrdU and GFP double-positive cells in the glomerular layer (Fig. 4oz), which amounted to 5% of the entire population of BrdU-positive cells that were present (Supplementary Fig. 8). Thus, a small population of adult-generated juxtaglomerular neurons is derived from a Neurog2-expressing lineage. Adult generation of glutamatergic short-axon cells In some cases, GFP expression in E1Neurog2-cre; Z/EG allowed us to assess the dendritic morphology of the GFP-labeled neurons that incorporated BrdU as adults. Three-dimensional reconstruction of these BrdU and GFP double-labeled cells from serial sections revealed dendritic arbors extending over 23 neighboring glomeruli (Fig. 4rt), a morphology that is typical for short-axon cells, a type of juxtaglomerular neuron mediating glutamatergic transmission between glomeruli31. To further corroborate the adult generation of these juxtaglomerular neurons, we used viral vectormediated lineage tracing. GFP-containing viral vectors were injected into the ventricle of 34-week-old mice (lentiviral vectors; Fig. 5al) or into the dorsal SEZ of adult mice (MLV-based retroviral vectors; Fig. 5mq), and GFPlabeled progeny were examined 6 weeks later in the glomerular layer (Fig. 5a,h,m). Similar to cells labeled by Neurog2-cre fate mapping, some of these virally transduced neurons settling in the vicinity of the glomeruli had long dendritic projections between different glomeruli. To examine whether the GFP-positive cells with this short-axon celllike morphology were indeed glutamatergic juxtaglomerular neurons, we stained for the vesicular glutamate transporter 2 (vGluT2; also
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

2009 Nature America, Inc. All rights reserved.

BrdU

BrdU 1-h pulse

a r t ic l e s
Figure 3 Presence of Tbr2 and Tbr1 defines a non-GABAergic subpopulation of neuroblasts. (ac) Virtually all of the GFP-positive cells in the SEZ (a,c) and RMS (b) colocalized with Gad67 (a,b) or Gad65 (c) mRNA in the respective Gad67::gfp and Gad65-gfp mouse lines, as indicated in the panels (see Online Methods). (dh) Notably, Tbr2 (d,h) and Tbr1 (e) were absent in cells expressing Gad65- (h) or Gad67- (dg) driven GFP. However, Tbr2- and Tbr1-positive cells were immunoreactive for the neuroblast marker Dcx in the dorsal SEZ (Tbr2, f) and RMS (Tbr1, g). (i) Consistently, Dlx transcription factors (pan-Dlx, green) were not coexpressed with Tbr1 (red). Boxed areas (orange) are shown at higher magnification for ae, h and i. Nuclei were visualized with DAPI (h). latSEZ, lateral subependymal zone. Scale bars represent 20 m.

Gad67::gfp mRNA

Tbr2

Tbr1 Tbr2 Gad65-gfp

known as Slc17a6). This transmitter transporter was predominantly located at presynaptic vesicles, as visualized by intense punctuate pattern via immunostaining32. Accordingly, immunostaining was strongest in the afferent fibers of the sensory neurons arriving in the center of the glomeruli (Supplementary Figs.4 and 5). However, we also observed vGluT2-immunoreactive cell somata of f virally traced GFP-positive cells with a shortaxon-like morphology (Fig. 5bd,il,nq). Although weak, this vGluT2 immuno reactivity was clearly specific, as neither juxtaglomerular cells with a different morphology (Fig. 5a,eg) nor Gad65- or Gad67-expressing neurons exhibited this somatic vGluT2 immunoreactivity (Supplementary Fig. 5). Notably, such cells were also reporter-positive 4 weeks after tamoxifen-induction of GLAST:: creERT2 mice25 that were crossed with the R26R-CFP33 reporter line (Supplementary Fig. 4). To further ensure the glutamatergic nature of the adult generated neurons, we combined in situ hybridization and BrdU labeling of adult-generated cells to detect the synthesis of the vesicular glutamate transporters in adult cells. Intense vGluT1 and vGluT2 (also known as Slc17a7 and Slc17a6, respectively) mRNA signals were observed in numerous cells located in the mitral cell layer, the external plexiform layer and the glomerular layer, but not in the GCL32,34 (Fig.6 and Supplementary Fig. 9). Consistent with the glutamatergic nature of the vGluT mRNAexpressing cells in the glomerular layer, none was found to express Gad67-driven GFP (data not shown and Supplementary Fig. 5). As expected, many vGluT-expressing cells colocalized with Tbr1 or Tbr2; however, a notable population of vGluT-expressing cells surrounding the glomeruli were not Tbr1 or Tbr2 positive (Supplementary Fig. 9). Indeed, consistent with the above observation that the adult-generated progeny of progenitors expressing both Neurog2 and Tbr2 downregulated Tbr protein expression in the olfactory bulb, some adult-born BrdU-positive nuclei were surrounded by vGluT2 mRNA (Fig. 6g), but BrdU and vGluT2 mRNA double-positive cells never colocalized with Tbr2 (Fig.6hj). BrdU seemed to not label dying cells, as we did not observe BrdU labeling in vGluT2-positive cells shortly after BrdU addition (data not shown). After sufficient time had passed to develop a glutamatergic phenotype, we found a small proportion of BrdU-positive cells (2 0.5%) in the glomerular layer that expressed vGluT2
Dcx Gad67::gfp

2009 Nature America, Inc. All rights reserved.

Tbr1 pan-Dlx

(Supplementary Fig. 8). None of the BrdU-labeled cells in the glomerular layer expressed vGluT1 (1,236 vGluT1-positive cells were analyzed; Fig. 6), suggesting that only vGluT2-expressing, but not vGluT1-expressing, neurons are adult generated. Thus, results from three independent techniques for labeling adult-generated neurons support the idea that glutamatergic juxtaglomerular neurons are generated in the adult olfactory bulb. To assess the functional integration of the newly generated glutamatergic neurons, we immunostained for the immediate early gene c-fos (also known as Fos), which is known to be regulated by neuronal activity 36 weeks after BrdU administration35. c-fos immunoreactivity was detected in the majority of cells that were positive for both BrdU and vGluT2 mRNA (86%, 60 cells, three mice; Fig. 6k,l), suggesting that most of the adult-generated glutamatergic neurons become functionally recruited about the same time as their GABAergic counterparts. To further scrutinize the functional integration of these adultgenerated glutamatergic neurons, we carried out ultrastructural analysis in E1Neurog2-cre; Z/EG fate-mapped cells in the glomerular layer that were stained for GFP with the electrondense DAB reaction product (Supplementary Fig. 10). Cells with glutamatergic morphology and at the position of the above described glutamatergic neurons were selected for electron microscopy and synapses formed by DABlabeled processes were examined (Supplementary Fig. 10). Consistent with previous evidence that vGluT expression parallels the formation of asymmetric synapses in the olfactory bulb32, GFP-positive presynaptic terminals established asymmetric contacts onto target cells in the glomerular layer (Supplementary Fig. 10), confirming that juxtaglomerular neurons derived from Neurog2-positive precursors are excitatory and synaptically connected in vivo.
1527

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

Gad65-gfp mRNA latSEZ

a r t ic l e s
Tbr2
BAC-gfp

a
GL

b
GFP RMS OB GL BrdU

RMS
GFP Dcx

c
OB

Tbr2

Merge

d
BrdU 3 weeks

e f

E1 Neurog2-cre; Z/EG

2009 Nature America, Inc. All rights reserved.

o
GFP BrdU

r
GFP BrdU

s
BrdU

u
GFP Tbr2 BrdU

v
GFP Tbr1 BrdU

x
GFP CR BrdU

Figure 4 Fate mapping of Tbr2- or Neurog2-derived progeny in the olfactory bulb. ( ac) Fluorescent micrographs depicting Tbr2BAC-gfp-positive cells in the RMS and olfactory bulb (green) and Dcx-positive neuroblasts (red). Boxed areas (white) are shown in higher magnifications in b and c, depicting cells in the RMS entering the olfactory bulb (OB, b) and on their way to the glomerular layer (GL, c). (di) Fluorescent micrographs of examples of BrdU-labeled Tbr2BAC-gfp-positive cells (arrows) that were apparent in the glomerular layer of the olfactory bulb 3 weeks after BrdU labeling. (jm) Consistently, the BrdU and Tbr2BAC-gfp fate-mapped cells (example indicated by arrow) lacked Tbr2 protein. ( n) Fate mapping of Neurog2-derived progeny in the glomerular layer of the olfactory bulb using E1Neurog2-cre; Z/EG mice revealed the presence of GFP-positive neuroblasts (Dcx, red). ( oq) BrdU-labeled juxtaglomerular cells expressing GFP derived from Neurog2 expression (E1Neurog2-cre; Z/EG ). BrdU was given in the mices drinking water for 3 weeks followed by a 3-week, BrdU-free period. Boxed area (gray) is shown at higher magnification as a Z projection ( p) and as a single optical section ( q). (rt) Three-dimensional reconstruction of the dendritic arborization of a Neurog2 -derived adult-labeled GFP and BrdU double-positive cell. ( s,t) The reconstructed cell is derived from serial sections of a cell labeled for GFP by E1Neurog2-cre; Z/EG fate mapping and for adult generation by BrdU incorporation (red). BrdU was given in the mices drinking water for 3 weeks followed by a 3-week, BrdU-free period. ( uz) Adult-born Neurog2 -derived GFP-positive cells that incorporated BrdU (blue) were negative for Tbr2 (red, u,v), Tbr1 (red, w,x) and calretinin (CR, red, y,z). The images shown in u, w and y are Z projections, and single optical sections are shown in v, x and z. Scale bars represent 20 m.

To examine the synapses formed by adult-generated glutamatergic neurons at the physiological level, we used an in vitro system, as the identification of the synaptic targets of a defined neuron by paired recordings in acute slices is virtually impossible. We took advantage of a previously established culture system of primary SEZ progenitors that allow for the maintenance of adult SEZ cells in the absence of mitogens; under these conditions, these cells give rise predominantly to neuronal progeny, thereby reflecting their endogenous lineage10. The majority of cells in this culture system are Dlx postive10, as observed in vivo and consistent with their GABAergic fate. However, consistent with our in vivo observations, we also detected a small percentage of Tbr2-positive cells (1.5%; Fig. 7) that could be transduced with a retroviral vector expressing GFP (Fig.7c). Notably, the number of Tbr2-positive cells increased when cells were isolated from the dorsal SEZ, whereas virtually no Tbr-positive cells were contained in the lateral SEZ (Fig. 7b). After a longer differentiation period, a small proportion of the cells was immunoreactive for vGluT, which became concentrated in puncta after several weeks
1528

GFP Dcx

in vitro, suggesting the formation of functional glutamatergic synapses (Fig. 7d,e and Supplementary Fig. 11). When mature neurons were monitored by synapsin (also known as Syn1) promoterdriven GFP36, we found that 3% (598 cells) of the cells differentiated into a glutamatergic (vGluT positive) phenotype. To unambiguously determine the functional nature of glutamatergic synapses, we performed perforated patch recordings37 (Fig. 7e and Supplementary Fig. 11). On the basis of our observation that vGluT-positive neurons typically possessed relatively large somata, we focused our analysis on large GFP-positive neurons (Fig. 7e and Supplementary Fig. 11) and found that nine out of ten had a 6-cyano-7-nitroquinoxaline2,3-dione (CNQX)-sensitive autaptic response (Fig. 7e). Notably, neurons with such a response were indeed vGluT immunopositive (Supplementary Fig.11). Furthermore, using paired recordings, we found synaptically connected excitatory neurons in these cultures (SupplementaryFig.11), indicating that adult SEZ progenitors generate glutamatergic neurons that form functional synapses in vitro. Taken together, the data from our ultrastructural analysis, c-fos
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 5 Viral vectormediated fate mapping of vGluT2-expressing juxtaglomerular neurons. (ai) GFP-encoding lentivirus injections into the murine ventricle at 3 weeks of age resulted in numerous GFP-positive cells in the glomerular layer of the olfactory bulb 6 weeks later. Some GFP-expressing cells had vGluT2-immunoreactive soma ( bd, red box in a, il and orange box in h), although the majority of GFP-expressing transduced cells lacked somatic vGluT2 immunoreactivity ( eg, greenblue box in a). (mq) Retroviral vector injection into the dorsal regions of the adult SEZ, analyzed 6 weeks postinjection, also resulted in some GFP-positive juxtaglomerular neurons with somatic vGluT2 (red) immunoreactivity. Nuclei were visualized with DAPI ( d,g,k,l,m,p,q). Insets in a and h show an overview of the juxtaglomerular location of the GFP-labeled neurons by counterstaining with DAPI. The images shown in bg, il and nq are single optical sections, whereas the images in a, h and m are collapsed stacks to show the morphology vGluT2-immunoreactive juxtaglomerular cells. Scale bars represent 20 m (a,h,m) and 5 m (bg,il,nq).
Lentiviral injection

a
vGluT2 GFP DAPI

2009 Nature America, Inc. All rights reserved.

ncorporation and in vitro physiology experiments support the idea i that glutamatergic olfactory neurons are generated in the adult. Tbr2+ cells migrate toward sites of neocortical injury One of the implications of endogenous glutamatergic progenitors being present in the adult SEZ is the possibility that they can be recruited to sites of neocortical injury. We therefore employed an injury model that has previously been shown to elicit repair of cortical projection neurons from endogenous sources of progenitors6,7. Chlorine e6coupled beads were injected into the cortex to retrogradely label callosal projection neurons in the contralateral hemisphere7. Indeed, 7 d after injection, neurons containing the beads were observed in layers 2, 3 and 5 of the contralateral site (Supplementary Fig. 12). Subsequent (7 d after beads injection) laser illumination of the contralateral hemisphere activated the chlorine e 6, killing the bead-containing neurons (as confirmed by staining for activated caspase 3; Supplementary Fig. 12). When we injected GFP-encoding retroviral vectors into the adult SEZ 2 d before laser illumination, some of the transduced GFP-positive cells migrating from the SEZ toward the lesion were Tbr2 immunoreactive (Fig. 8a). We found clusters of Tbr2 and Dcx double-positive neuroblasts in the corpus callosum 1 week after laser exposure (Fig. 8b,c), and some had entered the cortical gray matter (n = 3 mice; Fig. 8d), whereas such an invasion was never observed in control mice (n = 4 mice). Consistent with the lesion of callosal projection neurons, a fraction of Tbr2BAC-gfppositive cells were found in cortical layer 2 and 3 10 d after laser illumination. Accordingly, Foxp2 immunoreactivity (labeling lower cortical layers) was not detectable in these Tbr2-gfppositive cells (Supplementary Fig. 12). In contrast, some of them were Cux1 positive (Supplementary Fig. 12), which is indicative of an upper-layer neuron identity. The staining was specific, as other cells at similar positions were still Cux1 immunonegative (Supplementary Fig. 12). Notably, none of the adult-generated neurons in the olfactory bulb were Cux1 positive, suggesting that its expression in cells migrating into the injured cortex reflects the acquisition of a distinct glutamatergic subtype identity. Taken together, these data suggest that at least some of the Tbr2-positive cells in the SEZ and RMS serve as an endogenous source of progenitors that can be recruited to the cerebral cortex on injury. DISCUSSION To the best of our knowledge, this is the first description of adult generation of glutamatergic juxtaglomerular neurons of the olfactory
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

Retroviral injection

vGluT2

GFP

DAPI

bulb. These glutamatergic neurons arise from a lineage that is characterized by the sequential expression of the transcription factors Neurog2, Tbr2 and Tbr1 (expressed in that order) and that originates in the dorsal SEZ. These findings illustrate an unsuspected diversity of SEZ progenitors in regard to a glutamatergic transmitter phenotype of their progeny in the olfactory bulb that can be defined by their location in the dorsal SEZ and their combinatorial expression of different transcriptional cues. Glutamatergic neurogenesis in the adult olfactory bulb Thus far, the adult SEZ had been viewed as a region giving rise exclusively to GABAergic interneurons. We provide several independent lines of evidence that a subset of adult SEZ NSCs generates glutamatergic neurons. First, we found that some BrdU-labeled cells in the glomerular layer expressed vGluT2, a transporter for glutamate into synaptic vesicles. Previous electron microscopic analysis in the adult olfactory bulb has shown that cells containing vGluT2 are exclusively neurons forming asymmetric excitatory synapses32, indicating that the cells that we found are glutamatergic. Second, genetic and viral fate mapping showed that some adult-generated neurons in the glomerular layer had somatic immunoreactivity for vGluT2 protein32, consistent with mRNA expression, identifying these neurons
1529

a r t ic l e s a
vGluT2

b
GL

vGluT2

vGluT1

d
GL gm

vGluT2 gm

Ctx LV EPL OB MCL GCL MCL GCL EPL

EPL

vGluT1 gm

gm

EPL

f
3 weeks BrdU + 4 weeks

vGluT1 BrdU

vGluT2 BrdU Tbr2

h
vGluT2

vGluT2 c-fos BrdU

i
BrdU

GL

2009 Nature America, Inc. All rights reserved.

j
Merge EPL

Figure 6 A subpopulation of newly generated juxtaglomerular neurons expresses vGluT2 mRNA. (ae) In situ hybridization for the vGluT2 (a,b,d) or vGluT1 (c,e) mRNA in the adult olfactory bulb revealed that they were expressed in the mitral cell layer (MCL), external plexiform layer (EPL) and glomerular layer. In contrast, no mRNA signal for vGluT1 or vGluT2 was detected in the GCL (ac). Boxed areas (red) are shown at higher magnification in d and e. Note that only vGluT2 mRNA was detected between glomeruli (gm), whereas vGluT1 mRNA was largely restricted to the external plexiform layer underlying the glomerular layer. (f) Micrograph of BrdU (green) and vGluT1 mRNA (black) staining. vGluT1-expressing cells did not incorporate BrdU. (g) An example of a vGluT2expressing juxtaglomerular cell labeled by BrdU (green) that was Tbr2 negative (red). Boxed area (red) is shown in higher magnification in hj. (k,l) Two examples of vGluT2-expressing cells that incorporated BrdU (red) and were immunopositive for c-fos (green). Boxed area in k (red) is shown at higher magnification at the lower right corner of the panel. Z projections in f and g are shown below and to the right of the panels. Scale bars represent 20 m (b,c,l), 10 m (hj), 50 m (f,g) and 100 m (k).

as being glutamatergic. Third, by genetic fate mapping, we found that BrdU-positive neurons generated in the adult and derived from Neurog2-positive progenitors reached the glomerular layer and had the morphology of short-axon cells. This notion is further confirmed at the ultrastructural level by our observation of asymmetric synaptic

contacts established by the cells derived from the Neurog2-progenitor pool. Fourth, consistent with the existence of adult NSCs generating glutamatergic neurons, we found a small proportion of glutamatergic neurons that formed physiologically functional synapses in vitro. Thus, our in vitro and in vivo data suggest a previously unknown

Figure 7 Adult SEZ stem and progenitor 10 Adult SEZ, cells give rise to some glutamatergic neurons 9 3 DIV in vitro . (a) Two example SEZ neuroblasts 8 7 expressing Dcx (green) and Tbr2 (red) after 6 3 d in vitro (3 DIV). Boxed area shows 5 Tbr2-positive nuclei at higher magnification. 4 (b) Quantification of Tbr2-positive nuclei 3 in cultured SEZ stem and progenitor cells 2 isolated from the dorsal, lateral or both 1 0 walls of the lateral ventricle after 3 DIV. Adult SEZ, 3 DIV All SEZ Lateral Dorsal Tbr2-positive cells were enriched in cultures derived from the dorsal SEZ (n = 3 experiments; there were 471 cells in the dorsal cultures, 602 cells in the ventral cultures and 1,384 cells in the mixed cultures; error bars represent s.e.m.). ( c) Cells that were double positive for Tbr2 (red) and III-tubulin (blue) originated from proliferating progenitors, as indicated by transduction with a GFP-encoding retrovirus shortly after plating. Adult SEZ, 3 DIV Adult SEZ, 28 DIV (d) vGluT (red) immunoreactivity was also detected in GFP-positive cells 4 weeks after CNQX 5 M 100 pA retroviral transduction, indicating that they 10 ms originated from proliferating progenitors. (e) Left, micrograph shows an adult SEZ-derived glutamatergic neuron transduced with a lentiviral vector expressing GFP under the synapsin promoter that was patched by the recording electrode (inset). Middle, Adult SEZ, 28 DIV vGluT GFP Synapsin-GFP, 14 DPl immunostaining for vGluT and GFP after recording revealed that this cell was highly decorated with vGluT-positive puncta. Right, stimulation of the neuron evoked an autaptic response that was blocked by the AMPA/kainate receptor antagonist CNQX, revealing its glutamatergic nature. Nuclei were visualized with DAPI ( a,d). Scale bars represent 20 m.

Tbr2 GFP lll-Tubulin

1530

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

vGluT GFP DAPI

Percentage of DAPI-positive cells that were Tbr2 positive

Tbr2 Dcx DAPI

a r t ic l e s
Figure 8 Tbr2-positive cells migrate into the lesioned cerebral cortex. (a) Retroviral injections of GFP-encoding vectors 2 d before laser illumination resulted in GFP-positive cells that migrated away from the RMS 34 d after lesion, some of which were Tbr2 immunoreactive (red). (b) Fluorescence micrograph depicting Tbr2and Dcx-immunopositive cells in an overview of the lesioned cerebral cortex hemisphere and the RMS 1 week after chlorine e6induced lesion. The schematic drawing to the left depicts the location of the arrows in b on the right side and the location of the micrographs shown in c and d on the left side. (c,d) Tbr2-and Dcx-positive cells were present in dorsal regions of the SEZ and RMS (white arrows in c) and were sometimes found in small clusters (c) or in the cortical gray matter (d). Boxed areas (gray) are shown at higher magnifications. Nuclei were visualized with DAPI (ad). Scale bars represent 20 m (a,c,d) and 50 m (b).

Retroviral injection

Lesion site DAPI GFP

Tbr2 GFP

RMS LV

Tbr2

b
d c b Ctx

RMS
Tbr2 Dcx

2009 Nature America, Inc. All rights reserved.

source for adult-generated glutamatergic neurons in the adult SEZ. c The genetic and viral fate-mapping experiments permitted us to obtain more precise information on the morphology of the newly generated glutamatergic neurons in the glomerular layer. Distinct from GABAergic periglomerular neurons (PGNs), these neurons typically extended their dendritic arbors adjoining several glomeruli, characteristic of so-called short-axon cells31,38, a glutamatergic subpopulation of juxtaglomerular neurons. Notably, despite their small number, shortaxon cells are important for center-surround inhibition among glomeruli, providing excitatory glutamatergic drive to inhibitory PGNs of up to 30 glomeruli31. However, we cannot exclude that other subpopulations of juxtaglomerular glutamatergic neurons are also generated in the adult, potentially in a similar diversity of subtypes as observed for the GABAergic PGNs. The physiological maturation and integration of these cells is supported by the expression of the immediate-early gene c-fos, a member of a family of transcription factors that are rapidly regulated by neuronal activity35. Detection of c-fos has previously been used to demonstrate the integration of newborn neurons in the adult dentate gyrus39 and the olfactory bulb35. The expression of c-fos in the BrdU and vGluT2 double-positive neurons, as well as the presence of asymmetric synaptic contacts established by Neurog2-derived neurons, which we observed by electron microscopy, both support a functional integration of these newborn glutamatergic neurons into the neuronal network several weeks after their birth. One question is what is the functional role of this specific glutamatergic interneuron population. Indeed, even small subsets of olfactory neurons can have a profound functional effect on olfactory information processing40 and the specificity of adult generation of vGluT2expressing neurons in the glomerular layer suggests that they have a specific role in the glomerular layer neuronal network, the first synaptic station of the incoming afferents. Synapses using vGluT2 for loading synaptic vesicles generally have a higher release probability compared with those expressing vGluT1 (ref. 41). Given the importance of short-axon cells in center-surround inhibition, it can be speculated that the new addition of glutamatergic neurons, as with their
Tbr2 Dcx

LV

d
Dcx

Tbr2

DAPI

Tbr2

Dcx

GABAergic counterparts, is crucial for modifying the interglomerular circuitry during olfactory discrimination learning42. SEZ regionalization and neuronal subtype specification Recent evidence indicates that the SEZ in adult mice is highly regionalized, with stem cells in different locations having distinct embryonic origins and neuronal fates5,13,15,16. Here, we identified a previously unknown SEZ progenitor population that is specifically localized in dorsal regions and undergoes the same transcription factor expression sequence that characterizes many glutamatergic subtypes throughout the brain12,21. Sequential expression of Tbr2 and Tbr1 in developing neurons is a hallmark of glutamatergic lineage throughout the brain21. In the developing cerebral cortex and hippocampus, Tbr2-expressing progenitors arise from Neurog2-positive progenitors that themselves are generated by Pax6-expressing radial glia43. We found that this transcription factor sequence was well conserved in the adult dorsal SEZ (Supplementary Figs. 2 and 13). Although Pax6 was expressed in most Neurog2- or Tbr2-expressing cells, the latter never colocalized with Dlx. Notably, the Neurog2-expressing lineage also originated from Mash1-expressing progenitors, as many Neurog2-positive cells in the RMS are also Mash1 positive (Supplementary Fig. 2). Short-term fate mapping by Mash1BAC-gfp mice also revealed that Tbr2-positive progenitors originated from the Mash1 lineage (Supplementary Fig. 2). These data imply that there are at least two different lineages arising from Pax6- and Mash1-expressing progenitors (Supplementary Fig. 13), a larger population that leads to neuroblasts expressing both Dlx and Pax6
1531

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

a r t ic l e s
and a smaller population sequentially expressing Pax6, Neurog2 and Tbr2 (Supplementary Fig. 13). Thus, precursors expressing Pax6 and Mash1 can give rise to both GABA and glutamatergic lineages, raising the question of which signals regulate lineage progression. One candidate that might instruct dorsal SEZ progenitors toward a glutamatergic fate is p-cateninmediated Wnt signaling, which is involved in upregulation of Neurog1 via Lef1 that directly binds to the Neurog1 promoter in vitro in embryonic telencephalic progenitors44. Notably, Neurog1 and Neurog2 are typically expressed together 20. Once Neurog2 is upregulated, its functional relevance in the specification of glutamatergic neurons from adult SEZ cells is evident not only from developmental analysis20,45, but also from the fact that overexpression of Neurog2 in adult-derived neurosphere cells is sufficient to upregulate Tbr1 and to direct virtually all neurons toward a functional glutamatergic identity37. A source for regenerating glutamatergic neurons Although neurogenesis in most regions of the telencephalon, including the striatum and cerebral cortex, ceases at early postnatal stages, the ongoing olfactory neurogenesis serves as a potential source of neuro blasts after injury. Indeed, many new neuroblasts migrate into the striatum and generate GABAergic neurons after stroke46. Neuroblasts from the SEZ and RMS also migrate into the cerebral cortex, as has been observed after postnatal injury47,48 or adult cerebral cortex lesion with chlorine e6induced cell death6,7. Using the latter model in conjunction with retroviral lineage tracing, we found that phototoxic lesion of the cerebral cortex resulted in rerouting of some SEZ progenitors out of the RMS toward the damaged cortex. Some of the recruited SEZ progenitors expressed Tbr2, consistent with their derivation from the glutamatergic lineage. On settling in the damaged cortex, some of these neurons appear to acquire the appropriate layer identity, as suggested by expression of Cux1 in Tbr2-driven GFP-positive cells in upper cortical layers following lesion of callosal projection neurons. In contrast, none of the adult-generated glutamatergic neurons in the glomerular layer were found to express Cux1. These data suggest that progenitors destined to generate glutamatergic juxtaglomerular neurons of the olfactory bulb can be diverted toward a different, in this case a cortical projection neuron, fate6,7. Although the molecular mechanisms underlying such respecification remain unclear, they may involve local cues provided by the injured tissue. Notably, respecification of SEZ-derived cells toward a different neuronal identity has been described with cells migrating after stroke from the SEZ into the striatum, where they then acquire a Darp32-positive medium spiny neuronal phenotype that is normally not generated in adult olfactory bulb neurogenesis46. Taken together, our discovery of ongoing generation of glutamatergic juxtaglomerular neurons highlights not only the diversity of adult-generated olfactory neurons in this region, but also the importance of understanding the molecular code of the distinct subtypes of adult progenitors as a major step toward using these neurons for repair. Indeed, the recent discovery of molecular fate determinants for subsets of cortical projection neurons may allow directing the source of adult progenitors for glutamatergic neurons identified here more efficiently toward the repair of cortical projection neurons. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We would like to thank F. Bedogni for invaluable advice on methods and we are particularly grateful to Y. Yanagawa for the Gad67::gfp mice. We would also like to thank T. Simon-Ebert, A. Steiner, T. ztrk and S. Nestel for excellent technical assistance and S. Bauer for retroviral production. We would also like to thank A. Saghatelyan for invaluable discussions and comments on the manuscript. This work was supported by grants from the Deutsche Forschungsgemeinschaft, including the excellence cluster Center for Integrated Protein Science Munich, by European Transcriptome, Regulome and Cellular Commitment Consortium from the European Union and the Bundesministerium fr Bildung und Forschung to M.G., and the Bavarian State Ministry of Sciences, Research and the Arts (ForNeuroCell) to B.B. and M.G. E.W. was supported by a fellowship from the Medical Research Council, I.O. by a fellowship from the John & Lucille van Geest Foundation and O.R. by a grant from the Dr. Scholl Foundation. Research in the laboratory of F.G. is supported by funds from the Medical Research Council and grants from the European Commission Research and Technological Development program. C.P. is presently supported by an AVENIR-contract from Institut national de la sant et de la recherch mdicale and Fondation de France. R.D.H. is supported by a Heart and Stroke Fund of Canada Research Fellowship. Work done in the laboratory of R.F.H. was supported by grants from the US National Institutes of Health (R01 NS050248 and R01 MH080766). AUTHOR CONTRIBUTIONS M.S.B., O.R. and R.D.H. made the original observation of glutamatergic progenitors in the SEZ. M.S.B. conducted most of the experiments. J.N., E.W., O.R. and R.D.H. conducted experiments crucial for the identification of adult-generated glutamatergic neurons in the olfactory bulb. I.O. and R.Y. contributed to experiments using Neurog2+/gfp and Tbr2BAC-gfp mice, respectively. A.L. and S.G. designed and produced the viral vectors. F.E. and G.S. provided the Gad65-gfp mice. C.P. contributed with experiments using Ngn2+/gfp and Mash1BAC-gfp mice. F.G. provided the Neurog2gfp/+ and E1Neurog2-cre mice. M.F. designed the electron-microscopy experiments and analyzed the data. B.B. designed and conducted all of the electrophysiological experiments. R.F.H., O.R. and M.G. supervised the project and designed experiments. M.G. wrote the manuscript. M.S.B., B.B., R.D.H., R.F.H. and O.R. contributed to the writing of the manuscript.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
1. Kokaia, Z. & Lindvall, O. Neurogenesis after ischaemic brain insults. Curr. Opin. Neurobiol. 13, 127132 (2003). 2. Sohur, U.S., Emsley, J.G., Mitchell, B.D. & Macklis, J.D. Adult neurogenesis and cellular brain repair with neural progenitors, precursors and stem cells. Phil. Trans. R. Soc. Lond. B 361, 14771497 (2006). 3. Ming, G.L. & Song, H. Adult neurogenesis in the mammalian central nervous system. Annu. Rev. Neurosci. 28, 223250 (2005). 4. Gage, F.H. Mammalian neural stem cells. Science 287, 14331438 (2000). 5. Merkle, F.T., Mirzadeh, Z. & Alvarez-Buylla, A. Mosaic organization of neural stem cells in the adult brain. Science 317, 381384 (2007). 6. Chen, J., Magavi, S.S. & Macklis, J.D. Neurogenesis of corticospinal motor neurons extending spinal projections in adult mice. Proc. Natl. Acad. Sci. USA 101, 1635716362 (2004). 7. Magavi, S.S., Leavitt, B.R. & Macklis, J. D. Induction of neurogenesis in the neocortex of adult mice. Nature 405, 951955 (2000). 8. Lledo, P.M., Alonso, M. & Grubb, M.S. Adult neurogenesis and functional plasticity in neuronal circuits. Nat. Rev. Neurosci. 7, 179193 (2006). 9. Wonders, C.P. & Anderson, S.A. The origin and specification of cortical interneurons. Nat. Rev. Neurosci. 7, 687696 (2006). 10. Brill, M.S. et al. A dlx2- and pax6-dependent transcriptional code for periglomerular neuron specification in the adult olfactory bulb. J. Neurosci. 28, 64396452 (2008). 11. Waclaw, R.R. et al. The zinc finger transcription factor Sp8 regulates the generation and diversity of olfactory bulb interneurons. Neuron 49, 503516 (2006). 12. Bertrand, N., Castro, D.S. & Guillemot, F. Proneural genes and the specification of neural cell types. Nat. Rev. Neurosci. 3, 517530 (2002). 13. Hack, M.A. et al. Neuronal fate determinants of adult olfactory bulb neurogenesis. Nat. Neurosci. 8, 865872 (2005). 14. Roybon, L., Deierborg, T., Brundin, P. & Li, J.Y. Involvement of Ngn2, Tbr and NeuroD proteins during postnatal olfactory bulb neurogenesis. Eur. J. Neurosci. 29, 232243 (2009). 15. Young, K.M., Fogarty, M., Kessaris, N. & Richardson, W.D. Subventricular zone stem cells are heterogeneous with respect to their embryonic origins and neurogenic fates in the adult olfactory bulb. J. Neurosci. 27, 82868296 (2007). 16. Kohwi, M. et al. A subpopulation of olfactory bulb GABAergic interneurons is derived from Emx1- and Dlx5/6-expressing progenitors. J. Neurosci. 27, 68786891 (2007).

2009 Nature America, Inc. All rights reserved.

1532

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
17. Kroll, T.T. & OLeary, D.D. Ventralized dorsal telencephalic progenitors in Pax6 mutant mice generate GABA interneurons of a lateral ganglionic eminence fate. Proc. Natl. Acad. Sci. USA 102, 73747379 (2005). 18. Nikoletopoulou, V. et al. Neurotrophin receptormediated death of misspecified neurons generated from embryonic stem cells lacking Pax6. Cell Stem Cell 1, 529540 (2007). 19. Dellovade, T.L., Pfaff, D.W. & Schwanzel-Fukuda, M. Olfactory bulb development is altered in small-eye (Sey) mice. J. Comp. Neurol. 402, 402418 (1998). 20. Schuurmans, C. et al. Sequential phases of cortical specification involve Neurogenindependent and -independent pathways. EMBO J. 23, 28922902 (2004). 21. Hevner, R.F., Hodge, R.D., Daza, R.A. & Englund, C. Transcription factors in glutamatergic neurogenesis: conserved programs in neocortex, cerebellum, and adult hippocampus. Neurosci. Res. 55, 223233 (2006). 22. Hevner, R.F. et al. Tbr1 regulates differentiation of the preplate and layer 6. Neuron 29, 353366 (2001). 23. Geschwind, D. GENSAT: a genomic resource for neuroscience research. Lancet Neurol. 3, 82 (2004). 24. Kowalczyk, T. et al. Intermediate neuronal progenitors (basal progenitors) produce pyramidal-projection meurons for all layers of cerebral cortex. Cereb. Cortex 19, 24392450 (2009). 25. Ninkovic, J., Mori, T. & Gotz, M. Distinct modes of neuron addition in adult mouse neurogenesis. J. Neurosci. 27, 1090610911 (2007). 26. Mori, T. et al. Inducible gene deletion in astroglia and radial gliaa valuable tool for functional and lineage analysis. Glia 54, 2134 (2006). 27. Liu, X., Wang, Q., Haydar, T.F. & Bordey, A. Nonsynaptic GABA signaling in postnatal subventricular zone controls proliferation of GFAP-expressing progenitors. Nat. Neurosci. 8, 11791187 (2005). 28. Tamamaki, N. et al. Green fluorescent protein expression and colocalization with calretinin, parvalbumin and somatostatin in the GAD67-GFP knock-in mouse. J. Comp. Neurol. 467, 6079 (2003). 29. Lpez-Bendito, G. et al. Preferential origin and layer destination of GAD65-GFP cortical interneurons. Cereb. Cortex 14, 11221133 (2004). 30. Berger, J. et al. E1-Ngn2/Cre is a new line for regional activation of Cre recombinase in the developing CNS. Genesis 40, 195199 (2004). 31. Aungst, J.L. et al. Centre-surround inhibition among olfactory bulb glomeruli. Nature 426, 623629 (2003). 32. Gabellec, M.M., Panzanelli, P., Sassoe-Pognetto, M. & Lledo, P.M. Synapse-specific localization of vesicular glutamate transporters in the rat olfactory bulb. Eur. J. Neurosci. 25, 13731383 (2007). 33. Srinivas, S. et al. Cre reporter strains produced by targeted insertion of EYFP and ECFP into the ROSA26 locus. BMC Dev. Biol. 1, 4 (2001). 34. Ohmomo, H. et al. Postnatal changes in expression of vesicular glutamate transporters in the main olfactory bulb of the rat. Neuroscience 160, 419426 (2009). 35. Magavi, S.S., Mitchell, B.D., Szentirmai, O., Carter, B.S. & Macklis, J.D. Adult-born and preexisting olfactory granule neurons undergo distinct experience-dependent modifications of their olfactory responses in vivo. J. Neurosci. 25, 1072910739 (2005). 36. Gascn, S., Paez-Gomez, J.A., Diaz-Guerra, M., Scheiffele, P. & Scholl, F.G. Dualpromoter lentiviral vectors for constitutive and regulated gene expression in neurons. J. Neurosci. Methods 168, 104112 (2008). 37. Berninger, B., Guillemot, F. & Gotz, M. Directing neurotransmitter identity of neurones derived from expanded adult neural stem cells. Eur. J. Neurosci. 25, 25812590 (2007). 38. Pinching, A.J. & Powell, T.P. The neuron types of the glomerular layer of the olfactory bulb. J. Cell Sci. 9, 305345 (1971). 39. Jessberger, S. et al. Seizure-associated, aberrant neurogenesis in adult rats characterized with retrovirus-mediated cell labeling. J. Neurosci. 27, 94009407 (2007). 40. Eyre, M.D., Kerti, K. & Nusser, Z. Molecular diversity of deep short-axon cells of the rat main olfactory bulb. Eur. J. Neurosci. 29, 13971407 (2009). 41. Murphy, G.J. & Isaacson, J.S. Presynaptic cyclic nucleotidegated ion channels modulate neurotransmission in the mammalian olfactory bulb. Neuron 37, 639647 (2003). 42. Mouret, A. et al. Learning and survival of newly generated neurons: when time matters. J. Neurosci. 28, 1151111516 (2008). 43. Scardigli, R., Baumer, N., Gruss, P., Guillemot, F. & Le Roux, I. Direct and concentration-dependent regulation of the proneural gene Neurogenin2 by Pax6. Development 130, 32693281 (2003). 44. Israsena, N., Hu, M., Fu, W., Kan, L. & Kessler, J.A. The presence of FGF2 signaling determines whether beta-catenin exerts effects on proliferation or neuronal differentiation of neural stem cells. Dev. Biol. 268, 220231 (2004). 45. Parras, C.M. et al. Divergent functions of the proneural genes Mash1 and Ngn2 in the specification of neuronal subtype identity. Genes Dev. 16, 324338 (2002). 46. Arvidsson, A., Collin, T., Kirik, D., Kokaia, Z. & Lindvall, O. Neuronal replacement from endogenous precursors in the adult brain after stroke. Nat. Med. 8, 963970 (2002). 47. Goings, G.E., Sahni, V. & Szele, F.G. Migration patterns of subventricular zone cells in adult mice change after cerebral cortex injury. Brain Res. 996, 213226 (2004). 48. Faiz, M. et al. Substantial migration of SVZ cells to the cortex results in the generation of new neurons in the excitotoxically damaged immature rat brain. Mol. Cell. Neurosci. 38, 170182 (2008).

2009 Nature America, Inc. All rights reserved.

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1533

ONLINE METHODS
Animals. All animal procedures were performed in accordance with protocols approved by the state of Bavaria and in accordance with the UK (Scientific Procedures) Animals Act 1986. For wild-type analysis, we used C57BL/6 mice. In addition, we used mice with an enhanced GFP cassette inserted into either the Neurogenin2 locus (Neurog2+/gfp)49 or the Gad67 locus (Gad67::gfp)28 and mice that expressed the Cre recombinase in the GLAST locus25 crossed with the R26R-CFP reporter33 to track recombined cells (GLAST::creERT2 R26R-CFP). For induction of Cre-mediated recombination, we injected mice injected twice per day with 50 l tamoxifen (20 mg ml1, dissolved in corn oil, Sigma) for 5 d consecutively, followed by a 1-week interval and then another 5-d series of injections. We also used BAC transgenic mice expressing GFP under the Mash1 (also known as Ascl1) or Tbr2 genomic regulatory sequences (Mash1BAC-gfp, Tbr2BAC-gfp)23,50, transgenic mice that express GFP under the Gad65 promoter (Gad65-gfp)29 and mice expressing the Cre recombinase under the enhancer element 1 of Neurogenin2 (E1Neurog2-cre)30,43 crossed with the Z/EG25 reporter (E1Neurog2-cre; Z/EG) to follow progeny derived from the Ngn2 derived lineage. All of the mice that we used were adults of 23 months of age, with the exception of the mice that we used for lentiviral ventricular injections, which were 34 weeks old.

antibody staining was revealed by appropriate species- or subclass-specific secondary antibodies conjugated to Alexa-488 (1:1,000, Invitrogen), Alexa-568 (1:1,000, Invitrogen), Cy2 or Cy3 (1:1,000, Dianova). Biotinylated secondary antibodies (1:200, Vector Laboratories) were used in combination with the TSA fluorescent amplification kit (Perkin Elmer). Specific labeling was checked by omitting the primary antibody. BrdU immunohistochemistry was performed as described above after 2 M HCl pretreatment for 4560 min, followed by incubation in borate buffer (0.1 M, 10 min, pH 8.5). Antigen retrieval for detection of Neurog2 protein was carried out using citrate buffer (0.1 M, pH 6, 20 min, 80 C). Viral vector injections. The retro- or lentiviral vectors that we used encoded only GFP, which was driven by a CAG promoter or ubiquitin promoter, respectively. Viral preparations and injections were performed as described previously10,13. Briefly, mice were deeply anesthetized (ketamine (100 mg per kg, CP Pharma) and xylazine (5 mg per kg, Rompun, Bayer)) and injected with 0.5 l of viral suspension at 0.7 (anteroposterior), 1.2 (mediolateral) and 1.71.4 (dorsoventral) for adult SEZ, and on the level of bregma, 0.8 (mediolateral) and 2.0 (dorsoventral) (relative to bregma) for ventricular injections at the age of 3 weeks. Immunoelectron microscopy. For electron-microscopic analysis, mice were killed by perfusion with 4% PFA containing 0.1% glutaraldehyde (vol/vol) and postfixed overnight at 4 C. Free-floating, 50-m-thick vibratome sections were incubated in primary antibody overnight at 4 C. After thorough washing, primary antibody staining was detected with a biotin-conjugated secondary antibody (1:200, Vector labs), followed by the ABC kit (Vector labs) and DAB (Polysciences) labeling. The intensity of the staining reaction was monitored in the microscope. Control sections were stained with the secondary antibody alone and no staining could be observed. Sections containing DAB-stained cells were dehydrated in ascending series of ethanol and finally embedded in Durcupan. Ultrathin sections were cut on a Reichert Ultratome and viewed in a Philips 100 electron microscope. Statistical analysis. Stainings were analyzed using an Olympus FV1000, a Leica SPE or Zeiss LSM5 Pascal laser-scanning confocal microscope with optical sections of maximum 11.5-m intervals or with Zeiss AxioImager.Z1 with apotome component. Colocalization with cell typespecific antigens was quantified in single optical sections of the laser-scanning microscopes. Between five and ten sections per mouse were counted per experiment until comparable numbers of stained or GFP-positive cells per mouse or experiment were reached. The total number of cells counted in all of the mice is indicated in the text. One mouse or experiment represents one mean value and standard deviations were calculated between mice or cultures. For the three-dimensional reconstruction of the location of Neurog2+/gfp-positive and Tbr2+ cells and of the dendritic arborization of E1Neurog2-cre; Z/EG fate-mapped cells, we used the Neurolucida software (MBF Bioscience). Cultures and electrophysiology. Primary cultures of adult SEZ stem and progenitor cells were prepared as described previously10. Briefly, following dissection of the SEZ of adult mice (>8 weeks), cells were directly plated onto polyd-lysinecoated coverslips in defined medium containing DMEM/F12 (Gibco) and 10 mM HEPES (Gibco), supplemented with B27 (Gibco) in the absence of epidermal growth factor and fibroblast growth factor 2. Dorsal SEZ cultures were obtained by dissecting the upper third of the whole ventricular wall, including the white matter, whereas the lower two-thirds of the ventricular wall were referred to as lateral SEZ cultures. For retroviral labeling, cultures were transduced 2 h after plating. We replaced 50% of the medium with fresh medium 2 d after plating and subsequently maintained the cells for the indicated time periods until electrophysiological analysis or fixation for immunocytochemistry. For electrophysiology, perforated patch-clamp recordings were performed at 2025 C with amphotericin-B (Calbiochem) for perforation37. Micropipettes were made from borosilicate glass capillaries. Pipettes were tip-filled with internal solution and back-filled with internal solution containing 200 g ml1 amphotericin-B. The electrodes had resistances of 22.5 M. The internal solu tion contained 136.5 mM potassium gluconate, 17.5 mM KCl, 9 mM NaCl, 1 mM MgCl2, 10 mM HEPES and 0.2 mM EGTA (pH 7.4) at an osmolarity of 300mOsm. The external solution contained 150 mM NaCl, 3 mM KCl, 3 mM

2009 Nature America, Inc. All rights reserved.

BrdU treatment. For detection of proliferating cells, we injected BrdU (Sigma) intraperitoneally (50100 mg per kg of body weight) 12 h before perfusion to label fast-proliferating cells (short pulse) or placed it in the mices drinking water for 3 weeks (1 mg ml1, complete exchange of water twice per week), which was followed by BrdU-free water for 34 weeks to allow full neuronal differentiation of labeled progenitors. For the BrdU chase experiment, a single BrdU pulse was given to the pregnant mother at embryonic day 14 or postnatal day 3 (50 mg per kg) and littermates were killed at postnatal days 16 and 60. In situ hybridization. The vGluT1 and vGluT2 plasmids that we used for in situ hybridization were the kind gifts of L. Cheng (Chinese Academy of Sciences) and Q. Ma. (Harvard Medical School). The Mash1 and Neurog2 plasmids were provided by F.G. and the Tbr2 and gfp24 were provided by R.F.H.s laboratory. The Gad65 and Gad67 plasmids were obtained from W. Wurst (Max Plank Institute for Psychiatry). Digoxigenin-labeled RNA probes were generated by in vitro transcription (NTP labeling mix, Roche; T3, T7 or SP6 polymerase, Stratagene) and in situ hybridization was performed on 30-m-thick cryostate sections or 70-m-thick vibratome sections of perfused brains following standard protocols using -digoxigenin antibodies (Roche, 1:2,000). For fluorescent in situ hybridization, we used a tyramide signal amplification kit (Perkin Elmer). Immunohistochemistry. For immunohistochemistry, mice were deeply anaesthetized using 5% chloralhydrate (wt/vol) in phosphate-buffered saline (PBS, 0.1 ml per 10 g of body weight), and killed by transcardial perfusion with PBS and then 4% paraformaldehyde (wt/vol, PFA) in PBS. Brains were dissected and postfixed overnight in PFA at 4 C. For cryostate sections, brains were cryoprotected and cut, and immunostainings were carried out at a thickness of 20 m. Alternatively, free-floating vibratome sections were cut at a thickness of 60 m after postfixation. Primary antibodies were diluted in 0.1 M PBS containing 0.5% Triton X-100 (wt/vol) and 10% normal goat serum (vol/vol) or 2% bovine serum albumin (wt/vol). We used primary antibodies to BrdU (rat, 1:400, Abcam or Accurate; sheep, 1:1,000, Fitzgerald Industries), activated caspase-3 (Promega, rabbit, 1:200), c-fos (mouse, 1:200, Chemicon), pan-Dlx (1:500, rabbit, kindly provided by J. Kohtz, Northwestern University), Dcx (1:2,000, rabbit, Abcam; 1:500, goat, Santa Cruz), GFAP (1:250, guinea pig, Advanced Immunochemical; 1:500, mouse, Chemicon), GFP (1:500, rabbit, Invitrogen; 1:1,000 chicken, Abcam), Ki67 (goat, 1:500, Santa Cruz), Mash1 (mouse, 1:200, kindly provided by D. Anderson, California Institute of Technology; mouse, 1:100, BD Biosciences), Neurog2 (1:100, goat, Santa Cruz), Nestin (1:50, rabbit, Chemicon), Pax6 (mouse, 1:1,000, kindly provided by D. Schulte, Developmental Studies Hybridoma Bank; rabbit, 1:300, Covance), PSA-NCAM (1:1,000, mouse, Chemicon), Tbr1 (1:2,000, rabbit, Chemicon or Abcam), Tbr2 (1:2,000, rabbit, Chemicon or Abcam), vGluT1 (1:1,000, rabbit, Synaptic Systems) and vGluT2 (1:1,000, rabbit, Synaptic Systems). Specimens were incubated overnight at 4 C. After thorough washing,

nature NEUROSCIENCE

doi:10.1038/nn.2416

CaCl2, 2 mM MgCl2, 10 mM HEPES and 5 mM glucose (pH 7.4) at an osmolarity of 310 mOsm. The recording chamber was continuously perfused with external solution at a rate of 0.5 ml min1. Cells were visualized with a Zeiss Axioskop2. Signals were sampled at 10 kHz with Axopatch 200B patch-clamp amplifiers (Axon Instruments), filtered at 5 kHz and analyzed with Clampfit 9.2 software (Axon Instruments). To assess autaptic connections, we step-polarized single cells in voltage clamp for 1 ms from 70 to +30 mV and recorded responses in the same cell. Responses were considered to be autaptic when they occurred within 3 ms of the step-depolarization. To visualize neurons for electrophysiology, we transduced adult SEZ cultures 14 d after plating with a lentiviral SYN-SYN-GFP vector36 that encodes GFP under the control of the human Synapsin promoter, which allows for specific labeling of neurons. Chlorine e6induced cortical lesion. Chlorine e6 (Frontier Scientific) was coupled to latex beads (Lumaflour) as reported previously6,7. Briefly, 3 ml of 0.01M phosphate buffer (pH 7.4) were used to dissolve 1.79 mg of chlorine e6 and 5 mg of 1-ethyl-3-(3-dimethylaminopropyl)-carbodiimide (MP Biomedicals) was added. The solution was kept for 30 min at 4 C. We incubated 25 l of rhodamine latex beads (Lumafluor) on a shaker for 1 h at 2025 C with 1.5 ml of activated chlorine e6 solution. The reaction was stopped by adding 200 l of 0.1 M glycine buffer (pH 8.0). The coupled latex beads were centrifuged to remove

the supernatant (30 min, 50,000g), washed at least three times with 10ml of 0.01M phosphate buffer, resuspended in about 50 l of 0.01 M phosphate buffer and stored at 4 C. Injections were performed 57 times into the cerebral cortex (maximum depth 0.8 mm, around 100 nl each injection) using a glass capillary connected with an air pressure system (WPI). Retrograde transport of latex beads was checked in control mice that did not undergo laser illumination. Laser illumination (633 nm, Schfter and Kirchhoff) of the contralateral hemisphere was performed directly on the dura after skull removal 2, 3 or 7 d later (5 min) followed by transcardial perfusion 7 or 10 d after that. Different illumination times (20, 10, 5 and 2 min) and different laser intensities (100, 50, 20, 10 and 5 mW) were tested and best results (number and specificity of dying cells) were obtained with 2030 mW for 45 min. For the tracking experiments, retroviral injections were performed in one hemisphere, parallel with chlorine e6coupled latex beads on the contralateral side, followed by laser illumination 2 d later and perfusion 34 d later. Cell death was confirmed by staining for active caspase 3 and NeuN 3 d after laser illumination on the contralateral hemisphere.
49. Seibt, J. et al. Neurogenin2 specifies the connectivity of thalamic neurons by controlling axon responsiveness to intermediate target cues. Neuron 39, 439452 (2003). 50. Parras, C.M. et al. Mash1 specifies neurons and oligodendrocytes in the postnatal brain. EMBO J. 23, 44954505 (2004).

2009 Nature America, Inc. All rights reserved.

doi:10.1038/nn.2416

nature NEUROSCIENCE

ERRATa

Erratum: Adult generation of glutamatergic olfactory bulb interneurons


Monika S Brill, Jovica Ninkovic, Eleanor Winpenny, Rebecca D Hodge, Ilknur Ozen, Roderick Yang, Alexandra Lepier, Sergio Gascn, Ferenc Erdelyi, Gabor Szabo, Carlos Parras, Francois Guillemot, Michael Frotscher, Benedikt Berninger, Robert F Hevner, Olivier Raineteau & Magdalena Gtz Nat. Neurosci. 12, 15241533 (2009); published online 1 November 2009; corrected after print 11 December 2009 In the version of this article initially published, the email address of one of the corresponding authors was misspelled. It should be magdalena.goetz@helmholtz-muenchen.de. The error has been corrected in the HTML and PDF versions of the article.

2009 Nature America, Inc. All rights reserved.


nature neuroscience

a r t ic l e s

Glial precursors clear sensory neuron corpses during development via Jedi-1, an engulfment receptor
Hsiao-Huei Wu1,5, Elena Bellmunt2,6, Jami L Scheib1,6, Victor Venegas3,6, Cornelia Burkert1, Louis F Reichardt4, Zheng Zhou3, Isabel Farias2 & Bruce D Carter1
During the development of peripheral ganglia, 50% of the neurons that are generated undergo apoptosis. How the massive numbers of corpses are removed is unknown. We found that satellite glial cell precursors are the primary phagocytic cells for apoptotic corpse removal in developing mouse dorsal root ganglia (DRG). Confocal and electron microscopic analysis revealed that glial precursors, rather than macrophages, were responsible for clearing most of the dead DRG neurons. Moreover, we identified Jedi-1, an engulfment receptor, and MEGF10, a purported engulfment receptor, as homologs of the invertebrate engulfment receptors Draper and CED-1 expressed in the glial precursor cells. Expression of Jedi-1 or MEGF10 in fibroblasts facilitated binding to dead neurons, and knocking down either protein in glial cells or overexpressing truncated forms lacking the intracellular domain inhibited engulfment of apoptotic neurons. Together, these results suggest a cellular and molecular mechanism by which neuronal corpses are culled during DRG development. The extensive neuronal cell death that occurs during the ontogenesis of the peripheral ganglia was first described in the developing chick embryo, leading to the discovery of nerve growth factor (NGF)1,2. An important part of this tissue-sculpting process is to properly dispose of degenerated cellular components, thereby avoiding any inflammatory response3. Although much progress has been made in understanding the regulation of neuronal cell death4, little is known about how the vast pool of neuronal corpses is eliminated. In the developing mammalian CNS, glial cells and microglia have been implicated in the clearance of apoptotic neurons. Infiltration of F4/80-positive macrophages from the developing mouse vasculature into the retina and brain is associated with neuronal death. These invading macrophages further differentiate into microglia and engulf and degrade the apoptotic debris5,6. Early electron microscopy studies in the developing chick peripheral nervous system (PNS) suggested that macrophages, as well as satellite glial cells and their precursors, may be involved in clearing neuronal corpses 7,8; nonetheless, the potential function of these glial cells in engulfment and the molecular mechanism involved have not been explored. The engulfment process used by professional phagocytic cells, including macrophages and dendritic cells, is known to involve an array of receptors on the phagocytes that are able to sense find me and eat me cues exposed by dying cells and dont eat me signals from healthy cells912. Whether any of these receptors or cues is involved in clearing dead neurons during PNS development is not known. Recently, a Drosophila engulfment receptor, Draper, was identified that is structurally and functionally similar to CED-1, a
1The

2009 Nature America, Inc. All rights reserved.

phagocytic receptor found in Caenorhabditis elegans13,14. Draper is expressed exclusively in macrophages and glia, and Draper-deficient embryos had defects in the clearance of neuronal corpses and degenerating axons13,1517. Clearance of apoptotic cells is not just for waste disposal. Noningested apoptotic cells typically undergo secondary necrosis, which not only activates immature dendritic cells to become immunogenic, but also exposes normally sequestered self-antigens18, resulting in an increased risk for autoimmune disease later in life3. To gain insight into how apoptotic neurons are eliminated during DRG development, we investigated the cellular and the molecular mechanisms underlying this clearance process. We found that satellite glial cell (SGC) precursors were the primary cell type responsible for dead neuron clearance. We also identified two receptors that are homologous to CED-1 and Draper as mediators of this engulfment process: MEGF10, recently reported to be a CED-1 homolog19, and Jedi-1, an engulfment receptor (also known as PEAR1 or MEGF12). RESULTS Apoptotic DRG neurons are engulfed by SGC precursors In the mouse embryonic DRG, approximately 50% of the sensory neurons undergo apoptosis, starting around embryonic day 11 (E11), peaking at E13, then tapering off about E15 (refs. 20,21). To determine how these neuron corpses are cleared, we first considered the possibility that macrophages could be responsible. Notably, we found only a few sporadic macrophages, detected by the macrophagespecific antigen F4/80 (ref. 6), in the mouse DRG during the period of

Center for Molecular Neuroscience, Kennedy Center For Human Development, and Department of Biochemistry, Vanderbilt University Medical School, Nashville, Tennessee, USA. 2Department de Biologa Celular and CIBERNED, Universidad de Valencia, Burjassot, Spain. 3Verna and Marrs McLean Department of Biochemistry and Molecular Biology, Baylor College of Medicine, Houston, Texas, USA. 4Department of Physiology, University of California, San Francisco, California, USA. 5Present address: Department of Cell and Neurobiology, Keck School of Medicine of the University of Southern California, California, USA. 6These authors contributed equally to this work. Correspondence should be addressed to B.D.C. (bruce.carter@vanderbilt.edu). Received 21 June; accepted 7 October; published online 15 November 2009; doi:10.1038/nn.2446

1534

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s a b
Figure 1 Neuron corpses in developing DRG are engulfed by BFABP-positive SGC precursors. (a,b) Immunostaining with F4/80 antibody was used to detect the presence of macrophages in paraffin sections from E12.5 mouse DRG (a) or liver (b). Arrows point to cell corpses engulfed by macrophages, which were rare. Arrowheads point to several cell corpses that are not engulfed by macrophages. (c,d) Cryosections from E12.5 mouse DRG were immunostained with antibody to BFABP to label satellite glial cells and antibody to type III -tubulin (-IIITub) to label neurons. Cell nuclei were counterstained with TO-PRO3 iodide (642 of 661), a monomeric cyanine nucleic acid stain. Arrows indicate apoptotic cells engulfed by BFABP-positive cells (c). Dorsal is to the right. An enlarged view of nuclear fragments engulfed by BFABP-positive cells, indicated by the arrows, is shown in d. The arrowhead indicates a BFABP-positive SGC precursor undergoing mitosis. Scale bars represent 20 m in ad. (e) Immunohistochemical staining of E12.5 wild-type and Ntf3/ DRGs with antibody to BFABP (brown) on 5-m-thick sections counterstained with toluidine blue. S, SGC precursors; N, neurons. Arrows point to several apoptotic neurons engulfed/enveloped by BFABP-positive SGC precursors.

c
-IIITub -BFABP TO-PRO3 Overlay

d
-BFABP TO-PRO3 Overlay

2009 Nature America, Inc. All rights reserved.

e
S S N N N Wild type S S N S Ntf3
/

N S S Ntf3
/

aturally occurring cell death (Fig. 1a). Specifically, just 0.65 0.58% n (n = 3) of the total number of cells in the ganglia were F4/80 positive at E11 and 0.65 0.65% at E13 (n = 3). Moreover, even though F4/80-positive cells inside mouse DRG appeared to be encircling dead neurons (Fig. 1a), most of the apoptotic cells were not associated with macrophages (Fig. 1a). This was not a result of a lack of macrophages during this stage or a limitation of detection, as many F4/80-positive macrophages were present in the liver in the same section (Fig. 1b). The primary cell types in the embryonic DRG are neurons and SGC precursors2225 (Fig. 1c). SGC precursors arise after E10.5 in mice24,2628, corresponding to the time of cell death of DRG neurons, and eventually surround the mature neurons, thus positioning themselves for engulfment of neuronal corpses. Using an antibody to brain fatty acidbinding protein (BFABP), a marker of SGC precursors2225, we found that most of the apoptotic nuclei appeared to be engulfed by BFABP-positive cells (Fig. 1ce). We quantified the number of glial precursors surrounding dead cells during the period of normal neuronal apoptosis in the DRG and found that 86.2 5.4% of the apoptotic bodies were associated with BFABP-positive cells at E11, 73.4 2.8% at E12 and 77.1 3.6% at E13 (n = 3). Confocal images with serial optical sections acquired at a higher magnification revealed that several apoptotic nuclear remnants were present inside the BFABP-positive
Figure 2 Electron micrographs of apoptotic bodies engulfed and ingested by SGC precursors in embryonic E12.5 DRGs. (a) A healthy neuron (N) characterized by finely dispersed chromatin and abundant cytoplasm adjacent to a SGC precursor (S) with a characteristic pleomorphic nucleus with chromatin clumps. (b) An apoptotic cell (A) engulfed by a SGC precursor. (c) Debris of two ingested cells inside a SGC precursor. ( d) An ingested apoptotic cell present in a cell that is undergoing mitosis (M, mitotic figure, condensing chromosomes).

cytoplasm (Fig. 1d and data not shown). To be sure that macrophages were not expressing the glial marker, we performed double labeling with antibodies to BFABP and F4/80 and confirmed that these were separate cell populations (data not shown). To further demonstrate that SGCs were the primary cell type engulfing apoptotic bodies, we examined ganglia of E1113 embryos at the electron-microscopy level. Healthy neurons, morphologically characterized by a large nucleus with finely dispersed chromatin and abundant cytoplasm, were typically positioned near elongated cells with a polymorphic nucleus exhibiting a perinuclear chromatin ring and chromatin clumps, typical of satellite glia (Fig. 2a). In 100% of the cases in which a cell was surrounding an apoptotic body (53 apoptotic bodies total, in DRG sections from E11, n = 15; E12, n = 22; E13, n = 16),

a
S

c
A

A M A

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1535

a r t ic l e s
Figure 3 Glial cells engulf dead neurons induced by NGF withdrawal in vitro. Primary mixed cultures of sensory neurons and SGCs were derived from dissociated E15 rat DRG and cultured with NGF for 2 d. The NGF was then removed to induce neuronal apoptosis and the cultures were fixed and immunostained 2 d later. (a,b) Images were compiled from orthogonal optical sections showing apoptotic nuclear remnants (TO-PRO3, red) inside the cytoplasm of glial cells (-S100, green) (a) and neuron corpses (-IIITub, green) enveloped -S100/TO-PRO3 -IIITub/-S100 -GFP/TO-PRO3 by glial cells (-S100, red) (b). Nuclei were TO-PRO3 stained with TO-PRO3 (blue). (c) Apoptotic nuclear remnants (TO-PRO3, red) present in glial cells expressing membrane bound GFP (green). The window on the top of each panel shows the z axis stacks at the specific y axis location (indicated by a green line crossing the panel) and that on the right shows the z axis stacks at the specific x axis location (red line crossing the panel). Arrows and arrowheads indicated engulfed apoptotic bodies.

E12.5 mouse

Satellite

Brain Heart SpC DRG Neurons glia Figure 4 Putative Draper and CED-1 Jedi-1 homologs, Jedi-1 and MEGF10, are expressed in developing peripheral glial cells. (a) Schematic representation of the modular Megf10 architecture of Draper, CED-1 and possible mammalian homologs. A key for the predicted Megf11 domains and motifs is shown on the bottom (see Supplementary Fig. 1 for the sequence alignments of their predicted intracellular domains). (b) RT-PCR detection of Jedi-1, Megf10 or Megf11 mRNA in E13 mouse brain, heart, spinal cord (SpC), whole DRG and purified DRG neurons or satellite glial cells. Left, 1-kb DNA markers. (c) Jedi-1 and Megf10 transcripts were detected in mouse DRG and developing glial cells alongside axons. The developmental stages of the embryos are indicated on the left. For the sagittal sections, dorsal is on the left. Arrowheads indicate the nerves.

1536

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

E17.5

E15.5

E12.5

the ultrastructure of the engulfing cell resembled that of satellite cells were first grown in the presence of NGF for 23 d to keep the sensory (Fig. 2b,c). We did not find any macrophage-like cells engulfing cell neurons alive, followed by removal of NGF to induce apoptosis. The corpses in the electron micrographs. Moreover, we occasionally observed glial cells were detected with an antibody to BFABP or S100, which engulfing cells undergoing mitosis (Fig. 2d). Mitotic figures were also labels both SGC and Schwann cells, which most of the glia eventuobserved in BFABP-positive cells in the ganglia (Fig. 1d) and 29.3 1.8% ally convert to in culture31. Virtually all non-neuronal cells in the of the BFABP-positive cells at E12 colabeled with BrdU 1 h following cultures were BFABP and S100 positive; we did not detect any fibrobinjection. Because nearly all detectable apoptotic cells in the DRG at this lasts by Thy1.1 staining and we did not find any F4/80-positive cells time are neurons21 and this is the period during which SGC precursors (n = 3). Following 2 d of NGF withdrawal, 82% of the neurons were are proliferating, these observations indicate that SGC precursors are apoptotic and the engulfed and ingested dead neurons were exclusively detected inside the glial cells on the basis of S100 (Fig. 3a,b) responsible for engulfing the dying neurons. To further pursue this possibility, we asked whether elevated apop- and BFABP (Supplementary Fig. 1) immunostaining and confocal tosis in the developing DRG could be handled by SGC precursors or microscopy. We also expressed a membrane-bound form of GFP32 would lead to macrophage invasion. Using neurotrophin 3 null (Ntf3 /) (meGFP) in the glial cells to more clearly visualize the internalized mice, which lose 70% of their sensory neurons between E11 and E13 corpses. Under these conditions, engulfed nuclear remnants inside (refs. 20,29,30), we examined the clearing of dead neurons in the E12 phagosomes were clearly observed (Fig. 3c). Taken together with our DRG. As in the wild-type ganglia, the vast majority of apoptotic cells were surrounded a Draper by BFABP-positive cells (74.6 6.0%, n = 3 EMI EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF Ig EGF TM embryos; Fig. 1e), whereas only 3.1 2.5% EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF TM EMI CED-1 were associated with F4/80-positive cells (n = 3 embryos). This result indicates that SGC EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF TM EMI MEGF10 precursors are the primary engulfing cell type EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF TM EMI Jedi-1 during DRG development and are sufficient for the job even when the pool of dead neurons MEGF11 EMI EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF EGF TM is markedly increased. EMI EMI domain EGF EGF-like domain Ig Ig/MHC signature domain NIM domain The ability of glial precursors to ingest TM Transmembrane domain NPXY YXXL Proline-rich domain apoptotic neurons could also be demonstrated in vitro. We established an in vitro Jedi-1 Megf10 Control probe b c engulfment assay using dissociated ganglia from E15 rat or E13 mouse embryos. The cells

2009 Nature America, Inc. All rights reserved.

a r t ic l e s a
JediC-gfp

JediC-gfp

2009 Nature America, Inc. All rights reserved.

The number of germ cell corpses in wild-type C. elegans hermaphrodites expressing JediC-gfp Genotype Wild type Wild type Wild type Transgene Transgenic line None JediC-gfp JediC-gfp NA 1 2 Number of germ cell corpses n 2.9 2.6 9.5 4.0 8.2 5.1 39 22 18 P value NA < 0.005 < 0.005

Figure 5 The extracellular domain of Jedi-1 recognizes cell corpses when expressed in C. elegans engulfing cells. Transgenic worms expressed JediC-GFP in engulfing cells under the control of the ced-1 promoter. All of the worms that we analyzed were adult hermaphrodites aged 48 h post-mid L4 larval stage. (ad) GFP (a,b) and corresponding differential interference contrast microscopy (c,d) images of part of adult ced-1(e1735) hermaphrodite gonads to indicate the clustering of JediC-GFP around germ cell corpses. Arrows indicate a few germ cell corpses labeled with JediC-GFP on their surfaces. Arrowheads indicate examples of germ cell corpses not labeled by JediC-GFP. Dorsal is to the top. Midbody is to the left. Scale bars represent 10 m. The table indicates the number of germ cell corpses in wild-type C. elegans hermaphrodites expressing JediC-GFP scored under DIC optics in one gonadal arm of each adult hermaphrodite staged 48 h post L4 larval stage. Data are presented as mean s.d. n indicates the number of worms scored. The data obtained from transgenic and the wild-type control worms were compared and P values were obtained from two-tailed Students t tests.

in vivo findings, these results suggest that SGC precursors are the primary cell type responsible for clearing neuronal corpses during the period of naturally occurring cell death in the developing DRG. Jedi-1 and MEGF10 are expressed in SGC precursors The molecular mechanisms underlying apoptotic cell clearance in the mammalian PNS during development are not known. Draper, a Drosophila protein that is homologous to the C. elegans CED-1 receptor, was identified as an engulfment receptor that is expressed on glial cells and that is required for clearing degenerating neurons and axons1317,33; therefore, we speculated that a Draper/CED-1like engulfment receptor might exist in SGC precursors to mediate phagocytosis of dead neurons. Three mammalian proteins, MEGF10, MEGF11 and Jedi-1, were identified as being highly homologous to Draper and CED-1 using BLASTP (National Center for Biotechnology Information). Two regions in the intracellular domain of CED-1 are required for its engulfment function: an NPXY motif that may serve as a phosphotyrosine binding site and an YXXL motif, a Src Homology 2 (SH2) domain binding site14. Draper and MEGF10 have both NPXY and YXXL motifs, whereas Jedi-1 has an NPXY sequence and MEGF11 has an YXXL in their putative intracellular regions (Fig. 4a and Supplementary Fig. 2). To determine whether Jedi-1, MEGF10 or MEGF11 could mediate engulfment by SGC precursors, we examined their expression in these cells by reverse-transcription PCR (RT-PCR). The mRNAs for all of these proteins were present in E12.5 mouse brain and whole DRG (Fig. 4b); however, only MEGF10 and Jedi-1 were expressed in isolated SGC precursors, indicating that MEGF11 is unlikely to function as an engulfment receptor in DRG development. Curiously, the mRNA for all three proteins was detected in neurons, although their function there is not known. We then analyzed the expression pattern of Jedi-1 (also known as Pear1) and Megf10 in the developing mouse DRG at different developmental stages using
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

in situ hybridization (ISH; Fig.4c). At all of the ages that we examined (E12.5, E15.5 and E17.5), both Jedi-1 and Megf10 were observed in the ganglia and in the cells along the nerves, consistent with the location of SGC precursors and immature Schwann cells.

Jedi-1 and MEGF10 facilitate binding to apoptotic cells Recently, MEGF10 was proposed as a putative CED-1 homolog because it could promote dead thymocyte engulfment when ectopically expressed in HeLa cells19,34 and expression of an MEGF10-GFP fusion protein under the control of the ced-1 promoter in ced-1 (e1735) mutant C. elegans partially rescued the engulfment defect of ced-1 (e1735)19. Whether Jedi-1 exerts a function similar to CED-1 has never been examined. CED-1 and its truncated form, which lacks the intracellular domain (ICD) (CED-1C), similarly cluster around neighboring cell corpses, owing to their ability to recognize an apoptotic cell-surface signal(s)14. We expressed a Jedi-1 protein with a truncated intracellular domain tagged with GFP at its C terminus (JediC-GFP) in engulfing cells under the control of the ced-1 promoter in C. elegans and observed cell-surface presentation of a fraction of JediC-GFP molecules, although much of it remained inside the cells. Notably, the portion of JediC-GFP present on the surface of gonadal sheath cells, which are engulfing cells for apoptotic germ cells, was clustered around some of the germ cell corpses (Fig. 5). This localized enrichment of JediC-GFP around cell corpses suggests that the extracellular domain of Jedi-1 is capable of recognizing a signal on the surface of the dying cell, similar to CED-1. Furthermore, expression of JediC-GFP in wild-type worms resulted in the presence of excessive germ cell corpses ( Fig. 5). This result suggests that, among other possibilities, JediC-GFP might interfere with the normal engulfment of dead cells by associating with the eat me cue on the surface of cell corpses, thereby preventing endogenous CED-1 or other unknown engulfment receptors from binding these cues and transducing the normal engulfment signal. The expression of CED-1C similarly results in the inhibition of cell-corpse engulfment14. We also tested whether ectopic expression of full-length
1537

a r t ic l e s
Figure 6 Jedi-1 and MEGF10 expressed in Biotin No biotin Megf10 GFP Jedi + Megf10 Jedi HEK293 cells enable binding to dead neurons. 35% 35% (a) HEK293 cells were transfected with * * * 30% 30% GFP-tagged Jedi-1 or MEGF10 and their 25% 25% 20% expression on the cell surface analyzed by 20% 15% biotinylation of the surface proteins followed 15% Avidin 10% 10% by precipitation with avidin beads and western pulldown 5% 5% blotting with antibody to GFP. The biotin 0% 0% GFP Jedi Megf10 Jedi + reagent was not added to some cells (no biotin) Lysates 0 20,000 40,000 60,000 Megf10 to confirm the specificity of the pull down. Concentration of apoptotic neurons The expression of Jedi-GFP and MEGF10-GFP in total cell lysates is shown in the lower panel. ( b) Jedi-1 and/or MEGF10-transfected cells were incubated at 4 C with the indicated number of neuronal corpses, induced to undergo apoptosis by NGF withdrawal for 24 h and labeled with propidium iodide. After washing off the unbound dead cells, the cultures were fixed with 10% formalin. The percentages of GFP-positive cells with at least one propidium iodidepositive corpse bound are shown (one representative experiment of three is depicted). The lower panel shows the mean s.e.m. percentage binding at the highest concentration of neurons (* P < 0.01 relative to GFP-transfected cells, n = 4).
Megf10-GFP Megf10-GFP Percentage binding Percentage binding Jedi-GFP Jedi-GFP GFP GFP

2009 Nature America, Inc. All rights reserved.

Jedi-1 (Jedi-GFP) in C. elegans engulfing cells could also rescue the cell-corpse removal defects in ced-1 (e1735) mutants. Unfortunately, ectopically expressed Jedi-GFP was retained inside cells, forming protein aggregates and did not result in any rescue of the ced-1 mutant (data not shown). Although CED-1, Draper and MEGF10 are thought to be receptors for apoptotic cells, none of the ligands have been identified and it has not been directly demonstrated that these proteins specifically mediate binding, as opposed to just facilitating the engulfment process. To determine whether Jedi-1 or MEGF10 can act as receptors for dead neurons, we transiently expressed them in HEK293 cells and added various concentrations of apoptotic neurons, labeled with propidium iodide. We confirmed that Jedi-1 and MEGF10 were expressed and trafficked to the cell surface ( Fig. 6a) and then incubated the cells expressing these proteins with neuronal corpses at 4 C to prevent engulfment, a as we wanted to specifically assess binding
Figure 7 Ectopic expression of Jedi-1 or MEGF10 in glial cells promotes neuronal corpse engulfment. DRG from E13.5 mice were dissociated and grown in the presence of NGF for 2 d. The glial cells were then transfected with plasmids expressing MEGF10-GFP, Jedi-1Flag or meGFP and NGF was removed from the cultures. The cells were fixed after 2 d without NGF and immunolabeled with antibodies to GFP or Flag. (ad) z axis optical stacks of confocal images of glial cells expressing MEGF10-GFP (MEGF10, a,c,d), Jedi-1Flag (Jedi, bd) or meGFP (c,d) were acquired and the numbers of transfected cells containing at least one ingested nuclear remnant (condensed TO-PRO3 staining) was quantified. Arrows indicate the location of engulfed apoptotic nuclei in these cells. Arrowheads indicate the long processes in Jedi-1Flagexpressing cells. In c, the error bars indicate mean s.d. (P = 0.0002, one-way ANOVA). The number of engulfed nuclei per engulfing cell was also determined and expressed as the percentage of transfected cells containing the indicated number of apoptotic nuclei (open bar, meGFP; black bar, MEGF10-GFP; gray bar, Jedi-Flag; d). Based on a 2 analysis, there was a significant difference between all three groups of cells (expressing Jedi-1, MEGF10 and meGFP, P < 0.0001). Scale bars represent 20 m.

and not internalization. After rinsing, the cultures were fixed and the number of dead neurons attached to the transfected cells was scored. Some binding occurred to the control cells, most likely as a result of other endogenous proteins that can facilitate binding to dead cells, such as integrins or PSR 12; however, expression of either Jedi-1 or MEGF10 significantly increased binding to neuronal corpses (P < 0.01 relative to GFP-transfected cells, n = 4; Fig. 6b). This result indicates that these proteins can function as receptors, enhancing binding in the absence of any internalization. Expression of both proteins did not further increase the binding (Fig. 6b), which may indicate that they are part of a single binding complex, but the affinity and any cooperativity could not be accurately determined, as the ligand source is an entire dead neuron and not a small, freely diffusible molecule.

20 m

20 m

c
Percentage of transfected cells that are engulfing

90 80 70 60 50 40 30 20 10 0
GFP

Percentage of cells

60% 40%

20%

Jedi Megf10 Jedi + Megf10

0%

1 2 3 4 >5 Number of engulfed nuclei per transfected cell

1538

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s a
Percentage of GFP-positive cells that are engulfing 60

b
Percentage of transfected cells that are engulfing

80 60 40 20

40

20

0
m eG FP di C C f1 0 M eg Je

0 Scrambled MEGF10 Jedi

2X

1X 1X

2X

1X 1X 2X

1X 1X

Figure 8 Neuronal corpse engulfment by glial cells requires endogenous Jedi-1 and MEGF10. The neuronal engulfment assay described in Figure 7a was used to determine whether endogenous MEGF10 and Jedi-1 are required for engulfment. (a) C-terminal truncated forms of MEGF10 or Jedi-1 (Megf10C or JediC, respectively), as well as meGFP, were transfected into glial cells, and the number of GFP-positive cells with engulfed apoptotic nuclei were counted 2 d after NGF withdrawal. Error bars represent mean s.d. (P = 0.0005, one-way ANOVA). (b) Plasmids bi-cistronically expressing ZsGreen and shRNAs targeting Megf10, Jedi-1 or non-targeting shRNA (scrambled) were transfected into the glial cells. A total of 8 g of DNA was transfected, either in combinations or as a single plasmid, and the number of ZsGreen-positive cells containing engulfed corpses were counted after 2 d. Error bars represent mean s.d. (P = 0.0008, one-way ANOVA).

Modifying Jedi-1 or MEGF10 expression alters engulfment To investigate the involvement of Jedi-1 and/or MEGF10 in neuronal corpse engulfment by SGC precursors, we used the engulfment assay described above (Fig. 3). DRGs from E13.5 mouse embryos were dissociated and grown in the presence of NGF for 2 d. NGF was then removed from the culture and, at the same time, the glial cells were transfected with either a control (meGFP) or other transgene. Images of cells expressing these transgenes were acquired with full z axis optical sections and the numbers of transfected cells containing at least one fully internalized apoptotic nucleus were determined (Fig. 7 and Supplementary Fig. 3). Under these conditions, engulfed apoptotic nuclei were observed in about 50% of meGFP-expressing glial cells (Fig. 7c). In contrast, overexpression of MEGF10 or Jedi-1 enhanced engulfment of dead neurons by the glial cells; the number of Flag- or GFP-positive cells that had engulfed an apoptotic nucleus increased by approximately 50 and 80% when overexpressing either Jedi-1Flag or MEGF10-GFP, respectively (Fig. 7 and Supplementary Fig. 3). Transfection of both receptors into the cells did not further enhance phagocytosis (Fig. 7c), suggesting that there is an endogenous component in the pathway that is limiting or that the two proteins converge on a common pathway. MEGF10-GFP and Jedi-1Flag accumulated around vacuoles containing apoptotic nuclei (Fig. 7a,b and Supplementary Fig. 3) and in what appeared to be phagocytic cups (Supplementary Fig.4). The vacuoles were identified as lysosomes or late endosomes by LAMP-1 labeling (Supplementary Fig. 5). Notably, even though the overall number of cells that were engulfing increased when transfected with Jedi-1 or MEGF10, these cells exhibited some differences; cells expressing MEGF10-GFP contained more vacuoles with apoptotic nuclei than those expressing Jedi-1Flag, which typically had more long processes and lamellipodia (Fig. 7b). To quantify this difference in engulfment, we scored the number of vesicles with nuclear fragments in each transfected cell. Three or more apoptotic corpses were found in 70% of MEGF10-transfected cells, but only in ~30% of Jeditransfected cells, which more often contained one or two engulfed nuclei (Fig. 7d). These observations suggest that overexpression of either Jedi-1 or MEGF10 can ultimately increase the engulfment of dead cells, although there appear to be some differences in their mechanisms of action. The effect of expressing a mutant Jedi-1 that lacked the ICD in C. elegans suggested that this construct could act as an inhibitor of engulfment, preventing the endogenous phagocytic receptor(s) from binding and/or internalizing apoptotic cells (Fig. 5). Indeed, we found that transfection of glial cells with either JediC-GFP or MEGF10CGFP in the engulfment assay led to a ~30% decrease in the number of cells that were engulfing dead neurons when compared with those transfected with meGFP (Fig. 8a and Supplementary Fig. 6). The
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

2009 Nature America, Inc. All rights reserved.

ability of the truncated MEGF10 and Jedi-1 to reduce engulfment suggests that the mutants interfere with the action of the endogenous protein. Therefore, to directly determine whether endogenous Jedi-1 or MEGF10 is required for neuronal corpse clearance, we used short hairpin RNA (shRNA) to target Jedi-1 and Megf10 (Supplementary Fig. 7). Knocking down Jedi-1 or Megf10 in the glial cells did not alter their morphology (Supplementary Fig. 7); however, it did result in a 4050% decrease in the number of cells with internalized apoptotic neurons relative to those expressing a scrambled shRNA ( Fig. 8b). There was no difference between the number of engulfing glial cells transfected with the control shRNA (Scr1-1) and those transfected with meGFP (54.3 1.0 versus 53.4 4.3). Knocking down both Jedi-1 and MEGF10 did not further reduce the ability of the transfected cells to engulf apoptotic bodies, further suggesting that the two proteins may function in a common pathway ( Fig. 8b). These results indicate that endogenous Jedi-1 and MEGF10 are necessary for neuronal corpse clearance in the embryonic DRG. DISCUSSION Although it has long been recognized that there is extensive cell death in the developing PNS, the mechanisms responsible for disposing of the cellular waste have remained an open question. Our findings indicate that SGC precursors in the DRG are the primary cell type responsible for clearing neuronal corpses generated during the period of naturally occurring cell death. In addition, we identified Jedi-1, a previously unknown engulfment receptor, and MEGF10 as two CED-1 homologs that were expressed in the glial cells and were involved in phagocytosing neuronal corpses. Thus, these results reveal the cellular and molecular basis for clearing the neuronal waste generated during the development of sensory ganglia. Macrophages carry out the removal of cellular debris in many tissues during development or after injury and these cells are known to increase in the DRG after injury to the sciatic nerve35. However, we found few macrophages in the developing DRG, even in animals with unusually high numbers of apoptotic neurons (Ntf3/ mice). The phagocytic ability of glial cells in the developing nervous system has been known for some time, although the importance has largely been unrecognized. Axonal fragment ingestion by Schwann cells during Wallerian degeneration after injury was first described some 40 years ago8. However, macrophages subsequently invade the nerve and clear most of the debris, thereby overshadowing the contribution of the Schwann cells36. A more recent study demonstrated that Schwann cells have an important phagocytic role in synapse elimination during the development of the neuromuscular junction37. Early electron micro scopy studies of the developing chick embryo suggested the presence of degenerated axons and apoptotic neurons in astrocytes, satellite glial cells and Schwann cells7,8,38; however, the identity of the engulfing
1539

a r t ic l e s
cells was not confirmed because of the absence of immunological markers. Furthermore, in some cases, contradictory observations were reported, suggesting that macrophages were clearing the debris8. Our findings demonstrate that SGCs, rather than macrophages, are the primary phagocytes responsible for clearing the dead neurons generated during the normal development of the DRG. The physiological roles of SGCs are not well understood. They are found in sensory, sympathetic and parasympathetic ganglia, where they cluster around each neuron and regulate the extracellular environment, taking up neurotransmitters similar to astrocytes39. SGCs also produce numerous neuroactive agents, such as neurotrophins and bradykinin, although the functions of these in the ganglia are not clear39. Following injury, the SGCs undergo a morphological change, begin to proliferate and release many of these factors, leading some to suggest that they are involved in neuropathic pain39,40. The phagocytic ability characterized here provides a glimpse into an important function of SGCs. Whether the glial cells remain the primary cell type responsible for phagocytosing dead neurons in more mature animals, such as after axotomy or other inducers of neurodegeneration, remains to be determined. The molecular mechanisms involved in apoptotic neuron clearance in the developing mammalian PNS were previously unknown. In C.elegans, CED-1 was identified as a receptor required for engulfment of apoptotic cells14. Draper, the Drosophila homolog of CED-1, was shown to mediate dead neuron removal during development13 and to eliminate degenerating axons during metamorphosis and after injury15,16,33. We identified MEGF10 and Jedi-1 as possible homologs on the basis of their predicted structural organization and their expression pattern. MEGF10 was previously suggested to be an engulfment receptor19, but little is known about Jedi-1. It was shown to be phosphorylated on platelet activation, although its role in this process was not determined, and its overexpression in hematopoietic progenitors reduced the number of cells that committed to a myeloid lineage, suggesting that it is involved in the differentiation of these cells. We found that Jedi-1 functions as a phagocytic receptor and is involved in clearing dead sensory neurons. Notably, both Jedi-1 and MEGF10 promoted the engulfment of neuronal corpses by SGCs (Figs. 7 and 8); however, overexpression or knockdown of both proteins was no more effective than altering the expression of either alone. We therefore suggest that these receptors may converge on a common pathway or even form a complex. However, they also exerted somewhat different actions when overexpressed in glial cells, suggesting that they have some non-overlapping functions. Almost all of the cells transfected with MEGF10 had engulfed more than one apoptotic nuclei (90.6% 1.3) and most had numerous apoptotic nuclear remnants in lysosomes. This could be a result of an increase in the number of dead neurons taken up by a single cell or an increase in the number of lysosomes digesting the engulfed material. Increased vacuole formation has been described in HEK293 cells overexpressing MEGF10, even when no apoptotic cells were added34. On the other hand, most of the Jedi-1 overexpressing glial cells appeared to be slimmer with longer processes (Fig. 7b) and contained only one or two visible vacuoles with ingested nuclei, unlike those overexpressing MEGF10 (Fig. 7d). It is possible that Jedi-1 has the ability to promote both engulfment and degradation of apoptotic cells, similar to CED-1 (ref. 41), whereas MEGF10 only carries out the engulfment activity. Of course, we cannot rule out the possibility that overexpression of either protein indirectly increases engulfment, for example, by increasing process formation. In any case, these differences suggest that, on activation by the eat me signals, MEGF10 and Jedi-1 trigger, at least partially, non-overlapping molecular pathways for engulfment.
1540

2009 Nature America, Inc. All rights reserved.

Such a need for multiple receptors is typical of the phagocytic process. Multiple ligands and receptors are implicated in the recognition and uptake of apoptotic cells by professional phagocytes such as macrophages12,42. For phosphatidylserine alone, the most well-studied eat me cue, there are at least four transmembrane receptors, PSR, Tim4, BAI1 and Stabilin-2, that have been shown to bind this phospholipid and transduce an engulfment signal9. Even in C. elegans, there are two partially redundant pathways that mediate cell corpse removal, with the ced-1, ced-6 and ced-7 genes functioning in one pathway and the ced-2, ced-5, ced-10 and ced-12 genes acting in the other43. Recently, a second Drosophila receptor, SIMU, was identified and was proposed to function as the recognition receptor44, with Draper acting primarily in the engulfment and degradation process through recruitment of the src family kinases45. Programmed cell death and process elimination are essential for the development and maintenance of functional nervous systems. Consequently, considerable effort has been made to understand the molecular mechanisms involved in these events; however, less attention has been given to the resulting byproducts. During such regressive processes, large amounts of degenerated and excess cellular debris are generated that need to be efficiently eliminated. Several reports have provided convincing evidence for a link between defective clearance of apoptotic cells and the development of autoimmunity3,46,47. Thus, it is likely that defects in neural waste clearance during development predisposes an organism to autoimmune attack on the nervous system later in life, although this has yet to be demonstrated. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments The authors thank the Statistics and Methodology Services at the Vanderbilt Kennedy Center for assistance on statistical analysis and C. Yoon, C. Jones and other members of the Carter laboratory for technical assistance and helpful suggestions. This work was supported by grants from the US National Institutes of Health (NS048249 and NS064278 to B.D.C., GM067848 to Z.Z.), a Muscular Dystrophy Association Development grant (MDA4023) to H.-H.W., a US National Institutes of Health Minority Access to Research Careers Predoctoral Fellowship (GM079911) to V.V., and the Ministerio de Ciencia e Innovacin (SAF), Ministerio de Sanidad (TerCel and Ciberned), Fundacin la Caixa, and Generalitat Valenciana (Prometeo) to I.F. AUTHOR CONTRIBUTIONS H.-H.W. and B.D.C. initiated and developed the overall concept and design of the project. H.-H.W. also performed, analyzed and interpreted most of the experiments and prepared the initial version of the manuscript. E.B. performed the quantitative histological analysis of neuronal corpse engulfment in Ntf3+/+ and Ntf3/ mice. J.L.S. generated some of the Jedi-1 and MEGF10 constructs, performed the binding experiment and some of the immunostaining analyses. V.V. performed all of the experiments with C. elegans. C.B. assisted with the immunostaining on sections and generated the shRNA construct for MEGF10. L.F.R. provided technical expertise for electron microscopy analysis and critical intellectual input for this study. I.F. performed the electron microscopy analysis, supervised the quantitative histological analysis and provided intellectual input. Z.Z. designed and supervised the C. elegans study and provided intellectual input. B.D.C. directed the overall project and prepared the final version of the manuscript.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
1. Bennet, M.R., Gibson, W.G. & Lemon, G. Neuronal cell death, nerve growth factor and neurotrophic models: 50 years on. Auton. Neurosci. 95, 123 (2002). 2. Hamburger, V. & Levi-Montalcini, R. Proliferation, differentiation and degeneration in the spinal ganglia of the chick embryo under normal and experimental conditions. J. Exp. Zool. 111, 457501 (1949).

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
3. Savill, J., Dransfield, I., Gregory, C. & Haslett, C. A blast from the past: clearance of apoptotic cells regulates immune responses. Nat. Rev. Immunol. 2, 965975 (2002). 4. Yuan, J., Lipinski, M. & Degterev, A. Diversity in the mechanisms of neuronal cell death. Neuron 40, 401413 (2003). 5. Hume, D.A., Perry, V.H. & Gordon, S. Immunohistochemical localization of a macrophage-specific antigen in developing mouse retina: phagocytosis of dying neurons and differentiation of microglial cells to form a regular array in the plexiform layers. J. Cell Biol. 97, 253257 (1983). 6. Perry, V.H., Hume, D.A. & Gordon, S. Immunohistochemical localization of macrophages and microglia in the adult and developing mouse brain. Neuroscience 15, 313326 (1985). 7. OConnor, T.M. & Wyttenbach, C.R. Cell death in the embryonic chick spinal cord. J. Cell Biol. 60, 448459 (1974). 8. Pannese, E. The response of the satellite and other non-neuronal cells to the degeneration of neuroblasts in chick embryo spinal ganglia. Cell Tissue Res. 190, 114 (1978). 9. Bratton, D.L. & Henson, P.M. Apoptotic cell recognition: will the real phosphatidylserine receptor(s) please stand up? Curr. Biol. 18, R76R79 (2008). 10. Gregory, C.D. & Brown, S.B. Apoptosis: eating sensibly. Nat. Cell Biol. 7, 11611163 (2005). 11. Grimsley, C. & Ravichandran, K.S. Cues for apoptotic cell engulfment: eat-me, dont eat-me and come-get-me signals. Trends Cell Biol. 13, 648656 (2003). 12. Henson, P.M. & Hume, D.A. Apoptotic cell removal in development and tissue homeostasis. Trends Immunol. 27, 244250 (2006). 13. Freeman, M.R., Delrow, J., Kim, J., Johnson, E. & Doe, C.Q. Unwrapping glial biology: Gcm target genes regulating glial development, diversification, and function. Neuron 38, 567580 (2003). 14. Zhou, Z., Hartwieg, E. & Horvitz, H.R. CED-1 is a transmembrane receptor that mediates cell corpse engulfment in C. elegans. Cell 104, 4356 (2001). 15. Awasaki, T. et al. Essential role of the apoptotic cell engulfment genes draper and ced-6 in programmed axon pruning during Drosophila metamorphosis. Neuron 50, 855867 (2006). 16. MacDonald, J.M. et al. The Drosophila cell corpse engulfment receptor Draper mediates glial clearance of severed axons. Neuron 50, 869881 (2006). 17. Manaka, J. et al. Draper-mediated and phosphatidylserine-independent phagocytosis of apoptotic cells by Drosophila hemocytes/macrophages. J. Biol. Chem. 279, 4846648476 (2004). 18. Savill, J. & Fadok, V. Corpse clearance defines the meaning of cell death. Nature 407, 784788 (2000). 19. Hamon, Y. et al. Cooperation between engulfment receptors: the case of ABCA1 and MEGF10. PLoS One 1, e120 (2006). 20. Farias, I., Yoshida, C.K., Backus, C. & Reichardt, L.F. Lack of neurotrophin-3 results in death of spinal sensory neurons and premature differentiation of their precursors. Neuron 17, 10651078 (1996). 21. White, F.A. et al. Synchronous onset of NGF and TrkA survival dependence in developing dorsal root ganglia. J. Neurosci. 16, 46624672 (1996). 22. Kurtz, A. et al. The expression pattern of a novel gene encoding brain-fatty acid binding protein correlates with neuronal and glial cell development. Development 120, 26372649 (1994). 23. Schreiner, S. et al. Hypomorphic Sox10 alleles reveal novel protein functions and unravel developmental differences in glial lineages. Development 134, 32713281 (2007). 24. Taylor, M.K., Yeager, K. & Morrison, S.J. Physiological Notch signaling promotes gliogenesis in the developing peripheral and central nervous systems. Development 134, 24352447 (2007). 25. Woodhoo, A., Dean, C.H., Droggiti, A., Mirsky, R. & Jessen, K.R. The trunk neural crest and its early glial derivatives: a study of survival responses, developmental schedules and autocrine mechanisms. Mol. Cell. Neurosci. 25, 3041 (2004). 26. Britsch, S. et al. The transcription factor Sox10 is a key regulator of peripheral glial development. Genes Dev. 15, 6678 (2001). 27. Farias, I., Cano-Jaimez, M., Bellmunt, E. & Soriano, M. Regulation of neurogenesis by neurotrophins in developing spinal sensory ganglia. Brain Res. Bull. 57, 809816 (2002). 28. Maro, G.S. et al. Neural crest boundary cap cells constitute a source of neuronal and glial cells of the PNS. Nat. Neurosci. 7, 930938 (2004). 29. Ernfors, P., Lee, K.F., Kucera, J. & Jaenisch, R. Lack of neurotrophin-3 leads to deficiencies in the peripheral nervous system and loss of limb proprioceptive afferents. Cell 77, 503512 (1994). 30. Tessarollo, L., Vogel, K.S., Palko, M.E., Reid, S.W. & Parada, L.F. Targeted mutation in the neurotrophin-3 gene results in loss of muscle sensory neurons. Proc. Natl. Acad. Sci. USA 91, 1184411848 (1994). 31. Murphy, P. et al. The regulation of Krox-20 expression reveals important steps in the control of peripheral glial cell development. Development 122, 28472857 (1996). 32. Okada, A., Lansford, R., Weimann, J.M., Fraser, S.E. & McConnell, S.K. Imaging cells in the developing nervous system with retrovirus expressing modified green fluorescent protein. Exp. Neurol. 156, 394406 (1999). 33. Hoopfer, E.D. et al. Wlds protection distinguishes axon degeneration following injury from naturally occurring developmental pruning. Neuron 50, 883895 (2006). 34. Suzuki, E. & Nakayama, M. MEGF10 is a mammalian ortholog of CED-1 that interacts with clathrin assembly protein complex 2 medium chain and induces large vacuole formation. Exp. Cell Res. 313, 37293742 (2007). 35. Griffin, J.W., George, R. & Ho, T. Macrophage systems in peripheral nerves. A review. J. Neuropathol. Exp. Neurol. 52, 553560 (1993). 36. Hirata, K. & Kawabuchi, M. Myelin phagocytosis by macrophages and nonmacrophages during Wallerian degeneration. Microsc. Res. Tech. 57, 541547 (2002). 37. Bishop, D.L., Misgeld, T., Walsh, M.K., Gan, W.B. & Lichtman, J.W. Axon branch removal at developing synapses by axosome shedding. Neuron 44, 651661 (2004). 38. Aldskogius, H. & Arvidsson, J. Nerve cell degeneration and death in the trigeminal ganglion of the adult rat following peripheral nerve transection. J. Neurocytol. 7, 229250 (1978). 39. Hanani, M. Satellite glial cells in sensory ganglia: from form to function. Brain Res. Brain Res. Rev. 48, 457476 (2005). 40. Fenzi, F., Benedetti, M.D., Moretto, G. & Rizzuto, N. Glial cell and macrophage reactions in rat spinal ganglion after peripheral nerve lesions: an immunocytochemical and morphometric study. Arch. Ital. Biol. 139, 357365 (2001). 41. Yu, X., Lu, N. & Zhou, Z. Phagocytic receptor CED-1 initiates a signaling pathway for degrading engulfed apoptotic cells. PLoS Biol. 6, e61 (2008). 42. Ravichandran, K.S. & Lorenz, U. Engulfment of apoptotic cells: signals for a good meal. Nat. Rev. Immunol. 7, 964974 (2007). 43. Reddien, P.W., Cameron, S. & Horvitz, H.R. Phagocytosis promotes programmed cell death in C. elegans. Nature 412, 198202 (2001). 44. Kurant, E., Axelrod, S., Leaman, D. & Gaul, U. Six-microns-under acts upstream of Draper in the glial phagocytosis of apoptotic neurons. Cell 133, 498509 (2008). 45. Ziegenfuss, J.S. et al. Draper-dependent glial phagocytic activity is mediated by Src and Syk family kinase signaling. Nature 453, 935939 (2008). 46. Nagata, S. Autoimmune diseases caused by defects in clearing dead cells and nuclei expelled from erythroid precursors. Immunol. Rev. 220, 237250 (2007). 47. Silva, M.T., do Vale, A. & Dos Santos, N.M. Secondary necrosis in multicellular animals: an outcome of apoptosis with pathogenic implications. Apoptosis 13, 463482 (2008).

2009 Nature America, Inc. All rights reserved.

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1541

ONLINE METHODS
Animals. CD-1 mice, Sprague-Dawley rats or Ntf3+/+ and Ntf3/ mice20 were used for these studies. All experimental procedures using animals were approved by the Institutional Animal Care and Use Committee at Vanderbilt University, the committee on Animal Research at the University of Valencia and conformed to US National Institutes of Health and EU guidelines. ISH. Embryos were dissected in cold phosphate-buffered saline (PBS), fixed in 4% paraformaldehyde (wt/vol) overnight at 4 C and transferred to 30% sucrose/PBS (wt/vol) before being cryosectioned at 20 m. ISH on sections were performed as described previously48. We used digoxigenin-labeled cRNA probes to Jedi-1 (probes 1 and 2, 467931 bp and 8711,347 bp of GenBank #AF444274), megf10 (probes 1 and 2, 1,3261,828 bp and 2,3692,857 of GenBank # NM_001001979). Electron microscopy. Isolated E11.513.5 mouse embryos were fixed by immersion in 4% paraformaldehyde and 1% glutaraldehyde in 0.1 M phosphate buffer, pH 7.4, for 2 h and thoroughly washed with phosphate buffer. Body fragments containing lumbar DRGs were postfixed with 1% osmium tetroxide (wt/vol) in phosphate buffer for 90 min at 2025 C, dehydrated through a graded ethanol series followed by propylene oxide and embedded in araldite (Durcupan, Fluka). Samples were sectioned at 1 m and sections were stained with 1% toluidine blue (wt/vol) for ganglia identification. Ultrathin 60-nm-thick sections from lumbar ganglia were collected on formvar-coated slot grids, stained with uranyl acetate and lead citrate, and examined with a Jeol JEM-1010 electron microscope. RT-PCR. Total RNA from E12.5 CD-1 mouse brain, heart, spinal cord, DRG and cultured embryonic satellite cells19 or neurons was extracted with TRIzol reagent (Invitrogen) per manufacturers recommendation and reverse-transcribed using random primers and Superscript II reverse transcriptase (Invitrogen) following treatment with DNase I (DNA-free kit, Ambion). Resulting cDNAs were analyzed by PCR. We used primers to Jedi-1 (forward, 5- CCT GCA GCT GCC CAC CGG GCT GGA -3; reverse, 5- CCT GGC AGC CCG GGC CAT GCG TGT -3), Megf10 (forward, 5-GGC GCG CCT GTG TCC CGA GGG GCT TT-3; reverse, 5-CGG GCG CAC AGG TGC AGG TCC CAT CC-3) and Megf11 (forward, 5-GCG CGC CAC GGA AGC AGG CCC CGA TG-3; reverse, 5-CCA GCC TCG GAA GCC CGG GGC GCA CA-3). Plasmids and ISH probes. For information regarding cDNA constructs, truncated fusion constructs, ISH probes, and shRNA containing plasmids please see the Supplementary Methods. Neuronal engulfment assay. DRG from E13.5 mouse CD1 embryos or E15 rat embryos were dissociated with collagenase A (1 mg ml1, Roche) plus 0.05% trypsin (1:250, Gibco) and triturated. We then plated 50,000 cells onto a 25-mm round coverslip (#1.5, 0.17 mm, Warner Instruments) coated with collagen. For mouse cultures, cells were grown in mouse-engulfing medium (MEGM, 1:1 UltraCULTURE serum-free medium (BioWhittaker): Neural Basal medium (Invitrogen) supplemented with 3% fetal bovine serum (FBS, vol/vol, Hyclone), N2 and B27 (Invitrogen)). For rat cultures, cells were grown in Basal Medium Eagle (Invitrogen) supplemented with 0.4% glucose, (wt/vol) 3% FBS and N2. Cultures were first grown in the presence of 50 ng ml1 NGF (Harlan) for 2 d, then the NGF was removed by washing the cultures twice and refeeding with a 1:10,000 dilution of monoclonal antibody to mouse NGF (Chemicon) to remove any remaining NGF. For mouse cultures, 50 ng ml1 of glial growth factor (R&D Systems) were added to ensure the survival of immature peripheral glial cells. Cells were then transfected with the plasmids indicated using Effectene (Qiagen) according to the manufacturers recommendation. DRG SGCs isolation. DRG SGCs were isolated as previously described25 with the following modifications. E13.5 mouse DRGs were cultured as explants on 35-mm dishes coated with poly-d-lysine and laminin (Invitrogen) in MEGM with 25 ng ml1 NGF and 50 ng ml1 glial growth factor or 1 M insulin (Sigma) for 3 d to allow cell migration. Neuron soma and glia still associated with the explants were pinched out using fine forceps and dissecting scissors. The remaining satellite glial cells were dissociated by trypsinization and replated to a 10-cm dish coated with polyd-lysine/laminin in MEGM with 1 M insulin without NGF for 2 d.

2009 Nature America, Inc. All rights reserved.

Immunostaining and quantification. Embryos, processed as for in situ analysis, were cryosectioned at 12 m, permeablized and blocked with 0.5% Triton X-100 (wt/vol) in 10% bovine calf serum (wt/vol) and incubated with primary antibodies. For primary antibodies, we used rabbit antibody to BFABP (1:1,000, kind gift from T. Muller22, Max Delbruck Center, F4/80 (1:100, Serotec), rabbit antibody to S100 (1:4, ImmunoStar) and mouse antibody to neuronal class III -tubulin (Tuj1, 1:1,000, Convance). TO-PRO3 (Invitrogen) was used according to the manufacturers recommendation. For immunohistochemistry, Ntf3+/+ and Ntf3/ embryos were processed as described previously20. To quantify neuronal engulfment in vivo, we counted the proportion of apoptotic bodies that appeared to be surrounded by BFABP- or F4/80-positive cells in sections from at least six DRGs from at least three embryos at each age. For immunofluorescence staining of cultured cells, cultures were fixed with 4% paraformaldehyde or 3.5% formaldehyde (vol/vol) and incubated with primary antibodies to S100 (1:4), BFABP (1:1,000), Flag M2 (1:500, Sigma), TUJ1 (1:1,000), F4/80 (1:100), Thy1.1 (1:25, Serotec), LAMP1 (1D4B; 3.9 mgml1; the LAMP1 monoclonal antibody was 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) and chicken antibody to GFP (1:500, Abcam). Cultures were then incubated with 200 g ml1 of DNase-free RNase A in PBS for 30 min at 2025 C, rinsed, and nuclei were visualized with 1 M TO-PRO3 (Invitrogen). For secondary antibodies, we used Alexa488-conjugated goat antibody to rabbit (1:1,000) and Alexa488-conjugated goat antibody to mouse (1:200) from Invitrogen, and Rhodamine Xconjugated donkey antibody to mouse (1:200400) and Cy2conjugated donkey antibody to chicken (1:200) from Jackson ImmunoResearch Laboratories. Photomicrographs of z axis series were taken using a Zeiss LSM 510 inverted confocal microscope (Cell Imaging Shared Resource at Vanderbilt University Medical Center) and analyzed with LSM image Browser from Zeiss. For each experiment, at least 50 cells from each condition (done as duplicates or triplicates for each experiment) were counted. The results were obtained from 23 independent experiments. Quantitation of cell proliferation. Pregnant females were injected with 50 g per kg of body weight bromodeoxyuridine 1 h before being killed. Embryos were fixed in Carnoys solution and embedded in paraffin. Sections were treated with 2 N HCl for 30 min at 37 C, neutralized in 0.1 M sodium borate (pH 8.5) for 510 min and immunostained using mouse monoclonal antibodies to BrdU (1:300; Dako) and BFABP (1:1,000). Apoptotic neuron binding assay. DRG neurons from E13.5 CD1 mouse embryos were isolated as described above, plated on collagen-coated coverslips and grown in UltraCULTURE media with 50 ng ml1 NGF (Harlan). Non-neuronal cells were eliminated by two 48-h treatments with uridine (10 M) and fluorodeoxy uridine (10 M). NGF was removed to induce apoptosis by rinsing and addition of antibody to NGF (1:10,000). After 24 h, the neurons were harvested by scraping, stained with propidium iodide for 20 min, rinsed and added to transfected HEK293 cells. HEK293 cells were plated onto collagen-coated eight-well glass chamber slides at a density of 20,000 cells per well in DMEM with 10% FBS (Sigma). After 24 h, the cells were transfected using Effectene (Qiagen) and the indicated number of propidium iodidestained apoptotic neurons were added 48 h later for 1 h at 4 C. Unbound neurons were washed away with three rinses of PBS and fixed with 10% formalin. To quantify, we counted at least 100 GFP-positive HEK293 cells per condition. The percentage of GFP-positive cells with at least one propidium iodidepositive corpse bound was calculated. Surface biotinylation. HEK293 cells were transfected with GFP-tagged Jedi or MEGF10 and the surface expression analyzed 48 h later by biotinylation of the receptor at 4 C using EZ-Link Sulfo-NHS-Biotin (Pierce) according the manufacturers instructions. The biotinylated proteins were precipitated using avidin-agarose (Pierce) and subjected to SDS-PAGE and western blotting using antiserum to GFP (1:500, Roche). Analysis of the ectopic expression of JediC-GFP in C. elegans. The ced-1 JediC-gfp and ced-1Jedi-gfp fusion constructs were generated from full-length Jedi-1Flag or JediC-GFP and the ced-1 promoter14 and were introduced as

nature NEUROSCIENCE

doi:10.1038/nn.2446

extrachromosomal arrays into wild-type or ced-1(e1735) backgrounds. Transgenic lines were obtained using standard microinjection techniques49 and the transgenic worms were identified by GFP expression using a fluorescence microscope. Germ cell corpses in hermaphrodite gonads were analyzed using Axioplan 2 compound microscope (Carl Zeiss) with Nomarski DIC accessories, AxiocaCam camera and the AxioVision v4.3 imaging software. Germ cell corpses labeled with JediC-GFP were identified by fluorescence microscopy under the DeltaVision Deconvolution Microscope (Applied Precision). Fluorescence images were deconvolved using the SoftWoRx software (Applied Precision). Statistical analysis. The cell corpse data obtained from transgenic and the wild-type control C. elegans were compared and P values were obtained from unpaired, two-tailed Students t tests. The amount of dead neurons binding to

transfected HEK293 cells was compared using P values obtained from paired, two-tailed Students t tests. The data from over expressing Jedi-1 and/or MEGF10 in glial cells shown in Figure 7c was analyzed using a one-way ANOVA with a Dunnetts post-test and the data in Figure 7d were evaluated using a MantelHaenszel 2 analysis. To evaluate the effects of overexpressing the truncated mutants of Jedi-1 and MEGF10 and knocking them down in the engulfment assay shown in Figure 8, we subjected the data to a one-way ANOVA with a Dunnetts post-test.
48. Wu, H.H. et al. Autoregulation of neurogenesis by GDF11. Neuron 37, 197207 (2003). 49. Jin, Y., Jorgensen, E., Hartwieg, E. & Horvitz, H.R. The Caenorhabditis elegans gene unc-25 encodes glutamic acid decarboxylase and is required for synaptic transmission, but not synaptic development. J. Neurosci. 19, 539548 (1999).

2009 Nature America, Inc. All rights reserved.

doi:10.1038/nn.2446

nature NEUROSCIENCE

a r t ic l e s

Leucine-rich repeat transmembrane proteins instruct discrete dendrite targeting in an olfactory map
Weizhe Hong1, Haitao Zhu1,4, Christopher J Potter1, Gabrielle Barsh1, Mitsuhiko Kurusu2,3, Kai Zinn2 & Liqun Luo1
Olfactory systems utilize discrete neural pathways to process and integrate odorant information. In Drosophila, axons of first-order olfactory receptor neurons (ORNs) and dendrites of second-order projection neurons (PNs) form class-specific synaptic connections at ~50 glomeruli. The mechanisms underlying PN dendrite targeting to distinct glomeruli in a three-dimensional discrete neural map are unclear. We found that the leucine-rich repeat (LRR) transmembrane protein Capricious (Caps) was differentially expressed in different classes of PNs. Loss-of-function and gain-of-function studies indicated that Caps instructs the segregation of Caps-positive and Caps-negative PN dendrites to discrete glomerular targets. Moreover, Caps-mediated PN dendrite targeting was independent of presynaptic ORNs and did not involve homophilic interactions. The closely related protein Tartan was partially redundant with Caps. These LRR proteins are probably part of a combinatorial cell-surface code that instructs discrete olfactory map formation. Spatial representation of sensory stimuli in the brain is a fundamental organizational principle that facilitates the processing and integration of sensory information1. In the visual and auditory systems, neurons connect nearby spatial/frequency inputs to nearby target regions in the brain, thereby forming a spatially continuous neural map. In contrast, the olfactory system utilizes discrete channels to detect olfactory information, with each channel consisting of a group of ORNs that express a specific odorant receptor. The axons of ORNs expressing the same odorant receptor converge on an anatomically discrete glomerular unit in the insect antennal lobe or vertebrate olfactory bulb, the first olfactory processing center in the brain. In each glomerulus, a single class of ORN axons functionally synapses with the dendrites of a single class of second-order olfactory neurons: PNs in insects or mitral cells in vertebrates. Thus, from insects to mammals, olfactory input and output are spatially organized into distinct channels via glomeruli, forming a discrete neural map1. Studies of vertebrate visual map formation have supported a crucial role for continuous gradients of guidance molecules for instructing the formation of a continuous neural map during development. However, less is known about the mechanisms by which a discrete map, as exemplified by the olfactory system, is precisely constructed1,2. The Drosophila antennal lobe consists of ~50 glomeruli, which can be uniquely identified by their stereotypical size, shape and relative position3. Most PNs project their dendrites to a single glomerulus and synapse with the axons of a single ORN class. The dendrite targeting of PNs to a specific glomerulus is specified by their lineage and birth order4. Notably, the initial targeting of PN dendrites precedes the arrival of pioneering ORN axons5, suggesting that the coarse glomerular map arises independently of ORN input. The Drosophila olfactory system thus provides an attractive model system for studying mechanisms of PN dendrite and ORN axon targeting in the context of discrete map formation. PN dendrite targeting is presumably achieved by differential expression of cell-surface receptors in different classes of PNs so that they can respond differently to a common environment. Although combinatorial expression of intrinsic transcription factors in PNs has been shown to regulate their dendrite targeting 68, little is known about instructive cell-surface receptors during this process. The transmembrane proteins Dscam and N-cadherin are important for the elaboration and refinement of PN dendrites, respectively9,10, but they are expressed in and required for all PN classes equally. Differential expression of transmembrane Semaphorin-1a regulates coarse dendrite targeting along the dorsolateral-to-ventromedial axis during the initial formation of the antennal lobe11. However, no cellsurface molecules have been shown to instruct different classes of PN dendrites to select discrete glomerular targets in a three-dimensional neural map. We found that the LRR transmembrane protein Caps was differentially expressed in different classes of PNs and cell-autonomously instructed glomerulus-specific targeting of PN dendrites. Further analysis suggested that Caps-mediated dendrite targeting was independent of presynaptic ORNs and did not involve homophilic interactions. We propose that Caps mediates interactions among PN dendrites, leading to a mosaic segregation of Caps-positive and Caps-negative PNs to discrete glomerular targets.

2009 Nature America, Inc. All rights reserved.

1Howard Hughes Medical Institute and Department of Biology, Stanford University, Stanford, California, USA. 2Division of Biology, California Institute of Technology, Pasadena, California, USA. 3Structural Biology Center, National Institute of Genetics, and Department of Genetics, the Graduate University for Advanced Studies, Mishima, Japan. 4Present address: Department of Neurodegeneration, Genentech, South San Francisco, California, USA. Correspondence should be addressed to L.L. (lluo@stanford.edu).

Received 10 August; accepted 1 October; published online 15 November 2009; doi:10.1038/nn.2442

1542

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 1 Caps is differentially expressed in the developing antennal lobe. (ah) The developing antennal lobes at 48 h APF stained by antibodies against a synaptic marker nc82 (a, d and g), Caps (b, e and h) and caps-Gal4driven mCD8-GFP (c and f). Caps was differentially expressed in the wild-type developing antennal lobe, as shown in a single anterior (ac) and a single posterior section (df) from the same triple-labeled brain. Caps staining was absent in capsc28fs/Df(3L)Exel6118 trans-heterozygous mutant antennal lobes (g,h). (ik) Expression of the Flp-out GFP reporter UAS>stop>mCD8-gfp at the intersection of caps-Gal4 and PN-specific GH146-Flp in the adult antennal lobe. Three single sections in anterior, central and posterior regions of the antennal lobe are shown (magenta, nc82; green, mCD8-GFP). (l) Schematic representation of the glomerular innervation pattern of caps-Gal4expressing PNs across three different sections of the antennal lobe (see Supplementary Fig. 2 for details). Antennal lobes in this and all subsequent images are shown such that dorsal is up and medial is to the left. Scale bars represent 10 m.

a
Wild type anterior nc82

Caps

caps::CD8-gfp

d
Wild type posterior nc82

Caps

caps::CD8-gfp Dorsal Medial Lateral

g
Whole animal LOF nc82

Caps

Ventral

RESULTS Caps is differentially expressed in different PN classes To identify instructive cell-surface molecules for PN dendrite targeting, we used a recently established database that contains 976 transmembrane and secreted molecules that are potentially involved in cell-cell recognition12. We collected 462 transgenic lines with a UAS insertion in the 5 end of these genes. These transgenic lines could potentially drive expression of 410 of these 976 genes, covering ~40% of the repertoire of the potential cell-recognition molecules12. We expressed each line in a small subset of PNs using Mz19-Gal4 (ref.5). We identified P{GS6}10839 in the 5 end of caps as showing a strong PN dendrite mistargeting phenotype. caps encodes a transmembrane protein with 14 LRRs in its extracellular domain13. Previous studies have shown that caps is involved in regulating cell-cell interactions in a variety of developmental processes, including boundary formation in wing and leg discs14,15, organization of the morphogenetic furrow and ommatidial spacing16, and formation of branch interconnections in tracheal develop ment17. In the nervous system, caps has been shown to regulate the axon targeting of motor neurons to specific subsets of muscles12,13 and axon targeting of R8 photoreceptor neurons to the proper layer in the medulla18. Staining with polyclonal antibodies to Caps revealed that Caps protein was present in the developing antennal lobe (Fig. 1af and Supplementary Fig. 1). Around 48 h after puparium formation (APF), when individual glomeruli in the antennal lobe are just becoming identifiable, differential Caps expression was evident, with high expression in some glomeruli and low or undetectable expression in others (Fig. 1b,e). The distinct expression levels of Caps did not arise from a differential density of neurites, as the density of neurites was rather uniform between different glomeruli, as shown by staining of nc82, a presynaptic marker19 (Fig. 1a,d). The Caps staining was eliminated in a loss-of-function caps mutant (Fig. 1g,h), indicating that the antibody is specific to endogenous Caps protein. Furthermore, the expression of UASmCD8-gfp driven by the enhancer trap capsGal4 recapitulated the glomerular-specific Caps staining pattern (Fig. 1b,c,e,f), suggesting that caps-Gal4 is a faithful reporter of endogenous Caps expression. At 48 h APF, the antennal lobe consists of dendrites from PNs and axons from ORNs. To determine the contribution to caps expression by PN dendrites, we generated a PN-specific flipase line, GH146-Flp. Similar to the GH146-Gal4 expression pattern, GH146-Flp was expressed in the majority of PNs; these PNs innervated 40 out of the 46 glomeruli that we scored (Supplementary Fig. 2). Therefore, we used a Flp-out GFP reporter UAS>stop>mCD8-gfp to determine the
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

i
caps-Gal4 GH146-Flp UAS>stop>mCD8-gfp Anterior

2009 Nature America, Inc. All rights reserved.

Central
D

Posterior
DM3 DL5 DP1m DP1l DL1

l
GH146+, Caps+ GH146+, Caps GH146, Caps?

DM6 DA2 DM5 VM2 VA2

DA3 DL3 DA4 DA1

DM2 VM5

DL4 DL2d DC3 DL2v

DM1 DM4 VM7

DC1 DC2

VA6

VA7m

VA1d VA1lm

VA7lVA5 VA3 VM3

VC2 VA4

VL2a

VC1 VL2p VC3m VC3l VM1 VL1 VM4 VM6 V

Anterior

Central

Posterior

intersection of GH146-Flp and caps-Gal4, and thus Caps expression in PNs (Supplementary Fig. 2). Notably, caps-Gal4 was selectively expressed in a subset of PNs innervating 23 out of 40 GH146-positive glomeruli (Fig. 1il and Supplementary Fig. 2). Glomerular targets of Caps-positive and Caps-negative PNs did not segregate into broad domains and appeared to be intercalated (Fig. 1l). Loss of caps causes mistargeting of Caps-positive PNs The differential expression of Caps in different glomeruli raised the possibility that Caps instructs targeting of PN dendrites to specific glomeruli according to Caps expression patterns. This hypothesis predicts that a loss of Caps in Caps-positive PNs should cause their dendrites to mistarget to glomeruli that are normally innervated by Caps-negative PNs, whereas it may not affect the dendrite targeting of Caps-negative PNs. We first tested this prediction by performing mosaic analysis with a repressible marker (MARCM) using the null allele capsc28fs for the lateral neuroblast clone containing 12 classes of PNs, nine of which were Caps positive. Loss of caps in the lateral neuroblast clones resulted in two types of dendrite targeting defects: loss of innervation in glomeruli that are normally targets of lateral PNs and gain of innervation in glomeruli that are normally not the targets of lateral PNs (Fig. 2ac). In accordance with our prediction, quantitative analysis of dendrite distribution in caps mutants showed that all of the glomeruli that exhibited loss of innervation were targets of Caps-positive PNs, whereas the ectopically innervated glomeruli were mostly normal targets of Caps-negative PNs (Fig. 2d). This bias of mistargeting toward normal Caps-negative PN targets was highly significant (2, P < 0.001; Supplementary Fig. 3). Notably, loss of caps in Caps-positive PNs did not cause a random mistargeting of dendrites to all Caps-negative targets but instead caused a preferential mistargeting to specific ectopic targets (Fig. 2d). Our analysis of 12 PN classes in neuroblast clones suggests that caps is required in PNs for proper dendrite targeting. However, it is
1543

a r t ic l e s
Figure 2 Dendrite targeting phenotypes of 100% caps/ neuroblast clones. (a,b) Dendrite 75% Both loss and gain targeting of neuroblast MARCM clones of Loss of innervation / 50% wild-type and caps PNs. (a) Wild-type lateral Ectopic innervation neuroblast clones stereotypically targeted 25% dendrites to 12 glomeruli. (b) caps/ lateral 0% neuroblast clones exhibited selective dendrite c28fs mistargeting. The arrowhead indicates the Wild type caps ectopic innervation of VM7, which wild-type 100% Normal innervation PNs do not innervate, and the arrow indicates Loss of innervation 75% the loss of innervation of wild-type glomerular Ectopic innervation target VA4. (c) Quantification of dendrite 50% targeting defects in lateral neuroblast MARCM 25% clones of control and caps mutant. The y axis represents the percentage of the antennal lobes 100% that had mistargeting phenotypes (control, n = 9; / 75% caps , n = 12). (d) Glomerular innervation 50% specificity of lateral neuroblast MARCM clones of control and caps mutants analyzed in c. 25% The left 12 columns are the normal glomerular targets of lateral PNs; the green bars represent the percentage of antennal lobes in which an individual glomerulus was innervated and the Caps Caps+ Caps Caps+ ND white bars represent the percentage of antennal lobes in which a normally innervated glomerulus Glomeruli normally innervated by Glomeruli normally not innervated by was not innervated. The right 34 columns are labeled lateral neuroblast clones labeled lateral neuroblast clones glomeruli that are normally not innervated by lateral PNs; the black bars represent the percentage of antennal lobes in which an ectopic glomerulus was innervated. All glomeruli are color coded at the bottom on the basis of expression of Caps in the corresponding PNs as indicated. ND, not determined for Caps expression ( GH146-negative).

Mistargeting

ild

ty p

capsc28fs mistargeting

Wild type mistargeting

DA1 DA2 DL3 DM1 DM2 VA7m VC1 VC2 VM1 VA4 VA5 DM5

2009 Nature America, Inc. All rights reserved.

difficult to determine exactly which classes of PNs contribute to the ectopic innervations and whether the loss of innervation is caused by mistargeting rather than by gross defects in dendrite arborization. In addition, it is unclear whether the phenotype is caused by a cell-autonomous requirement for caps. To address these questions, we performed MARCM analysis of specific PN classes, including single-cell clones. Using GH146-Gal4 and MZ19-Gal4, along with additional information of neuroblast lineage, heat-shock window and axon-branching pattern, we sampled four Caps-negative (DL1, DA1, DC3 and VA1d) and four Caps-positive (VC1, VC2, VA4 and DM1) PN classes innervating different regions in the antennal lobe (Fig. 3, see Online Methods). Consistent with our prediction, the four Caps-negative PN classes (DL1, DA1, DC3 and VA1d) did not have detectable dendrite targeting defects ( Fig. 3ac,hj). However, loss of caps in four Capspositive PN classes (VC1, VC2, VA4 and DM1) resulted in innervation of additional ectopic glomeruli that were normally targeted by Caps-negative PNs (Fig. 3dg,kn). All caps/ VC1 PNs exhibited strong ectopic innervation, and 92% of this ectopic innervation occurred in the DA2, DC2, VM7, DC1 and VM5 glomeruli, all of which are normally innervated by Caps-negative PNs (Fig. 3s,t). Similarly, loss of caps in the Caps-positive VC2, VA4 and DM1 PNs resulted in strong ectopic innervation of the VM7, DA2 and VM5 glomeruli, all of which are normally innervated by Caps-negative PNs (Fig. 3u,v). In both cases, the bias of mistargeting toward normal Caps-negative PN targets was highly significant (2, P < 0.01; Supplementary Fig. 3). Therefore, the loss-of-function analysis in both neuroblast and single-cell clones suggested that Caps instructs the targeting of Caps-positive and Caps-negative PN dendrites to different glomeruli. In the absence of Caps, Caps-positive PNs retained part of their dendrites in their normal glomeruli but mistargeted part of their dendrites preferentially into glomeruli that are normal targets of Caps-negative PNs. We also noticed that caps/ PNs mistargeted to ectopic glomeruli that tended to be in close
1544

proximity to their normal glomerular targets (for example, Fig. 3w for VC1), suggesting that Caps regulates local dendrite targeting. To test the cell autonomy of Caps function, we used MARCM to express a UAS-caps transgene only in single cells that were homozygous mutant for caps and labeled by mCD8-GFP. This resulted in a rescue of the dendrite mistargeting phenotype of all four Caps-positive PNs (Fig. 3os,u). Because the GH146-Gal4 used for rescue was expressed only in postmitotic PNs8, this result indicates that Caps is cellautonomously required in postmitotic neurons for the dendrite targeting of these Caps-positive PNs. In contrast with dendrite mistargeting, lateral horn axon terminal arborization patterns of caps mutant PNs still followed their classspecificity, as previously described2022 (Supplementary Fig. 4, see Online Methods). This suggests that caps regulates the targeting of dendrites as opposed to the general fate determination of PNs and that dendrite targeting and axon terminal arborization are separable processes. Misexpression causes mistargeting of Caps-negative PNs We found that loss of Caps in Caps-positive PNs caused dendrite mistargeting preferentially to glomerular targets of Caps-negative PNs. Next, we tested whether ectopic expression of Caps in Caps-negative PNs would cause dendrite mistargeting and whether such mistargeting would be preferential to glomeruli that are normally innervated by Caps-positive PNs. We found that MARCM overexpression of Caps in PN neuroblast clones caused severe mistargeting, resulting in a deformation of the entire antennal lobe structure (Fig. 4ad). Next, we misexpressed Caps using Mz19-Gal4 in three classes of PNs, DA1, VA1d and DC3, which were all Caps negative and sent their dendrites to three adjacent glomeruli (Fig. 4e,f). Caps misexpression by Mz19-Gal4 resulted in a mistargeting of dendrites to nearby VA1lm (81%) and VA4 (31%), both of which were targets of Caps-positive PNs (Fig. 4f). We further misexpressed Caps in a single DL1 PN, which normally did not express (Fig. 1l) or require Caps for targeting (Fig.3h).
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

VM7 DA4 DC2 DC1 VM5 D DC3 DL1 DP1m VA1d VA3 VA7l VL1 VM2 DL2d DL2v DL4 DL5 DM3 DM4 DM6 VA1lm VA2 VA6 VC3l VL2a VL2p VM3 DA3 DP1l V VC3m VM4 VM6

ca

ps

c2 8f s

a r t ic l e s
Figure 3 Cell-autonomous requirement of Caps in Caps-positive PNs for dendrite targeting. (ac,hj) Normal dendrite targeting of caps/ MARCM clones in DL1 single cells (h), DA1 neuroblasts (i) and VA1d/DC3 neuroblasts (j), compared with wild type (ac). These PNs were all Caps negative (n = 20 for both wild-type and caps/). (dg,kn) Defective dendrite targeting of VC1, VC2, VA4 and DM1 PNs in single-cell caps/ clones (kn) compared with wild type (dg). These PNs were all Caps positive. Their identities were determined as described in the Online Methods. (or) Rescue of dendrite targeting of four Caps-positive PNs by expressing a UAS-caps transgene only in the single-cell clones. (sv) Quantification (s,u) and glomerular innervation specificity (t,v) of dendrite targeting defects in single-cell clones of control and caps mutant analyzed in dg and kr. VC2, VA4 and DM1 were analyzed together (see Online Methods). y axes in s and u represent the percentage of PNs in particular classes that have dendrite mistargeting phenotypes. Each black bar in t and v indicates the percentage of antennal lobes in which an ectopic glomerulus was innervated. The glomeruli are color coded as in Figure 2d. The VC1 and VC2/VA4/DM1 classes are omitted in t and v, respectively, as they are the ones from which clones were made (wild type: VC1, n = 15; VC2, n = 15; VA4, n = 15; DM1, n = 10; caps/: VC1, n = 7; VC2/VA4/DM1, n = 19; rescue: VC1, n = 6; VC2, n = 5; VA4, n = 6; DM1, n = 4). (w) Schematic representation of the glomerular innervation pattern of individual caps/ VC1 PNs across three different sections of the antennal lobe. Red glomeruli are the ectopic targets of caps/ VC1 dendrites.
Caps-negative PNs DL1 DA1 VA1d/DC3 VC1 Caps-positive PNs VC2 VA4 DM1

a
WT

b
WT

c
WT

d
WT

e
WT

f
WT

g
WT

h
caps/

i
caps/

j
caps/

k
caps/

l
caps/

m
caps/

n
caps/

s
100% VC1 Mistargeting 75% 50% 25% 0% Wild type capsc28fs Rescue

t
Mistargeting

o
90% 60% 30% 0% Rescue

p
Rescue

q
Rescue

r
Rescue

2009 Nature America, Inc. All rights reserved.

u
VC2/VA4/DM1 Mistargeting

75% 50% 25% 0% Wild type capsc28fs Rescue

Mistargeting

100%

v 40%
30% 20% 10% 0%

w
DM5 VM2

D DM6 DA2 DA3 DL3 DA4 DA1 DM2 VM5 VC2 VM3 VA4 DC1 DC2

VA6 VA2

VA7m

VA1d VA1lm

DC3 DL2v VL2a

VA7lVA5 VA3

Caps misexpression produced partial loss of innervation of the DL1 glomerulus and ectopic innervation of a selective subset of other glomeruli (Fig. 4gi); these ectopic glomerular targets were normally innervated by Caps-positive PNs, except for DP1l whose Caps expression status was undetermined (Fig. 4j). This bias of mistargeting toward normal Caps-positive PN targets was highly significant (2, P < 0.001; Supplementary Fig. 3), supporting the hypothesis that Caps instructs dendrite targeting of Caps-positive and Caps-negative PNs to different glomeruli. Notably, over two-thirds of the mistargeting events occurred at glomeruli near DL1 (DL2d, DL2v, VL2a and VL2p; Fig. 4k), suggesting that mistargeting is preferentially local. We also noticed that mistargeted dendrites avoided two glomeruli DL4 and DL5, which were adjacent to DL1 and innervated by Capspositive PNs (Fig. 4k), suggesting that mistargeting is not random among local ectopic targets (see Discussion). Caps misexpression in single DL1 PNs did not affect PN axon targeting in the lateral horn (Supplementary Fig. 4), further confirming that caps specifically regulates the specificity of dendrite targeting, as opposed to the general determination of cell fate. caps is not required for ORN axon targeting Our expression pattern, loss-of-function and gain-of-function data suggest that Caps instructs dendrite targeting of Caps-positive and Caps-negative PNs to discrete glomeruli. We next explored the cellular mechanisms by which Caps functions to regulate dendrite targeting. Caps has been proposed to determine axon-target connectivity by regulating the interaction between photoreceptor or motor axons and their postsynaptic targets13,18. This model is further supported by a recent
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

observation that Caps appears to mediate direct interaction between postsynaptic filopodia of muscles and presynaptic growth cones23. PNs send their dendrites to the developing antennal lobe before the arrival of pioneering ORN axons5. The dendrites subsequently elaborate and refine their processes while ORNs extend their axons into the antennal lobe9,10. The glomerular positioning of dendrites and axons eventually requires pre- and postsynaptic interactions to achieve the proper matching specificity between PNs and ORNs 9. If Caps has a function in the olfactory system that is analogous to its role in motor neurons and photoreceptors, it might mediate interactions between ORN axons and PN dendrites. However, the following experiments strongly argue against this model. To determine whether caps is also expressed in presynaptic ORNs, we examined the expression intersection between caps-Gal4 and ey-Flp. ey-Flp is expressed in precursors of ORNs but not in central neurons including PNs24, thereby allowing us to specifically visualize caps-Gal4 expression in ORNs using a Flp-out reporter. We found that caps-Gal4 was expressed in a subset of ORNs selectively innervating 28 out of 46 glomeruli (Fig. 5ad and Supplementary Fig. 2). However, glomeruli innervated by Caps-expressing PNs and ORNs exhibited only partial overlap (Fig. 5e and Supplementary Fig. 2) and the correlation between Caps expression in PNs and ORNs was not statistically significant (2, P > 0.3). To test whether caps is required in ORNs for their axon targeting, we removed caps from about half of ORNs using the ey-Flp MARCM strategy24 and analyzed targeting of nine different ORN classes using Or-Gal4 or AM29-Gal4 lines to label specific classes of ORN axons. These nine classes of ORN-PN pairs sample four pairs of Caps-positive
1545

DA2 DC2 VM7 DC1 VM5 DP1m D DA1 DA4 DC3 DL1 DL3 VA1d VA3 VA7l VL1 VM2 DL2d DL2v DL4 DL5 DM2 DM3 DM4 DM5 DM6 VA1lm VA2 VA5 VA6 VA7m VC1 VC3l VL2a VL2p VM1 VM3 DA3 DP1l V VC3m VM4 VM6

DA2 DC2 VM7 DC1 VM5 DP1m D DA1 DA4 DC3 DL1 DL3 VA1d VA3 VA7l VL1 VM2 DL2d DL2v DL4 DL5 DM1 DM2 DM3 DM4 DM5 DM6 VA1lm VA2 VA4 VA5 VA6 VA7m VC2 VC3l VL2a VL2p VM1 VM3 DA3 DP1l V VC3m VM4 VM6

Caps

Caps+

ND

Caps

Caps+

ND

DL4 DL2d

DM1 DM4 VM7

DM3 DL5 DP1m

DL1

GH146+, Caps+ GH146+, Caps GH146, Caps? Ectopic targets (>30%) Ectopic targets (<30%)

DP1l VL2p VL1 V

VC3m VC3l VM1 VM4 VM6

VC1

a r t ic l e s
adPN lPN Mz19 PNs DL1

a
Wild type

Misexpression

j
Mistargeting

80% 60% 40% 20% 0%


DL2d DL2v VC2 VL2a VA4 DM1 VA2 VL2p DM2 VA6 DL4 DM6 DL5 DM3 DM4 DM5 VA1lm VA5 VA7m VC1 VC3l VM1 VM3 DP1m D DA1 DA2 DA4 DC1 DC2 DC3 DL3 VA1d VA3 VA7l VL1 VM2 VM5 VM7 DA3 DP1l V VC3m VM4 VM6

2009 Nature America, Inc. All rights reserved.

Caps+

Caps DL1
D

ND

k
DM6 DA2 DM5 VM2 VA2 DA3 DL3 DA4 DA1 DM2 VM5 VC2 VM3 VA4 DL4 DL2d DC3 DL2v VL2a DM1 DM4 VM7 DC1 DC2 VA6 VA1d VA1lm

DM3 DL5 DP1m DP1l

GH146+, Caps+
+ GH146 , Caps

VA7m

VA7lVA5 VA3

VC1 VL2p VC3m VC3l VM1 VL1 VM4 VM6 V

GH146 , Caps? Ectopic targets (>30%) Ectopic targets (<30%)

Figure 4 Dendrite targeting phenotypes of Caps misexpression in Caps -negative PNs. (ad) MARCM misexpression of Caps in anterodorsal ( b) and lateral ( d) neuroblast clones resulted in severe mistargeting phenotypes compared with wild-type neuroblast clones ( a,c) (control, n = 9; Caps misexpression, n = 10). (e,f) Caps misexpression in DA1, DC3 and VA1d PNs using Mz19-Gal4 resulted in strong mistargeting of dendrites to VA1lm and VA4 (arrowheads), both of which were normally innervated by Caps-positive PNs (control, n = 40; Caps misexpression, n = 48). (gi) Misexpression of Caps in single DL1 PNs using GH146-Gal4 caused mistargeting of dendrites to ectopic glomeruli (arrowheads in h and i). (j) Quantification of glomerular innervation pattern of individual DL1 PNs misexpressing Caps. Each black bar indicates the percentage of antennal lobes in which an ectopic glomerulus was innervated. All glomeruli are color coded as in Figure 2d (control, n = 20; Caps misexpression, n = 19). (k) Schematic representation of the glomerular innervation pattern of individual Caps-misexpressing DL1 PNs across three different sections of the antennal lobe. Red glomeruli indicate ectopic targets of DL1 dendrites misexpressing Caps. adPN, PNs derived from the anterodorsal neuroblast; lPN, PNs derived from the lateral neuroblast.

ORNs and Caps-positive PNs (Or22a-DM2, Or47a-DM3, Or47b-VA1lm and AM29-DM6), four pairs of Caps-positive ORNs and Caps-negative PNs, (Or46a-VA7l, Or59c-VM7, Or67b-VA3 and Or88a-VA1d), and one pair of Caps-negative ORNs and Caps-positive PNs (AM29-DL4). None of these nine ORN classes exhibited any obvious axon-targeting defects (Fig. 5fi and data not shown), indicating that caps is not cellautonomously required in ORNs for their proper axon targeting.

In addition, when the glomerular position of PN dendrites was shifted as a result of the loss of caps in ventral VA1lm PNs, the axons of presynaptic Or47b ORNs shifted accordingly without compromising the matching between ORN axons and PN dendrites (Supplementary Fig. 5). Thus, loss of Caps in PNs does not appear to disrupt the proper targeting of ORN axons, at least for the specific ORN-PN pair that we tested.

a
Figure 5 caps is not required in ORNs for their axon targeting. (ac) Expression of the intersectional reporter UAS>stop>mCD8-gfp for caps-Gal4 and ey-Flp. ey-Flp was expressed in ORN precursors but not in PNs or their precursors (magenta, nc82; green, mCD8-GFP). (d) Schematic representation of the glomerular innervation pattern of caps-Gal4expressing ORNs across three different sections of the antennal lobe. Gray glomeruli indicate targets of caps-Gal4positive ORNs and white glomeruli indicate targets of caps-Gal4negative ORNs. (e) Schematic comparison of the innervation pattern of Caps-positive PNs and ORNs (see also Supplementary Fig. 2). (fi) Axon targeting of ORNs in the antennal lobe was not affected in caps/ ORNs. Or67b ORNs innervating VA3 and Or22a ORNs innervating DM2 are shown. In these experiments, only caps/ axons of a particular class of ORNs were visualized by the Gal4 lines, whereas all of the cells in the central brain, including all PNs, are heterozygous because ey-Flp restricts recombination in the olfactory system to the peripheral organs.

Anterior

Central

Posterior

WT, Or67b (VA3)

caps-Gal4, ey-Flp, UAS>stop>mCD8-gfp

caps/, Or67b (VA3)

e
DM6 DA2 DM5 VM2 VA2 DA3 DL3 DA1 DA4 VA6 VA1d VA1lm VA7m VA5 VA7l VA3 VM3 DM2 VM5

h
D DC1 DC2 DL4 DL2d DC3 DL2v VL2a DM1 DM4 VM7 DM3 DL5 DL1 DP1m DP1l

VC2 VA4

VC1 VL2p VC3m VC3l VM1 VL1 VM4 VM6 V

WT, Or22a (DM2)

Both PN and ORN ORN only PN only caps/, Or22a (DM2)

1546

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 6 Caps-mediated PN dendrite targeting is independent of ORNs. (ad) Mistargeting of DL1 PN dendrites caused by Caps misexpression was observed at early developmental time points. Dotted lines bisect the antennal lobe into dorsolateral and ventromedial parts. Arrowheads in b and d point to DL1 dendrites that crossed the dotted line and invaded the ventromedial part of the antennal lobe. (e) Quantification of mistargeting across the dorsomedial-ventrolateral midline of the antennal lobe (dotted lines in ad) at different developmental stages (adult, n = 12; 16 h APF, n = 10; 48 h APF, n = 5). (fj) Caps misexpression by Mz19-Gal4 caused a segregation of dendritic field as a consequence of dendrite mistargeting to VA1lm and VA4 (arrowheads in g) compared with continual dendritic field from control Mz19-Gal4labeled PNs (f). Caps misexpression by Mz19-Gal4 in ORN-ablated flies caused dendrite segregation ( i) compared with control Mz19 PNs in ORN-ablated flies, which formed a continual field (h). (j) Quantification of the dendrite segregation phenotypes caused by Mz19-Gal4 misexpression of Caps, with and without ORN ablation (Caps misexpression without ORN ablation, n = 48; Caps misexpression with ORN ablation, n = 18).
Wild type Misexpression

a
DL1 16 h APF

e
Mistargeting

100% 75% 50% 25% 0% Adult 16 h APF ORN ablated 48 h APF

c
DL1 48 h APF

f
Mz19 PNs

j
Mistargeting

100% 75% 50% 25% 0% Control

Caps-mediated PN dendrite targeting is independent of ORNs Even though Caps is not required for ORN axon targeting, Capsdependent PN dendrite targeting could, in principle, still depend on the interaction with cues from ORNs. Two lines of evidence argue against the possibility that Caps itself provides the putative ORNderived cue. First, there was only partial overlap between Caps expression pattern in PNs and ORNs (Fig. 5e and Supplementary Fig.2). Second, both loss of innervation and ectopic innervation of PN dendrites occurred in glomerular targets of both Caps-positive and Caps-negative ORNs with no obvious preference (Supplementary Fig. 3). These observations argue against a specific hypothesis that PN dendrite targeting is dependent on homophilic interactions of Caps between PN dendrites and ORN axons. Below we provide two lines of evidence further suggesting that Caps-mediated PN dendrite targeting is independent of ORN axons (Fig. 6). We have previously shown that PNs start to elaborate dendrites shortly after puparium formation and that PN dendrites localize to their initial stereotypical target region of the developing antennal lobe before pioneering ORN axons arrive at 18 h APF5. Therefore, an examination of dendrites at 16 h APF would allow us to examine the ORN-independent phase of early PN dendrite targeting. Developmental time-course analysis of Caps expression in the antennal lobe indicated that the developing antennal lobe was already positive for both Caps-specific antibody staining and caps-Gal4 expression at 1216 h APF (Supplementary Fig. 1), confirming that Caps was expressed in developing PN dendrites. At 16 h APF, all wild-type DL1 PN dendrites localized to the dorsal-lateral corner of the developing antennal lobe (Fig. 6a); however, Caps mis expression in DL1 caused dendrites to extend across the midline of the dorsomedial-ventrolateral axis of the antennal lobe in ~40% of the samples (Fig. 6b), arguing against the possibility that Caps mediates interactions between PN dendrites and ORN axons, at least for PN initial targeting. The medial mistargeting phenotype persisted at 48 h APF (Fig. 6c,d) and the penetrance of mistargeting was comparable among different developmental stages (Fig. 6e), suggesting that adult defects are likely caused by defects in early PN dendrite targeting. To further test whether ORN axons are involved in Caps-instructed PN dendrite targeting at a later developmental stage, we compared Caps misexpression phenotypes in an otherwise normal or ORNablated background. If Caps instructs PN dendrite targeting by mediating the interactions between ORN axons and PN dendrites, we expect that ORN ablation would suppress the mistargeting phenotypes of PN dendrites caused by PN misexpression of Caps. To ablate all ORNs during early development, we used Pebbled-Gal4 and ey-Flp to express the flip-out toxin UAS>stop>RicinA. Pebbled-Gal4 and
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

2009 Nature America, Inc. All rights reserved.

h
Mz19 PNs in ORN ablation

ey-Flp are expressed in all ORNs during early development24 and this strategy therefore ablated almost all ORNs before their axons entered the developing antennal lobe (Supplementary Fig. 6). ey-Flp is not expressed in PNs or their progenitors, allowing us to simultaneously use a PNspecific Mz19-Gal4 line to assess PN dendrite development in ORN-ablated flies. Mz19-Gal4 labeled three PN classes innervating three adjacent glomeruli, DA1, VA1d and DC3, located in the dorsolateral region of the antennal lobe (Fig. 6f). When ORNs are ablated during develop ment, the dendrites of these three PN classes still converged to discrete regions and remained adjacent to each other, although they were located in the ventrolateral region as a result of a shift of the antennal lobe orientation in the absence of ORN axons ( Fig. 6h). These results suggest that PNs retain their intrinsic ability to converge their dendrites to specific glomerular regions at both early and late developmental stages without the contribution of ORN axons. Caps misexpression by Mz19-Gal4 resulted in mistargeting of PN dendrites to VA1lm and VA4 and the dendrites were frequently segregated to nonadjacent regions in the antennal lobe as a consequence of this mistargeting (Fig. 6g). Notably, this dendrite segregation also occurred when Caps was misexpressed by Mz19-Gal4 after ablating all ORNs (Fig. 6i), and the penetrance was comparable to Caps misexpression without ORN ablation (Fig. 6j). Taken together, these data strongly suggest that Caps instructs PN dendrite targeting independently of ORN axons throughout develop ment. Besides ORNs, PNs are likely the only cell type that can provide class-specific positional cues. Given the mosaic distribution of Capspositive and Caps-negative glomeruli, we suggest that Caps regulates PN dendrite targeting through PN-PN interactions, as opposed to responding to a global cue, leading to a segregation of Caps-positive and Caps-negative PNs to different glomeruli (see Discussion). Caps does not mediate homophilic interactions Caps has been proposed to act as a homophilic recognition molecule on the basis of its ability to promote S2 cell aggregation, although this can only be seen when the expression level is very high13,14,18 (A. Nose, personal communication). Here, we used a genetic approach
1547

a r t ic l e s
100% Figure 7 Caps does not mediate homophilic Full interactions for PN dendrite targeting. 75% Partial (a) Normal dendrite targeting of a single DL1 labeling in Misexpression in Misexpression in Normal caps/ fly wild-type fly caps/ fly DL1 PN in a caps whole-animal mutant 50% (capsc28fs/Df(3L)Exel6118). (b,c) Dendrite 25% mistargeting of a single DL1 PN misexpressing Caps in an otherwise wild-type background 0% (b) or an otherwise capsc28fs/Df(3L)Exel6118 mutant background (c). (d) Quantification of DL1 targeting defects for the genotypes in ac. (DL1 labeling in wild-type background, n = 20; DL1 labeling in trans-heterozygous mutant background, n = 20; Caps misexpression in an otherwise wild-type background, n = 8; Caps misexpression in an otherwise trans-heterozygous mutant background, n = 9). Gray bars represent the percentage of partial mistargeting events in which PN dendrites still partially innervated DL1, and black bars represent the percentage of full mistargeting events in which no dendrites innervated DL1.
Mistargeting DL1 labeling in wild type DL1 labeling in LOF Misexpression in wild type

2009 Nature America, Inc. All rights reserved.

to functionally test whether Caps mediates homophilic interactions in vivo during PN dendrite targeting (Fig. 7). We found that Caps misexpression in a single Caps-negative DL1 PN resulted in a preferential mistargeting (Fig. 4h,i and 7b). If homophilic interactions among Caps-expressing cells underlie this misexpression phenotype, we would expect that eliminating endogenous Caps expression in the entire fly would suppress this phenotype. The caps homozygous mutants die primarily as embryos, but a few (<0.1%) survive until adulthood. DL1 PN dendrites still targeted properly to the DL1 glomerulus in these caps homozygous mutants (Fig. 7a). However, ectopic expression of Caps in a single DL1 PN in these caps mutant flies still caused mistargeting of DL1 dendrites to ectopic glomeruli in the antennal lobe (Fig. 7c). Quantification

showed that Caps misexpression in single DL1 PN caused a similar degree of mistargeting in a whole-animal caps/ background as in the wild type (Fig. 7d), indicating that Caps-dependent dendrite targeting does not use Caps in other cells as a cue, at least in the gain-offunction context. Overall, these data suggest that Caps uses a heterophilic ligand to instruct dendrite targeting. Partially redundant function of Caps and Tartan Caps shares 67% sequence identity in its extracellular domain with another LRR transmembrane protein, Tartan (Trn)13,25, the closely related paralog of Caps. trn and caps have redundant functions in regulating boundary formation in wing imaginal dics 14,26, leg segmentation15 and the architecture of the morphogenetic furrow and ommatidial spacing16. In the nervous system, trn and caps also have redundant functions in regulating motor axon targeting12,23. VC2/VA4/DM1 Indeed, Trn overexpression also resulted in dendrite mistargeting phenotypes in
Figure 8 trn enhances caps phenotypes in PN dendrite targeting. (a) Quantification of dendrite targeting defects in lateral neuroblast MARCM clones of control, single and double mutants of caps and trn, as indicated. The y axis and the colored bars are represented as described in Figure 2c (wild type, n = 31; capsc28fsFRT2A, n = 30; trn28.4FRT2A, n = 18; trn17capsc28fsFRT2A, n = 17; trn28.4caps1FRT80B, n = 11). (b) Quantification of dendrite targeting defects in single-cell MARCM clones of control, single and double mutants of caps and trn, as indicated. The y axis and the bars are represented as described in Figure 7d (for VC1, wild type, n = 20; capsc28fsFRT2A, n = 7; trn28.4FRT2A, n = 5; trn17 capsc28fsFRT2A, n = 5; trn28.4caps1FRT80B, n = 5; for VC2/VA4/DM1, wild type, n = 30; capsc28fsFRT2A, n = 19; trn28.4FRT2A, n = 13; trn17capsc28fsFRT2A, n = 16; trn28.4caps1FRT80B, n = 13). (c) Glomerular innervation specificity of the lateral neuroblast MARCM clones of trn single mutant and trn caps double mutants analyzed in a. The y axis and the colored bars are represented as described in Figure 2d. (dh) Dendrite targeting of single-cell VC1 MARCM clones of control, single and double mutants of caps and trn, as indicated. Arrowheads indicate the ectopic innervation of DA2, which wild-type VC1 PNs did not innervate.

a
100% Mistargeting 75% 50% 25% 0%

Both loss and gain Loss of innervation Ectopic innervation

b
Mistargeting

Full Partial 100% 75% 50% 25% 0%

VC1

pe

.4

pe

pe

8f

8f

28

8f

8f

8f

c2

c2

28

28

ty

c2

c2

ps

ty

ty

c2

c2

8f

trn

ps

ps

ps

trn

ps

trn

ca

ps

ps

ild

ps

ild

ild

ca

ca

ca

ca

ca

ca

ca

.4

28

.4

28

trn

trn

trn

trn28.4 mistargeting

100% 75% 50% 25% 100% 75% 50% 25% 100%

Normal innervation Loss of innervation Ectopic innervation

trn28.4 caps1 mistargeting

trn17 capsc28fs mistargeting

75% 50% 25%


VM7 DA4 DC2 DC1 VM5 D DC3 DL1 DP1m VA1d VA3 VA7l VL1 VM2 DL2d DL2v DL4 DL5 DM3 DM4 DM6 VA1lm VA2 VA6 VC3l VL2a VL2p VM3 DA3 DP1l V VC3m VM4 VM6 DA1 DA2 DL3 DM1 DM2 VA7m VC1 VC2 VM1 VA4 VA5 DM5

Caps

Caps+

Caps

trn

Caps+

trn

trn

28

.4

ca

Glomeruli normally innervated by labeled lateral neuroblast clones

Glomeruli normally not innervated by labeled lateral neuroblast clones

Wild type

capsc28fs

trn28.4

trn17 capsc28fs

trn28.4 caps1

1548

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

ps

.4

.4

ND

Misexpression in caps LOF

a r t ic l e s
neuroblast clones and DL1 single PN clones (Supplementary Fig. 7). Moreover, the expression of an enhancer trap trn-lacZ, along with caps-Gal4, GH146-Flp and UAS>stop>mCD8gfp, indicated that trn was expressed in PNs and partially overlapped with caps expression (Supplementary Fig. 7). To test the requirement of trn in PN dendrite targeting and its potential redundant function with Caps, we carried out loss-of-function studies of trn single and trn caps double mutants, analogous to the caps experiments described above (Fig. 8). We found that the loss of trn in the lateral neuroblast clone resulted only in ectopic innervation, but two independent trn caps double-mutant pairs showed a significantly higher percentage of combined loss of innervation and ectopic innervation compared with either of the single mutants (Fig. 8a). A glomerulus is usually innervated by multiple PNs so that ectopic innervation reflects a partial mistargeting of these PNs, whereas loss of innervation indicates that all of these PNs completely mistarget away from the normal region. The loss of innervation of VA7m, VC1 and VC2 were not observed in either caps (Fig. 2d) or trn single mutants but occurred frequently in trn caps double mutants (Fig. 8c), indicating that trn caps double mutants have more severe mistargeting phenotypes. Furthermore, single-cell loss-of-function analysis of VC1 and VC2/VA4/DM1 PNs consistently revealed more severe mistargeting phenotypes for double mutants than for either of the single mutants (Fig. 8b). For example, caps single-mutant VC1 PNs always retained a part of their dendrites in the VC1 glomerulus (Fig. 8b,e); however, a large percentage of trn caps double-mutant VC1 PNs no longer innervated VC1 at all (Fig. 8b,g,h), consistent with the strong loss of innervation of VC1 that we observed in lateral neuroblast clones of trn caps double mutants. Given that Caps and Trn have high sequence similarity, similar overexpression phenotypes, overlapping expression patterns and enhancement of PN dendrite mistargeting in double mutants compared with either of the single mutants alone, we conclude that Caps and Trn have partially redundant functions in PN dendrite targeting. However, similar to the caps single mutant, neither trn single nor trn caps double mutants had any obvious targeting defects in the axons of nine different ORN classes tested for caps single mutants (data not shown), suggesting that Trn and Caps are dispensable for ORN axon targeting. DISCUSSION Discrete cell-surface codes in a discrete neural map Graded expression of guidance molecules is widely used in patterning continuous, topographic representation of sensory inputs. A classic example is graded Ephrin/Eph signaling in regulating retinotopic projections in vertebrates1,27. Graded molecules may also be used in the initial coarse patterning of discrete olfactory maps in flies and mice11,28, but selective targeting to discrete glomerular units requires additional mechanisms. Here, we found that the LRR transmembrane protein Caps was expressed in a subset of PNs whose dendrites targeted to a subset of glomeruli intermingled with other glomeruli targeted by Caps-negative PNs. Loss of caps selectively affected targeting of Caps-positive PNs, causing them to preferentially mistarget to glomeruli that were normally innervated by Caps-negative PNs. Conversely, misexpression of Caps in Caps-negative PNs caused them to selectively target dendrites to glomeruli that are normal targets of Caps-positive PNs. These data indicate that Caps instructs the targeting of Caps-positive and Caps-negative PN dendrites to discrete glomeruli in the Drosophila antennal lobe. How does Caps-mediated targeting act together with molecular gradients used for establishing the coarse dendrite map? A general conceptual problem of making a discrete map using gradient
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

cues is that nearby discrete glomerular units have to be distinguished with a minimal difference in gradient cues. One potential solution would be to express a second type of cell-surface molecules that form a spatially intercalated expression pattern with a tight correlation with each discrete class of neurons. Indeed, recent studies in mouse ORN axon targeting have suggested a hierarchical model in which molecular gradients help coarse targeting of axons, followed by local refinement mediated by activity-dependent expression of cellrecognition molecules1,2. Neuropilin-1 and Semaphorin-3A are expressed in broad domains in olfactory bulb and disruption of Semaphorin-3A causes global axon mistargeting2933. However, other cell-surface molecules, including Ephrin-A2/A5 and EphA5, as well as immunoglobulin superfamily proteins Kirrel-2, Kirrel3 and BIG-2, are differentially expressed in a mosaic pattern and likely regulate local axon targeting through axon-axon interactions among ORNs3436. Drosophila PN dendrite targeting appears to utilize a conceptually similar strategy even though details may differ regarding types of molecules, cells (ORNs versus PNs), neuronal processes (axons versus dendrites) and dependence on activity; this hierarchical regulation may be a general strategy for discrete map formation. Specifically, we propose that molecular gradients such as Semaphorin-1a may set up a coarse map for PN dendrites. This is followed by Caps-dependent local dendrite targeting to discrete glomeruli. In support of this notion, loss of semaphorin-1a causes a directional shift of dendrites11, whereas loss of caps causes a specific mistargeting to local ectopic targets. Both loss-of-function and misexpression analyses indicate that the mistargeted dendrites are restricted to only a subset of ectopic glomeruli following the Caps code. For example, Caps-misexpressing DL1 PNs mistargeted to the DL2v and DL2d glomeruli but avoided DL4 and DL5 glomeruli, even though they were all in the vicinity of DL1 and were normal targets of Caps-positive PNs. Similarly, caps loss of function in VC1 PNs caused dendrites to mistarget to VM7, DC2, DC1 and VM5, and to avoid VL1, DP1m and DC3, even though they were all in the vicinity of VC1 and were normal targets of Capsnegative PNs. These observations suggest that additional molecules must work together with Caps to distinguish targeting specificity among different Caps-positive PNs or among different Caps-negative PNs. Because these additional molecules are still intact in PNs with an altered Caps code, their actions might explain the mistargeting specificity. Thus, Caps likely acts in parallel with additional molecules to specify dendrite targeting of the 50 PN classes into areas that eventually develop into 50 glomeruli. Cellular and molecular mechanisms The matching expression of Caps in motor neurons and their target muscles12,13, as well as in R8 photoreceptor axons and their target layers18, suggest that Caps functions in synaptic matching through interactions between pre- and postsynaptic partners. During the assembly of the olfactory system, 50 classes of ORN axons need to match with 50 classes of PNs. However, our data strongly argue against the possibility that the PN dendrite targeting defect is caused by a predominant role of Caps in mediating synaptic matching between ORN axons and PN dendrites. First, caps is required for targeting of PN dendrites but not ORN axons. This cannot be caused by the redundancy with its close paralog Trn, as trn caps double mutants enhanced caps phenotypes in PN dendrite targeting but still exhibited no phenotypes in ORNs. Second, expression patterns of Caps in PNs and ORNs did not match with regard to their glomerular targets. Third, Caps-mediated PN dendrite targeting was independent of ORNs. Fourth, Caps misexpression phenotype was not affected by
1549

2009 Nature America, Inc. All rights reserved.

a r t ic l e s
the loss of Caps in all other cells, suggesting that it interacts with a heterophilic ligand for PN dendrite targeting. The use of a heterophilic ligand is consistent with studies of Caps in boundary formation of the wing imaginal disc14,26. Given the mosaic distribution of Caps-positive and Caps-negative glomeruli, as well as the similarities with the axon-axon interactions in refining local targeting of mouse ORN axons, we propose that the heterophilic Caps ligand is most likely derived from other PNs and that the interactions between Caps and its ligand mediate PN-PN interactions to segregate Caps-positive and Caps-negative PNs to specific glomeruli. How do Caps and its heterophilic ligand segregate Caps-positive and Caps-negative PNs via PN-PN interactions? The simplest hypothesis consistent with our data is that a repulsive ligand is expressed in Caps-negative PNs, forming a mosaic pattern complementary to Caps. Caps and its ligand mediate repulsive interactions among PN dendrites to segregate Caps-positive and Caps-negative PNs. When Caps is lost in a Caps-positive PN, its dendrites innervate ectopic glomeruli that are preferentially targets of Caps-negative PNs because of a loss of repulsion from Caps-negative PN dendrites; additional molecules that normally instruct Caps-positive PN dendrites to select a specific Caps-positive glomeruli are still intact, thereby preventing caps mutant PNs from invading other Caps-positive glomeruli. When Caps is misexpressed in a Caps-negative PN, the repulsive ligand would force dendrites to mistarget to glomerular targets of Caps-positive PNs. Future identification of the Caps ligand will be essential to test this hypothesis and will shed further light on the mechanisms of discrete olfactory map formation. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank A. Nose (University of Tokyo), S. Cohen (Temasek Life Sciences Laboratory), M. Freeman (Medical Research Council Laboratory of Molecular Biology), S. Hayashi (RIKEN Center for Developmental Biology) and M. Milan (Icrea and Parc Cientific de Barcelona) for fly stocks and reagents; the Bloomington, Szeged, Kyoto and Harvard Stock Centers for fly stocks; M. Spletter for making antennal lobe schemes; and T. Clandinin, K. Miyamichi, M. Spletter, L. Sweeney, J. Wu, X. Yu, D. Berdink and other laboratory members for comments and discussions. This work was supported by US National Institutes of Health grant R01-DC005982. L.L. is an investigator of the Howard Hughes Medical Institute. AUTHOR CONTRIBUTIONS W.H. performed most of the experiments and analyzed the data. H.Z. initiated the overexpression screen. C.J.P. provided the GH146-Flp transgenic fly line. G.B. assisted in some experiments. M.K. and K.Z. provided the database and collection of fly strains for the overexpression screen. W.H. and L.L. designed the experiments and wrote the paper.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
4. Jefferis, G.S., Marin, E.C., Stocker, R.F. & Luo, L. Target neuron prespecification in the olfactory map of Drosophila. Nature 414, 204208 (2001). 5. Jefferis, G.S. et al. Developmental origin of wiring specificity in the olfactory system of Drosophila. Development 131, 117130 (2004). 6. Komiyama, T., Johnson, W.A., Luo, L. & Jefferis, G.S. From lineage to wiring specificity. POU domain transcription factors control precise connections of Drosophila olfactory projection neurons. Cell 112, 157167 (2003). 7. Komiyama, T. & Luo, L. Intrinsic control of precise dendritic targeting by an ensemble of transcription factors. Curr. Biol. 17, 278285 (2007). 8. Spletter, M.L. et al. Lola regulates Drosophila olfactory projection neuron identity and targeting specificity. Neural Dev. 2, 14 (2007). 9. Zhu, H. et al. Dendritic patterning by Dscam and synaptic partner matching in the Drosophila antennal lobe. Nat. Neurosci. 9, 349355 (2006). 10. Zhu, H. & Luo, L. Diverse functions of N-cadherin in dendritic and axonal terminal arborization of olfactory projection neurons. Neuron 42, 6375 (2004). 11. Komiyama, T., Sweeney, L.B., Schuldiner, O., Garcia, K.C. & Luo, L. Graded expression of semaphorin-1a cell-autonomously directs dendritic targeting of olfactory projection neurons. Cell 128, 399410 (2007). 12. Kurusu, M. et al. A screen of cell-surface molecules identifies leucine-rich repeat proteins as key mediators of synaptic target selection. Neuron 59, 972985 (2008). 13. Shishido, E., Takeichi, M. & Nose, A. Drosophila synapse formation: regulation by transmembrane protein with Leu-rich repeats, CAPRICIOUS. Science 280, 21182121 (1998). 14. Miln, M., Weihe, U., Prez, L. & Cohen, S.M. The LRR proteins capricious and Tartan mediate cell interactions during DV boundary formation in the Drosophila wing. Cell 106, 785794 (2001). 15. Sakurai, K.T., Kojima, T., Aigaki, T. & Hayashi, S. Differential control of cell affinity required for progression and refinement of cell boundary during Drosophila leg segmentation. Dev. Biol. 309, 126136 (2007). 16. Mao, Y., Kerr, M. & Freeman, M. Modulation of Drosophila retinal epithelial integrity by the adhesion proteins capricious and tartan. PLoS One 3, e1827 (2008). 17. Krause, C., Wolf, C., Hemphl, J., Samakovlis, C. & Schuh, R. Distinct functions of the leucine-rich repeat transmembrane proteins capricious and tartan in the Drosophila tracheal morphogenesis. Dev. Biol. 296, 253264 (2006). 18. Shinza-Kameda, M., Takasu, E., Sakurai, K., Hayashi, S. & Nose, A. Regulation of layer-specific targeting by reciprocal expression of a cell adhesion molecule, capricious. Neuron 49, 205213 (2006). 19. Wagh, D.A. et al. Bruchpilot, a protein with homology to ELKS/CAST, is required for structural integrity and function of synaptic active zones in Drosophila. Neuron 49, 833844 (2006). 20. Jefferis, G.S. et al. Comprehensive maps of Drosophila higher olfactory centers: spatially segregated fruit and pheromone representation. Cell 128, 11871203 (2007). 21. Wong, A.M., Wang, J.W. & Axel, R. Spatial representation of the glomerular map in the Drosophila protocerebrum. Cell 109, 229241 (2002). 22. Marin, E.C., Jefferis, G.S., Komiyama, T., Zhu, H. & Luo, L. Representation of the glomerular olfactory map in the Drosophila brain. Cell 109, 243255 (2002). 23. Kohsaka, H. & Nose, A. Target recognition at the tips of postsynaptic filopodia: accumulation and function of Capricious. Development 136, 11271135 (2009). 24. Sweeney, L.B. et al. Temporal target restriction of olfactory receptor neurons by Semaphorin-1a/PlexinA-mediated axon-axon interactions. Neuron 53, 185200 (2007). 25. Chang, Z. et al. Molecular and genetic characterization of the Drosophila tartan gene. Dev. Biol. 160, 315332 (1993). 26. Miln, M., Prez, L. & Cohen, S.M. Boundary formation in the Drosophila wing: functional dissection of Capricious and Tartan. Dev. Dyn. 233, 804810 (2005). 27. McLaughlin, T. & OLeary, D.D. Molecular gradients and development of retinotopic maps. Annu. Rev. Neurosci. 28, 327355 (2005). 28. Imai, T., Suzuki, M. & Sakano, H. Odorant receptorderived cAMP signals direct axonal targeting. Science 314, 657661 (2006). 29. Walz, A., Rodriguez, I. & Mombaerts, P. Aberrant sensory innervation of the olfactory bulb in neuropilin-2 mutant mice. J. Neurosci. 22, 40254035 (2002). 30. Taniguchi, M. et al. Distorted odor maps in the olfactory bulb of semaphorin 3A-deficient mice. J. Neurosci. 23, 13901397 (2003). 31. Schwarting, G.A. et al. Semaphorin 3A is required for guidance of olfactory axons in mice. J. Neurosci. 20, 76917697 (2000). 32. Schwarting, G.A., Raitcheva, D., Crandall, J.E., Burkhardt, C. & Pschel, A.W. Semaphorin 3A-mediated axon guidance regulates convergence and targeting of P2 odorant receptor axons. Eur. J. Neurosci. 19, 18001810 (2004). 33. Imai, T. et al. Pre-target axon sorting establishes the neural map topography. Science 325, 585590 (2009). 34. Cutforth, T. et al. Axonal ephrin-As and odorant receptors: coordinate determination of the olfactory sensory map. Cell 114, 311322 (2003). 35. Kaneko-Goto, T., Yoshihara, S., Miyazaki, H. & Yoshihara, Y. BIG-2 mediates olfactory axon convergence to target glomeruli. Neuron 57, 834846 (2008). 36. Serizawa, S. et al. A neuronal identity code for the odorant receptorspecific and activity-dependent axon sorting. Cell 127, 10571069 (2006).

2009 Nature America, Inc. All rights reserved.

1. Luo, L. & Flanagan, J.G. Development of continuous and discrete neural maps. Neuron 56, 284300 (2007). 2. Imai, T. & Sakano, H. Roles of odorant receptors in projecting axons in the mouse olfactory system. Curr. Opin. Neurobiol. 17, 507515 (2007). 3. Laissue, P.P. et al. Three-dimensional reconstruction of the antennal lobe in Drosophila melanogaster. J. Comp. Neurol. 405, 543552 (1999).

1550

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

ONLINE METHODS
Fly stocks. GH146-Gal4 (ref. 37) and Mz19-Gal4 (ref. 5) were used to label PNs. Or-Gal4 lines (Or22a-Gal4, Or46a-Gal4, Or47a-Gal4, Or47b-Gal4, Or59c-Gal4, Or67b-Gal4 and Or88a-Gal4)24,3841, AM29-Gal4 (ref. 42) and Pebbled-Gal4 (ref. 24) were used to label ORNs. The enhancer trap line caps-Gal4 (NP3294)18 is used to visualize Caps-expressing cells and trn-lacZ25 is used to visualize Trnexpressing cells. P{GS6}10839 (caps) and P{GS6}10885 (trn) were generated by the Drosophila Gene Search Project (Metropolitan University)43. The single and double null-mutant alleles of caps and trn that we used were capsc28fsFRT2A15,18, trn28.4FRT2A15,25, trn17capsc28fsFRT2A15 and trn28.4caps1FRT80B16, as well as a small caps deficiency allele, Df(3L)Exel6118. capsc28fs contains a sevenbase pair deletion that results in a frame shift after the first 28 amino acids15,18. trn28.4 carries a small deletion of trn gene generated by P-element excision25. trn17capsc28fs was generated on the basis of the null allele capsc28fs and carries an additional deletion covering the entire trn coding sequence15. trn28.4caps1 was generated on the basis of the null allele trn28.4 and carries an additional deletion covering the entire caps coding sequence16. A UAS-caps transgene13 inserted on the third chromosome was used for rescue and misexpression experiments. The flies that we used in the overexpression screen were previously described12. Clonal analysis. The MARCM method was applied as described previously44,45. Briefly, caps and/or trn alleles were placed trans-heterozygous to Gal80 on the FRT chromosome arm15,18. Flipase activity caused mitotic recombination of the FRT chromosome arm such that one of the daughter cells became homozygous for caps or trn and simultaneously lost Gal80. This cell (and its progeny) can therefore be labeled by the Gal4/UAS system. For ORN analysis, ey-Flp was used to induce clones in nearly half of the ORNs and Or-specific Gal4 lines were used to label these clones. For PN analysis, clones were generated by hsFlp and labeled by GH146-Gal4 or Mz19-Gal4. For generating neuroblast clones, flies were heat shocked between 2436 h after egg laying for 1 h at 37 C. For generating class-specific clones, flies were heat shocked during a unique time window after egg laying for 1 h at 37 C and the clone identities were determined by the combination of Gal4 lines used, heat shock window, neuroblast lineage and axon branching pattern. Four Caps-negative (DL1, DA1, DC3 and VA1d) and four Caps-positive (VC1, VC2, VA4 and DM1) PN classes were investigated here. Mz19-Gal4 labels DC3 and VA1d from the anterodorsal neuroblast and DA1 from the lateral neuroblast5. We used MZ19-Gal4 to analyze dendrite targeting of DC3/VA1d and DA1 by generating neuroblast clones at 2436 h after egg laying. The other five classes were analyzed as single-cell clones using GH146-Gal4. To analyze DL1 PNs, we heat shocked flies between 2436 h after egg laying; 100% of the anterodorsal PN single-cell clones generated at this time are DL1 (ref. 4). To analyze the four Caps-positive PNs in the lateral lineage, we identified a time window of clone induction (3648 h after egg laying) in which all of the GH146-Gal4positive single-cell PN clones generated from the lateral neuroblast were caps-positive and projected their dendrites to VA4, DM1, VC1 or VC2 glomerulus in wild-type controls (Fig. 3dg). The axon branching pattern of VC1 in the lateral horn was uniquely identifiable among the four groups and was unchanged in caps mutants (Supplementary Fig. 4). We performed a blind test of mixing axon patterns of single-cell MARCM clones of four PN classes each with three genotypes, control, mutant and rescue. All of the VC1 PNs were identified in the blind test with 100% accuracy; the mutant VC1 samples were also validated by their partial dendrite innervation of VC1 in addition to ectopic glomeruli. These criteria allowed us to identify VC1 PNs unequivocally. The other three Caps-positive PN classes (VC2, VA4 and DM1) also had characteristic lateral horn axon branching patterns that were largely unchanged in mutants. However their axon branching patterns were more similar to each other and were not 100% identified in our blind test. Therefore, we analyzed dendrite mistargeting of these three classes together (Fig. 3u,v). We assigned a class to each representative image in Figure 3ln on the basis of their partial innervation of VC2, VA4 or DM1. Flies were raised at 25 C after heat shock and dissected 57 d after eclosion. For analyzing DL1 single-cell misexpression during development, flies were raised at 29 C after heat shock to increase expression level and overcome Gal80 perdurance. Flies were dissected at 12 h APF and 36 h APF at 29 C, which are equivalent to 16 h APF and 48 h APF at 25 C, respectively.

To determine whether the preference of mistargeting events that occur in glomerular targets of Caps-positive or negative PNs (or ORNs) is statistically significant, each experiment was tested by 2 on the basis of the null hypothesis that mistargeting events occur randomly in Caps-positive and negative glomeruli. P values were determined for each experiment and used to accept or reject the null hypothesis. For PN MARCM analysis, we used the hsFlp122; UAS-mCD8-gfp, GH146-Gal4; UASmCD8-gfp, and caps (and/or trn) FRT2A (or FRT80B) / G80 FRT2A (or FRT80B) genotypes. For ORN analysis, we used the ey-Flp; UASmCD8-gfp, Or-Gal4; UASmCD8-gfp, and caps (and/or trn) FRT2A/G80 FRT2A genotypes. New transgenes. The p[Gal4, w+]GH146 enhancer trap was inserted into the 5 upstream region of the oaz gene (CG17390). The cloned GH146 genomic enhancer was a 12.6-kb fragment that included all of the genomic DNA between oaz and the upstream gene CG17389, as well as all oaz introns. The Gal4 coding region from pBAC-3xPDsRed-GH146-Gal4 (ref. 46) was replaced with Gal80 to generate pBAC-3xPDsRed-GH146-Gal80 (C.J.P. and L.L., unpublished reagent). Site-directed mutagenesis of pBAC-3xPDsRed-GH146-Gal80 was used to introduce FseI and AvrII restriction sites flanking the Gal80 coding region. A linker oligonucleotide was ligated into the FseI/AvrII site to generate pBAC3xPDsRed-GH146-MCS with multicloning site FseI-AscI-AvrII. Mammalian optimized FLPase (FLPo, Addgene plasmid 13792)47 was PCR amplified to include the FseI/AvrII restriction sites and cloned into pBAC-3xPDsRed-GH146-MCS to generate pBAC-3xPDsRed-GH146-FLPo (GH146-Flp). The FRT-stop-FRT cassette, with additional BglII/NotI restriction sites, was PCR amplified from genomic DNA of UAS-FRT-stop-FRT-shibirets flies48 and cloned into pUASTmCD8-gfp45 to generate pUAST-FRT-stop-FRT-mCD8-gfp (UAS>stop>mCD8-gfp). All constructs were sequence verified. Transgenic flies were generated as previously described49. Immunochemistry. The procedures that we used for fixation, immunochemistry and imaging were described previously50. For primary antibodies, we used mouse nc82 (1:30), rat antibody to N-cadherin (1:40), rat antibody to mCD8 (1:100), chicken antibody to GFP (1:300) and rabbit antibody to Caps (1:100)13.

2009 Nature America, Inc. All rights reserved.

37. Stocker, R.F., Heimbeck, G., Gendre, N. & de Belle, J.S. Neuroblast ablation in Drosophila P[GAL4] lines reveals origins of olfactory interneurons. J. Neurobiol. 32, 443456 (1997). 38. Couto, A., Alenius, M. & Dickson, B.J. Molecular, anatomical, and functional organization of the Drosophila olfactory system. Curr. Biol. 15, 15351547 (2005). 39. Fishilevich, E. & Vosshall, L.B. Genetic and functional subdivision of the Drosophila antennal lobe. Curr. Biol. 15, 15481553 (2005). 40. Komiyama, T., Carlson, J.R. & Luo, L. Olfactory receptor neuron axon targeting: intrinsic transcriptional control and hierarchical interactions. Nat. Neurosci. 7, 819825 (2004). 41. Kreher, S.A., Kwon, J.Y. & Carlson, J.R. The molecular basis of odor coding in the Drosophila larva. Neuron 46, 445456 (2005). 42. Endo, K., Aoki, T., Yoda, Y., Kimura, K. & Hama, C. Notch signal organizes the Drosophila olfactory circuitry by diversifying the sensory neuronal lineages. Nat. Neurosci. 10, 153160 (2007). 43. Toba, G. et al. The gene search system. A method for efficient detection and rapid molecular identification of genes in Drosophila melanogaster. Genetics 151, 725737 (1999). 44. Wu, J.S. & Luo, L. A protocol for mosaic analysis with a repressible cell marker (MARCM) in Drosophila. Nat. Protoc. 1, 25832589 (2006). 45. Lee, T. & Luo, L. Mosaic analysis with a repressible cell marker for studies of gene function in neuronal morphogenesis. Neuron 22, 451461 (1999). 46. Berdnik, D., Fan, A.P., Potter, C.J. & Luo, L. MicroRNA processing pathway regulates olfactory neuron morphogenesis. Curr. Biol. 18, 17541759 (2008). 47. Raymond, C.S. & Soriano, P. High-efficiency FLP and PhiC31 site-specific recombination in mammalian cells. PLoS One 2, e162 (2007). 48. Stockinger, P., Kvitsiani, D., Rotkopf, S., Tirin, L. & Dickson, B.J. Neural circuitry that governs Drosophila male courtship behavior. Cell 121, 795807 (2005). 49. Schuldiner, O. et al. piggyBac-based mosaic screen identifies a postmitotic function for cohesin in regulating developmental axon pruning. Dev. Cell 14, 227238 (2008). 50. Wu, J.S. & Luo, L. A protocol for dissecting Drosophila melanogaster brains for live imaging or immunostaining. Nat. Protoc. 1, 21102115 (2006).

doi:10.1038/nn.2442

nature NEUROSCIENCE

a r t ic l e s

Structural requirements for the activation of vomeronasal sensory neurons by MHC peptides
Trese Leinders-Zufall1, Tomohiro Ishii2,3,5, Peter Mombaerts2,3, Frank Zufall1 & Thomas Boehm4
In addition to their role in the immune response, peptide ligands of major histocompatibility complex (MHC) molecules function as olfactory cues for subsets of vomeronasal sensory neurons (VSNs) in the mammalian nose. How MHC peptide diversity is recognized and encoded by these cells is unclear. We found that mouse VSNs expressing the vomeronasal receptor gene V2r1b (also known as Vmn2r26) detected MHC peptides at subpicomolar concentrations and exhibited combinatorial activation with overlapping specificities. In a given cell, peptide responsiveness was broad, but highly specific; peptides differing by a single amino-acid residue could be distinguished. Cells transcribing a V2r1b locus that has been disrupted by gene targeting no longer showed such peptide responses. Our results reveal fundamental parameters governing the response to MHC peptides by VSNs. We suggest that the peptide presentation system provided by MHC molecules co-evolves with the peptide recognition systems expressed by T cells and VSNs. In the immunological literature, the function of proteins encoded by MHC in peptide presentation is viewed in the context of the generation of a self-tolerant and diverse repertoire of T cells and the immune response. T cells are considered to be a kind of sensory organ that is trained to ignore self-peptide/MHC complexes and is primed to detect nonself-peptide/MHC complexes1. The presentation mechanisms of intracellular peptides at the cell surface for recognition by T-cell receptors (TCRs) emerged 500 million years ago and exist in all jawed vertebrates, from sharks to humans24. Recently, it has become apparent that MHC molecules have additional, unconventional functions in the CNS, where they are involved in neuronal development and plasticity5,6. A nonclassical, neuronal role has also been proposed for the peptide ligands (antigens) that are presented by MHC molecules. These peptides function as chemosensory cues in the mouse nose, where they activate subsets of sensory neurons in the vomeronasal organ (VNO)7,8 as well as some olfactory sensory neurons (OSNs) in the main olfactory epithelium9. MHC peptide ligands can mediate olfactory imprinting in adult mice, as has been observed by the induction of selective pregnancy failure, a procedure that is indicative of the formation of a social memory following the detection of specific chemosensory stimuli7. MHC peptide ligands can also change the attractiveness of social chemostimuli in mice9. Thus, in addition to T cells, some VSNs and OSNs have a receptive modality that evaluates MHC peptide ligands and thereby mediates a sensory evaluation of MHC diversity10,11. Very little is known as to how MHC peptide diversity is recognized and encoded by VSNs and OSNs. Specifically, it is unclear how well sensory neurons and their receptors discriminate related peptide sequences and which ligand features are critical for sensory neuron activation. In contrast, a large body of work has defined the rules underlying the binding preferences of distinct MHC molecules for peptide ligands12. In addition to the coding rules, the molecular basis of peptide recognition in the nose remains a mystery. None of the components of the relevant receptor(s) on sensory neurons have been identified. VSNs activated by MHC peptides express members of the V2r superfamily of genes7, which code for candidate receptors with seven putative transmembrane domains. Described in 1997 (refs. 1315), the 121 intact V2r genes1618 of the mouse remain orphan receptors. We explored whether peptide sensing in VSNs depends on the same or different structural features as those governing the interaction of peptides and MHC molecules. We focused on a genetically defined subset of VSNs in which the green fluorescent protein (GFP) is coexpressed with the V2r gene V2r1b19. We found that these neurons were extremely sensitive detectors of MHC peptide ligands, with activation thresholds in the subpicomolar range. Peptide detection was broad, but highly specific, and peptides differing by only a single amino-acid residue could be distinguished by these VSNs. V2r1b-positive VSNs were functionally heterogeneous with regard to peptide specificities. They had combinatorial patterns with overlapping specificities for peptide recognition, rather than a one ligandone receptorone cell scheme. By using a mouse strain with a targeted disruption of V2r1b, we found that this receptor is required for activation of these VSNs by MHC peptides. RESULTS V2r1b-GFP VSNs show heterogeneous response profiles Our basic assay used acute tissue slices of the mouse VNO and highresolution confocal microscopy to image cellular Ca2+ responses7,20,21. VSNs in these slices were loaded with the Ca2+-indicator dye fura-red/AM

2009 Nature America, Inc. All rights reserved.

1Department of Physiology, University of Saarland School of Medicine, Homburg, Germany. 2The Rockefeller University, New York, New York, USA. 3Max Planck Institute of Biophysics, Frankfurt, Germany. 4Max Planck Institute of Immunobiology, Freiburg, Germany. 5Present address: Department of Cell Biology, Graduate School of Medical and Dental Science, Tokyo Medical and Dental University, Bunkyo-ku, Tokyo, Japan. Correspondence should be addressed to F.Z. (frank.zufall@uks.eu) or T.B. (boehm@immunbio.mpg.de).

Received 31 August; accepted 21 October; published online 22 November 2009; doi:10.1038/nn.2452

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1551

a r t ic l e s
Figure 1 V2r1b-expressing VSNs have transient Ca2+ responses to MHC peptide ligands. (a) Confocal imaging in a coronal VNO slice identified a V2r1b-GFP VSN (arrow, green cell body). Scale bar indicates 10m. (b) Ca2+ fluorescence at rest (red) observed after loading the VNO epithelium with fura-red/AM. (c) Merged image shows that the V2r1b-GFP cell depicted in a took up the Ca2+ dye (yellow). (d) Superposition of a spatial V2r1b expression map (green) and a reconstructed F/F Ca2+ response map (red). Peptide-responsive V2r1b-GFP cells appear in yellow. Two V2r1b-GFP cells were seen in the imaged region (cells #34 and #37). Cell #34 responded to the AAPDNRETF peptide at 1011 M, whereas cell #37 did not. A V2r1bnegative neuron (#N95) also recognized AAPDNRETF. Cell numbers refer to corresponding lists shown in Supplementary Figures 1 and 2. Scale bar represents 10 m. (e) Ca2+ responses from the three cells that are depicted in d. Note that the two V2r1b-GFP cells differed in their response profiles. Although V2r1b-GFP cell #34 detected both the AAPDNRETF and SYFPEITHI peptides (1011 M), as well as diluted urine stimuli of the relevant haplotypes, V2r1b-GFP cell #37 recognized only SYFPEITHI and BALB/c urine. DU-C57BL/6, diluted urine of C57BL/6 mice (1:100); DU-BALB/c, diluted urine of BALB/c mice (1:100). Previous experiments established that a mutant version of AAPDNRETF (scrambled sequence) does not elicit a sensory response in the VNO7.

a
V2r1b-GFP

b
Fura-red

c
Merged

d
V2r1b-GFP AAPDNRETF Merged N95

e
AAPDNRETF DU-C57BL/6 SYFPEITHI DU-BALB/c N95 34
37

34

37 10% 20 s

2009 Nature America, Inc. All rights reserved.

and analyzed for their degree of activation in response to peptide or urine stimulation. This assay enabled us to determine whether a given VSN responded to multiple, structurally distinct peptides. In wild-type mice, ~1% of all VSNs (~3,000 cells per mouse, see Online Methods) respond to a given MHC peptide7. We employed the V2r1bIRES-tauGFP (of MHC haplotype b) mouse strain to identify, in live tissue slices, the population of VSNs that express V2r1b (refs. 19,22). These cells, henceforward referred to as V2r1b-GFP VSNs, express tauGFP owing to the insertion, via homologous recombination in embryonic stem (ES) cells, of an IRES-tauGFP cassette immediately downstream of the stop codon of the V2r1b gene. The gene targeting design did not disrupt the coding region of V2r1b and allowed for co-translation of V2r1b polypeptides and tauGFP. V2r1b-GFP cells constitute ~0.1% of all VSNs (see below). Their cell bodies reside in the H2-Mvnegative compartment of the basal layer of the VNO epithelium23. They project their axons to the anterior part of the caudal accessory olfactory bulb, where the axons terminate in six to ten glomeruli19. The V2r1b gene exhibits monoallelic expression19; we used homozygous mice, enabling us to identify all of the ~400 VSNs in a mouse that expressed V2r1b. We found that 79% of V2r1b-GFP VSNs (56 of 71) recognized at least one peptide from a panel of eight nonamers (Fig. 1, see Supplementary Fig. 1 for complete dataset). For comparison, we analyzed Ca2+ responses in V2r1b-negative VSNs located in the same VNO slices in parallel (n = 132; see Supplementary Fig. 2 for complete dataset of peptide-responsive, V2r1b-negative VSNs). In a first experiment, we used two peptide ligands that are characteristic of two distinct MHC molecules: AAPDNRETF, an H2-Db ligand, and SYFPEITHI, an H2-Kd ligand. These MHC molecules are encoded by the MHC haplotypes b and d, which are carried by, for instance, C57BL/6 and BALB/c mice, respectively. We identified V2r1b-expressing VSNs (Fig. 1ac) that responded to either or both peptides (Fig. 1d,e); 20 of 57 (35%) cells tested with AAPDNRETF responded, 33 of 60 (55%) cells tested with SYFPEITHI responded, and 4 of 49 (8%) cells tested with both nonamers responded (Supplementary Fig. 1). Thus, similar to randomly chosen VSNs7, V2r1b-expressing VSNs comprise two subsets that are roughly equal in size and largely nonoverlapping in their responses to two peptides. We also observed activation of these neurons by urine from mouse strains with the cognate MHC haplotype; cells that responded to AAPDNRETF also responded to diluted urine from C57BL/6 mice and cells that responded to SYFPEITHI also responded to diluted urine from BALB/c mice (12 of 12, 100%; Fig. 1e and Supplementary Fig. 1).
1552

We thus made four basic observations. First, a given MHC peptide was able to activate V2r1b-positive and V2r1b-negative VSNs. Second, a given V2r1b-positive VSN could be activated by several MHC peptides, including those that were associated with a different haplotype. Third, V2r1b-positive VSNs differed in their receptive properties for MHC peptides and constituted a heterogeneous population. Finally, in contrast with a response rate of ~1% for randomly selected VSNs, the proportion of V2r1b-positive cells that were responsive to a given peptide was ~50-fold higher; that is, 50% for some of the peptides of the panel. This difference suggests that expression of V2r1b is preferentially associated with the peptide-responsive subset of VSNs. To estimate the fraction of peptide-responsive cells that expressed the V2r1b gene, we determined that the total number of VSNs per mouse was ~300,000 (see Online Methods). The number of V2r1b-GFP VSNs per homozygous mouse was ~400. Given that a single, selected MHC peptide activates ~1% of all VSNs (~3,000)7, and that about 50% of all V2r1b-GFP VSNs (~200) responded to a given peptide (Fig. 1 and Supplementary Fig. 1), it follows that ~7% (200 of 3,000) of VSNs that were activated by a single MHC peptide expressed the V2r1b gene. Conversely, these calculations indicate that responsiveness of the VSN population to MHC peptides does not depend on V2r1b expression in absolute terms. Specificity of VSN peptide recognition We next compared the activation of individual VSNs by successive exposure to selected, structurally different peptides: SYFPEITHI (H2-Kd ligand), AAPDNRETF (H2-Db ligand) and AYKDNRETI (anchor residues known to be involved in binding to MHC molecules are indicated in bold). The AYKDNRETI sequence is designed to be a chimaeric representation of SYFPEITHI and AAPDNRETF, with the anchor residues at positions 2 and 9 from SYFPEITHI and the anchor residue at position 5 from AAPDNRETF. In this experiment, 13 V2r1b-GFP and 39 V2r1b-negative VSNs responded to at least one of these three peptides (Fig. 2a,b and Supplementary Figs. 1 and 2). Seven response patterns (denoted as IVII) are theoretically possible, reflecting the activation of a VSN by one, two or all three peptides. We observed six of these seven patterns in the 13 V2r1b-GFP cells (Fig. 2) and all seven in the 39 V2r1b-negative cells (Supplementary Fig. 3). These seven response patterns provide information about the identity of the amino-acid residues that determine specificity to the responses. Cells responding only to the chimaeric peptide AYKDNRETI (response pattern I; Fig. 2a,b) were likely to recognize the lysine residue at position three (Fig. 2c,d), as the three peptides
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 2 V2r1b-expressing VSNs have V2r1b-GFP neurons heterogeneous response profiles as a population. 20 25 30 (a) Ca2+ response of a single cell (#20, see b) 5% 50 s to three distinct MHC peptides. This pattern was also observed in 11 of 39 peptide-responsive Response No response V2r1b-negative cells (Supplementary Fig. 3). Arrows indicate the time points at which stimulus application was turned on. (b) Summary Pattern III: Pattern I: Pattern II: of response profiles of V2r1b-positive cells stimulated successively with three different IV (n = 1) Pattern VII: I (n = 1) II (n = 1) peptides. Stimuli were delivered in random order. Cell numbers correspond to the list VII (n = 0) Pattern V: Pattern VI: Pattern IV: V (n = 7) VI (n = 1) shown in Supplementary Figure 1. (c,d) Analysis of the results shown in b identified amino-acid residues (red) that were likely to determine the specificity of III (n = 2) 019% 2049% peptide responses. In principle, the distribution 50100% of V2r1b-positive cells responding to these peptides can be described by at least seven possible response patterns (denoted I to VII) exhibiting overlapping peptide specificities; only six patterns were observed in V2r1b-positive cells, whereas all seven patterns were seen in the analysis of V2r1b-negative cells (Supplementary Figs. 1 and 2). The number of cells (n) is indicated in the Venn diagram. The gray scale in c indicates the probability that a given pattern occurred. Note that for patterns IV, single specificities can be defined, although patterns IV and V could also be explained by coexpression of two specificities (shown in light gray). Patterns VI and VII required coexpression of two specificities (light gray); for pattern VII, two possibilities are indicated. Peptides were delivered at concentrations of 10 11 M.

2009 Nature America, Inc. All rights reserved.

differed from each other only at position 3, whereas two of three peptides shared the same residue at all other positions (Fig. 2c,d). Thus, for pattern I, a single amino-acid residue may determine response specificity. Applying the same reasoning, pattern II can be explained by residues 3 and/or 9 of AAPDNRETF, as they are the only unique residues among the three peptides. Furthermore, the specificity of cells responding uniquely to SYFPEITHI (pattern III) cannot be determined precisely from this comparison, because only positions 2 and 9 are shared with at least one other peptide. Hence, patterns IIII can be explained by three specificities expressed in distinct nonoverlapping groups of V2r1b-positive and V2r1b-negative VSNs. Pattern IV can be explained in at least two ways. Such VSNs could either coexpress the specificities found in patterns I and II, or, alternatively, could express another type of specificity, recognizing one or more

a
40 s 5%

64

68 Response No response Not tested

c
(n = 18)

residues at positions 1 and 48 of the AYKDNRETI and AAPDNRETF peptides. Coactivation by AYKDNRETI and SYFPEITHI (but not by AAPDNRETF, pattern V) can be explained by a specificity for one or both of the common residues at positions 2 and 9, tyrosine and isoleucine, respectively. Pattern VI can be explained by coexpression of at least two specificities. VSNs responding to AAPDNRTEF and SYFPEITHI, but not the chimaeric peptide AYKDNRETI, must recognize the AAPDNRTEF peptide through the proline residue at position 3 and/or the phenylalanine residue at position 9, where this peptide differs from the chimaeric peptide; the recognition of the SYFPEITHI peptide could formally occur through one or more residues at positions 1 and 38, which are unique among the three peptides. Because the critical residues in AAPDNRETF and SYFPEITHI (shown in bold) are not identical, coexpression of two specificities is required to explain this reaction pattern. Furthermore, comparison of the amino-acid residues in the SYFPEITHI peptide that determine the specificity of responses in patterns V and VI indicates that SYFPEITHI could be recognized in two ways. In pattern V, specificity seems to depend on residues at positions 2 and/or 9, whereas these residues can be excluded in pattern VI. Finally, pattern VII, which we saw only in V2r1b-negative VSNs (Supplementary Fig. 3), again suggests the coexpression of two peptide specificities.
Figure 3 Specificity of the peptide responses of V2r1b-expressing VSNs. (a) Example showing Ca2+ responses to the nonamer SYIPSAEKI and its three derivative tripeptides, SYI, PSA and EKI. Arrows indicate the time points at which stimulus application was turned on. ( b) Summary of response profiles of V2r1b-positive VSNs stimulated successively with the nonamer, the corresponding trimers and a mixture of free amino acids (S, Y, I, P, A, E and K). Corresponding cell numbers listed in Supplementary Figure 1 are indicated. Only cells that reacted to SYIPSAEKI were used for this analysis. Note that there was complete concordance of VSN activation by SYIPSAEKI and SYI. (c) Complete concordance of VSN activation by two H2-Kd ligands, among 18 V2r1b-positive VSNs tested. ( d) Response patterns observed after successive stimulation with two peptides. The specificity determining residues were deduced in consideration of the data in c and indicated for each pattern as in Figure 2. (e) Response patterns observed after successive stimulation with the three unrelated peptides, AAPDNRETF, SYFPEITHI and SFVDTRTLL, indicating that specificities are determined by amino acids at different positions. The amino-acid residues critical for the specificity of the response are depicted in red or white. Stimuli were delivered at 1011 M.

(n = 3)

(n = 7)

(n = 2) 019% 2049% 50100%

e
(n = 0) X (n = 4) (n = 0) (n = 1) (n = 1) (n = 0)

Pattern X:

(n = 0)

019% 2049% 50100%

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1553

a r t ic l e s
Figure 4 Dose dependency of peptide-induced Ca2+ responses imaged in individual V2r1b-expressing VSNs. (a) Ca2+ responses of a V2r1b-positive cell to increasing concentrations of AYKDNRETI peptide. Arrows in a and b indicate the time points at which stimulus application was turned on. (b) Example of a V2r1b-positive cell in which stimuli where delivered in random order. Repeated presentations indicate that Ca2+ peak responses were highly reproducible and did not change over time under these conditions. Because the results from repeated stimulations were identical, response heterogeneity is a feature that is intrinsic to each neuron. (ch) Normalized peak responses of V2r1b-positive cells to different peptides: AYKDNRETI (n = 4 cells), SYFPEITHI (n = 3 cells), AAPDNRETF (n = 4 cells), AAKDNRETA (n = 5 cells) and SYI (n = 4 cells). EC50 values and s.d. are shown in each panel. Each curve is from a different cell. (g) Comparison of dose-response curves to different peptides revealed that peptide sensitivity of V2r1b-positive VSNs depends on peptide sequence. Included are published data obtained with randomly chosen VSNs of wild-type mice using the Ca2+ indicator dye fluo-4 (dotted curves taken from ref. 7), showing that the high sensitivity of peptide responses was independent of the Ca2+ imaging method and the expression of GFP.

a
10

AYKDNRETI
15

10

14

10

13

10

12

10

11

M 40 s 5%

AAPDNRETF
10
12

10

14

10

11

310

11

10

12

10

14

10

13

10

11

M 40 s 5%

c
Normalized response 1.0 0.5 9.0 1.2 10
15

d
Normalized response 1.0 M 0.5 2.8 1.3 10
14

e
Normalized response 1.0 M 0.5 8.2 0.2 10
13

0 10
15

10

15

10 10 [AYKDNRETI] (M)

13

11

10

10 10 [SYFPEITHI] (M)

13

11

10

10

15

10 10 [AAPDNRETF] (M)

13

11

10

f
Normalized response 1.0 0.5 5.8 2.5 1012 M

g
Normalized response 1.0 0.5 5.7 1.7 10
15

h
Normalized response 1.0 AYKDNRETI M 0.5

Fura-red, V2r1b-GFP Fluo-4, VSNs AAKDNRETA AAPDNRETF

2009 Nature America, Inc. All rights reserved.

SYI

SYFPEITHI

0 15 10

10 10 [AAKDNRETA] (M)

13

11

10

0 15 10

10

13

10 [SYI] (M)

11

10

0 15 10

10 10 [Ligand] (M)

13

11

10

Positions of amino-acid residues conferring specificity Our analyses indicate that residues at several positions along the peptide sequence determine the various patterns of VSN activation. In most cases, more than one amino-acid residue appears to be involved in conferring specificity to VSN activation; however, pattern I (Fig. 2) could be explained by just one specificity-determining amino-acid residue. A single peptide can also be recognized on the basis of different amino-acid residues. To substantiate these findings, we performed two sets of additional experiments. First, we applied successive exposure to SYIPSAEKI (a ligand for H2-Kd, similar to SYFPEITHI) and three derivative trimers, SYI, PSA and EKI, respectively. VSNs that responded to SYIPSAEKI also responded to these trimers (n = 17; Fig. 3a,b and Supplementary Fig. 3), indicating that some short peptides contain sufficient information for sequence-specific activation. Some VSNs (3 of 15, 20%), however, were activated only by the nonamer and the SYI trimer, suggesting that the C-terminal isoleucine residue (occurring also in the EKI trimer) is not essential for this activation. Second, we examined activation patterns with a panel of up to five peptide ligands. Exposure to SYFPEITHI and SYIPSAEKI peptides (both ligands for H2-Kd) revealed that all of the 18 V2r1b-positive cells that we examined that responded to one peptide also responded to the other peptide (Fig. 3c). The residues that may confer specificity in this comparison are at positions 1, 2, 4 and/or 9. In contrast, activation by both SYFPEITHI and AYKDNRETI (the latter being a mimic of a H2-Kd ligand, possessing identical residues at positions 2 and 9) occurred only in 7 of 12 cells (P = 0.0056, Fisher exact probability test, two tailed; Fig. 3d). This dichotomy was also observed in V2r1b-negative cells (6 of 6 concordant responses for the first and 7 of 23 for the second case, P = 0.0036, Fisher exact probability test, two tailed; Supplementary Fig. 3). A specificity for residues occupying positions 2 and/or 9 accounts for the co-recognition of SYFPEITHI and SYIPSAEKI and of SYFPEITHI and AYKDNRETI (that is, the positions where all three peptides possess the same residues); note that this conclusion was also drawn for pattern V (Fig. 2). The second specificity for SYFPEITHI required
1554

one or more residues at positions 1 and/or 38; however, again considering the concordance of responses to SYFPEITHI and SYIPSAEKI, the specificity probably lies in residues 1 and/or 4 and then accounts for concordant responses to SYFPEITHI and SYIPSAEKI, while discriminating against AYKDNRETI (Fig. 3d). In a comparison of SYFPEITHI (a ligand for H2-Kd), SFVDTRTLL (also a known ligand of H2-Kd) and AAPDNRETF (a ligand of H2-Db), two patterns were particularly informative (Fig. 3e). VSNs coactivated by the first two, but not by the third peptide, could be explained by specificities focusing on positions 1 and/or 7; activation by the last two peptides, but not by the first peptide, can be explained by either specificity for positions 4 and/or 6 or at least two specificities for one of the unique residues in each peptide (Fig. 3e). Thus, VSNs can distinguish between different ligands of the same MHC molecule, attesting to their discriminatory power. Additional activation patterns resulting from the successive stimulation by up to five peptides support our conclusions (Supplementary Tables 13). These results indicate that V2r1b-positive and V2r1b-negative VSNs are heterogeneous in their responses to structurally distinct peptides. Because of this heterogeneity and the similarity of the patterns that we observed in V2r1b-positive and V2r1b-negative VSNs, V2r1b by itself may not be sufficient to explain the specificity of the peptide responses. Nonetheless, at least one pattern (pattern V; Fig. 2) was significantly more frequent in V2r1b-positive cells (P = 0.00036, Fisher exact probability test, two tailed), suggesting that expression of distinct specificities may not be random with respect to expression of V2r1b. Furthermore, a given VSN can detect multiple peptides; in some cases, responses might occur by recognizing the same residues and, in others, by coexpressing distinct specificities. V2r1b-GFP VSNs have exquisite peptide sensitivity We next determined the dose-response relationships of V2r1b-GFP VSNs (Fig. 4). First, we compared sensitivity to three related peptides: the chimeric AYKDNRETI peptide (Fig. 4a,c), the H2-Kd ligand SYFPEITHI (Fig. 4d) and the H2-Db ligand AAPDNRETF (Fig. 4b,e). Although stimulation with SYFPEITHI occurred with an EC50 value
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 5 Concentration independence of spatial 1011 M AAPDNRETF 109 M AAPDNRETF 105 M AAPDNRETF 1:100 DU-BALB/c peptide activation patterns. Reconstructed peptide activation patterns (red, F/F images) superimposed L L L L onto a V2r1b-GFP expression map (green) are shown. Scale bar in d, 10m. (ac) Successive stimulation using increasing concentrations of the AAPDNRTEF peptide (H-2b ligand), from 1011 M to 105 M, evoked an identical activation pattern V2r1b-tGFP consisting of one V2r1b-GFP cell (green/yellow F/F fluorescence, arrow) and three V2r1b-negative cells d (red fluorescence) in the imaged region. (d) Diluted BALB/c urine (H-2 ) stimulated a different V2r1b-GFP cell (arrowhead), together with other V2r1b-negative VSNs. This experiment is representative of the analysis of 27 V2r1b-positive cells in 28 VNO slices. As stimuli, we used SYI (10151012 M, n = 4 cells), AAPDNRETF (1014105 M, n = 7 cells), SYFPEITHI (1014105 M, n = 4 cells), KLYEQGSNK (1014109 M, n = 3 cells), AAKDNRETA (1014109 M, n = 5 cells) and AYKDNRETI (10151011 M, n = 4 cells). Analysis of an additional set of 39 V2r1b-negative VSNs in these slices produced the same results.

(half-maximal effective concentration) of ~3 1014 M, stimulation with AAPDNRETF yielded an EC50 value of 8 1013 M, indicating that the sensitivity of V2r1b-positive VSNs differs for structurally distinct peptides. The sensitivity of V2r1b-positive and V2r1b-negative cells is similar for these two peptides ( Fig. 4h). Activation with AYKDNRETI yielded an EC50 value of 9 1015 M, which is somewhat lower than that of SYFPEITHI, suggesting that residues other than those at positions 2 and 9 influence the responses to these peptides. To examine this question further, we mutated both H2-Kd anchor residues in the AYKDNRETI peptide (that is, tyrosine at position 2 and isoleucine at position 9) to alanine residues. The resulting peptide, AAKDNRETA, showed a strongly reduced activity, with an EC 50 value of 6 1012 M (Fig. 4f), which is 650-fold higher than that of the original peptide. This difference indicates that positions 2 and/or 9 are important, but not absolutely essential, for peptide activity. The EC 50 value of AAKDNRETA is sevenfold higher than that of the related AAPDNRETF peptide, indicating that, in the latter case, residues at positions 3 and/or 9 are important determinants of activity. Finally, the EC50 value for the trimer SYI (Fig. 4g) was 6 1015 M, in the range of nonamers sharing the tyrosine at position 2 and the isoleucine at the C-terminal position. Thus, the sensitivity of V2r1b-expressing VSNs depends on the peptide sequence, and a trimeric peptide is sufficient to evoke ultrasensitive responses in V2r1b-expressing VSNs. Spatial VNO activation patterns evoked by MHC peptides were invariant over a concentration range of several orders of magnitude, both for V2r1b-positive and V2r1b-negative VSNs (Fig. 5). In this example, stimulation with AAPDNRETF activated one of two V2r1b-positive neurons at 1011 M (Fig. 5a). The second V2r1b-positive cell was not activated by this peptide, even when the concentration was raised by six orders of magnitude to 105 M (Fig. 5ac), which is much higher than the saturation level of V2r1b-positive cells for this peptide (Fig. 4e). In contrast, this cell responded to diluted urine from BALB/c mice, as did two V2r1b-negative cells (Fig. 5d). Among the many V2r1b-negative VSNs that were located in the imaged region, three were activated by AAPDNRETF, and this activation pattern was also concentration invariant over an increase by six orders of magnitude (Fig. 5ac). Lack of peptide response in V2r1b-IRES-tauGFP cells The above experiments on V2r1b-GFP cells do not enable us to imply the involvement of V2r1b in the peptide responses. We therefore generated, via gene targeting, a mouse strain that carries a targeted disruption of the V2r1b gene. We replaced exon 6 of V2r1b, which encodes the seven putative transmembrane helices, by IRES-tauGFP (Fig. 6a,b). VSNs that chose the targeted allele, abbreviated V2r1bGFP, expressed tauGFP, but not the V2r1b protein. GFP-positive cells could be readily identified in tissue of these mice (Fig. 6c).
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

2009 Nature America, Inc. All rights reserved.

We prepared VNO tissue slices from homozygous mice and tested the responses of 34 V2r1b-GFP cells to one to four peptides. None of these cells responded to the peptides at concentrations as high as 107 M, whereas other, GFP-negative cells in the same region of the VNO epithelium were activated by these stimuli (Fig. 6dg). This complete lack of peptide responsiveness in V2r1b-GFP VSNs was highly significant (for example, 33 of 60 V2r1b-GFP cells were activated by SYFPEITHI, as compared with 0 of 34 V2r1b-GFP cells (P < 0.0001, 2 test). Likewise, no urine responses were observed in V2r1b-GFP VSNs (n = 7), although all these VSNs showed normal Ca2+ transients in response to high K+ solution (Fig. 6dg). To demonstrate that the lack of peptide responsiveness in V2r1bGFP VSNs does not reflect a loss of downstream signaling elements in the transduction pathway, we applied the diacylglycerol analog 1-stearoyl-2-arachidonoyl-sn-glycerol (SAG, 100M), a potent activator of TRPC2-dependent cation channels21,24. SAG treatment produced a Ca2+ rise in V2r1b-GFP VSNs that was indistinguishable in terms of magnitude and time course from SAG-evoked Ca2+ signals in both V2r1b-GFP and GFP-negative VSNs (Fig. 6f,g and Supplementary Fig. 4). Thus, an intact V2r1b receptor is required for responses to MHC peptides and urine in VSNs that transcribe the V2r1b locus. DISCUSSION This study provides a quantitative foundation for understanding coding and recognition of MHC peptides by VSNs. This information is important for gaining insight into the mechanisms that mediate the sensory evaluation of genotype25, which, in turn, may underlie a variety of processes as diverse as speciation and the regulation of immunological fitness of individuals. We drew four main conclusions regarding the structural characteristics of peptide activation by VSNs. First, a given VSN can detect a range of peptides, even those that are associated with different haplotypes. Despite a relatively broad response profile, the peptide responses are highly specific and maintain high sensitivity. For some peptides, we obtained EC50 values of 1014 M, placing these neurons among the most sensitive chemodetectors known to exist in mammals (Fig. 4). Given that the volume of the vomeronasal lumen is on the order of 0.5 mm3, a few thousand molecules may be sufficient to elicit a half-maximal response; as each VNO harbors ~1,500 VSNs that are responsive to a given peptide, a few molecules may thus be sufficient to activate a single VSN. Second, a given peptide activates VSNs with heterogeneous peptide specificities, indicating that these cells express different receptors or receptor combinations. A given peptide could be detected by both V2r1b-positive and V2r1b-negative VSNs, providing direct evidence for different peptide receptor expression in these neurons. In contrast
1555

a r t ic l e s a
Targeting vector Wild-type V2r1b E1 E2 E3 E4 E5 ES genome Mouse genome K
taugfp ACNF

RI

b
Removed

IRES loxP RI K E6 RI RI K K RI RI RI RI 1 kb RI 565 K

c
c1 c2

d
D5

SYFPEITHI

SEIDLILGY

AAKDNRETA DU-C57BL/6

KCI

c1 D5

c3

c2

c3

2009 Nature America, Inc. All rights reserved.

e
SYFPEITHI SFVDTRTLL SEIDLILGY AYKDNRETI AAPDNRETF AAKDNRETA DU-C57BL/6 KCI

D1

D5

D10

V2r1b-GFP neurons D15 D20

5% 25 s D25 D30

Response

No response

Not tested

f
20 (33) (4) (5) 10

V2r1b-GFP (9) (20)

(19)

g
20 (3)

V2r1b-GFP (3) (34)

F/F (%)

F/F (%)

(5)

10

Figure 6 Lack of peptide responses in V2r1bGFP VSNs. (a,b) Schematic descriptions of targeting vector (top, a) and germline configuration of the V2r1b locus (second line); the structure of the targeted locus before (third line) and after (bottom) removal of the neo selection cassette is shown. The design removed the last exon of V2r1b (exon 6, E6), which encodes the seven transmembrane helices (b), and replaced it with an IRES-tauGFP cassette. The other exons, E1E5, encode the N-terminal region of V2r1b. The blue bar between the EcoRI (RI) and KpnI (K) restriction sites indicates the position of the Southern blot probe used for screening of ES clones. (c) Confocal fluorescence image of fura-red loaded VSNs (red) superimposed onto a V2r1b expression map (green) identified V2r1b-GFP VSNs (exemplified by cell D5, yellow, see arrow) among other VSNs (exemplified by c13) in homozygous V2r1b-GFP mice. Scale bar represents 10m. L, lumen. (d) Time course of Ca2+ responses from the cells indicated in c. Peptide concentration was 109 M, urine was diluted at 1:100 (DU-C57BL/6) and the KCl concentration was 60 mM. (e) Summary chart of response patterns in V2r1b-GFP mice. (f,g) Average fluorescence peak responses in V2r1bGFP (f) and V2r1b-GFP VSNs (g) in response to stimulation with different peptides, dilute urine, the diacylglycerol analog SAG or KCl. The number of cells tested is shown in brackets above each bar with s.d. indicated. Note that although V2r1b-GFP VSNs were unresponsive to peptide and urine stimuli, they were viable, as shown by a normal uptake of Ca2+ dye and KCl-induced Ca2+ transients. They also showed normal Ca2+ responses to stimulation with SAG. Peptide concentrations were 1011107 M, urine was diluted at 1:100, and SAG and KCl concentrations were 100M and 60 mM, respectively.

(34) (10) (6)

(10) (16)

(7)

(7)

with what is known about OSNs that express a given endogenous odorant receptor2629 or VSNs that express a given endogenous V1r (ref. 30), we found that VSNs expressing a given endogenous V2r gene (V2r1b) were qualitatively heterogeneous in their response profiles. Although our experiments did not allow us to determine the precise number of distinct peptide reactivities among VSNs, this number may be high. Note that among the 13 VSNs expressing V2r1b that we examined (Fig. 2), six of seven theoretically possible patterns were observed. Third, the specificity of the response to a given peptide, as determined by the spatial activation pattern of VSNs in the VNO epithelium, was invariant over a vast range of ligand concentrations (Fig. 5). This concentration constancy ensures the stable representation of a given ligand in the ensemble of VSNs. Hence, the logic identified here differs from the combinational coding scheme employed by OSNs, which show less-specific tuning curves as ligand concentration is increased, resulting in the recruitment of additional OSNs at higher ligand concentrations31,32. Finally, our study establishes some of the sequence rules by which peptides are recognized by VSNs. Perhaps the most
1556

substantial finding is that, in contrast with the binding mode of MHC molecules, VSNs may be able to discriminate a single aminoacid among peptides ( Figs. 2 and 3). The peptide-recognition mode of VSNs appears to combine features of both MHC molecules and TCRs. In addition, our results offer new insights concerning the receptors underlying MHC peptide detection in the VNO. Because a given VSN can be activated by peptides that are ligands of distinct MHC molecules, it seems unlikely that classical MHC molecules themselves are directly involved in peptide recognition. Furthermore, H2-Mv genes, which encode nonclassical class I MHC molecules, can be ruled out because V2r1b-expressing VSNs do not express any of the nine known H2-Mv genes23, but they respond to MHC peptide ligands. The essential role of V2r1b in peptide responses is revealed by the absence of responses to six nonamers and to diluted urine in V2r1b-GFP neurons (Fig. 6). Our results assign for the first time, to the best of our knowledge, a specific function to a V2R and are consistent with the hypothesis that V2r gene products function as chemosensory receptors. The unexpected heterogeneity of peptide responses in V2r1b-expressing VSNs, together with the dependence of their responsiveness on the V2r1b protein, can be explained by coexpression of V2r1b with other V2rs, or with another group of molecules. In support of this hypothesis, we found that Vmn2r1 was
A U G ET SA KC D I

I ET R R N PD AA

TH

TH

ET

TL

TL

ET

LG

SA

EI

EI

TR

TR

LG

ET

ET AA KD

LI

LI

FP

FP

ID

ID

KD

VD

PD

KD

SY

VD

SY

SE

SE

AY

SF

AA

AA

AY

SF

KD

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s a
Frequency of anchor function (%) 100 80 60 40 20 0 1 2 3 4 5 6 7 8 9 Amino-acid position in ligand 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 MHC class I (n = 60)

b
Frequency of anchor function (%)

100 80 60 40 20 0

MHC class II (n = 24)

1 2 3 4 5 6 7 8 9 10 Amino-acid position in ligand 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Figure 7 Distribution of anchor residues in human MHC ligands. (a) Frequency of anchor function assigned to individual positions in MHC class I ligands (n = 60, ref. 12). (b) Frequency of anchor function assigned to individual positions in MHC class II ligands ( n = 24, ref. 12). (c) Position of amino-acid residues determining specificity of VSN responses (see Figs. 2, 3 and Supplementary Fig. 3) mirror the position of anchor residues in MHC class I or class II ligands.

2009 Nature America, Inc. All rights reserved.

coexpressed in 80% of V2r1b-positive VSNs and 80% of V2r1b-GFP cells (T.I. and P.M., unpublished observations). These observations also imply that Vmn2r1 itself cannot confer responsiveness to the peptides tested here. Collectively, our results argue against a one cell one receptorone ligand strategy33 in MHC peptide ligandsensing VSNs. A combinatorial assembly of V2R proteins in hetero-oligomeric complexes, perhaps akin to T1R taste receptors34, may underlie the heterogeneous response profiles observed here. The discovery of the nonclassical function of MHC peptides as sensory cues for VSNs and OSNs gave rise to an explanation of the role of MHC molecules in guiding behavioral decisions in a variety of contexts and species35. Given this functional convergence, it is interesting to compare the rules that govern the interaction of peptide ligands with MHC molecules and VSN peptide receptors. Extensive work has identified the structure of different MHC molecules and their binding preferences for peptide ligands12. A typical MHC class I peptide is a nonamer, and side chains of two amino-acid residues, known as anchors and which are typically located at positions 2 and 9, strongly interact with an MHC molecule (Fig. 7a). Therefore, the chemical nature of the anchor residues, collectively referred to as a motif, most clearly defines the identity of the cognate ligand. MHC class II peptide ligands are usually longer than MHC class I peptides, but the binding cleft also harbors a central stretch of nine amino acids. Of those, three to four are anchored in the binding pockets of the MHC molecules, most often at positions 1, 4, 6 and 9 (Fig. 7b). Although the ragged ends of MHC class II ligands are prone to exoproteolytic degradation, the core sequence is protected from degradation by virtue of its interaction with MHC class II molecules36. Hence, it is conceivable that the peptides that are eventually liberated from both MHC class I and class II molecules all finally become the same length, nine amino acids. Discriminatory amino-acid residues occur at several positions. In some instances, positions 2 and/or 9, 3 and/or 9, or 3 alone are relevant for specificity; in other cases, specific residues are located at positions 1 and/or 4, 1 and/or 7, or 4 and/or 6 (Fig. 7c). Variability at positions 2 and 3 of peptides is a distinguishing feature of MHC class I ligands, whereas variability at positions 1, 4, 6 and 7 is found in MHC class II ligands (Fig. 7a,b). Hence, VSNs are likely to be capable of distinguishing peptide sequences representing either MHC class I or class II ligands. This capacity would be consistent with some studies implicating genetic diversity at MHC class II, rather than MHC class I, loci in determining the outcome of MHC-dependent behavioral procedures37.
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

Our results have evolutionary implications. V2r genes are known to be a unique feature of jawed vertebrates, as they are not found in the genome of jawless fishes and other chordates18. Hence, V2r genes and the genes encoding MHC molecules and TCRs have all arisen in basal vertebrates 500 million years ago24, although the temporal order of their evolutionary emergence is not known. This timing suggests the possibility of co-evolution between peptide recognition by MHC molecules, TCRs and the receptors expressed on VSNs. If so, functional similarities between the sensory modalities employed by T cells and VSNs can be expected. T-cell recognition is distinguished by an evolutionarily selected affinity of invariant regions of the TCRs for conserved structures of MHC molecules38; there is also a conserved mode of interaction between TCRs and peptide/MHC complexes39. Similarly, VSN peptide receptors appear to be tuned to specific properties of MHC class I and II ligands defined by the chemical nature of amino acids at certain positions (anchor sites) in the peptide. Several other remarkable features are shared by these functionally diverse peptide recognition systems. First, both are exquisitely sensitive. A few peptide molecules may be sufficient to activate a given VSN; similarly, T cells can be activated by a single agonist peptide/MHC complex on the surface of an antigen-presenting cell40. Second, the sensitivity profile of VSNs and T cells40 to specific ligands is invariant over a wide dynamic range. This invariance ensures faithful representation of the chemical nature of a given peptide in the nervous system and a precisely tailored immune response to cells presenting a specific peptide/MHC complex. Third, in some instances, the activation of VSNs appears to depend on a single amino-acid residue in the peptide; likewise, TCRs can distinguish single amino-acid residues in peptides bound to MHC molecules41. Fourth, at least some VSNs appear to express more than one type of specificity for MHC peptides; likewise, T cells can express two distinct TCRs on their surface owing to incomplete allelic exclusion at the Tcra locus42. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank M. Konzmann for excellent technical assistance, R. Escher for peptide synthesis, K. Del Punta for a backbone plasmid of the gene targeting vector, W. Tang for blastocyst injections of ES cells and B. Bufe for comments and discussion. This work was supported by the Volkswagen Foundation (T.L.-Z.), Deutsche Forschungsgemeinschaft grants Sonderforschungsbereich 530 (F.Z.) and Schwerpunktprogramm 1392 (F.Z. and P.M.), the Max Planck Society (T.I., P.M. and T.B.), and the US National Institutes of Health (F.Z. and P.M.). AUTHOR CONTRIBUTIONS T.L.-Z., T.I., P.M., F.Z. and T.B. designed the study. T.L.-Z. and T.I. carried out the experiments. T.L.-Z., T.I., P.M., F.Z. and T.B. analyzed the data. T.L.-Z., T.I., P.M., F.Z. and T.B. contributed reagents and analytic tools. T.L.-Z., P.M., F.Z. and T.B. wrote the paper.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.

1. Davis, M.M. et al. T cells as a self-referential, sensory organ. Annu. Rev. Immunol. 25, 681695 (2007). 2. Cooper, M.D. & Alder, M.N. The evolution of adaptive immune systems. Cell 124, 815822 (2006). 3. Davis, M.M. & Bjorkman, P.J. T-cell antigen receptor genes and T-cell recognition. Nature 334, 395402 (1988). 4. Eason, D.D. et al. Mechanisms of antigen receptor evolution. Semin. Immunol. 16, 215226 (2004).

1557

a r t ic l e s
5. Huh, G.S. et al. Functional requirement for class I MHC in CNS development and plasticity. Science 290, 21552159 (2000). 6. McConnell, M.J., Huang, Y.H., Datwani, A. & Shatz, C.J. H2-Kb and H2-Db regulate cerebellar long-term depression and limit motor learning. Proc. Natl. Acad. Sci. USA 106, 67846789 (2009). 7. Leinders-Zufall, T. et al. MHC class I peptides as chemosensory signals in the vomeronasal organ. Science 306, 10331037 (2004). 8. He, J., Ma, L., Kim, S., Nakai, J. & Yu, C.R. Encoding gender and individual information in the mouse vomeronasal organ. Science 320, 535538 (2008). 9. Spehr, M. et al. Essential role of the main olfactory system in social recognition of major histocompatibility complex peptide ligands. J. Neurosci. 26, 19611970 (2006). 10. Boehm, T. Quality control in self/nonself discrimination. Cell 125, 845858 (2006). 11. Brennan, P.A. & Zufall, F. Pheromonal communication in vertebrates. Nature 444, 308315 (2006). 12. Rammensee, H.G., Bachmann, J. & Stefanovic, S. MHC Ligands and Peptide Motifs, (Landes Bioscience, Georgetown, TX, 1997). 13. Herrada, G. & Dulac, C. A novel family of putative pheromone receptors in mammals with a topographically organized and sexually dimorphic distribution. Cell 90, 763773 (1997). 14. Matsunami, H. & Buck, L.B. A multigene family encoding a diverse array of putative pheromone receptors in mammals. Cell 90, 775784 (1997). 15. Ryba, N.J. & Tirindelli, R. A new multigene family of putative pheromone receptors. Neuron 19, 371379 (1997). 16. Young, J.M. & Trask, B.J. V2R gene families degenerated in primates, dog and cow, but expanded in opossum. Trends Genet. 23, 212215 (2007). 17. Silvotti, L., Moiani, A., Gatti, R. & Tirindelli, R. Combinatorial co-expression of pheromone receptors, V2Rs. J. Neurochem. 103, 17531763 (2007). 18. Grus, W.E. & Zhang, J. Origin of the genetic components of the vomeronasal system in the common ancestor of all extant vertebrates. Mol. Biol. Evol. 26, 407419 (2009). 19. Del Punta, K., Puche, A., Adams, N.C., Rodriguez, I. & Mombaerts, P. A divergent pattern of sensory axonal projections is rendered convergent by second-order neurons in the accessory olfactory bulb. Neuron 35, 10571066 (2002). 20. Leinders-Zufall, T. et al. Ultrasensitive pheromone detection by mammalian vomeronasal neurons. Nature 405, 792796 (2000). 21. Spehr, J. et al. Ca2+-calmodulin feedback mediates sensory adaptation and inhibits pheromone-sensitive ion channels in the vomeronasal organ. J. Neurosci. 29, 21252135 (2009). 22. Ukhanov, K., Leinders-Zufall, T. & Zufall, F. Patch-clamp analysis of gene-targeted vomeronasal neurons expressing a defined V1r or V2r receptor: ionic mechanisms underlying persistent firing. J. Neurophysiol. 98, 23572369 (2007). 23. Ishii, T. & Mombaerts, P. Expression of nonclassical class I major histocompatibility genes defines a tripartite organization of the mouse vomeronasal system. J. Neurosci. 28, 23322341 (2008). 24. Lucas, P., Ukhanov, K., Leinders-Zufall, T. & Zufall, F. A diacylglycerol-gated cation channel in vomeronasal neuron dendrites is impaired in TRPC2 mutant mice: Mechanism of pheromone transduction. Neuron 40, 551561 (2003). 25. Boehm, T. & Zufall, F. MHC peptides and the sensory evaluation of genotype. Trends Neurosci. 29, 100107 (2006). 26. Bozza, T., Feinstein, P., Zheng, C. & Mombaerts, P. Odorant receptor expression defines functional units in the mouse olfactory system. J. Neurosci. 22, 30333043 (2002). 27. Feinstein, P., Bozza, T., Rodriguez, I., Vassalli, A. & Mombaerts, P. Axon guidance of mouse olfactory sensory neurons by odorant receptors and the beta2 adrenergic receptor. Cell 117, 833846 (2004). 28. Grosmaitre, X., Vassalli, A., Mombaerts, P., Shepherd, G.M. & Ma, M. Odorant responses of olfactory sensory neurons expressing the odorant receptor MOR23: a patch clamp analysis in gene-targeted mice. Proc. Natl. Acad. Sci. USA 103, 19701975 (2006). 29. Mombaerts, P. Genes and ligands for odorant, vomeronasal and taste receptors. Nat. Rev. Neurosci. 5, 263278 (2004). 30. Boschat, C. et al. Pheromone detection mediated by a V1r vomeronasal receptor. Nat. Neurosci. 5, 12611262 (2002). 31. Malnic, B., Hirono, J., Sato, T. & Buck, L.B. Combinatorial receptor codes for odors. Cell 96, 713723 (1999). 32. Firestein, S. How the olfactory system makes sense of scents. Nature 413, 211218 (2001). 33. Luo, M. & Katz, L.C. Encoding pheromonal signals in the mammalian vomeronasal system. Curr. Opin. Neurobiol. 14, 428434 (2004). 34. Chandrashekar, J., Hoon, M.A., Ryba, N.J. & Zuker, C.S. The receptors and cells for mammalian taste. Nature 444, 288294 (2006). 35. Slev, P.R., Nelson, A.C. & Potts, W.K. Sensory neurons with MHC-like peptide binding properties: disease consequences. Curr. Opin. Immunol. 18, 608616 (2006). 36. Mouritsen, S., Meldal, M., Werdelin, O., Hansen, A.S. & Buus, S. MHC molecules protect T cell epitopes against proteolytic destruction. J. Immunol. 149, 19871993 (1992). 37. Milinski, M. The major histocompatibility complex, sexual selection, and mate choice. Annu. Rev. Ecol. Evol. Syst. 37, 159186 (2006). 38. Marrack, P., Scott-Browne, J.P., Dai, S., Gapin, L. & Kappler, J.W. Evolutionarily conserved amino acids that control TCR-MHC interaction. Annu. Rev. Immunol. 26, 171203 (2008). 39. Garcia, K.C., Adams, J.J., Feng, D. & Ely, L.K. The molecular basis of TCR germline bias for MHC is surprisingly simple. Nat. Immunol. 10, 143147 (2009). 40. Krogsgaard, M., Juang, J. & Davis, M.M. A role for self in T-cell activation. Semin. Immunol. 19, 236244 (2007). 41. Wucherpfennig, K.W. The structural interactions between T cell receptors and MHCpeptide complexes place physical limits on self-nonself discrimination. Curr. Top. Microbiol. Immunol. 296, 1937 (2005). 42. Padovan, E. et al. Expression of two T cell receptor alpha chains: dual receptor T cells. Science 262, 422424 (1993).

2009 Nature America, Inc. All rights reserved.

1558

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

Generation of the V2r1b-GFP mouse. A stop codonPacI site was generated by recombinant PCR in the V2r1b-IRES-tauGFP targeting vector19 at the position corresponding to cysteine 566. The fragment between this new PacI site and the PacI site located immediately after the stop codon of V2r1b in the original targeting vector was removed and the IRES-tauGFP-ACNF cassette26 was inserted to produce the V2r1b-IRES-tauGFP-ACNF targeting vector. The vector was linearized with AscI and electroporated into embryonic day 14 ES cells as described previously43. G418-resistant clones were screened by Southern blot hybridization with a probe that was 5 external to the targeting vector. Homologous recombinant ES cells were injected into C57BL/6 blastocysts and chimeras were bred with wild-type C57BL/6 mice. The ACNF gene was automatically removed during transmission of the targeted allele through the male germline. The V2r1b-IREStauGFP strain is in a mixed (129 C57BL/6) background (MHC haplotype b). The strain has been deposited as cryopreserved sperm and will be publicly available from Jackson Laboratory as stock #6728. VNO preparation, solutions and stimuli. Imaging experiments were carried out in coronal VNO tissue slices freshly prepared from homozygous V2r1b-GFP or V2r1b-GFP mice (26 months old, either sex) using methods described previously7,20. Mice were killed using CO2 followed by decapitation. The two VNOs were quickly removed and submerged in oxygenated extracellular solution 1 (95% O2/5% CO2) containing 120 mM NaCl, 25 mM NaHCO3, 5 mM KCl, 5 mM N,N-Bis(2-hydroxyethyl)-2-aminoethanesulfonic acid, 1 mM MgSO4, 1 mM CaCl2 and 10 mM glucose (300 mOsm). The vomer bones were then removed, the VNO was embedded in 3% low gellingtemperature agarose, and coronal slices (250m) were cut on a vibratome (Microm HM 650 V). Slices were transferred to a submerged, chilled and oxygenated chamber until use. Unless otherwise stated, chemicals were obtained from Sigma. Peptides were designed according to known binding specificities of MHC molecules and were chemically synthesized, purified and verified by mass spectroscopy (matrix-assisted laser desorption/ionization time of flight). They were initially dissolved in phosphate-buffered saline at a concentration of 110 mM and stored frozen. Peptide preparations did not contain free amino acids. Immediately before use, peptides were diluted in HEPES-buffered extracellular solution 2, consisting of 145 mM NaCl, 5 mM KCl, 10 mM HEPES, 1 mM MgCl2 and 1 mM CaCl2, adjusted to pH 7.2 (NaOH) and 300 mOsm (glucose). Fresh urine was collected daily from male or female mice (C57BL/6 or BALB/c) by holding a mouse over a funnel and gently massaging the bladder area, taking specific care that no fecal contamination occurred. Ca2+ responses to urine (diluted 1:100) were observed in V2r1b-GFP VSNs with both male and female urine, irrespective of whether recordings were obtained from male or female animals (data not shown). Experiments were performed using a specified laminar flow microchamber (Warner Instruments). Stimuli were applied using separate reservoirs in combination with a dedicated valve system to ensure constant speed and height of the bath flow. Near-zero dead space manifolds minimized the delay on switching between different solutions (for further details see http://www. warneronline.com/Documents/uploader/perfusion_strategies.pdf). Peptide and urine stimuli each were applied for 10 s (ref. 7). The interstimulus interval was 4 min. Stock solutions of SAG were prepared in DMSO and aliquots were stored at 20 C. Final dilutions in solution 2 were made immediately before use as described previously21,24. Calcium imaging, calibration and data analyses. All of the GFP-expressing cells analyzed here exhibited an intact cytoarchitecture with a single dendrite projecting toward the lumen of the VNO ending in a distal dendritic knob and a thin axon projecting from the soma toward the basal lamina (Fig. 1a). Intracellular

ONLINE METHODS

2009 Nature America, Inc. All rights reserved.

Ca2+ was monitored with fura-red/AM (Invitrogen) using a confocal scanning microscope and dedicated laser and filter systems (Radiance 2100, Zeiss). Fura-red/AM was dissolved in a solution consisting of DMSO and 20% Pluronic F-127 (wt/vol), which was then added to oxygenated saline solution, sonicated briefly and added to the experimental chamber to give a final concentration of 30M fura-red/AM, 0.065% Pluronic F-127 and 0.33% DMSO (vol/vol). Slices were incubated in dye solution for 60 min at 37 C using an O2/CO2 incubator (Binder) and then perfused with dye-free oxygenated saline for 10 min before beginning the experiments. The extensive Stokes shift of fura-red permits multicolor analysis of red and green fluorescence in GFP-labeled neurons using a single excitation wavelength. Excitation illumination was 488 nm; emitted fluorescence was collected using a 522 DF 35-nm filter for GFP and a longpass filter for wavelengths greater than 600 nm for fura-red. All scanning head settings except gain (that is, black level, detector aperture and laser neutral density filter) were kept constant during each experiment. The gain was usually adjusted slightly, but only once at the beginning of an experiment during acquisition of baseline images. High-resolution images of baseline fluorescence were taken for subsequent comparison with time-series images and several images were acquired at the beginning of each time series before applying experimental solutions. Images were acquired at a rate of 0.5 Hz using LaserSharp2000 software (Zeiss) and analyzed using ImageJ (US National Institutes of Health) and Igor Pro (Wavemetrics). Ca2+ signals are presented as arbitrary fluorescence units or as absolute values (F/F) of the relative changes in fluorescence intensity (F) normalized to baseline fluorescence (F). This procedure results in a positive deflection when the intracellular Ca2+ concentration rises, despite the fact that fura-red fluorescence decreases on binding of Ca2+. A response was defined as a baseline deflection after the onset of stimulation that exceeded twice the s.d. of the mean of the baseline noise. No spontaneous Ca2+ waves were observed. Cell counting. Mice were anesthetized and killed by transcardial perfusion with 4% paraformaldehyde (vol/vol) in phosphate-buffered saline. After cryopreservation, 25-m coronal VNO cryostat sections were collected. Primary antibody to V2R-expressing VSNs (rabbit antibody to V2R2, 1:10,000) was kindly provided by R. Tirindelli (University of Parma). Bound antibody was detected with fluorescently labeled secondary antibody. Omission of either primary or secondary antibody abolished staining7. In five adult mice (three females, two males), we measured the areas of the entire VNO sensory epithelium and the fluorescently labeled V2R regions (five slices in each VNO). Corresponding slices were obtained from caudal to rostral at equal distances from each other. The area of the central section of the sensory epithelium was measured using ImageJ software and the number of VSNs found in this area was counted. In this way, the number of cells per area was obtained for each coronal slice, thus presenting an estimate of the total amount of VSNs in the sensory epithelium and an estimate for the amount of V2R-expressing VSNs in the antibody to V2R2labeled region. The estimated length of the VNOs was 2,308 212m (n = 5). The number of VSNs was determined to be 3.05 0.32 105 for the two VNOs of a mouse (n = 7), of which 1.52 0.21 105 VSNs expressed V2R-type receptors. These cells were typically located in the basal, Go-positive expression layer of the VNO epithelium. The number of V2r1b-GFP VSNs per mouse was determined to be 403 92 (n = 4). Thus, only a small fraction of V2R-positive VSNs (~0.3%) expressed V2r1b. No difference was found between male and female mice. Statistical analysis. Data were analyzed using NCSS 2004 statistical software (NCSS). Unless otherwise stated, results are presented as means s.d.
43. Mombaerts, P. et al. Visualizing an olfactory sensory map. Cell 87, 675686 (1996).

doi:10.1038/nn.2452

nature NEUROSCIENCE

a r t ic l e s

Dynamic DNA methylation programs persistent adverse effects of early-life stress


Chris Murgatroyd, Alexandre V Patchev, Yonghe Wu, Vincenzo Micale, Yvonne Bockmhl, Dieter Fischer, Florian Holsboer, Carsten T Wotjak, Osborne F X Almeida & Dietmar Spengler
Adverse early life events can induce long-lasting changes in physiology and behavior. We found that early-life stress (ELS) in mice caused enduring hypersecretion of corticosterone and alterations in passive stress coping and memory. This phenotype was accompanied by a persistent increase in arginine vasopressin (AVP) expression in neurons of the hypothalamic paraventricular nucleus and was reversed by an AVP receptor antagonist. Altered Avp expression was associated with sustained DNA hypomethylation of an important regulatory region that resisted age-related drifts in methylation and centered on those CpG residues that serve as DNA-binding sites for the methyl CpGbinding protein 2 (MeCP2). We found that neuronal activity controlled the ability of MeCP2 to regulate activity-dependent transcription of the Avp gene and induced epigenetic marking. Thus, ELS can dynamically control DNA methylation in postmitotic neurons to generate stable changes in Avp expression that trigger neuroendocrine and behavioral alterations that are frequent features in depression. Epigenetic regulation of gene expression allows the integration of intrinsic and environmental signals in the genome1. Greater emphasis is being placed on the role of epigenetic mechanisms in facilitating the adaptation of organisms to changing environments through alterations in gene expression. Evidence that dietary or pharmacological interventions have the potential to reverse environment-induced modification of epigenetic states25 has provided an additional impetus for understanding the epigenetic basis of disease, including disorders of the brain. It has been suggested that epigenetic mechanisms underlie brain plasticity, a process requiring stable modulation of gene expression6,7. DNA methylation is one of the most intensely studied epigenetic mechanisms8 and recent work has suggested that this form of gene regulation may determine risk for psychiatric disorders3,9,10. Exposure to stress during neurodevelopment has an effect on the quality of physical and mental health11,12. Periodic infant-mother separation during early postnatal life is one of the most commonly used procedures for inducing ELS in rodents. It is characterized by lifelong elevated glucocorticoid secretion, heightened endocrine responsiveness to subsequent stressors and disruption of the homeostatic mechanisms that regulate the activity of the hypothalamo-pituitaryadrenal (HPA) axis, all of which are considered to be pathogenetic factors in disorders of mood and cognition1315. Here we examined the coupling of experience-driven neuronal activity with DNA methylation and gene expression6. We focused on the expression of the two hypothalamic secretagogues that regulate HPA axis activity by increasing the synthesis and release of pituitary adrenocorticotropin, namely, AVP and corticotropin-releasing hormone (CRH). Abundant evidence links AVP and CRH to mood and cognitive behaviors16,17, making their receptors the targets of psychopharmacological agents18,19. In addition to being important in the postnatal development and functional maturation of the pituitary-adrenal axis, AVP potentiates the actions of CRH under circumstances that demand sustained activation of the pituitary and adrenal glands20. We found that ELS induces persistent hypomethylation of the Avp enhancer, accompanied by sustained upregulation of Avp expression, increased HPA axis activity and behavioral alterations. In the course of exploring the molecular mechanisms underlying these changes, we found that MeCP2 is important in the epigenetic programming of neuroendocrine and behavioral functions. RESULTS ELS-induced phenotypes Consistent with previous studies 15,21, ELS during the first 10 d of life led to sustained hyperactivity of the HPA axis, characterized by corticosterone hypersecretion under basal conditions, hyperresponsiveness to acute stressors applied later in life and escape from the inhibitory constraints of dexamethasone (Fig. 1a). Although it had no effect on body mass, ELS induced involution of the thymus, hypertrophy of the adrenals (Supplementary Fig. 1) and increased expression of pituitary pro-opiomelanocortin (Pomc) mRNA, which encodes the adrenocorticotropin pro-hormone ( Fig. 1b). Pomc expression is induced by the hypothalamic neuropeptides AVP and CRH, and all of them are under negative feedback control by the glucocorticoid receptor. Conspicuously, levels of Nr3c1 (the gene that encodes the glucocorticoid receptor) mRNA in the hippocampus, hypothalamic paraventricular nucleus (PVN) and pituitary were either unchanged or upregulated in ELS-treated mice (Supplementary Fig. 1), arguing against impaired corticosterone feedback as the primary cause of the observed increases in Pomc

2009 Nature America, Inc. All rights reserved.

Max Planck Institute of Psychiatry, Munich, Germany. Correspondence should be addressed to D.S. (spengler@mpipsykl.mpg.de). Received 2 June; accepted 22 September; published online 8 November 2009; corrected after print 3 December 2009; doi:10.1038/nn.2436

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1559

a r t ic l e s a
Serum corticosterone (ng ml )
250 200 150 100 50 0 Peak Stress Peak Stress Peak Stress Dex

b
* *
Pomc mRNA (relative expression)

Control 2.0 1.5 1.0 0.5 0

ELS

c
*
AVP mRNA (Ci g )
1

d
20 16 12 8 4 0 6 weeks 3 months 1 year PVN (parvocellular) 0 Naive

* * *

*
Latency (s)

300

***

***

*** ***

200

100

Naive

+ + SSR

+ + SSR

6 weeks

3 months

1 year

2009 Nature America, Inc. All rights reserved.

Figure 1 Endocrine and behavioral consequences of ELS depend on sustained AVP expression. ( a) ELS mice had higher serum corticosterone levels at 6 weeks, 3 months and 1 year of age under basal conditions, at the daily nocturnal peak and after exposure to an acute stressor applied 30 min before sampling. An injection of dexamethasone (dex, 6 h before sampling) suppressed corticosterone secretion more effectively in 3-month-old control as compared with ELS mice. (b) Pomc mRNA (measured by qualitative PCR, qPCR) was significantly higher in the pituitaries of 3-month-old ELS mice. Treatment of 3-month-old control and ELS mice with the AVP V1b receptor antagonist SSR149415 (SSR) reduced Pomc mRNA levels (measured by qPCR) in ELS mice. (c) Avp mRNA expression (detected by ISH) was significantly higher in parvocellular PVN neurons of 6-week-old (+31%), 3-monthold (+13%) and 1-year-old (+15%) ELS mice. (d) ELS mice showed shorter step-down latencies 24 h after learning but not during training (control (19 2 s) versus ELS (21 2 s), P > 0.05). Treatment with the AVP V1b receptor antagonist partially reversed memory deficits by increasing step-down latencies from 47 3% to 78 7% of vehicle-treated controls, without affecting the behavioral performance of control mice (101 6%). Data are presented as mean s.e.m. (n = 816 mice per group). *P < 0.05, ** P < 0.01 and *** P < 0.001 (t test or ANOVA followed by Newman-Keuls post hoc test).

expression and glucocorticoid secretion. Although ELS did not influence hypothalamic Crh mRNA expression (Supplementary Fig. 1), the procedure resulted in a significant upregulation of Avp mRNA (P < 0.05; Fig. 1c). The changes in Avp expression persisted for at least 1 year and were restricted to the parvocellular subpopulation of neurons in the PVN, that is, in those neurons that drive the pituitary-adrenal axis (Supplementary Fig. 2). AVP exerted its regulatory role on Pomc expression levels via activation of pituitary AVP V1b receptors. Application of SSR149415, a selective V1b receptor antagonist, normalized the elevated Pomc mRNA levels (Fig. 1b) and corticosterone secretion (data not shown), verifying the critical role of AVP in driving the disturbed endocrine phenotype in ELS mice. ELS also produced long-lasting behavioral changes. Adult ELSexposed mice showed memory deficits in an inhibitory avoidance task (Fig. 1d). In addition, they had increased immobility in the forced swim test ( Supplementary Fig. 3 ). In contrast, anxietylike behavior was unaffected by ELS in the elevated plus-maze, novelty-induced hypophagia and light-dark avoidance tests (data not shown). The ELS-induced behavioral phenotypes were reproduced in two further independent replications. Treatment with the SSR149415 partially reversed the impaired memory in ELS mice (Fig. 1d) and abolished the changes in behavioral stress coping (Supplementary Fig. 3) without influencing the behavioral performance of control mice. Differential methylation of the Avp gene Methylation of cytosine residues in CpG dinucleotides can result in epigenetic gene silencing; such CpGs are conspicuously under-represented in mammalian genomes and typically cluster in GC-rich regions called CpG islands (CGIs)22. Computa-tional analysis and a recent genome-wide classification of promoter CGIs23 predicted 4 CGIs in Avp: CGI1 (intermediate CpG frequency in the promoter region), CGI2 (high CpG frequency covering the second and third exons), CGI3 and CGI4 (intermediate CpG frequency in the ~3.6-kb downstream region) (Fig. 2a and Supplementary Fig. 4). The latter region, also referred to as the intergenic region (IGR), separates the neighboring, tail-to-tailorientated Avp and oxytocin genes and includes a composite enhancer region in the first 2.1 kb proximal to Avp that is important for expression24.
1560

Sequence analysis of bisulfite-converted DNA isolated from the PVN of naive C57BL/6N mice showed sparse methylation in the promoter CGI1 and exonic CGI2. In contrast, we found high levels of CpG methylation clustered at the more distal enhancer encompassing CpG7 to CpG32 and spanning CGI3 (Fig. 2b). The latter region is highly conserved between species and is important for Avp regulation24. Less-dense methylation was observed in CGI4 and the adjacent oxytocin tissue-specific enhancer region had only a few, irregularly spaced and highly methylated CpG residues (Fig. 2b). A similar methyl ation of CpG residues at the Avp locus was found in unrelated CD1 mice, supporting the idea that this pattern is unlikely to be strain specific (data not shown). Together, these results support the idea that CGIs in intergenic regions are more likely to be methylated than those at gene promoters and that CGIs with intermediate CpG densities are methylated more frequently25. Persistent hypomethylation of CGI3 after ELS We compared PVN tissue from ELS and control mice aged 6 weeks, 3 months and 1 year and found hypomethylation of multiple CpG residues throughout the downstream Avp enhancer region in ELS mice (Fig. 3ac). Analysis of overall methylation of the enhancer revealed substantial reductions in methylation in ELS mice of all ages (Fig.3d). Significantly marked (P < 0.05) CpG residues largely mapped to CGI3 of the enhancer. For many of these, the degree of ELS-induced hypomethylation was consistently greater than that observed for overall CGI3 hypomethylation; for example, CpG10 showed uniformly strong reductions in methylation (by 37% at 6 weeks, (P < 0.005), 21% at 3 months (P < 0.05) and 66% at 1 year (P < 0.005)). This finding reveals that ELS triggers a heterogeneous response in CpG hypomethylation and indicates a functional role for marked changes. To obtain a functional measure of those CpG residues that are likely to control Avp expression, we sought correlations between Avp mRNA levels and the degree of methylation of individual CpG residues that were significantly hypomethylated (P < 0.05) in ELS mice. Therefore, in situ hybridization (ISH) and DNA methylation analyses were performed on tissues from the same individual mice. Of the 11 CpGs that were significantly hypomethylated (P < 0.05) in 6-week-old ELS mice (Fig. 3a), only seven (CpGs 7, 10, 12, 13, 14, 15 and 17) had methylation patterns that were strongly correlated with Avp mRNA levels (Fig.3e). For example,
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 2 Selective methylation of the intergenic region of the Avp gene. (a) Schematic diagram of the Avp and oxytocin genes orientated tailto-tail and separated by the IGR. Exons are indicated by open (numbered) boxes and CGIs by numbered bars, and the distribution of CpG residues and size and position of the respective PCR amplicons used for their analyses are shown. CpG10 is marked with an asterisk. (b) CpG methylation profile in the IGR of 6-weekold naive mice. Residues belonging to the Avp-Oxt tissue-specific enhancer (Avp/Oxt TSE), CpG islands 3 or 4 (CGI3 or CGI4) or the Oxt tissue-specific enhancer (Oxt TSE) are color coded (inset). CGI3, lying ~0.5 kb downstream of the Avp gene, had the highest level of methylation. Data are presented as mean s.e.m. (n = 8 mice per group).

a
Exons CGI1 CpGs PCR amplicons 1

1 kb

Avp 2 CGI2 3
Avp/Oxt TSE

Oxt IGR CGI3 CGI4 Oxt TSE 3 2 1

b 100
Percentage methylation

80 60 40 20 0 1.0 1.5 kb 2.0 2.5

Avp/Oxt TSE CGI3 CGI4 Oxt TSE

2009 Nature America, Inc. All rights reserved.

0 0.5 although residues CpG10 and CpG22 had similar levels of methylation (methylation at CpG10 in controls and ELS was 60.3 5.9% and 37.6%, respectively; methylation at CpG22 in controls and ELS was 69.5 8.6% and 39 9.8%, respectively; Fig.3a), the methylation status of CpG10, but not of CpG22, correlated strongly with differences in Avp expression (CpG10, r2 = 0.44, P < 0.05; CpG22, r2 = 0.07, P > 0.1).

3.0

3.5

a
Percentage methylation

CGI3 100 80 60 40 20 0 10 20 30 40 CpGs in IGR (AVP OXT) Control ELS

CGI4

* * * * ** *

** *

Thus, ELS-induced alterations in CpG methylation appear to be important for Avp mRNA levels, although individual CpG residues located in CGI3 seem to contribute, in different degrees, to altered expression. Those CpGs that failed to show a priori significant differences (P > 0.05) in their methylation status in response to ELS correlated poorly with Avp mRNA levels (data not shown). Notably, differential methylation of CGI3 was not evident when DNA from the hypothalamic supraoptic nucleus of 6-week-old control and ELS mice were compared (Supplementary Fig. 5); the
50 60

Percentage methylation

100 80 60 40 20 0

* *

* * *

d
Percentage overall methylation

50 40 30 20 10 0

*** ***

*** *** **

Percentage methylation

10 20 CpGs in AVP enhancer

30

100 80 60 40 20 0

* * *
10 20 CpGs in AVP enhancer
7# 10 * 12 # 13 # 14 # 15 # 17 # 21 22 23 32 r2 0.28 0.44 0.21 0.22 0.24 0.17 0.31 0.07 0.07 0.08 0.08

6 weeks 3 months

1 year

30

e
Avp mRNA (Ci g1)

18 16 14 12 10 0

Avp mRNA (Ci g1)

f
6 week 3 month 1 year Sig. diff.
CpGs 7 10 11 12 13 14 15 17 20 21 22 23 31 32

g 16
14 12 10 0

10 ** 12 # 13 * 14 # 15 #

r2 0.42 0.30 0.34 0.32 0.26

>25% diff.

<25% diff.

25 50 75 100 CpG methylation (%)

25 50 75 100 CpG methylation (%)

CpG 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Control ELS Increased methylation

* *

*** *

**

Decreased methylation

**** *

Unchanged methylation

* **

Figure 3 ELS induces hypomethylation of the Avp enhancer. (ac) CpG methylation profiles of 6-week-old (a), 3-month-old (b) and 1-year-old (c) control and ELS mice. The entire IGR is shown in a, whereas b and c focus on the enhancer. (d) Overall methylation of the enhancer decreased significantly in ELS mice at all ages. Note the significant age-related hypomethylation in control, but not ELS, mice. (e) CpG methylation inversely correlated with Avp expression, revealed by correlating all significantly marked CpGs between 6-week-old control and ELS mice (shown in a) with respective Avp mRNA levels. (f) All CpG residues (shown in ac) that differed significantly in methylation at least once () and by more than 25% () at the other two ages were defined as methylation landmarks (boxed CpGs 10, 12, 13, 14 and 15). (g) The composite methylation status of the methylation landmarks, in particular CpG10, correlated negatively with Avp expression over all ages. (h) Differences in methylation (>10%) between 6-week-old and 1-year-old control and ELS mice for each CpG in the enhancer revealed significant changes in both control and ELS mice. Methylation landmarks were not influenced by age in either group. Data are presented as means s.e.m. (n = 810 mice per group). # P < 0.01, * P < 0.05, ** P < 0.005 and *** P < 0.0001 (t test, ANOVA followed by Newman-Keuls post hoc test or Pearsons correlation coefficient).

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1561

a r t ic l e s
PV N N 6 N 2 3T a 3

Relative difference

a
AVP GAPDH

b
Percentage methylation 173 bp 326 bp

80 60 40 20

Methylation AVP expression

8 6 4 2 0

c Avp
1 2

Luc
3

Enhancer E E 1025 enh

0. 0 05 0. 5 5

Percentage activity

80 60 40 20 0

Percentage activity

Percentage activity

100

0. 0 05 0. 5 5

0 5-aza (M) 100 80 60 40 20 0

Unmethylated Methylated

100 80 60 40 20 0

En

g
1 Rel. dif. MeCP2/IgG 3 2 1 0 1 CGI1

Vector

Site specific OXT

AVP 23 4 5 6 7

h
Rel. dif. MeCP2/IgG

4 3 2 1 0

32 24 16 8 0 MeCP2 Rpol

2009 Nature America, Inc. All rights reserved.

3 CGI2

5 CGI3

6 CGI4

ds1

atatggtttcagaataagcgctctaagtttaagaaatt

CpG 10 CpG 10 A/T mutant CpG12 CpG14

acctaaaggagacaacgccacctgcatt

ds1

m u

CpG 10m 10u 10m 12m 14m 10m

Acctgctggagacaacgccacctgcatt gtgttcaaactgccacgcccttactccc cacggaaatagacaagataggctggagt

k
Bound/free

0.1 0.08 0.06 0.04 0.02 0 0.02 0 0.1 0.2 0.3 0.4 0.5 Free probe (nM) 0.6 CpG 10 CpG10 A/T mutant CpG 12 CpG 14

KD(nM) 2.6 54 9.6 10.5

Percentage DB 100 18.2 MeCP2

100 6.9 1.7 27.6 37.9 10.3

MeCP2 GST MeCP2

Rel. dif. Rpol/IgG

Control 5-aza

48 40

Ab

Figure 4 Enhancer methylation represses Avp expression as a result of MeCP2 occupancy. (a) Reverse-transcription PCR (RT-PCR) analysis of Avp expression in mouse PVN, hypothalamic N6, neuroblastoma Neuro2a and fibroblast 3T3 cells. (b) Treatment of N6 cells with 5-azacytidine (5-aza, 5 d) prevents CpG methylation and induces Avp expression. (c) The parent Avp-Gaussia construct contains the entire 2.1-kb enhancer, 1025 is devoid of CpGs 10 to 25 and enh lacks the entire enhancer. (d) Deletion of either CpGs 1025 or the entire enhancer reduced reporter activity by 37% and 90%, respectively, in N6 cells. (e) Entire vector or site-specific enhancer (CpGs 125) methylation reduced reporter activity by 90% and 50%, respectively. (f) MeCP2 strongly repressed the site-specific methylated CGI3 vector; transfection with MBD1, MBD2 and MBD3 resulted in weaker repression. (g) ChIP analysis revealed that MeCP2 selectively occupied CGI3 at the Avp locus in N6 cells. (h) Treatment with 5-azacytidine (5 M, 5 d) relieved MeCP2 binding and enhanced activated RNA polymerase II (Rpol) occupancy at the promoter. ChIP data (g,h) are presented as means s.d. (four independent experiments). (i) The oligonucleotides used in EMSAs encoded CpG 10, 12 and 14, a mutant form of CpG10, and the high-affinity MeCP2binding site ds1. (j) MeCP2 bound strongly to methylated (m), but not to unmethylated (u), CpG10 and ds1. Compared with methylated CpG10, MeCP2 bound less to methylated CpG12 (12m) and CpG14 (14m). A representative autoradiogram and mean DNA-binding values (percentage DB, four independent experiments) are shown. (k) Dissociation constants (KD) for MeCP2 binding.

latter is consistent with the observation that Avp transcript levels in this nucleus were not influenced by ELS (Supplementary Fig. 2). Methylation landmarks correlate with Avp expression These data led us to hypothesize that ELS-induced changes in the methylation status of relevant CpG residues are persistent and sustain elevated Avp expression, whereas changes in functionally less-significant CpGs (P > 0.05) wane over time. To identify CpGs predictive of persistently increased Avp expression, we examined each CpG residue in detail. Residues considered to be of predictive value were those that were significantly methylation (P < 0.05) marked by ELS at one age at least and nominally altered by more than 25% at two other ages. By these criteria, CpG residues 10, 12, 13, 14 and 15 were revealed as methylation landmarks in the Avp enhancer (Fig. 3f). The number of CpG residues that were significantly marked by ELS decreased with age (11 in 6-week-old, 7 in 3-month-old and 3 in 1-year-old mice; Fig. 3f). With the exception of CpG10, which localized to the upstream boundary (Fig. 2a), all of the emerging methylation landmarks mapped to the center of CGI3 (CpGs 12, 13, 14 and 15). We corroborated the functional role of these residues by correlating their individual methylation status with Avp mRNA levels in control and ELS 6-week-old, 3-month-old and 1-year-old mice (Fig. 3g). This revealed that the composite methylation status of these residues
1562

faithfully reflected longitudinal Avp expression. Poor, if any, correlations were found between Avp mRNA expression (at any age) and those CpG residues that showed either initial (CpGs 7, 17, 21, 22, 23 and 32) or otherwise transient (CpGs 11, 20 and 31) differences in methylation status (data not shown). This set of findings highlights the functional importance of CpG residues 10, 12, 13, 14 and 15 in the regulation of Avp expression. Although significant hypomethylation ( P < 0.0001) of the Avp enhancer occurred with age in control mice ( Fig. 3d), we did not observe age-dependent changes in Avp mRNA levels (Fig. 1c). In contrast, ELS mice did not show age-related hypomethylation of the Avp enhancer (Fig. 3d), but nevertheless maintained higher levels of Avp mRNA, as compared to controls (Fig. 1c). This raised the question of whether single CpG residues might be differentially sensitive to age- versus ELS-induced hypomethylation. Analysis of the effects of aging on hypomethylation of all 32 CpGs of the Avp enhancer in 6-week-old and 1-year-old control and ELS mice showed that ageassociated hypomethylation only occurred in 16% of the CpGs in the Avp enhancer region of ELS mice, as compared with 38% in control mice (Fig. 3h). Notably, those CpG residues with an assigned regulatory role (methylation landmarks 10, 12, 14 and 15) did not show significant hypomethylation (P > 0.05) with aging (Fig. 3h). Enhancer methylation directs Avp expression An AVP-expressing N6 mouse hypothalamic cell line26 was used to examine whether Avp enhancer methylation modulates Avp expression
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

M BD 1 M BD 2 M BD 3 M eC P2

25 e nh

a r t ic l e s a
Percentage activity

120 100 80 60 40 20

b
Rel. dif. Flag/IgG

0 MeCP2 CaMKII*

+ +

0 MeCP2 S438A CaMKII*

+ +

+ +

AVP mRNA (relative expression)

2.0

d 10
Rel. dif. Rpol or MeCP2/IgG

1.5

8 6 4 2

1.0

0.5

2009 Nature America, Inc. All rights reserved.

0 MeCP2 S438A CaMKII*

+ +

+ + MeCP2

0 KCl + + + KN-93 + DMSO + + Rpol pS438 +

Figure 5 CaMKII relieves MeCP2 occupancy and repression of the Avp enhancer. (a) CaMKII precludes repression by MeCP2. Transfection of CaMKII* and the site-specific methylated CGI3 vector stimulated Avp expression and completely reversed repression by transfected MeCP2 in N6 cells. (b) CaMKII abolished MeCP2 occupancy at CGI3. Flag-tagged forms of MeCP2 or S438A (S438 nonphosphorylatable MeCP2) were transfected singly or together with CaMKII* in N6 cells. ChIP experiments with an antibody to Flag showed that CaMKII* mediated the release of MeCP2 from the Avp enhancer and that enhancer occupancy was maintained when MeCP2 (S438A) was transfected. ( c) MeCP2 repressed endogenous Avp expression in a CaMKII-regulated manner. Flag-tagged forms of MeCP2 or S438A (S438A) and CaMKII* were cotransfected in N6 cells. RT-PCR analysis of Avp expression revealed that CaMKII* attenuated MeCP2-mediated repression; CaMKII* had only minor effects in the presence of MeCP2 S438A. (d) Membrane depolarization relieved MeCP2 occupancy at the Avp enhancer. N6 cells were depolarized with 55 mM KCl (30 min). ChIP experiments showed reduced MeCP2 occupancy at CGI3, paralleled by increased activated Rpol occupancy at the promoter (CGI1). Pretreatment of N6 cells with a CaMKII inhibitor (KN-93) reversed these effects. (e) Immunoblot analysis of MeCP2-S438 phosphorylation. Compared with controls, depolarized N6 cells showed increased MeCP2-pS438 immunoreactivity; depolarization did not influence MeCP2 immunoreactivity. Data (ad) are presented as means s.d. (four independent experiments).

+ + + + + + MeCP2

KCl

~75 kDa

(Fig. 4a). We analyzed of the methylation profile in the CGI3 region that spans CpG1014 and found a pattern that was similar to the one that we observed in the mouse PVN (Supplementary Fig. 6). Treatment of N6 cells with 5-azacytidine, a potent inhibitor of DNA methylation, reduced the level of methylation of the Avp enhancer and, concomitantly, increased Avp expression (Fig. 4b and Supplementary Fig. 6). In transfection assays (Fig. 4c), deletion of the CGI3 region reduced reporter activity by 37% and deletion of the entire enhancer resulted in almost complete abolition of reporter activity (Fig. 4d). We examined Avp gene reporter activity after in vitro methylation of the entire Avp vector, including the promoter and transcribed regions. Methylation led to a tenfold decrease in reporter activity (Fig. 4e). Moreover, reporter activity was reduced by 50% when methylation was targeted specifically to CGI3 (Fig. 4e). Together with the results obtained in hypothalamic tissue, this finding suggests that CGI3specific methylation is critical for the control of Avp expression. MeCP2 selectively binds CGI3 and represses Avp expression DNA methylation is interpreted by a family of methyl CpGbinding domain (MBD) proteins comprising MeCP2, MBD1, MBD2, MBD3 and MBD4, the first three of which couple DNA methylation to transcriptional repression. Although hypothalamic N6 cells expressed MeCP2, MBD1 and MBD2 (Supplementary Fig. 7), we found MeCP2 to be the most potent repressor of the CGI3 methylated vector in co-transfection experiments (Fig. 4f). We directly assessed the binding of MeCP2 at the Avp locus by immunoprecipitation of cross-linked chromatin from N6 cells with antibodies to MeCP2 or control IgG, followed by PCR analysis of the recovered DNA using seven primer pairs bracketing the Avp locus (Fig. 4g). As expected, MeCP2 did not occupy the poorly methylated Avp promoter and exonic CpG islands (CGI1 and 2); moreover, MeCP2 was also absent at CGI4, which is methylated to a relatively high
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

extent. In contrast, MeCP2 was strongly enriched at the CGI3 region (Fig. 4g). Notably, CGI3, but not CGI4, was poorly recovered when the same chromatin samples were immunoprecipitated with antibodies to MBD1 and MBD2 (Supplementary Fig. 7). Thus, MeCP2 preferentially and selectively occupies CGI3 of the Avp enhancer. Pre-treatment of N6 cells with 5-azacytidine robustly decreased MeCP2 occupancy at CGI3 and increased promoter binding of activated (pSer5) RNA polymerase II and Avp transcription in parallel (Fig. 4b,h). Therefore, MeCP2 occupancy at the Avp locus is DNA methylation-dependent and, once bound, MeCP2 acts to repress transcription. High-affinity binding of MeCP2 to methylated DNA requires a local sequence context, namely a symmetrical methyl-CpG dinucleotide that localizes close to a run of four or more A/T bases that facilitate DNA binding27. We identified four CpG dinucleotides (CpGs 13, 14 and 21, as well as the highly relevant CpG10) that matched the latter criterion in the CGI3 sequence (Fig. 4i). Their function in the context of MeCP2 binding was tested by in vitro DNAbinding electrophoretic mobility shift assays (EMSAs), using recombinant MeCP2 and oligonucleotides that spanned CpG10, CpG12 or CpG14. As anticipated, methylation of these motifs proved to be essential for MeCP2 binding and effective self-competition (Fig. 4j and Supplementary Fig. 8). MeCP2 specifically bound to the key motif CpG10 with a KD of 2.6 nM (comparable to that previously reported27) and DNA binding was strongly impaired (KD > 50 nM) after mutation of the A/T run adjacent to the CpG10 dinucleotide (CpG10A/Tmut) (Fig. 4k). Compared with CpG10, the neighboring motifs CpG12 and CpG14 bound MeCP2 with lower affinity (KD 9.6 and 10.5 nM, respectively) and competed poorly with CpG10 for forming a complex with MeCP2 (Supplementary Fig.8). Together, these results indicate that the Avp enhancer contains context-specific, high-affinity MeCP2 DNA-binding sites that are important for the regulation of Avp. Phosphorylation of MeCP2 prevents Avp enhancer occupancy Neuronal depolarization has been shown to trigger Ca2+-dependent phosphorylation of MeCP2, causing dissociation of MeCP2 from the Bdnf promoter and increased Bdnf transcription28,29. Recently, de novo Ca2+/calmodulin-dependent protein kinase II (CaMKII) was shown to mediate phosphorylation of rat MeCP2 at serine 421 (S438 in mouse)30.
1563

a r t ic l e s
6 weeks 10 d To explore whether this mechanism might be AVP AVP/pS438 AVP pS438 AVP/pS438 pS438 responsible for regulating MeCP2 occupancy at the Avp enhancer, we transfected N6 cells with either MeCP2 and/or a constitutively active form of CaMKII (CaMKII*) together with the CGI3-methlyated Avp vector. Transfection Control of CaMKII* increased Avp reporter activity slightly, and completely reversed MeCP2mediated repression (Fig. 5a). Furthermore, we observed that CaMKII* markedly reduced the occupancy of Flag-tagged MeCP2 at the 50 m Avp enhancer, but failed to release DNAbound nonphosphorylatable MeCP2 (S438A; Fig. 5b). Subsequently, MeCP2-S438A and (to a lesser degree) MeCP2 prevented CaMKII* ELS activitydependent increases in Avp expression (Fig. 5c); this result is consistent with a repressive role of MeCP2. Additional experiments showed that K+-induced depolarization of N6 cells faithfully reproduced the effects of CaMKII* Figure 6 ELS induces phosphorylation of MeCP2 in parvocellular PVN neurons. ELS led to transfection, that is, relieved MeCP2 occu- increased immunostaining of MeCP2-pS438 (pS438) and AVP in the PVN of 10-d-old mice. pancy at the Avp enhancer. A role of CaMKII Colocalization of AVP and MeCP2-pS438 in parvocellular neurons in the PVN was apparent. in mediating MeCP2-S438 phosphoryla- Comparable levels of MeCP2-pS438 staining were detected in 6-week-old control and ELS mice. tion was confirmed by the complete reversal The images that are shown are representative of five mice per group and age. of this regulation after pretreatment with the CaMKII inhibitor KN-93. Moreover, K+-induced depolariza- was prominently increased in parvocellular AVP-expressing neurons tion increased the presence of activated RNA polymerase II at the in the PVN of 10-d-old ELS mice (Fig. 6). In addition, the PVN of Avp promoter (Fig. 5d), verifying the ability of MeCP2 to repress 10-d-old ELS mice had increased phospho-CaMKII immunoreactivity in activity-dependent gene expression6. Lastly, an antibody to the AVP-positive neurons, a finding that is compatible with a role for this regulatory MeCP2-S438 phosphorylation site (MeCP2-pS438; kinase in the mediation of activity-dependent MeCP2-S438 phosphoSupplementary Fig. 9) reacted strongly with extracts from membrane- rylation (Supplementary Fig. 10). The extents to which CaMKII and depolarized N6 cells, but only weakly with extracts from nonstimu- MeCP2-S438 were phosphorylated in the parvocellular division of the lated N6 cells (Fig. 5e). Thus, membrane depolarization directly leads PVN did not differ between 6-week-old control and ELS mice (Fig. 6 to phosphorylation of MeCP2 at S438. and Supplementary Fig. 10). In addition, neither Mecp2 mRNA nor total MeCP2 and CaMKII protein expression differed between the ELS reduces MeCP2 occupancy at Avp enhancer two groups (Supplementary Fig. 10). Lastly, MeCP2-pS438 immuno We next asked whether the sustained increased expression of Avp after reactivity in the supraoptic nucleus did not differ between control and ELS is triggered by MeCP2-S438 phosphorylation and subsequent ELS mice at all ages, demonstrating the site-specificity of the effects relief of MeCP2 occupancy at the Avp enhancer. This hypothesis was (data not shown). These results indicate that there is an age- and cell supported by the observation that MeCP2-S438 phosphorylation typespecific role for ELS-induced MeCP2-S438 phosphorylation, prompting us to examine its relevance for enhancer occupancy. Control a 20 * 8 ELS Although RNA polymerase II occupancy * 16 6 at the Avp promoter and Avp expression * Avp mRNA (Ci g1) 12 8 4 0 10 d 10 d 6 week 4 2 0 OXT 5 CGI2 * CGI3 * 6 7 CGI4 Rel. dif. Rpol/IgG

2009 Nature America, Inc. All rights reserved.

b
1 CGI1

AVP 2 3 4

Rel. dif. MeCP2/IgG

4 3 2 1 0

10 d 6 week

10 d 6 week

10 d 6 week

10 d 6 week

10 d 6 week

10 d 6 week

10 d 6 week

Figure 7 ELS reduces MeCP2 occupancy at the Avp enhancer. (a) ELS upregulated Avp expression in the parvocellular PVN (measured by ISH) of 10-d-old mice. In vivo ChIP analysis revealed increased activated Rpol occupancy at the Avp promoter (CGI1) of ELS mice, reflecting increased Avp transcription. (b) ELS reduced MeCP2 occupancy at CGI3. An in vivo ChIP scan of the Avp locus (schematized above) revealed selective MeCP2 occupancy at CGI3. ELS significantly reduced MeCP2 occupancy at this region in both 10-d-old and 6-week-old mice. Data are presented as means s.e.m. (n = 8 mice per group for ISH analysis; ChIP analysis based on five groups of pooled PVN). * P < 0.05 (t test).

1564

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
were markedly increased in 10-d-old ELS mice (Fig. 7a), CGI3 methylation did not differ between control and ELS mice of this age (Supplementary Fig. 11). Notably, in vivo chromatin immunoprecipitation (ChIP) experiments on PVN tissue revealed that, of the various MBDs, MeCP2 was selectively enriched at CGI3 in 10-d-old and 6-week-old control mice (Fig. 7b and data not shown) and that binding of MeCP2 to the Avp enhancer was reduced in ELS mice of both ages (Fig. 7b). Given that 10-d-old control and ELS mice have identical methylation patterns, the measured differences in MeCP2 occupancy suggest that ELS-induced MeCP2-S438 phosphorylation results in relief of MeCP2 occupancy at the Avp enhancer. The repressive function of MeCP2 at the Avp enhancer was substantiated by sequential in vivo ChIP experiments, which revealed strong coupling of MeCP2 to transcriptionally inactive chromatin marks (data not shown), comparable with those reported for Crh occupancy31. In sum, de-repression of Avp transcription in 10-d-old ELS mice appears to involve increased MeCP2-S438 phosphorylation, whereas reduced enhancer occupancy in 6-week-old mice most likely reflects ELS-induced CGI3 hypomethylation (Figs. 3a and 7b). DISCUSSION Adverse experiences during early life contribute to the etiology of psychiatric conditions in later life14,32. Our results suggest that ELS in mice leads to epigenetic marking (hypomethylation) of a key regulatory region of the Avp gene in the PVN. These epigenetic events are accompanied by persistent upregulation of Avp expression in the parvocellular subdivision of the PVN and, consequently, sustained hyperactivity of the HPA axis. Notably, the ELS-induced endocrine phenotype lasted for at least 1 year following the initial adverse event and could be normalized through administration of an AVP V1b receptor antagonist. Studies in humans and in animal models suggest that stress or elevated glucocorticoid secretion are important for the function of interdependently regulated behavioral domains33,34. Here, ELStreated mice showed increased immobility in the forced-swim test, which assesses stress-coping ability35, and had deficits in step-down avoidance learning. Although acute rises in glucocorticoid secretion can facilitate inhibitory avoidance learning13, our data support the notion that sustained elevated glucocorticoid levels impair memory performance in ELS mice. Notably, the behavioral phenotypes induced by ELS were shown to be partly reversible after antagonism of AVP V1b receptors, thus highlighting AVP as an important mediator of these processes, but not necessarily the only14. Our results identify CpG residues in the CGI3 region of the Avp enhancer whose persistent hypomethylation after ELS is critical for the regulation of Avp expression. Recent work defined these residues as being high-affinity, context-specific MeCP2 DNA-binding sites27. On the basis of previous reports28,29, we hypothesized that signaling mechanisms controlling MeCP2 occupancy are critical for gene repression and the dynamic methylation of CGI3 in response to ELS. Supporting this, we found that depolarization of hypothalamic cells can regulate MeCP2 function by inducing its site-specific phosphorylation via CaMKII activity. Taken together, our results indicate that phosphorylation of MeCP2 at S438 is critical for MeCP2 to function as a reader and interpreter of the DNA methylation signal at the Avp enhancer. That experience-dependent stimuli dynamically control the methylation of CGI3 is supported by the observation that ELS induced contemporaneous increases in CaMKII activation, MeCP2S438 phosphorylation and Avp expression in 10-d-old mice. On the other hand, MeCP2-S438 and CaMKII were phosphorylated to similar extents in adult control and ELS mice, indicating that
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

2009 Nature America, Inc. All rights reserved.

ELS-induced MeCP2 phosphorylation is important for the establishment of epigenetic marks. Once established, the observed differences in Avp enhancer methylation centered on MeCP2 binding sites, which appeared to be actively maintained in ELS and control mice. This interpretation is compatible with the view that MeCP2 serves as an epigenetic integration platform on which synergistic cross-talk between histone deacetylation, H3K9 methylation and DNA methylation act to confer gene silencing22. From a physiological perspective, it is conceivable that increased methylation of the Avp enhancer during early postnatal life serves to restrain the HPA axis in critical periods when homeostatic thresholds are set; this would facilitate adaptation of the endocrine system to future environmental stimuli. Our data suggest that ELS tilts the balance toward persistent hypomethylation and Avp overexpression by inducing reductions in MeCP2 binding. Thus, phosphorylation of MeCP2 appears to be a conduit of experience-driven changes in gene expression, serving as an important mediator of the persistent effects of ELS. In this respect, certain parallels may be drawn between the mechanisms underlying ELS and Rett syndrome; the latter, caused by mutations in Mecp2, also presents with altered cognitive, mood and HPA axis function36. Together with other recent work3,10, our results suggest that adverse events in early life can leave persistent epigenetic marks on specific genes that may prime susceptibility to neuroendocrine and behavioral dysfunction. Focusing on DNA methylation, our results provide evidence for postmitotic epigenetic modifications in neuronal function; such modifications can serve to facilitate (or disfavor) physiological and behavioral adaptations3,37. These marks and their initiators, mediators and readers (for example, MeCP2) provide new inroads for understanding the molecular basis of stressrelated disorders of the brain. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank A. Hoffmann and R. Stoffel for their excellent technical assistance, A. Varga and B. Wrle for help with animal care, N. Sousa, J.-P. Schlke, T. Bettecken, F. Roselli and R. Spanagel for support. We thank S. Aventis for supplying SSR149415. This work was funded by the European Union (CRESCENDO European Union contract number LSHM-CT-2005-018652 to O.F.X.A. and D.S.) and the Deutsche Forschungsgemeinschaft (SP 386/4-2 to D.S.). AUTHOR CONTRIBUTIONS The study was conceived and designed by D.S. and O.F.X.A. C.M. and D.S. designed and interpreted the molecular studies that were carried out by C.M., Y.W., Y.B. and D.F., A.V.P. and O.F.X.A. were responsible for the neuroendocrine studies and A.V.P. and V.M. carried out the behavioral experiments under the guidance of C.T.W. C.M., A.V.P., F.H., O.F.X.A. and D.S. wrote the paper, with input from all of the other authors.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.

1. Jaenisch, R. & Bird, A. Epigenetic regulation of gene expression: how the genome integrates intrinsic and environmental signals. Nat. Genet. 33 Suppl, 245254 (2003). 2. Jirtle, R.L. & Skinner, M.K. Environmental epigenomics and disease susceptibility. Nat. Rev. Genet. 8, 253262 (2007). 3. Weaver, I.C. et al. Epigenetic programming by maternal behavior. Nat. Neurosci. 7, 847854 (2004). 4. Tsankova, N.M. et al. Sustained hippocampal chromatin regulation in a mouse model of depression and antidepressant action. Nat. Neurosci. 9, 519525 (2006).

1565

a r t ic l e s
5. Renthal, W. et al. Histone deacetylase 5 epigenetically controls behavioral adaptations to chronic emotional stimuli. Neuron 56, 517529 (2007). 6. Flavell, S.W. & Greenberg, M.E. Signaling mechanisms linking neuronal activity to gene expression and plasticity of the nervous system. Annu. Rev. Neurosci. 31, 563590 (2008). 7. Tsankova, N., Renthal, W., Kumar, A. & Nestler, E.J. Epigenetic regulation in psychiatric disorders. Nat. Rev. Neurosci. 8, 355367 (2007). 8. Reik, W. Stability and flexibility of epigenetic gene regulation in mammalian development. Nature 447, 425432 (2007). 9. McGowan, P.O. et al. Promoter-wide hypermethylation of the ribosomal RNA gene promoter in the suicide brain. PLoS One 3, e2085 (2008). 10. McGowan, P.O. et al. Epigenetic regulation of the glucocorticoid receptor in human brain associates with childhood abuse. Nat. Neurosci. 12, 342348 (2009). 11. Fumagalli, F., Molteni, R., Racagni, G. & Riva, M.A. Stress during development: Impact on neuroplasticity and relevance to psychopathology. Prog. Neurobiol. 81, 197217 (2007). 12. Gluckman, P.D., Hanson, M.A., Cooper, C. & Thornburg, K.L. Effect of in utero and early-life conditions on adult health and disease. N. Engl. J. Med. 359, 6173 (2008). 13. de Kloet, E.R., Jels, M. & Holsboer, F. Stress and the brain: from adaptation to disease. Nat. Rev. Neurosci. 6, 463475 (2005). 14. Lupien, S.J., McEwen, B.S., Gunnar, M.R. & Heim, C. Effects of stress throughout the lifespan on the brain, behavior and cognition. Nat. Rev. Neurosci. 10, 434445 (2009). 15. Levine, S. Developmental determinants of sensitivity and resistance to stress. Psychoneuroendocrinology 30, 939946 (2005). 16. Charmandari, E., Tsigos, C. & Chrousos, G. Endocrinology of the stress response. Annu. Rev. Physiol. 67, 259284 (2005). 17. Engelmann, M., Landgraf, R. & Wotjak, C.T. The hypothalamic-neurohypophysial system regulates the hypothalamic-pituitary-adrenal axis under stress: an old concept revisited. Front. Neuroendocrinol. 25, 132149 (2004). 18. Holmes, A., Heilig, M., Rupniak, N.M., Steckler, T. & Griebel, G. Neuropeptide systems as novel therapeutic targets for depression and anxiety disorders. Trends Pharmacol. Sci. 24, 580588 (2003). 19. Serradeil-Le Gal, C. et al. An overview of SSR149415, a selective nonpeptide vasopressin V(1b) receptor antagonist for the treatment of stress-related disorders. CNS Drug Rev. 11, 5368 (2005). 20. Aguilera, G. & Rabadan-Diehl, C. Vasopressinergic regulation of the hypothalamicpituitary-adrenal axis: implications for stress adaptation. Regul. Pept. 96, 2329 (2000). 21. Meaney, M.J. Maternal care, gene expression and the transmission of individual differences in stress reactivity across generations. Annu. Rev. Neurosci. 24, 11611192 (2001). 22. Allis, C., Jenuwein, T. & Reinberg, D. Epigenetics (Cold Spring Harbor Laboratory Press, Cold Spring Harbor, New York, 2007). 23. Weber, M. et al. Distribution, silencing potential and evolutionary impact of promoter DNA methylation in the human genome. Nat. Genet. 39, 457466 (2007). 24. Gainer, H., Fields, R.L. & House, S.B. Vasopressin gene expression: experimental models and strategies. Exp. Neurol. 171, 190199 (2001). 25. Suzuki, M.M. & Bird, A. DNA methylation landscapes: provocative insights from epigenomics. Nat. Rev. Genet. 9, 465476 (2008). 26. Belsham, D.D. et al. Generation of a phenotypic array of hypothalamic neuronal cell models to study complex neuroendocrine disorders. Endocrinology 145, 393400 (2004). 27. Klose, R.J. et al. DNA binding selectivity of MeCP2 due to a requirement for A/T sequences adjacent to methyl-CpG. Mol. Cell 19, 667678 (2005). 28. Chen, W.G. et al. Derepression of BDNF transcription involves calcium-dependent phosphorylation of MeCP2. Science 302, 885889 (2003). 29. Martinowich, K. et al. DNA methylation-related chromatin remodeling in activitydependent BDNF gene regulation. Science 302, 890893 (2003). 30. Zhou, Z. et al. Brain-specific phosphorylation of MeCP2 regulates activity-dependent Bdnf transcription, dendritic growth and spine maturation. Neuron 52, 255269 (2006). 31. McGill, B.E. et al. Enhanced anxiety and stress-induced corticosterone release are associated with increased Crh expression in a mouse model of Rett syndrome. Proc. Natl. Acad. Sci. USA 103, 1826718272 (2006). 32. Malaspina, D. et al. Acute maternal stress in pregnancy and schizophrenia in offspring: a cohort prospective study. BMC Psychiatry 8, 71 (2008). 33. Bessa, J.M. et al. A trans-dimensional approach to the behavioral aspects of depression. Front. Behav. Neurosci. 3, 1 (2009). 34. Kalueff, A.V., Wheaton, M. & Murphy, D.L. Whats wrong with my mouse model? Advances and strategies in animal modeling of anxiety and depression. Behav. Brain Res. 179, 118 (2007). 35. Cryan, J.F. & Slattery, D.A. Animal models of mood disorders: recent developments. Curr. Opin. Psychiatry 20, 17 (2007). 36. Chahrour, M. & Zoghbi, H.Y. The story of Rett syndrome: from clinic to neurobiology. Neuron 56, 422437 (2007). 37. Miller, C.A. & Sweatt, J.D. Covalent modification of DNA regulates memory formation. Neuron 53, 857869 (2007).

2009 Nature America, Inc. All rights reserved.

1566

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

ONLINE METHODS
ELS. Maternal separation stress11,15 was used to induce ELS. Briefly, pups delivered (postnatal day 0 (P0) on day of birth) by timed-pregnant C57BL/6N mice (Charles River) were placed, as individual litters, in a clean cage (with heating pad) for 3 h each day on P110, having no physical contact with their mothers. Control (non-ELS) pups remained undisturbed in the maternal nest throughout. Pups remained with their mothers until weaning (P21), when they were housed in sex-matched groups (35 mice per cage); only males were used for analyses. Standard laboratory animal housing conditions were maintained throughout, with 12-h daily illumination (lights on at 06:00). All procedures were approved by the Regierung von Oberbayern and were in accordance with European Union Directive 86/609/EEC. Behavioral phenotyping. At 3 months of age, control and ELS mice were housed singly and randomly assigned to the following long-term treatment groups: naive (no injections), vehicle (5% DMSO (vol/vol), 5% Chremophor EL (vol/vol), saline, intraperitoneal injection), and SSR149415 (20 mg per kg of body weight per d). Treatments started 4 weeks before behavioral phenotyping and continued for the duration of the experiments; injections were administered 1 h before behavioral testing. An investigator who was blind to the treatments carried out the behavioral analyses; an interval of 2 d was allowed between each test procedure. Anxiety-like behavior was assessed in the elevated plus-maze38, light-dark avoidance39 and novelty-induced hypophagia40 tests. The forced swim test was used to evaluate passive stress coping behavior and was performed essentially as described previously41. Briefly, each mouse was placed into a glass beaker (5 l 23.5 16.5 cm) that was filled with water (25 1 C) up to a height of 15 cm, for 6 min. Floating (immobility) was scored during the last 4 min of the exposure. A mouse was considered to be immobile when it floated passively in an upright position, making only small movements to keep its head above water surface. Memory was evaluated using the step-down avoidance learning test. Training sessions involved placing mice onto a platform (2.5 10 10 cm3) and administering a scrambled electric foot shock (0.7 mA) when they stepped off the platform onto a metal grid; mice were thereafter immediately returned to their home cages. Passive avoidance memory was tested by placing mice back onto the platform 24 h later and step-down latencies (four-paws criterion) were measured in three consecutive trials. Trials were terminated after 5 min in cases of failure to step down; a step-down latency of 301 s was ascribed to the trial. The mean of the three trials served as a measure of memory performance. Tissue preparation and hormone assays. Serum corticosterone was measured in adulthood by radioimmunoassay in blood samples. Samples were collected at 6 p.m. (peak) and 30 min following application of a previously described acute psychological stressor (911 a.m.)42. At the age of 3 months, mice received an intraperitoneal injection of dexamethasone (10 g per 100 g) at 12 a.m. noon and blood was collected at 6 p.m. for determination of corticosterone; the latter mea surements were compared to values obtained at the nocturnal sampling on the previous day. At various ages, mice were killed by cervical dislocation and tissues (brain, pituitary, thymus and adrenal) were collected. Brains for ISH and micropunching were cryosectioned (10 m) at the level of the rostral PVN (bregma 0.75 to 0.85) and the hippocampus (bregma 1.70 to 1.90). Punches of the PVN were obtained by in loco microdissection under histological control. ISH. Avp and Crh transcripts were detected with 48/50-mer 35S-labeled antisense probes, complementary to the murine Avp (bases 1,4931,540, accession number M88354) and Crh (bases 1,6851,732 accession number AY128673) genes, respectively. Nr3c1 and Mecp2 transcripts were measured using ribonucleotide probes (Nr3c1, bases 81528, accession number M14053; Mecp2, bases 6121,604, accession number NM010788) and previously published protocols42. Avp, Crh and Mecp2 transcript signal intensities were measured in the ventromedial compartment of the PVN, representing the parvocellular division. Nr3c1 hybridization signals were measured in the PVN and hippocampal subfields CA13 and dentate gyrus. Bisulfite sequencing. Genomic DNA (200400 ng) isolated from PVN tissue punches was digested with EcoRI, sodium bisulfite converted (Qiagen DNA methylation kit), aliquoted and used for PCR reactions. Primers used are listed in

Supplementary Table 1. Products were cloned into pGEM-T vector; at least 20 independent recombinant clones per PCR and mouse were analyzed on an ABI Prism 3700 capillary sequencer. Overall methylation levels (Fig. 3d) were calculated for the entire enhancer region from mean levels of individual CpG residues. RT-PCR and RNA extractions. Primer sequences are listed in Supplementary Table 2. The expression levels of the housekeeping genes Hprt and Gapdh were used for normalization. Total RNA was extracted with Trizol (Invitrogen) and reverse-transcription reactions were performed on 1 g of total cell culture extracted RNA or 100 ng of tissue-derived RNA with SuperscriptII (Invitrogen) and poly-dT primer. qPCR was carried out on a LightCycler (Roche) using LightCycler FastStart DNA Master Plus SYBR Green (Roche). In vitro methylation. Vectors were methylated with SssI and S-adenosylmethionine (New England Biolabs). For site-specific methylation, an 888-bp fragment containing CpGs 1025 in the Avp enhancer was excised by digestion with Eco81I; the vector was further digested with XbaI and BamHI to prevent re-ligation. Methylated or control unmethylated digests were ligated into the dephosphorylated reporter construct, cleaved with Eco81I. Completeness of in vitro methylation and maintenance (until cell harvesting) was confirmed by bisulfite sequencing. Recombinant proteins and EMSA. GST- or His-MeCP2 fusion proteins were grown in DH5, purified and quantified as described previously43. For in vitro DNA-binding assays (Fig. 4j and Supplementary Fig. 8), recombinant MeCP2 (0.5 g) was incubated with 20,000 cpm of double-stranded 32P end-labeled naive or in vitro methylated oligonucleotides27. Reactions were fractionated on 8% polyacrylamide gels. Although GST protein itself does not recognize methylated CpG10, inclusion of a GST antibody abolished MeCP2 binding, verifying the identity of the shifted complex (Fig. 4j). Dissociation constants (KD) were deduced by Scatchard analysis of saturation binding isotherms43. Plasmids. The AVP expression vectors (kindly gifted by H. Gainer and R.L. Fields, US National Institutes of Health) were modified by exchanging the egfp reporter gene in the third exon of the Avp gene44, with a cDNA for Gaussia luciferase KDEL encoding intracellular Gaussia luciferase (Targeting Systems) (see Fig. 4c). The parent Avp-Gaussia construct contained 288 bp of the promoter region, all exons (numbered) and introns, and the entire 2.1-kb enhancer. The Avpenhancer construct has the entire enhancer sequence removed, whereas the Avp1025 was generated by deletion of an 888-bp fragment of the enhancer containing CpGs125 by digestion with Eco81I and subsequent vector religation. The Mecp2 expression vector and His-tagged MeCP2 1205 (kindly provided by A. Bird, University of Edinburgh) consisted of the mouse MeCP2 variant45 in pRL-SV40 (Promega) and of a cDNA for the first 205 amino acids of human MeCP2 with a C-terminal His-tag46 in a pet30b vector (Novagen), respectively. For prokaryotic expression (pGEx2tk-Mecp2), the MeCP2 cDNA was PCR amplified (forward primer, aag gga tcc gta gct ggg atg tta gg; reverse primer, tct gat atc ctc agt ggt gga gga gga g) and inserted into the BamHI and SmaI sites of pGEx2tk (Pharmacia). N-terminal Flag-tagged forms of different Mecp2 constructs were obtained by PCR cloning of wild type (forward primer, aag gga tcc gcc gcc gct gcc gcc acc gc; reverse primer, tct gat atc ctc agc taa ctc tct cgg tc) or of a form lacking the 45 C-terminal amino acids of MeCP2 (forward primer, aag gga tcc gcc gcc gct gcc gcc acc gc; reverse primer, tct gat atc ctc agc taa ctc tct cgg tc) into the BamHI and EcoRI sites of pRK7-Flag43. The phosphor-acceptor residue Ser 438 was replaced by Ala in MeCP2 (S438A) by site-directed mutagenesis (forward primer, ccc gag gag gcc gac tgg aaa gcg atg gc; reverse primer, gcc atc gct ttc cag tcg gcc tcc tcg gg). The MeCP2 riboprobe (nucleotides 6121,604, accession number NM_ 010788) contains the conserved sequence in exons 3 and 4 of the mouse Mecp2 gene. A corresponding PCR product (forward primer, aaa ggt ggg aga cac ctc ct; reverse primer, tcc aca ggc tcc tct ctg tt) was cloned in the pGEM-T vector (Promega) for generation of riboprobes. Expression vectors for MBD2 and MBD3 (kindly provided by S.T. Jacob, Ohio State University) contain the mouse MBD2 or MBD3 cDNAs47 in the pcDNA3.1 vector (Invitrogen). The MBD1 expression vector contains the full-length cDNA for mouse MBD1 (accession number NM_013594; forward primer, tac ctc tag

2009 Nature America, Inc. All rights reserved.

doi:10.1038/nn.2436

nature NEUROSCIENCE

aat ggc tga gga ctg gct gga ctg; reverse primer, ttt cta gaa aca att tgc aaa gaa ttt tca gg) inserted in the pRK7 expression vector. CaMKII expression vectors contained either full-length CaMKII (1317) or CaMKII (1290), a constitutively active form resulting from the absence of the calmodulin-binding domain (kindly provided by A.R. Means, Duke University Medical Center). The constitutively active CaMKII(T286D) contains a replacement of Thr286, which is located in the autoinhibitory domain48, by Asp (kindly provided G. Turrigiano, Brandeis University). All constructs used in this study were entirely sequence verified. Cell culture and transfection experiments. Mouse hypothalamic cells (N6 line)26 were grown using standard conditions (DMEM supplemented with 10% fetal calf serum, vol/vol). Cells (105) were treated for 5 consecutive days with different concentrations of 5-azacytidine (Calbiochem), which was replenished in fresh growing medium every other day. N6 cells were transfected using Lipofectamine 2000 (Invitrogen). Briefly, 8 105 cells were seeded 24 h earlier in 6-well plates. DNA was mixed with 4 l of Lipofectamine, incubated at 25 C for 20 min and then added to the cells, which were grown for 18 h. Epithelial kidney cells (LLC-PK1, ATTC CL-101) were transfected by electroporation as described previously49. For cotransfection, we used 0.1 g of the Camk2a expression constructs, 1 g of the pRK7-Flag Mecp2 constructs and 1 g of the Mbd1, Mbd2 and Mbd3 expression vectors. Luciferase values were normalized against -galactosidase values42. Immunohistochemistry, immunoblots and ChIP experiments. Brains were extracted from microwave-fixed heads, placed overnight in 4% paraformaldehyde (wt/vol), cryo-preserved and sectioned (20 m) at the level of the PVN before immunostaining. Immunoblot analysis was carried out on whole-cell extracts (50 g) after fractionation by PAGE gel electrophoresis43. For ChIP experiments, chromatin from N6 cells or mouse PVN punches (individual pools formed from groups of three or five mice for 6-week-old and 10-d-old mice, respectively) was cross-linked50, disrupted by sonification (Diagenode Bioruptor, and purified with the Magna ChIP G kit (Millipore). The ChIP primers that we used for qPCR are listed in Supplementary Table 3. Antibodies. The antibodies that we used are listed in Supplementary Table 4. The polyclonal antibody to MeCP2, which recognizes MeCP2 (accession number GI:123122664) irrespective of its phosphorylation status, was generated by injecting New Zealand White rabbits with the KLH-conjugated peptide NH2CSMPRPNREEPVDSRTPV-CONH2, corresponding to amino acids 480496. The antiserum was purified by affinity chromatography on a column containing coupled MeCP2 480496 peptide and the affinity-purified antibody to MeCP2 was eluted. The polyclonal antibody to phosphorS438-MeCP2 (MeCP2-pS438) was generated by injecting New Zealand White rabbits with the KLH-conjugated peptide NH2-CMPRGGpSLES-CONH2. The antiserum was purified by affinity chromatography on a column containing coupled nonphosphorylated MeCP2S438 peptide. The flow through was then passed over a second column containing coupled phosphorylated MeCP2-S438 peptide and the affinity-purified antibody to MeCP2-pS438 was eluted.

Characterization of the MeCP2 antibodies was performed by transfection of pRK7-FLAG MeCP2 and pRK7-FLAG MeCP2 (S438A) (0.1 g each), singly or together, with the different CaMKII expression vectors (0.5 g each) into LLCPK1cells. Mock transfections were performed using an equal amount of filling plasmid. Whole-cell extracts (20 g) were fractionated on 8% SDS-PAGE gels, immunoblotted and tested with either antibody to MeCP2 (1:1,000), the UPMeCP2 (Upstate, 07013, 1:1,000), antibody to MeCP2-pS438 (1:1,000) or antibody to Flag (1:1,000). In a parallel set of experiments, the same cellular lysates were treated for 1 h at 37C with calf intestine phosphatase (10 units) or assay buffer alone followed by PAGE gel electrophoresis. Antibody to MeCP2-pS438 and antibody to MeCP2-pS421 (kindly provided by Z. Zhou and M.E. Greenberg30, Harvard Medical School) produced similar results when tested on PVN sections (Supplementary Fig. 8). Statistical analysis. Numerical data were analyzed by t tests or ANOVA, followed by Newman-Keuls post hoc test. In all cases, the nominal level of significance was P 0.05. Correlations between AVP expression and CpG methylation were tested by Pearsons correlation coefficient.

38. Touma, C. et al. Mice selected for high versus low stress reactivity: a new animal model for affective disorders. Psychoneuroendocrinology 33, 839862 (2008). 39. Mller, M.B. et al. Limbic corticotropin-releasing hormone receptor 1 mediates anxiety-related behavior and hormonal adaptation to stress. Nat. Neurosci. 6, 11001107 (2003). 40. Siegmund, A. & Wotjak, C.T. A mouse model of post-traumatic stress disorder that distinguishes between conditioned and sensitised fear. J. Psychiatr. Res. 41, 848860 (2007). 41. Bchli, H., Steiner, M.A., Habersetzer, U. & Wotjak, C.T. Increased water temperature renders single-housed C57BL/6J mice susceptible to antidepressant treatment in the forced swim test. Behav. Brain Res. 187, 6771 (2008). 42. Patchev, A.V. et al. Insidious adrenocortical insufficiency underlies neuroendocrine dysregulation in TIF-2 deficient mice. FASEB J. 21, 231238 (2007). 43. Hoffmann, A. et al. Transcriptional activities of the zinc finger protein Zac are differentially controlled by DNA binding. Mol. Cell. Biol. 23, 9881003 (2003). 44. Fields, R.L., House, S.B. & Gainer, H. Regulatory domains in the intergenic region of the oxytocin and vasopressin genes that control their hypothalamus-specific expression in vitro. J. Neurosci. 23, 78017809 (2003). 45. Kriaucionis, S. & Bird, A. The major form of MeCP2 has a novel N-terminus generated by alternative splicing. Nucleic Acids Res. 32, 18181823 (2004). 46. Fuks, F. et al. The methyl-CpG-binding protein MeCP2 links DNA methylation to histone methylation. J. Biol. Chem. 278, 40354040 (2003). 47. Ghoshal, K. et al. Role of human ribosomal RNA (rRNA) promoter methylation and of methyl-CpG-binding protein MBD2 in the suppression of rRNA gene expression. J. Biol. Chem. 279, 67836793 (2004). 48. Hanson, P.I., Meyer, T., Stryer, L. & Schulman, H. Dual role of calmodulin in autophosphorylation of multifunctional CaM kinase may underlie decoding of calcium signals. Neuron 12, 943956 (1994). 49. Murgatroyd, C. et al. Impaired repression at a vasopressin promoter polymorphism underlies overexpression of vasopressin in a rat model of trait anxiety. J. Neurosci. 24, 77627770 (2004). 50. Barz, T., Hoffmann, A., Panhuysen, M. & Spengler, D. Peroxisome proliferatoractivated receptor gamma is a Zac target gene mediating Zac antiproliferation. Cancer Res. 66, 1197511982 (2006).

2009 Nature America, Inc. All rights reserved.

nature NEUROSCIENCE

doi:10.1038/nn.2436

e r r at u m

Erratum: Dynamic DNA methylation programs persistent adverse effects of early-life stress
Chris Murgatroyd, Alexandre V Patchev, Yonghe Wu, Vincenzo Micale, Yvonne Bockmhl, Dieter Fischer, Florian Holsboer, Carsten T Wotjak, Osborne F X Almeida & Dietmar Spengler Nat. Neurosci. 12, 15591566 (2009); published online 8 November 2009; corrected after print 3 December 2009 In the version of this article initially published, on page 2, left column, the phrase and typically cluster in glucocorticoid-rich regions called CpG islands (CGIs) should be and typically cluster in GC-rich regions called CpG islands (CGIs). The error has been corrected in the HTML and PDF versions of the article.

2009 Nature America, Inc. All rights reserved.


nature neuroscience

a r t ic l e s

Amyloid- as a positive endogenous regulator of release probability at hippocampal synapses


Efrat Abramov1,3, Iftach Dolev1,3, Hilla Fogel1, Giuseppe D Ciccotosto2, Eyal Ruff1 & Inna Slutsky1
Accumulation of cerebral amyloid- peptide (A) is essential for developing synaptic and cognitive deficits in Alzheimers disease. However, the physiological functions of A, as well as the primary mechanisms that initiate early A-mediated synaptic dysfunctions, remain largely unknown. Here we examine the acute effects of endogenously released A peptides on synaptic transfer at single presynaptic terminals and synaptic connections in rodent hippocampal cultures and slices. Increasing extracellular A by inhibiting its degradation enhanced release probability, boosting ongoing activity in the hippocampal network. Presynaptic enhancement mediated by A was found to depend on the history of synaptic activation, with lower impact at higher firing rates. Notably, both elevation and reduction in A levels attenuated short-term synaptic facilitation during bursts in excitatory synaptic connections. These observations suggest that endogenous A peptides have a crucial role in activity-dependent regulation of synaptic vesicle release and might point to the primary pathological events that lead to compensatory synapse loss in Alzheimers disease. It is now more than a decade since A was first shown to be a normal, soluble product of neuronal metabolism 13. However, its physiological role in neuronal function remains elusive. A is produced by sequential limited proteolysis of the amyloid precursor protein (APP) conducted by two aspartyl proteases, - and -secretase4,5. The production and subsequent release of A positively correlate with the level of neuronal and synaptic activity6,7. It has been suggested that synaptic vesicle recycling through coupled endo-exocytosis is the primary mechanism that mediates activitydependent A production and release7,8. The concentration of A in the synaptic cleft varies, and at any given time is determined by the balance of A production, release and degradation. Several members of the M13 zinc-dependent metallopeptidase family, such as neprilysin, insulin-degrading enzyme and endothelin-converting enzymes 1 and 2, have been implicated in the degradation of endogenous A in the rodent brain9,10. Of these enzymes, neprilysin might be the best candidate for regulating A in the synaptic cleft because of its presynaptic localization11 and the extracellular position of its catalytic site12. Synapse loss is the strongest structural correlate of cognitive decline in patients with Alzheimers disease13,14. Studies on mouse models of Alzheimers disease provide evidence that soluble A oligomers contribute to the downregulation of synapse density 1518. Although these data support a homeostatic role of A in the control of synapse strength and number on a long timescale6,15, early modifications mediated by A that initiate synaptic dysfunctions remain controversial. Much effort has been directed toward elucidating synaptotoxic A species18,19. The ectopic application of specific A peptides has generated highly heterogeneous data, ranging from an increase in spontaneous synaptic activity and intrinsic excitability of neurons20,21 to a lack of effect on synaptic transmission18 or even its depression6,22. Such inconsistencies suggest that the concentration, molecular composition and oligomeric states of A peptides differentially affect synaptic function. However, it is still not understood how endogenously released A, comprising peptides of different lengths and molecular conformations, regulates synaptic transfer in normal, non-transgenic hippocampal circuits on a fast timescale. In an attempt to answer this question, we used optical and electrophysiological tools to assess the acute effects of changes in endogenous extracellular A concentration ([A]o) on synaptic transmission during different patterns of neuronal activity in rat hippocampal cultures and in acute hippocampal slices. Our results show that endogenously released A peptides positively regulate the release probability of synapses, but do not alter postsynaptic function or intrinsic neuronal excitability. The relationships between endogenous [A ]o, shortterm synaptic plasticity, the excitation-inhibition (E/I) balance, and ongoing neuronal activity are examined here. RESULTS Inhibition of neprilysin increases release probability As synaptic vesicle recycling positively regulates [A ]o on a rapid timescale7,8, we investigated whether endogenously released A would in turn modulate vesicle recycling through a feedback mechanism. To achieve an acute increase in endogenous A in the synaptic cleft we used thiorphan, a competitive inhibitor of the presynaptic rate-limiting peptidase neprilysin11,23,24 and neprilysin-like metalloendopeptidases25. Thiorphan (500 nM, 1 h) significantly increased

2009 Nature America, Inc. All rights reserved.

1Department of Physiology and Pharmacology, Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel. 2Department of Pathology, Bio21 Molecular Science and Biotechnology Institute and Mental Health Research Institute, The University of Melbourne, Parkville, Victoria, Australia. 3These authors contributed equally to this work. Correspondence should be addressed to I.S. (islutsky@post.tau.ac.il) .

Received 19 May; accepted 25 September; published online 22 November 2009; doi:10.1038/nn.2433

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1567

a r t ic l e s
Figure 1 Acute inhibition of neprilysin 30 action potentials at 0.2 Hz increases basal synaptic vesicle recycling. 100 300 High Th Cnt (a) [Ax40]o and [Ax42]o were elevated by 75 AF1126 Th application of thiorphan (500 nM, 1 h; n = 4, *** 200 50 ** P < 0.001 for A42 and P < 0.01 for A40) and ** *** AF1126 (5 g ml1, 1 h; n = 4, P < 0.001 for 25 100 A42 and P < 0.01 for A40). (b) Representative 0 high-magnification F images before and 0 250 500 A40 A42 Low F (a.u) 10 min after application of 500 nM thiorphan. Stimulation during FM1-43 staining: 30 1,000 10 150 *** *** *** 60 action potentials at 0.2 Hz. Scale bar: 2 m. 0.4 r = 0.66 7 100 Fluorescence intensities (arbitrary units, a.u.) 2.9 0.07 40 500 are coded using a pseudocolor transformation. 4 0.2 50 20 (c) F histograms before (black bars, Cnt) and 1 0 0 0 0 after (red bars, Th) thiorphan application in a 0 500 1,000 0 250 500 single experiment. The median F increased FCnt (a.u.) FCnt (a.u.) from 86 to 142 a.u. and D increased from 254 to 433. (d) Thiorphan effect at the AF1126 boiled 150 0.4 30 single-synapse level. Slope of linear fit is AF1126 *** *** *** *** Th 2.9 0.07. (e) Thiorphan-induced augmentation 200 100 20 *** *** 0.2 is inversely proportional to the initial presynaptic 50 10 strength (Spearman r = 0.66, P < 0.0001, 100 the same data as in d). (f) Effects of thiorphan 0 0.0 0 F on F, D and S across synaptic populations D S in 12 experiments. (g) Average magnitude of the effects produced by a 1-h application of thiorphan (n = 43, P < 0.001), anti-neprilysin AF1126 antibody (n = 10, P < 0.001), and heat-denatured AF1126 antibody (n = 4, P > 0.2) on presynaptic strength. (h) Effect of AF1126 antibody on F, D and S (n = 10, P < 0.0001). Error bars represent s.e.m. * P < 0.05, ** P < 0.01, *** P < 0.001.
Percentage of basal [A]o Cnt

Th

FTh / FCnt

FTh (a.u.)

2 D ( m )

F (a.u.)

Number of synapses

Th

Th

S (a.u.)

C nt

C nt

Percentage of control

D (m )

F (a.u.)

2009 Nature America, Inc. All rights reserved.

C AF nt 11 26

AF Cnt 11 26

S (a.u.)

the concentrations of Ax40 and Ax42 in the extracellular culture medium ([A40]o and [A42]o, measured by sandwich ELISA, n = 4, P < 0.01; Fig. 1a). We investigated whether thiorphan acutely affects synaptic vesicle recycling at individual hippocampal synapses using the activity-dependent dye FM1-43 as a marker of synaptic vesicle turnover26. We evaluated the functional properties of synapses by estimating the release probability (Pr) and spatial distribution of active presynaptic terminals27,28. To this end, we quantified the total amount of releasable fluorescence at each bouton (F) and the density of FM+ puncta per image (D) following stimulation by 30 action potentials at a rate of 0.2 Hz in the presence of 10 M FM1-43 (Supplementary Fig. 1). Figure 1b illustrates a typical high-magnification F image (a sub-region of the total imaged area, see Supplementary Fig. 2) before and after thiorphan application. The total presynaptic strength within a given region of the hippocampal network ( S) can be estimated as the product of F and D (S = F D). Single vesicle resolution ( Supplementary Fig. 3 and Supplementary Discussion) enables us to convert F to Pr and ensures that we detect active terminals that release at least one vesicle per applied 30 action potentials in our experimental system (corresponding to the detection limit of Pr ~0.04). Acute application of thiorphan (500 nM, 10 min) induced a substantial increase in both F and D across synaptic populations, resulting in 2.8-fold increase in the total presynaptic strength S ( Fig.1c). Analysis of the effect of thiorphan at the single synapse level in the same experiment revealed considerable variability across synapses (Fig. 1d,e). The degree of thiorphan-induced presynaptic facilitation was inversely correlated to initial F (Fig. 1e, Spearman r = 0.66, P < 0.0001). No correlation was found between presynaptic changes and initial F in a control experiment with two identical successive staining-destaining runs (Spearman r = 0.02, P > 0.5, Supplementary Fig. 1 ), indicating that the inverse correlation detected for the thiorphan effect was not a statistical artifact. The collective results of these experiments show a profound increase in F, D and, subsequently, S (n = 12, P < 0.0001, Fig. 1f). The
1568

EC50 of thiorphan for presynaptic enhancement was 47 7 nM (Supplementary Fig. 2), of the same order of magnitude as the EC50 to increase [A42]o in our experimental system (28 2 nM; Supplementary Fig.2). The presynaptic effect of thiorphan was reversible and long-lasting (Supplementary Fig. 2). Given that fluorescence quantification based on FM staining depends on exocytosis-endocytosis coupling, we determined the effect of thiorphan on vesicle exocytosis per se by measuring the rate of FM destaining during low-frequency stimulation29. The experimental protocol described above was modified for staining of the total pool of recycling vesicles by 600 action potentials at 10 Hz and destaining by 1-Hz stimulation (Supplementary Fig. 4). Thiorphan increased the destaining rate constant (k = 1/decay,, where decay is an exponential time course) by ~180% (n = 4, P < 0.001, Supplementary Fig. 4), suggesting acceleration of the exocytosis rate. Furthermore, thiorphan was ineffective in increasing presynaptic strength during low-frequency stimulation at a high [Ca2+]o (four-fold higher than physiological), which caused saturation of synaptic vesicle release in our preparation (Supplementary Fig. 4). Together, these results suggest that the acute effect of thiorphan is attributable to an increase in release probability. To examine whether a selective blockage of neprilysin activity affects synaptic vesicle release, we used the polyclonal AF1126 antibody, which was raised against rodent neprilysin and neutralizes the enzymes activity. Application of AF1126 (5 g ml1, 1 h) triggered an increase in [A]o (128 14% and 190 20% for [A40]o and [A42]o, respectively, n = 4, P < 0.001; Fig. 1a) and an increase in presynaptic strength by 209 6% due to an increase in both F and D (n = 8, P < 0.0001; Fig. 1h). The magnitude of the effect of AF1126 was similar to that of thiorphan for an identical application time (1 h, P > 0.3, Fig. 1g). Heat-denaturated AF1126 had no significant influence on presynaptic activity (n = 4, P > 0.2; Fig. 1g), confirming the molecular specificity of AF1126 in increasing release probability. These results indicate that neprilysin activity negatively regulates basal synaptic vesicle recycling in hippocampal synapses.
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

AF Cnt 11 26

C nt

Th

a r t ic l e s
Figure 2 A mediates thiorphan-induced A-depleted neurons App/ neurons presynaptic enhancement. (a) Left: YFP Cn Cnt Th 4G8: 30 action potentials at 0.2 Hz High representative images of a YFP-expressing 100 / neuron in App culture. Scale bar: 20 m. 4G8 80 HJ5.1 Right: high-magnification F images 60 -s.i. 40 -s.i. (FM4-64, 15 M) from axon of the same neuron 20 High (white rectangle) before and after thiorphan Low 0 application. Scale bar: 2 m. (b) As a, but for A40 A42 YFP / in App cultures. neurons expressing APP (c) Single synapse analysis of thiorphan effect in APPYFP Cnt Th 0.95 0.02 125 non-transfected App/ boutons (gray symbols), 200 4G8 100 YFP Th APP -expressing boutons (blue symbols), Low 75 100 non-transfected boutons in the proximity 50 (50 m) of the APPYFP-expressing dendrites 25 0 0 (red symbols) and YFP-only expressing boutons 0 200 400 0 100 200 (yellow symbols). (d) Average effect of thiorphan F (a.u.) F4G8 (a.u.) (n = 5) on presynaptic strength in App/ cultures 4G8 4G8-boiled for the groups shown in c. (e) Effects of 4G8 HJ5.1 HJ5.1-boiled *** -sec.inh. antibody (5 g ml1, 1 h), HJ5.1 antibody 400 400 Non-transf. -sec.inh. 200 1 *** (5 g ml , 1 h), -secretase inhibitor (BACE1 300 300 YFP-only *** inhibitor IV, 0.5 M, 8 h) and -secretase inhibitor *** ** 200 100 APPYFP 200 *** (L-685,458, 0.2 M, 8 h) on [Ax40]o and boutons 100 100 Near APPYFP [Ax42]o (n = 4, P < 0.0001). (f) Representative 0 0 F D S F dendrites D S 0 100 200 F images (FM1-43, 10 M) before and after FCnt (a.u.) thiorphan application, following pretreatment with 4G8 (5 g ml1, 2 h). Scale bar: 2 m. (g) F histograms before and after thiorphan application. (h) Single synapse analysis. Slope of linear fit is 0.95 0.02. (i) Average effects of thiorphan on F, D and S in A-depleted cultures following pretreatment with 4G8 (5 g ml1, n = 11, P > 0.3) and HJ5.1 (5 g ml1, n = 6, P > 0.2) antibodies, -secretase inhibitor (0.5 M, n = 16, P > 0.2) or -secretase inhibitor (0.2 M, n = 13, P > 0.2). Heat-denatured 4G8 and HJ5.1 antibodies did not block the thiorphan effect (5 g ml1, n = 4, P < 0.001). Error bars represent s.e.m. *P < 0.05,**P < 0.01, ***P < 0.001.

Percentage of basal [A]o

30 action potentials at 0.2 Hz

Th

Cnt

Number of synapses

Th (% control)

2009 Nature America, Inc. All rights reserved.

A mediates thiorphan-induced presynaptic enhancement To determine whether thiorphan-induced presynaptic potentiation was attributable to the activity of endogenously released A or other neprilysin substrates, we examined its effects in cultures with reduced A content (Fig. 2). Several methods have been used to reduce [A]: hippocampal cultures lacking A precursor protein (APP) produced from APP knockout30 (App/) mice; chelation of extracellular A by specific 4G8 or HJ5.1 (ref. 8) monoclonal antibodies that recognize the intermediate A domain; and reduction of A production by inhibition of - or -secretases. We first tested whether thiorphan affects presynaptic function in App/ hippocampal cultures30. Thiorphan was ineffective in increasing either F or D in neurons lacking APP (n = 8, P > 0.6; Fig. 2a,c,d). However, transient expression of human APP695 fused to YFP fluorophore (AppYFP) restored the thiorphan effect in App/ hippocampal cultures (n = 7, Fig. 2bd). We used FM4-64 to test the effect of thiorphan on presynaptic activity in APPYFP-expressing neurons (Fig. 2b) compared with neurons expressing YFP only (Fig. 2a). Single synapse (Fig. 2c) and population (Fig. 2d) analysis revealed that thiorphan induced a four-fold increase in presynaptic strength at boutons expressing APPYFP (P < 0.001), but did not produce presynaptic enhancement at non-transfected App/ boutons (P > 0.5) or in boutons expressing YFP-only (P > 0.3, Fig. 2c,d). Interestingly, thiorphan also triggered three-fold potentiation in non-transfected App/ boutons in the vicinity ( 50 m) of APPYFP-expressing dendrites (n = 7, P < 0.001, Fig. 2c,d). These data indicate that human A might be released from either a pre- or post-synaptic compartment and can enhance presynaptic strength in a cell-autonomous fashion6. To further examine whether A, a proteolytic derivative of APP, mediates thiorphan-induced presynaptic enhancement, we tested the ability of 4G8 antibody to prevent the effect of thiorphan. Incubation of hippocampal cultures with 4G8 (5 g ml1, 1 h) reduced [A40]o and [A42]o by more than 80% (n = 4, P < 0.001, Fig. 2e) and prevented thiorphan from inducing an increase in presynaptic strength
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

(Fig. 2fi). Analysis of 224 FM+ boutons revealed that thiorphan could not increase presynaptic strength in 4G8-pretreated neurons (Fig. 2g,h). On average (n = 15), neither F nor D was changed following thiorphan application in 4G8-treated neurons (P > 0.5, Fig. 2i). Thiorphan-induced presynaptic potentiation was also precluded by the use of another A monoclonal antibody, HJ5.1, to reduce [A]o (n = 6, P > 0.2, Fig. 2i). Heat-denatured 4G8 and HJ5.1 antibodies (5 g ml1, 1 h) failed to prevent the effect of thiorphan, confirming the molecular specificity of these antibodies in rescuing release probability ( n = 4, P < 0.001 compared to control; Fig. 2i). To verify that A mediates APP-dependent, thiorphaninduced presynaptic enhancement, we inhibited A production using specific blockers of -secretase (BACE1 inhibitor IV, 500 nM) or -secretase (L-685,458, 200 nM) ( n = 4, P < 0.0001; Fig. 2e). Thiorphan had no effect on synaptic vesicle turnover in neurons with inhibited A production (n = 1316, P > 0.2; Fig. 2i). These results indicate that endogenous A positively regulates the presynaptic strength of hippocampal synapses. Inhibition of A degradation increases mEPSC frequency To test whether a thiorphan-induced increase in [A]o affects quantal synaptic transmission, we measured miniature vesicle release in the presence of tetrodotoxin (TTX) at the level of presynaptic terminals using FM1-43 dye, and at the level of miniature AMPAR-mediated miniature excitatory postsynaptic currents (mEPSCs) using wholecell voltage-clamp recordings. Analysis of the mEPSC recordings before and 10 min after thiorphan application (Fig. 3ac) revealed an increase of 172 28% in mEPSC frequency (n = 11, P < 0.01), but no change in mEPSC amplitude (n = 11, P > 0.1). No correlation was detected between series resistance and mEPSC frequency (r2 = 0.0002, Pearsons correlation test, Supplementary Fig. 6), confirming that the differences across conditions are not attributable to variability in the quality of the recordings. Analysis of miniature vesicle recycling detected by FM1-43 under non-spike conditions
1569

Th (% control)

FTh (a.u.)

FTh (a.u.)

a r t ic l e s a b
Fraction of mEPSCs
Control

Thiorphan

Cnt 0.5

Fraction of mEPSCs

1.0

Th

1.0

10 pA

0.5

20 ms

0 4 0 2 6 8 Inter-event interval (s)

0 0 50 100 150 Peak amplitude (pA) FM1-43 +TTX 250 Cntr HJ5.1

c
250

mEPSC Cntr HJ5.1

d
Th (% control)
200 150 100 50 0

Th (% control)

200 150 100 50 0

**

***

***

***

2009 Nature America, Inc. All rights reserved.

Amp

Freq

HJ5.1

Figure 3 Inhibition of A degradation increases miniature synaptic vesicle release and mEPSC frequency. Effect of thiorphan (500 nM, 10 min) in control (a,b) and HJ5.1-treated cultures (e,f). (a) Representative mEPSCs before (control) and after thiorphan application. (b) Cumulative distributions of the inter-event intervals (left) and mEPSC peak amplitude (right) before (black) and after (gray) thiorphan application. Differences between the conditions for the displayed mEPSC inter-event interval cumulative distributions are statistically significant (P = 0.003, K-S test). Differences between mEPSC amplitude distributions ( P = 0.1, K-S test) are not significant. (c) Average effects of thiorphan on the frequency and peal amplitude of mEPSCs in control ( n = 11) and HJ5.1-treated (n = 9) cultures. Thiorphan increased mEPSC frequency only in control cultures (P < 0.01). (d) Average effect of thiorphan on miniature vesicle recycling measured by FM1-43 in the presence of TTX (loading for 90 s without external stimulation). Thiorphan increased the total presynaptic strength of miniature vesicle exocytosis by 208 13% ( n = 11, P < 0.001) in control, but not in HJ5.1-treated ( n = 5, P > 0.3) cultures. (e) Representative recordings of mEPSCs before and after thiorphan application in HJ5.1-treated (5 g ml 1, 2 h) cultures. (f) Cumulative distributions of the inter-event intervals (left) and mEPSC peak amplitudes (right) before (black) and after (gray) thiorphan application in HJ5.1-treated cultures. No statistical difference has been found between mEPSC amplitude ( P = 0.6, K-S test) or inter-event interval (P = 0.5, K-S test) distributions. Error bars represent s.e.m. Scale bars: 40 pA, 64 ms ( a,e). * P < 0.05, ** P < 0.01, *** P < 0.001.

HJ5.1 + thiorphan

f
Fraction of mEPSCs

0.5

Fraction of mEPSCs

1.0

1.0

0.5
10 pA
20 ms

0 0 2 4 6 8 Inter-event interval (s)

0 0 50 100 150 Peak amplitude (pA)

revealed a significant increase in presynaptic strength by 208 13% (n = 11, P < 0.0001; Fig. 3d) after acute thiorphan application. To test whether the thiorphan-induced increase in mEPSC frequency is mediated by A peptides, we repeated the experiments described above in cultures pretreated with anti-A antibodies. In HJ5.1-treated cultures, thiorphan affected neither the frequency nor the amplitude of mEPSCs (n = 9, P > 0.5; Fig. 3cf), and nor did it affect miniature vesicle recycling (n = 5, P > 0.3; Fig. 3d). These results suggest that on a rapid timescale, endogenously released A peptides enhance the probability of quantal transmitter release, without altering the number of postsynaptic AMPA receptors. A-mediated presynaptic enhancement is history dependent Having established a positive relationship between [A]o and basal synaptic vesicle release, we investigated how inhibition of A degradation affects vesicle release and synaptic transmission evoked by high-frequency discharges. Given the inverse relationship between release probability and paired-pulse facilitation31,32, A should trigger a reduction in short-term facilitation during periods of correlated activity. To test this prediction, we measured the total number of recycled synaptic vesicles stimulated by a constant number of spikes (30 action potentials) at various frequencies and temporal patterns. The following stimulation parameters were modified: frequency of spikes; temporal pattern (continuous versus burst); and inter-burst interval. To prevent the induction of long-term plasticity by bursts, we applied a postsynaptic glutamate receptor blocker (kynurenic acid).
1570

First, we assessed the effect of thiorphan on vesicle recycling triggered by spikes at gradually increasing rates (10100 Hz). The magnitude of thiorphan-induced presynaptic potentiation was inversely proportional to stimulation frequency: it was ~35% lower at 10 Hz than at 1 Hz, and was completely abolished at higher frequencies of 50 Hz (Fig. 4a,b). Furthermore, the temporal spike pattern affected the thiorphan-induced potentiation at intermediate stimulation frequencies of 1020 Hz; the magnitude of potentiation was higher for continuous stimulation than for short bursts (n = 79, P < 0.05). However, at spike rates 50 Hz, thiorphan did not influence presynaptic activity for either pattern. Moreover, decreasing the inter-burst interval from 5 to 0.2 s did not enhance the susceptibility of vesicle recycling to thiorphan during 100-Hz bursts (Fig. 4c, n = 911, P > 0.3). These data cannot be explained by saturation of synaptic vesicle release or optical signal, as a similar reduction in the inter-burst interval under control conditions enhanced the average fluorescence intensity of boutons (Supplementary Fig. 7). Finally, we assessed the effect of thiorphan on natural stimulation patterns derived from CA1 hippocampal place-cell firing in freely moving rats (the sequence was provided by A. Lee and M. Wilson, unpublished data). Periods of high-frequency discharges from two CA1 cells with a mean inter-spike interval of 46 3 ms (3237 ms) corresponding to the cells place field were selected as stimulation patterns (Fig. 4d). Thiorphan induced a small presynaptic enhancement during natural high-frequency discharges (126 5% for S, n = 11, P < 0.01, 4 epochs from 2 cells; Fig. 4e). Together, these data suggest that there is an inverse relationship between spike frequency and the magnitude of A-mediated presynaptic effects with some contribution of the temporal spike pattern at intermediate firing rates. The above analysis of synaptic vesicle recycling enables us to estimate presynaptic activity at the single synapse and network levels, but lacks high temporal (millisecond) resolution. To investigate the effect of thiorphan on the input/output relationship for bursts, we measured evoked AMPAR-mediated EPSCs between pairs of excitatory neurons, using a double-perforated patch-clamp technique to preserve the intracellular environment of neurons. Bursts were evoked in presynaptic neurons held in the current-clamp mode, whereas the EPSCs from postsynaptic neurons were recorded at
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 4 Potentiation of vesicle release by thiorphan depends on the history of synaptic activation. (a) Representative F images before and after application of thiorphan stained during stimulation by bursts (30 action potentials in 6 bursts; each burst contains 5 action potentials; inter-spike interval, 10 ms). Scale bar: 2 m. (b) Average effect of thiorphan (n = 69) on S for two stimulation patterns with varying (1100 Hz) frequencies: (i) 30 action potentials delivered continuously at a given frequency (gray bars); (ii) 30 action potentials distributed in 6 bursts (inter-burst interval, 5 s; black bars). (c) Average effect of thiorphan on S for 30 action potentials distributed in 6 bursts with varying inter-burst intervals (0.25 s), but constant inter-spike interval of 10 ms (n = 911, P > 0.3). (d) Inter-spike interval histogram for natural stimulation pattern (A. Lee and M. Wilson, unpublished data) used for the experiments presented in e. Insert: magnification of small inter-spike-intervals. (e) Average effect of thiorphan on presynaptic strength for 30 action potentials delivered in natural patterns (n = 11, P < 0.01 for S). (f) Examples of the effect of thiorphan on synaptic currents in two pairs of neuronal connections. (g) Effect of thiorphan on the first EPSC amplitude in the burst (n = 6, P < 0.01). (h) Averaged relative amplitudes of EPSCs within the burst, normalized to the first peak. Thiorphan increased synaptic depression during the burst (n = 6, P < 0.01). Error bars represent s.e.m. *P < 0.05, ** P < 0.01, *** P < 0.001.

Th (% control S)

200 100 0

** *
ns ns

Th (% control S)

Bursts: 5 action potentials at 100 Hz Cnt Th

300

c 150
100 50 0 1 10 20 50 100 Action potential frequency (Hz)

Bursts: 5 action potentials at 100 Hz

d
Number of spikes

60 40 20 0 0

Cell1 Cell2

e 150
Th (% control) 100 50 0

0.2 1 5 Inter-burst interval (s) "Natural" pattern

**

**

0 5 10 15 20

Low

High

50 100 150 200 250 Inter-spike interval (ms)

f
Pair #1

g
40

h
Relative amplitude (%)

2009 Nature America, Inc. All rights reserved.

**

200 150

Pair #2

50 0

Cnt Th 0 1 2 3 4 5 Number of action potentials

nt

a holding potential of 70 mV. As shown in the representative recordings depicted in Figure 4f, acute thiorphan application resulted in an increase in amplitude of the first EPSC within the burst, but its effect on the amplitude of the last EPSC varied from no change (pair 1; Fig. 4f) to a decrease (pair 2; Fig. 4f). On average (n = 6), thiorphan increased the first EPSC amplitude by 171 33% (P < 0.01; Fig. 4g), without a significant change in the integrated burst charge transfer (P > 0.1). Consequently, thiorphan decreased short-term synaptic facilitation and/or increased short-term depression in excitatory synaptic connections, as estimated by the

reduction in the relative EPSC amplitude within the burst (P < 0.01; Fig. 4h). These results suggest that inhibition of A degradation induces history-dependent enhancement of synaptic transmission, converting excitatory synaptic connections into low-pass filters. A positively regulates ongoing neuronal activity To assess the effect of an A-mediated increase in release probability on ongoing neuronal activity, we recorded the spontaneous neuronal firing rate using perforated whole-cell current-clamp (Fig. 5a), and imaged spontaneous vesicle exocytosis in the absence of external stimulation
Figure 5 Inhibition of A degradation increases ongoing spontaneous activity through a shift in E/I balance. ( a) Representative recordings of spontaneous firing of pyramidal hippocampal neurons before and after thiorphan application. Neurons were kept in current-clamp mode. Scale bars: 20 mV, 3 s. (b) Effect of thiorphan on spontaneous vesicle recycling measured by FM1-43 (loading for 90 s without external stimulation). Scale bar: 2 m. (c) Effects of thiorphan on sEPSC and sIPSC at the level of single pyramidal neurons. Scale bars: 300 pA, 5 s. Neurons were held sequentially at 70 mV (reversal potential for GABA AR-mediated currents) to isolate sEPSCs and at +4 mV (reversal potential for AMPAR-mediated currents) to isolate sIPSCs. (d) Average effect of thiorphan on mean firing rate (FAP: n = 14, P < 0.0001), sEPSC charge transfer (E: n = 6, P < 0.05), sIPSC charge transfer (I: n = 6, P > 0.4), and E/I balance (E/I: n = 6, P < 0.01). (e) Average effect of thiorphan on spontaneous vesicle recycling (n = 12, P < 0.0001 for S). (f) Frequency-current relationship before and after thiorphan application ( n = 9). Right, current clamp recordings for the injected DC current (100 ms, 120 pA) before (black line) and after (gray line) thiorphan application. Scale bars: 20 mV, 40 ms. In all the experiments ( af) the effect of thiorphan (500 nM) was quantified 10 min after its application. Error bars represent s.e.m. *P < 0.05, ** P < 0.01, *** P < 0.001.

Control

Thiorphan

b c
sIPSC sEPSC
Low High

d 500
Th (% control) 400 300 200 100 0

*** *

300 200 100 0

Firing rate (Hz)

Th (% control)

**

e 400
*** **

***

60 40 20 0 0

Cnt Th

FAP E

E/I

F D S

40 80 120 160 I injection (pA)

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

Th

**
1571

+Th

First amplitude (pA)

20

100

a r t ic l e s
Figure 6 Neutralization of [A]o reduces 30 action potentials at 0.2 Hz 1.5 presynaptic activity of hippocampal synapses. r = 0.05 0.40 0.009 200 (a) Representative high-magnification 1.0 F images obtained before and 30 min after application of HJ5.1 antibody (5 g ml1). 100 0.5 Stimulation during FM1-43 staining: 30 action potentials at 0.2 Hz. Scale bar: 2 m. (b) Effect of HJ5.1 at the level of 0 0 0 100 200 100 200 single synapses. Slope of linear fit is Low High FCnt (a.u.) FCnt (a.u.) 0.4 0.01 (for 240 synapses). (c) HJ5.130 action potentials induced presynaptic inhibition does not F 1 Hz at 0.2 Hz 100 125 D depend on the initial presynaptic strength 100 S Cnt (the same data as in b, Spearman r = 0.06, 100 Cnt HJ5.11 75 P > 0.3). (d) Histograms of F before (black HJ5.1 50 75 bars) and after (gray bars) HJ5.1 application in 50 ** ** a single experiment. The median F decreased 50 25 from 102 to 94 arbitrary units (a.u.), and the number of FM+ boutons decreased from 0 25 0 0 200 400 0 200 400 600 267 to 173. (e) Average effect of HJ5.1 on Time (s) F (a.u.) presynaptic strength across population of synapses (n = 6, P < 0.01) was reversible upon 30 min of washout (n = 4). (f) Effect of HJ5.1 on the destaining rate of 264 synapses during 1 Hz stimulation. Destaining rate constants were 0.0044 5 105 and 0.0031 9 106 s1 in control and HJ5.1, respectively. Error bars represent s.e.m. *P < 0.05, ** P < 0.01, *** P < 0.001.

HJ5.1

Number of synapses

Percentage of control

.1

2009 Nature America, Inc. All rights reserved.

using FM1-43 (Fig. 5). Thiorphan application significantly increased the neuronal firing rate by 418 68% from 0.5 0.15 to 2.1 0.4 Hz (n = 14, P < 0.001; Fig. 5) and spontaneous vesicle exocytosis by 308 28% (n = 12, P < 0.0001; Fig. 5e). The observed increase in spontaneous firing rate could be achieved by shifting E/I balance or by increasing the intrinsic excitability of neurons. To assess the effect of thiorphan on E/I balance directly, we isolated the spontaneous excitatory and inhibitory postsynaptic currents (sEPSC and sIPSC, respectively; Fig. 5c) at the same cell, based on the reversal potentials of AMPAR- and GABAAR-mediated currents, respectively. We found that thiorphan

a b
HJ5.1

30 action potentials 30 action potentials at single at 6 bursts Unload 1 Unload 2 Single Burst

Th

increased the sEPSC charge transfer (from 21 4 to 59 15 nC, n = 10, P < 0.05,) without significantly altering the sIPSC charge transfer (51 13 versus 53 15 nC, n = 10, P > 0.7), resulting in an increase of 339 61% in E/I ratio (0.52 0.07 versus 1.6 0.3, n = 10, P < 0.01; Fig. 5d). In the HJ5.1-treated cultures, thiorphan did not affect either sEPSC or sIPSC charge transfer (n = 9, P > 0.2, Supplementary Fig. 8), and subsequently did not change the E/I ratio (n = 9, P > 0.7, Supplementary Fig. 8), suggesting that the thiorphan-induced increase in excitatory drive is mediated by extracellular A peptides. To assess the effect of thiorphan on the intrinsic excitability of neurons, we measured the frequency of action potential firing in response to different levels of depolarizing current injection (FI curve) in the presence of synaptic transmission blockers. On average (n = 9), there was no significant change in the FI curve after acute thiorphan application (P > 0.1; Fig. 5f). Neither the threshold current for action potential generation (69.8 7.9 versus 73.7 8.6 pA, P > 0.6), nor the maximal action potential rate (49.3 6.5 versus 46.3 4.5, P > 0.2) was affected by thiorphan. Input resistance (237.8 24.4 versus 242.6 26.2 M, P > 0.6) and resting membrane potential (71.4 1.0 versus 71.7 1.3 mV, P > 0.4) were not significantly altered. These data suggest that inhibition
Figure 7 Optimal [A]o enables maximal short-term presynaptic facilitation by bursts. (a) Experimental protocol used to determine shortterm plasticity indexed through Sburst/Ssingle (Ssingle: 30 action potentials at 0.2 Hz; Sburst: 30 action potentials in 6 bursts, where each burst contained 5 action potentials; inter-spike interval, 10 ms; inter-burst interval, 5 s). (b) Representative high-magnification F images for single and burst stimulations under control conditions ([Ca 2+]o/[Mg2+]o = 1), in the presence of thiorphan (500 nM), and in the presence of HJ5.1 (5 g ml1, 1 h). Scale bar: 2 m. (c) F, D and S for single versus burst inputs over synapse population in control ( n = 16, P < 0.0001), thiorphan-treated (n = 13) and HJ5.1-treated (n = 10) cultures. (d) The magnitude of short-term plasticity as a function of [A 40+42]o. Sburst/Ssingle under control conditions (n = 16, black symbols); following reduction in [A40+42]o by 65 4% (HJ5.1, 5 g ml1, 30 min; n = 4, P < 0.001, light blue symbol) and by 86 4% (HJ5.1, 5 g ml1, 1 h; n = 4, P < 0.001, dark blue symbol); after a thiorphan-induced increase in [A 40+42]o by 125 15% (50 nM thiorphan; n = 3, P < 0.001, orange symbol) and by 145 8% (500 nM thiorphan; n = 4, P < 0.001, red symbol). Error bars represent s.e.m. *P < 0.05, ** P < 0.01, *** P < 0.001.

Cnt

c
F (a.u.)

HJ5.1 200 100 0 ns

Cnt

***

Th ns

Low

High

d
ns

D (m2)

0.4 0.2

***

ns 3

Cnt Th (50 nM) Th (500 nM)

Sburst / Ssingle

0 60 S (a.u.) 40 20 0 ns

***

ns

1 HJ5.1 (0.5 h) HJ5.1 (1 h) 0 0 50 100 150 [A40+42]o (% control)

Si ng l Bu e rs t

Si ng l Bu e rs t

Si n

gl Bu e rs t

1572

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

H J5

W as

F (%)

F HJ5.1 / FCnt

FHJ5.1 (a.u.)

Cnt

a r t ic l e s a
fEPSP amplitude (mV) +Th 0.4

b
fEPSP slope (V ms1) 200 150 100 50 0 0 20 40 Time (min) 60 0

Control

c
1 fEPSP slope (V ms )

HJ5.1-treated 200 150 100 50 0 0 HJ5.1 +Th 50 100 150 200 Fiber volley (V) Relative amplitude (%)

d
Cntr +Th

e
HJ5.1 +Th

*** * *

***

0.3

Cnt +Th

50 100 150 Fiber volley (V)

200

2009 Nature America, Inc. All rights reserved.

* Figure 8 Acute inhibition of A degradation increases basal synaptic 200 200 +Th +Th ** transmission and decreases short-term synaptic facilitation in CA3-CA1 * 150 connections in acute hippocampal slices. (a) Time course of the thiorphan 150 ** effect and its washout in control slices. (b,c) Effect of 20 M thiorphan on 100 100 input-output relationship between the amplitude of fiber volley and the slope of fEPSP for gradually increasing stimulation intensities in control (b, n = 10, 1 2 3 4 5 1 2 3 4 5 P < 0.05) and in HJ5.1-treated (c, n = 7, P > 0.3) slices. Inset: representative Number of action potentials Number of action potentials recordings of fEPSPs before (black) and 15 min after (gray) application of thiorphan. Scale bars: b, 0.2 mV, 20 ms; c, 0.1 mV, 20 ms. (d,e) Representative recordings of fEPSPs evoked by a burst (each burst contains 5 action potentials; inter-spike interval, 10 ms; inter-burst interval, 30 s) before (black) and 15 min after (gray) application of thiorphan in control ( d, scale bars: 0.2 mV, 20 ms) and in HJ5.1-treated (e, scale bars: 0.1 mV, 20 ms) slices. (f,g) Average relative amplitudes of fEPSPs within the burst, normalized to the first peak. Thiorphan reduced synaptic facilitation during the burst in control ( f, n = 6, P < 0.01), but had no effect in HJ5.1-treated slices (g, n = 6, P > 0.8). Error bars represent s.e.m. *P < 0.05, ** P < 0.01, *** P < 0.001.

Cntr

Relative amplitude (%)

HJ5.1

of A degradation results in over-excitation of the hippocampal network through an increase in the E/I ratio. Endogenous A maintains basal presynaptic activity To assess the involvement of [A]o in tonic regulation of release probability, we tested the effects of the HJ5.1 antibody on basal presynaptic activity (Fig. 6a). Single synapse analysis revealed that a reduction in [A]o (5 g ml1 HJ5.1, 1 h) induced a ~60% decrease in F values across the boutons (Fig. 6b). No correlation was detected between initial F and the degree of presynaptic inhibition (Spearman r = 0.05, P > 0.4; Fig. 6c). Population synapse analysis of the effect of HJ5.1 revealed a decrease in presynaptic activity mainly through reduction in the density of FM+ puncta (Fig. 6d,e). On average, treatment with HJ5.1 reduced presynaptic strength to 53 6% of control (n = 6, P < 0.01; Fig. 6e). This presynaptic suppression was reversible upon 30 min washout of the antibody (n = 4; Fig. 6e). To confirm that a reduction in [A]o below normal levels suppresses synaptic vesicle exocytosis, we tested the effect of the HJ5.1 antibody on the rate of FM1-43 destaining induced by 1-Hz stimulation. Chelation of A by HJ5.1 caused a 30% decrease in the destaining rate constant (264 synapses, n = 3, P < 0.001; Fig. 6f). In addition, a decrease in presynaptic strength was detected following a reduction in A production (0.5 M BACE1-inhibitor IV, S = 48 6% of control, n = 20, P < 0.0001; 0.2 M L-685,458, S = 49 3% of control, n = 13, P < 0.001). These results suggest that extracellular A might be involved in the maintenance of basal synaptic vesicle release in hippocampal synapses. Optimal [A]o enables maximal presynaptic facilitation Although our results indicate that an increase in [A]o reduces shortterm presynaptic facilitation, it is still unclear whether facilitation monotonically relates to [A]o. To address this issue, we determined the magnitude of presynaptic facilitation during bursts over a wide range of [A]o (Fig. 7). The magnitude and the sign of short-term presynaptic plasticity (Sburst /Ssingle) was calculated by dividing the total number of vesicles recycled in response to bursts by the number of vesicles recycled by a similar number of single spikes in the same population of synapses (Fig. 7a). Sburst /Ssingle > 1 reflects short-term
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

facilitation, whereas Sburst /Ssingle < 1 reflects short-term depression. On average, hippocampal synapses exhibited 2.3 0.2 facilitation over the synaptic population at a physiological [Ca]o/[Mg]o ratio (1.2/1.2 mM; n = 16, P < 0.0001; Fig. 7bd), displaying stronger facilitation at low-Pr synapses (n = 4, Spearman r = 0.67, P < 0.0001; Supplementary Fig. 9). Strikingly, facilitation of vesicle release was attenuated at both increased and reduced [A]o (Fig. 7bd). Increasing [A40+42]o by ~25% (thiorphan, 50 nM) resulted in a decrease of synaptic facilitation to 1.4 0.1 (n = 5, P < 0.01; Fig. 7d), whereas further boosting [A40+42]o by ~45% (thiorphan, 500 nM) completely abolished synaptic facilitation (0.9 0.07, n = 13, P < 0.0001; Fig. 7c,d) due to an increase in release probability (Fig. 1). On the other hand, reducing [A40+42]o by ~60% (5 g ml1 HJ5.1, 0.5 h) triggered a slight decrease in presynaptic facilitation compared to control (2.0 0.1, n = 5; P < 0.05; Fig. 7d), whereas further reduction in [A40+42]o by ~86% (5 g ml1 HJ5.1, 1 h) resulted in a marked decrease in Sburst /Ssingle (1.2 0.07, n = 10, P < 0.001; Fig. 7c,d). An increase in [Ca2+]o rescued Sburst /Ssingle in HJ5.1-treated cultures (Supplementary Fig. 9), suggesting that Ca2+-dependent mechanisms underlie the decrease in presynaptic facilitation produced by A chelation. In addition, a substantial reduction in short-term facilitation was detected when A production was reduced by >90% (Supplementary Fig. 9). Together, our data suggest that there is a bell-shaped relationship between short-term facilitation of presynaptic terminals and [A]o, indicating that physiological A levels keep hippocampal synapses in a high-pass filter mode. A regulates synaptic transmission in hippocampal slices Finally, we investigated whether acute inhibition of A degradation by thiorphan affects synaptic transmission in a more intact preparation; namely, in acute hippocampal slices (Fig. 8). Extracellularly recorded field EPSPs (fEPSPs) evoked by low frequency spikes (0.1 Hz) were used to assess the strength of basal synaptic transmission between hippocampal CA3 and CA1 cells (Schaffer collateral-commissural pathway). Acute application of thiorphan induced a reversible increase in fEPSP amplitude ( Fig. 8a ). On average, thiorphan increased the slope of the input (amplitude of
1573

a r t ic l e s
fiber volley)/output (slope of fEPSP) curve (I/O curve) by ~35% (from 0.8 0.015 to 1.1 0.026, n = 10, P < 0.01; Fig. 8b), pointing to an increase in basal synaptic transmission. To test whether an increase in the basal transmitter release could underlie the positive effect of thiorphan on fEPSPs, we measured its effect at a high [Ca 2+] o/[Mg 2+] o ratio. Thiorphan failed to increase the fEPSP at high [Ca 2+] o/[Mg 2+] o (3.6/1.2 mM, 101 2%, n = 7, P > 0.3) compared to its effect at physiological concentration (1.2/1.2 mM; 145 9%, n = 7, P < 0.005). Analysis of the effect of thiorphan on short-term synaptic plasticity during bursts revealed a significant increase in the peak amplitude of the first fEPSP within the burst, but no change in the last one ( Fig. 8d , f ). Thiorphan decreased the ratio of the last peak amplitude to the first peak amplitude by ~25% ( n = 6, P < 0.01; Fig. 8f ), suggesting a reduction in shortterm facilitation due to an increase in basal transmitter release. To determine whether the effect of thiorphan on CA3-CA1 synaptic transmission in acute hippocampal slice was attributable to the action of endogenously released A, we tested the effect of thiorphan in slices pre-incubated with HJ5.1 antibody (5 g ml1, 1 h). Chelation of [A]o by HJ5.1 occluded the effect of thiorphan on the fiber volley/ fEPSP relationship (n = 7, P > 0.3; Fig. 8c). Notably, the I/O slope was significantly reduced after pre-treatment with A antibody (0.51 0.01 versus 0.08 0.015 in control, P < 0.001), suggesting a reduction in basal synaptic transmission after A chelation. Moreover, pretreatment with HJ5.1 occluded the effect of thiorphan on burst-evoked synaptic transmission (n = 6, P > 0.8; Fig. 8e,g). These data confirm that endogenously released A positively regulates basal synaptic transmission in a history-dependent manner, resulting in reduction of short-term facilitation by bursts in the CA3-CA1 pathway in acute hippocampal slices. DISCUSSION Extensive experimental efforts over the past decade have identified the effects of chronic elevation in [A] on the synaptic failure that precedes cognitive decline in transgenic mouse models of Alzheimers disease33. However, the initial steps underlying the very early phases of synaptic dysfunction, as well as the physiological functions of A, remain obscure. In this study we focused on identification of the primary, fast-timescale effects mediated by endogenous A peptides on synaptic activity at the level of individual presynaptic terminals, neuronal connections and synaptic networks. Our results indicate that endogenously released A peptides have a critical role in the regulation of synaptic transfer function under physiological and pathological conditions. A and synaptic transfer function The dependency of synaptic transmission on spike pattern is a universal property of central and peripheral synapses34,35. Input/output relationships in synaptic connections show a high degree of heterogeneity and depend primarily on the basal release probability of synapses3436. Interestingly, there is a rich repertoire of negative endogenous regulators of neurotransmitter release. Presynaptic G-protein-coupled receptors (GPCRs), such as the GABAB, metabotropic glutamate, muscarinic, adenosine and cannabinoid receptors, typically mediate the feedback inhibition of basal neurotransmitter release. This diversity of negative endogenous regulatory pathways might make possible the marked synaptic plasticity in hippocampal networks. Our results reveal that A acts as a positive endogenous modulator of release probability in hippocampal synapses. The acute effects induced by endogenous A peptides in our studies were exclusively presynaptic, without any detectable change in postsynaptic function or intrinsic neuronal
1574

excitability. The enhancement of release probability by A was nonuniform, and negatively correlated to basal presynaptic strength, suggesting that unreliable low-Pr terminals might be more susceptible to [A]o fluctuations. The electrophysiological data indicate that excitatory synapses are highly sensitive to changes in A levels, whereas inhibitory drive remained relatively immune to immediate effects of A. However, future studies are required to determine whether A differentially affects excitatory-excitatory, excitatory-inhibitory, inhibitoryinhibitory, and inhibitory-excitatory synaptic connections. A-mediated presynaptic potentiation, like GPCR-mediated presynaptic inhibition, depends on the history of synaptic activation, reducing its impact at higher firing rates. Several mechanisms might underlie this frequency-dependent phenomenon. First, it might stem from a decrease in A release during high-frequency stimulation. However, [A]o positively correlates to the level of neuronal activity in APP-transgenic6,7 and wild-type8 mice, making this unlikely. Alternatively, the inability of thiorphan to increase presynaptic activity during bursts with long inter-burst intervals might be explained by the short life-time of A in the synaptic cleft. Nevertheless, thiorphan effectively enhanced presynaptic activity for single spikes with interspike intervals equal to the applied inter-burst intervals (Fig. 4b,c). Therefore, a reduction in the responsiveness of presynaptic terminals to A during bursts seems to be the most plausible explanation for the observed phenomenon. Further studies should be conducted to identify the molecular mechanism involved in frequency-dependent enhancement of synaptic vesicle release by endogenous A. An intriguing finding in this study was that short-term synaptic facilitation, a mechanism that is thought to be involved in information processing and memory function37, was impaired not only at increased but also at significantly (>60%) decreased [A]o (Fig. 7d). Although a reduction in synaptic facilitation at increased [A]o might simply occur due to a greater probability of vesicle release, as suggested by the residual calcium hypothesis 34,38, a decrease in facilitation at lowered [A]o could be attributed to the reduced size of the readily releasable pool of vesicles27,32 or reduced Ca2+ cooperativity in triggering vesicle fusion39. Regardless of the precise molecular mechanism, A peptides might maintain release probability in the optimal range, enabling the efficient transfer of temporospatially correlated inputs in hippocampal networks. Rodent A, which differs from human A at three amino acids, is less susceptible to oligomerization. Therefore, it was essential to examine the effects of human endogenously released A on presynaptic strength. Thiorphan triggered a pronounced enhancement of vesicle recycling in neurons expressing APPYFP in App/ cultures, indicating that the presynaptic effects of A do not depend on the primary peptide structure. Presynaptic potentiation was also observed for two main human A isoforms, A1-40 and A1-42 (Supplementary Fig. 5). However, the effective concentrations of ectopically applied synthetic peptides were several orders of magnitude higher than the steady-state endogenous [A]o elevated by neprilysin inhibition, thus seriously impeding the interpretation of these results. Therefore, at this stage, we were unable to identify the isoforms and conformations of the endogenously released A peptides that induce presynaptic enhancement. Although the impairment of synaptic and cognitive functions in Alzheimers disease is widely believed to be triggered by the rise in [A], the relationship between reduced A levels and synaptic state is less clear. Impaired synaptic plasticity, spatial learning and memory have been observed in App/ (refs. 40,41), BACE1/ (ref. 42) and PS1/PS2 conditional knockout43 mice. Furthermore, the injection of small amounts of A42 can enhance long-term synaptic potentiation and memory, in contrast to the synaptic impairments produced by
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

2009 Nature America, Inc. All rights reserved.

a r t ic l e s
injection of high A42 concentrations44. The bell-shaped relationship between [A]o and synaptic plasticity might have several clinical implications. For example, boosting of ongoing neuronal activity induced by inhibition of A degradation might have potential significance for epileptic activity in the early stages of Alzheimers disease45. Furthermore, various pathological conditions associated with the reduction in neuronal activity might lead to decrease in [A ]o and, consequently, impairment of synaptic transmission capability. A recent study in human subjects with acute brain injury provided the first evidence of a correlation between the amount of A in brain interstitial fluid and neurological status46. Finally, the fact that a reduction in [A]o decreases synaptic facilitation should be taken into account when developing potential therapies for Alzheimers disease based on inhibition of A production or neutralization of A. A and synaptic homeostasis Synaptic loss is the best-known correlate of cognitive decline in Alzheimers disease patients 14. Therefore, identification of the cellular mechanisms that trigger the reduction in synapse number is of critical importance. Previous studies have shown that downregulation in the density and strength of APP-overexpressing synapses occurs slowly, over months in vivo47 and days in vitro15. Such time courses are typical of the expression of homeostatic and metaplastic changes in a wide range of synapses. The question then arises: what are the primary versus compensatory changes induced by an increase in [A]o? Having established a negative relationship between A degradation and ongoing neuronal activity on a rapid timescale, we examined whether long-term inhibition of A degradation triggers a reduction in the number of functional presynaptic terminals. Hippocampal cultures were pre-incubated with thiorphan for 48 h, and the density of functional synaptic terminals capable of vesicle recycling was determined by a maximal stimulation protocol (600 action potentials at 20 Hz). The number of FM+ functional synapses obtained by this protocol correlates highly with structural findings28. Our data demonstrated a reduction in the density of functional termi nals, accompanied by an increase in the area of the fluorescent puncta after prolonged thiorphan application ( Supplementary Fig. 10). These results are important for several reasons. They show that long-term inhibition of A degradation leads to synaptic loss, similar to APP overexpression15,16,47. Second, they demonstrate that accumulation of rodent A, displaying a different primary structure and lower amyloidogenic potential than human A, triggers a loss of presynaptic terminals, a prominent feature of the brains of patients with Alzheimers disease13,14. Third, the data suggest that the loss of functional terminals triggers a compensatory increase in the area of the remaining active terminals, similar to the compensatory increase in postsynaptic density size in patients with Alzheimers disease48. Finally, they provide evidence that thiorphan-sensitive A-degrading peptidases, such as neprilysin, might be critical for regulating synaptic density. The increase in neuronal activity triggered by the deficit in A degradation might represent the first step in a cascade that leads to a compensatory reduction in the number of synapses to balance the ongoing activity of the hippocampal network. Given that A overproduction has been implicated in less than 1% of the cases of Alzheimers disease, a deficit in A clearance might contribute to the development of synaptic loss and memory decline in the most common cases (late-onset Alzheimers disease)9. Together, the results of this study suggest that endogenously released A peptides are essential for maintaining synaptic vesicle release in a functional range, optimizing the high-pass filter properties
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

of excitatory hippocampal synapses. Given the remarkable correlation between a high default brain activity in young adults and a spatial pattern of amyloid depositions in elderly individuals with Alzheimers disease49, it is worth examining whether an increase in ongoing neuronal activity and a decrease in synapse capacity for burst transfer might represent a basic feature of the crucial early pathological phase that leads to synapse loss in the common form of late-onset Alzheimers disease. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank A. Lee and M. Wilson for sharing unpublished data on natural spike sequence, D.M. Holtzman for providing the HJ5.1 hybridoma cell line, A.I. Bush, R. Cappai and H. Zheng for App/ mice, C. Kaether for the human APP695-YFP cDNA construct, and the members of our laboratory for comments on the manuscript. We thank A.I. Bush and E. Gazit for discussions. This work was supported by a Rosalinde and Arthur Gilbert Foundation/American Federation for Aging Research research grant (I.S.), the Legacy Heritage Biomedical Program of the Israel Science Foundation (I.S.), the Israel Ministry of Health (I.S.), the National Institute of Psychobiology in Israel founded by the Charles E. Smith family (I.S.), and the Center for Nanoscience and Nanotechnology of Tel Aviv University (I.D.). AUTHOR CONTRIBUTIONS E.A., I.D., H.F., E.R. and I.S. designed, performed and analyzed experiments. H.F. wrote the program for image processing. G.D.C. provided the App/ mice colony. I.S. designed and supervised the project. I.S., E.A. and I.D. wrote the manuscript.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.

2009 Nature America, Inc. All rights reserved.

1. Haass, C. et al. Amyloid -peptide is produced by cultured cells during normal metabolism. Nature 359, 322325 (1992). 2. Seubert, P. et al. Isolation and quantification of soluble Alzheimers -peptide from biological fluids. Nature 359, 325327 (1992). 3. Shoji, M. et al. Production of the Alzheimer amyloid protein by normal proteolytic processing. Science 258, 126129 (1992). 4. De Strooper, B. & Annaert, W. Proteolytic processing and cell biological functions of the amyloid precursor protein. J. Cell Sci. 113, 18571870 (2000). 5. Haass, C. Take fiveBACE and the gamma-secretase quartet conduct Alzheimers amyloid -peptide generation. EMBO J. 23, 483488 (2004). 6. Kamenetz, F. et al. APP processing and synaptic function. Neuron 37, 925937 (2003). 7. Cirrito, J.R. et al. Synaptic activity regulates interstitial fluid amyloid- levels in vivo. Neuron 48, 913922 (2005). 8. Cirrito, J.R. et al. Endocytosis is required for synaptic activity-dependent release of amyloid- in vivo. Neuron 58, 4251 (2008). 9. Iwata, N., Higuchi, M. & Saido, T.C. Metabolism of amyloid- peptide and Alzheimers disease. Pharmacol. Ther. 108, 129148 (2005). 10. Selkoe, D.J. Clearing the brains amyloid cobwebs. Neuron 32, 177180 (2001). 11. Iwata, N. et al. Presynaptic localization of neprilysin contributes to efficient clearance of amyloid- peptide in mouse brain. J. Neurosci. 24, 991998 (2004). 12. Devault, A. et al. Amino acid sequence of rabbit kidney neutral endopeptidase 24.11 (enkephalinase) deduced from a complementary DNA. EMBO J. 6, 13171322 (1987). 13. Terry, R.D. et al. Physical basis of cognitive alterations in Alzheimers disease: synapse loss is the major correlate of cognitive impairment. Ann. Neurol. 30, 572580 (1991). 14. DeKosky, S.T. & Scheff, S.W. Synapse loss in frontal cortex biopsies in Alzheimers disease: correlation with cognitive severity. Ann. Neurol. 27, 457464 (1990). 15. Hsieh, H. et al. AMPAR removal underlies A-induced synaptic depression and dendritic spine loss. Neuron 52, 831843 (2006). 16. Jacobsen, J.S. et al. Early-onset behavioral and synaptic deficits in a mouse model of Alzheimers disease. Proc. Natl. Acad. Sci. USA 103, 51615166 (2006). 17. Shankar, G.M. et al. Natural oligomers of the Alzheimer amyloid- protein induce reversible synapse loss by modulating an NMDA-type glutamate receptor-dependent signaling pathway. J. Neurosci. 27, 28662875 (2007). 18. Shankar, G.M. et al. Amyloid- protein dimers isolated directly from Alzheimers brains impair synaptic plasticity and memory. Nat. Med. 14, 837842 (2008).

1575

a r t ic l e s
19. Lesn, S. et al. A specific amyloid- protein assembly in the brain impairs memory. Nature 440, 352357 (2006). 20. Hartley, D.M. et al. Protofibrillar intermediates of amyloid -protein induce acute electrophysiological changes and progressive neurotoxicity in cortical neurons. J. Neurosci. 19, 88768884 (1999). 21. Ye, C., Walsh, D.M., Selkoe, D.J. & Hartley, D.M. Amyloid -protein induced electrophysiological changes are dependent on aggregation state: N-methyl-daspartate (NMDA) versus non-NMDA receptor/channel activation. Neurosci. Lett. 366, 320325 (2004). 22. Nimmrich, V. et al. Amyloid {} oligomers (A{}142 globulomer) suppress spontaneous synaptic activity by inhibition of P/Q-type calcium currents. J. Neurosci. 28, 788797 (2008). 23. Iwata, N. et al. Identification of the major A142-degrading catabolic pathway in brain parenchyma: suppression leads to biochemical and pathological deposition. Nat. Med. 6, 143150 (2000). 24. Iwata, N. et al. Metabolic regulation of brain A by neprilysin. Science 292, 15501552 (2001). 25. Shirotani, K. et al. Neprilysin degrades both amyloid peptides 140 and 142 most rapidly and efficiently among thiorphan- and phosphoramidon-sensitive endopeptidases. J. Biol. Chem. 276, 2189521901 (2001). 26. Ryan, T.A. et al. The kinetics of synaptic vesicle recycling measured at single presynaptic boutons. Neuron 11, 713724 (1993). 27. Murthy, V.N., Sejnowski, T.J. & Stevens, C.F. Heterogeneous release properties of visualized individual hippocampal synapses. Neuron 18, 599612 (1997). 28. Slutsky, I., Sadeghpour, S., Li, B. & Liu, G. Enhancement of synaptic plasticity through chronically reduced Ca2+ flux during uncorrelated activity. Neuron 44, 835849 (2004). 29. Zakharenko, S.S., Zablow, L. & Siegelbaum, S.A. Visualization of changes in presynaptic function during long-term synaptic plasticity. Nat. Neurosci. 4, 711717 (2001). 30. Zheng, H. et al. -Amyloid precursor protein-deficient mice show reactive gliosis and decreased locomotor activity. Cell 81, 525531 (1995). 31. Debanne, D., Guerineau, N.C., Gahwiler, B.H. & Thompson, S.M. Paired-pulse facilitation and depression at unitary synapses in rat hippocampus: quantal fluctuation affects subsequent release. J. Physiol. (Lond.) 491, 163176 (1996). 32. Dobrunz, L.E. & Stevens, C.F. Heterogeneity of release probability, facilitation, and depletion at central synapses. Neuron 18, 9951008 (1997). 33. Selkoe, D.J. Alzheimers disease is a synaptic failure. Science 298, 789791 (2002). 34. Zucker, R.S. & Regehr, W.G. Short-term synaptic plasticity. Annu. Rev. Physiol. 64, 355405 (2002). 35. Abbott, L.F. & Regehr, W.G. Synaptic computation. Nature 431, 796803 (2004). 36. Lisman, J.E., Raghavachari, S. & Tsien, R.W. The sequence of events that underlie quantal transmission at central glutamatergic synapses. Nat. Rev. Neurosci. 8, 597609 (2007). 37. Lisman, J.E. Bursts as a unit of neural information: making unreliable synapses reliable. Trends Neurosci. 20, 3843 (1997). 38. Katz, B. & Miledi, R. The role of calcium in neuromuscular facilitation. J. Physiol. (Lond.) 195, 481492 (1968). 39. Lou, X., Scheuss, V. & Schneggenburger, R. Allosteric modulation of the presynaptic Ca2+ sensor for vesicle fusion. Nature 435, 497501 (2005). 40. Dawson, G.R. et al. Age-related cognitive deficits, impaired long-term potentiation and reduction in synaptic marker density in mice lacking the -amyloid precursor protein. Neuroscience 90, 113 (1999). 41. Seabrook, G.R. et al. Mechanisms contributing to the deficits in hippocampal synaptic plasticity in mice lacking amyloid precursor protein. Neuropharmacology 38, 349359 (1999). 42. Ohno, M. et al. BACE1 deficiency rescues memory deficits and cholinergic dysfunction in a mouse model of Alzheimers disease. Neuron 41, 2733 (2004). 43. Saura, C.A. et al. Loss of presenilin function causes impairments of memory and synaptic plasticity followed by age-dependent neurodegeneration. Neuron 42, 2336 (2004). 44. Puzzo, D. et al. Picomolar amyloid- positively modulates synaptic plasticity and memory in hippocampus. J. Neurosci. 28, 1453714545 (2008). 45. Palop, J.J. et al. Aberrant excitatory neuronal activity and compensatory remodeling of inhibitory hippocampal circuits in mouse models of Alzheimers disease. Neuron 55, 697711 (2007). 46. Brody, D.L. et al. Amyloid- dynamics correlate with neurological status in the injured human brain. Science 321, 12211224 (2008). 47. Hsia, A.Y. et al. Plaque-independent disruption of neural circuits in Alzheimers disease mouse models. Proc. Natl. Acad. Sci. USA 96, 32283233 (1999). 48. Scheff, S.W., DeKosky, S.T. & Price, D.A. Quantitative assessment of cortical synaptic density in Alzheimers disease. Neurobiol. Aging 11, 2937 (1990). 49. Buckner, R.L. et al. Molecular, structural, and functional characterization of Alzheimers disease: evidence for a relationship between default activity, amyloid, and memory. J. Neurosci. 25, 77097717 (2005).

2009 Nature America, Inc. All rights reserved.

1576

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

ONLINE METHODS
Hippocampal cell culture. Primary cultures of CA3-CA1 hippocampal neurons were prepared from newborn Wistar rats, App/ and App+/+ mice on postnatal days 02, as described28. The generation of the App/ mice has been described30. The experiments were performed in mature (1528 days in vitro (DIV)), highdensity (synaptic density > 1.5 synapses per m2 of dendritic surface area) cultures. The cultures were grown in a culture medium (Gibco, 51200-038) with modified [Mg2+]o (1.2 mM instead of 0.8 mM)28. All animal experiments were approved by the Tel Aviv University Committee on Animal Care. Estimation of synaptic vesicle release based on FM dye staining. Synaptic vesicle release at single synapses was determined by counting the number of presynaptic vesicles turned over by a fixed number of action potentials, using the activitydependent FM dye as a marker27,28. Briefly, action potentials in neurons were initiated by field stimulation (30 action potentials at 0.2 Hz) during dye loading, and the terminals, after undergoing vesicle exocytosis coupled to endocytosis, were stained by FM1-43 (see details in Supplementary Fig. 1). 10 M FM1-43 (or 15 M FM4-64, Fig. 2ad) was present 5 s before and 20 s after the electrical stimulation. During FM loading and unloading, the extracellular solution contained (in mM): NaCl, 145; KCl, 3; glucose, 15; HEPES, 10; MgCl2, 1.2; CaCl2, 1.2; pH adjusted to 7.4 with NaOH. Kynurenic acid (0.5 mM) was added to prevent recurrent activity through blockage of excitatory postsynaptic responses during loading and unloading. After dye loading, external dye was washed away in Ca2+-free solution containing ADVASEP-7 (0.1 mM; Sigma). To confirm that the fluorescent spots corresponded to release sites, we evoked activity at 25 Hz for 4 min during the unloading step to obtain release of dyefilled vesicles. The total amount of releasable fluorescence at each bouton (F) was calculated from the difference between fluorescence after loading and after unloading (F = FloadingFunloading). The release probability (Pr) of individual terminals was calculated as Pr = F/(NAP FQ), where FQ is the estimated releasable fluorescence of a single synaptic vesicle, and NAP is the number of action potentials applied during loading (see Supplementary Fig. 3). Under these conditions, application of 30 action potentials would make it possible to detect functional terminals with Pr > 0.04. The density of active terminals (D; Figs. 17) was calculated as N/A, where N is the number of FM+ terminals per FM image, and A is the imaged area (except chronic thiorphan incubation, Supplementary Fig.10, where A was dendrite area). The total presynaptic strength has been calculated as S = F D. To determine the total presynaptic strength during burst patterns (Sburst), 30 action potentials were delivered in bursts consisting of 5 action potentials at regular frequency (10100 Hz, at inter-burst intervals of 0.25 s) or in natural burst patterns derived from extracellular recordings in the CA1 region of the hippocampus of running rats (provided by A. Lee and M. Wilson, unpublished data). To determine the sign and magnitude of short-term plasticity, we calculated Sburst/Ssingle over the same image area, whereas Ssingle was measured for the loading of 30 action potentials at 0.2 Hz. Functional imaging and analysis. Images were obtained with an Olympus (FV300) confocal laser inverted microscope. The 488-nm line of an argon laser was used for excitation, and the emitted light was filtered using a 510-nm long-pass filter and detected by a photomultiplier. A 60 1.2 NA water-immersion objective was used for imaging. The confocal aperture was partially open and image resolution was 5792 nm per pixel. The gain of the photomultiplier was adjusted to maximize the signal-to-noise ratio without causing saturation by the strongest signals. The image after FM dye unloading was subtracted from the initial image; thus, only those terminals containing activity-dependent releasable FM dye (~90% of total staining) were analyzed. FM+ puncta were selected for further analysis by means of custom scripts written in ImagePro Plus (Media Cybernetics) and MATLAB (Mathworks) programs based on the following criteria: the fluorescence intensity (F) was 2 s.d. above the mean background and the area of puncta was 0.12 m2. The FM+ puncta were detected by applying a binary mask that was created from the F image using adaptive thresholding. To analyze total presynaptic strength over synaptic population (S = F D), we created an individual binary mask for each image. To perform single synapse analysis comparing two images before and after treatment, we created a binary mask using the F image that has the most synapses and applied it to both images. Centers of mass of detected FM+ puncta were compared between two F images to ensure analysis of the same synapses.

Transfections. Transient (1824 h) cDNA transfections of APP695YFP (ref. 50) fusion protein were performed using Lipofectamine-2000 reagents in 1113 DIV cultures. Whole-cell recordings in hippocampal culture. Experiments were performed at room temperature in a recording chamber on the stage of a Zeiss Axiovert S100 microscope. Extracellular Tyrode solution contained (in mM): NaCl, 145; KCl, 3; glucose, 15; HEPES, 10; MgCl2, 1.2; CaCl2, 1.2; pH adjusted to 7.4 with NaOH. For measurement of mEPSCs (Fig. 3) and I/F curve (Fig. 5f), whole-cell patches with low-access resistance (<10 M) were performed by membrane rupture. Whole-cell patches were recorded using the following intracellular solution (in mM): KGluconate, 120; KCl, 3; HEPES, 10; NaCl, 8; CaCl2, 0.5; EGTA, 5; Mg2+-ATP, 2; and GTP, 0.3; pH adjusted to 7.25 with KOH. Serial resistance was not compensated. For mEPSC recordings, tetrodotoxin (TTX; 1 M, Alamon Labs), amino-phosphonopentanoate (AP-5; 50 M, Sigma) and gabazine (30 M, Tocris) were added to the Tyrode solution. Frequency versus current intensity curves were plotted by measuring the average rate of action potentials in current clamp during 100-ms long depolarizing steps of increasing intensity in the presence of synaptic blockers (50 M AP-5, 20 M DNQX and 30 M gabazine). For the experiments shown in Figure 4fh and 5a,c, we obtained perforated patch-clamp recordings. Perforated patch pipettes (23 M resistance) were front-filled with a solution containing (in mM): KGluconate, 130; KCl, 4; HEPES, 10; NaCl, 8, EGTA, 0.4; pH adjusted to 7.2 with KOH and then back-filled with the same solution containing 150200 ng ml1 amphotericin B (Sigma). Dual perforated whole-cell data were recorded from two interconnected cultured hippo campal pyramidal neurons (Fig. 4f). 0.1 M CNQX, blocking EPSCAMPA by 10%, was added to reduce spontaneous network activity in dual-patch recordings. This low concentration did not affect short-term plasticity measurements. The presynaptic cell was held in current-clamp mode, and a 10 ms current pulse (80120 pA) induced action potential firing. The postsynaptic cells were held in voltage-clamp mode at 70 mV. Only neurons with monosynaptic connections were used. The access resistances of both pre- and post-synaptic neurons were monitored online and were typically 720 M. Recordings with access resistances that exceeded 20 M or that varied substantially were excluded from analysis. To measure the excitation/inhibition (E/I) balance (Fig. 5c), sEPSCs and sIPSCs were isolated in the same cell based on reversal potentials of GABAA R-mediated and AMPAR-mediated currents, respectively. For these experiments, perforated patch pipettes were front-filled with a solution containing (in mM): CsOH, 127; D-gluconic acid, 127; CsCl, 4; HEPES, 10; NaCl, 8; EGTA, 0.4; pH adjusted to 7.25 with CsOH and then backfilled with the same solution containing 150220 ng ml1 amphotericin B. Given the intracellular and extracellular solutions used in the present study, the reversal potential for E (VE) was close to +4 mV and VI was close to 70 mV. Membrane holding potentials were corrected following data acquisition for the experimentally determined liquid junction potential (13 2 mV). Signals were recorded using a MultiClamp 700A amplifier, digitized by DigiData1440A (Molecular Devices) at 10 kHz, and filtered at 2 kHz. Electrophysiological data were analyzed using MiniAnalysis (Synaptosoft) for miniature and spontaneous currents or potentials, and in pClamp10 (Molecular Devices) for EPSCs evoked presynaptically. Electrophysiology of hippocampal slices. Coronal hippocampal slices (400 m) were prepared from 1- to 2-month-old Wistar rats in a cold (4 C) storage buffer containing (in mM): sucrose, 206; KCl, 2; MgSO4, 2; NaH2PO4, 1.25; NaHCO3, 26; CaCl2, 1; MgCl2, 1; glucose, 10. The slicing procedure was preformed using a Leica VT1200 vibratome. Slices were transferred to a submerged recovery chamber at room temperature containing oxygenated (95% O2 and 5% CO2) artificial cerebrospinal fluid (ACSF) for 1 h before recording. The ACSF contained, in mM: NaCl, 125; KCl, 2.5; CaCl2, 1.2; MgCl2, 1.2; NaHCO3, 25; NaH2PO4, 1.25; glucose, 25. Experiments were performed at room temperature in a recording chamber on the stage of a Zeiss Axiovert S100 microscope. Extracellular fEPSPs were recorded with a glass pipette containing ACSF (12 M) from the CA1 stratum radiatum, using a MultiClamp700A amplifier (Molecular Devices). Stimulation was evoked in the Schaffer collateral-commissural pathway and delivered through a glass suction electrode (1020 m tip) filled with ACSF. fEPSPs were induced by repetitive stimulations at 0.033 Hz or by bursts (each burst contains 5 action potentials, inter-spike interval 10 ms, inter-burst interval 30 s). The relationship between input (peak amplitude of the fiber volley, 0.033 Hz) and output (fEPSP amplitude

2009 Nature America, Inc. All rights reserved.

doi:10.1038/nn.2433

nature NEUROSCIENCE

and slope) was calculated to estimate basal synaptic transmission. fEPSPs were analyzed by pClamp10 software (Molecular Devices). ELISA of A. Concentrations of rat A40 and A42 in fresh extracellular medium were determined by sandwich ELISA using highly sensitive commercial kits (Wako) according to the manufacturers instructions. BNT77 was used as a capture antibody, BA27 as antibody for A40, and BC05 for A42. Only fresh samples were used. In each experiment, media from four coverslips were mixed and analyzed. Chemical reagents. FM1-43 and Advasep-7 were purchased from Biotium, thiorphan and AP-5 from Sigma, 4G8 from Signet, AF1126 from R & D Systems, BACE1 inhibitor IV and L-685,458 from Calbiochem, A1-40 and A1-42 from Bachem, TTX from Alamon Labs, and gabazine and CNQX from Tocris. Thiorphan solution was stored at 1 mM in ACSF solution containing 1 mM ascorbic acid to prevent thiorphan oxidation23. Equal amounts of ascorbic acid were added to control samples before thiorphan was applied.

Statistical analysis. Error bars shown in the figures represent s.e.m. (s.e.m.). Students paired t-tests were used in all the experiments where the effect of thiorphan (or other compounds) was tested in the same cell/synapse (*P < 0.05; **P < 0.01; ***P < 0.001). Unpaired t-tests were used to compare different populations of synapses. A one-way ANOWA Kruskal-Wallis non-parametric test was used to compare several populations of synapses. The nonparametric Spearman test was used for correlation analysis. For comparison of mEPSC amplitude or frequency under different conditions, 200 mEPSCs were randomly selected for each cell and pooled for each condition. A Kolmogorov-Smirnov (K-S) test was used to compute differences in mEPSC amplitude and frequency across the pooled datasets.

50. Kaether, C., Skehel, P. & Dotti, C.G. Axonal membrane proteins are transported in distinct carriers: a two-color video microscopy study in cultured hippocampal neurons. Mol. Biol. Cell 11, 12131224 (2000).

2009 Nature America, Inc. All rights reserved.


nature NEUROSCIENCE

doi:10.1038/nn.2433

a r t ic l e s

Input normalization by global feedforward inhibition expands cortical dynamic range


Frdric Pouille1,2, Antonia Marin-Burgin1,2, Hillel Adesnik1, Bassam V Atallah1 & Massimo Scanziani1
The cortex is sensitive to weak stimuli, but responds to stronger inputs without saturating. The mechanisms that enable this wide range of operation are not fully understood. We found that the amplitude of excitatory synaptic currents necessary to fire rodent pyramidal cells, the threshold excitatory current, increased with stimulus strength. Consequently, the relative contribution of individual afferents in firing a neuron was inversely proportional to the total number of active afferents. Feedforward inhibition, acting homogeneously across pyramidal cells, ensured that threshold excitatory currents increased with stimulus strength. In contrast, heterogeneities in the distribution of excitatory currents in the neuronal population determined the specific set of pyramidal cells recruited. Together, these mechanisms expand the range of afferent input strengths that neuronal populations can represent. A characteristic of cortical excitatory neurons is their widely divergent axonal projection. This property enables cortical neurons to contact a large number of postsynaptic cells and allows each postsynaptic cell to receive inputs from many presynaptic neurons. In a circuit constructed with this excitatory divergence alone, the number of active presynaptic neurons (input strength) that is sufficient to recruit all neurons in the postsynaptic population is only slightly larger than the input strength required to recruit any postsynaptic neuron at all. In other words, the input range that can be faithfully represented by the postsynaptic population is restricted. For example, if presynaptic neurons connect to a postsynaptic population with a probability of 15%1 and each postsynaptic cell requires 40 active inputs to be recruited2, then 2% of the postsynaptic cells would be recruited by the activity of 200 presynaptic neurons and almost all (>99%) would be recruited by simply doubling the number of active presynaptic neurons (as determined by binomial statistics). Thus, in the absence of control mechanisms, small fluctuations in the fraction of presynaptically active neurons results in all-or-none recruitment of the postsynaptic population36 (this all-or-none behavior is qualitatively similar for a wide range of connectivity values and number of inputs necessary to reach threshold). However, both spontaneous and sensory-evoked cortical activity involves large fluctuations in the fraction of active neurons (for example, refs. 79). What mechanisms does the cortex use to expand the range of input strengths over which it faithfully responds? One could imagine at least two distinct mechanisms. Reducing the gain of individual neurons (that is, the change in spiking probability as a function of input strength) would allow each neuron in the population to respond over a wider range of input strengths; this gain modulation could be achieved through GABAA receptormediated conductances1013. Alternatively, staggering the recruitment of individual neurons over a wide range of input strengths would allow the population as a whole, rather than
1Howard

2009 Nature America, Inc. All rights reserved.

individual neurons, to represent a wider input range. This could be achieved by varying the amplitude of the excitatory postsynaptic currents (EPSCs) necessary to reach threshold for spike generation as a function of input strength. We found that hippocampal and neocortical feedforward inhibitory circuits staggered the recruitment of individual pyramidal cells over a wide range of input strengths. Feedforward inhibition (FFI) acted homogeneously across the postsynaptic population of pyramidal cells to rapidly adjust their excitability to the strength of incoming presynaptic activity. As a result, the amplitude of the EPSC necessary for a pyramidal cell to reach spike threshold was dynamic and varied with the strength of the input. Heterogeneities in the amplitudes of EPSCs across the postsynaptic population determined the specific subset of pyramidal cells that would spike in response to the presynaptic input. Through this coordinated action of direct excitation and FFI, pyramidal cell populations can remain sensitive to weak inputs, but will not saturate in response to stronger activity. RESULTS EPSC necessary to spike pyramidal cell is dynamic We established the range of stimulus strengths over which the CA1 pyramidal cell population responds, that is, the dynamic range. We recorded from individual pyramidal cells in the loose-patch configuration and stimulated Schaffer collaterals over a range of intensities, from those that failed to trigger any spike to those that triggered spikes on every trial (Fig. 1a). The relationship between spiking probability of individual pyramidal cells and input strength (input strength is proportional to the number of activated Schaffer collateral; for details see Online Methods and Supplementary Fig. 1) was fitted with a sigmoid to interpolate the threshold input strength, where pyramidal cells spiked in 50% of the trials (Fig. 1a). The cumulative distribution of threshold input strengths for all recorded pyramidal cells

Hughes Medical Institute and Neurobiology Section, Division of Biology, University of California San Diego, La Jolla, California, USA. 2These authors contributed equally to this work. Correspondence should be addressed to M.S. (massimo@ucsd.edu). Received 5 August; accepted 25 September; published online 1 November 2009; doi:10.1038/nn.2441

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1577

a r t ic l e s
Figure 1 The stronger the stimulus, the larger the excitation necessary to recruit a pyramidal cell. (a) Left, spiking probability plotted against input strength for one CA1 pyramidal cell (PC, sigmoidal fit, dashed lines indicate 50% spiking probability). Inset, loose-patch recording at two different input strengths, five consecutive sweeps. Successes are shown in black and failures in gray. Right, black data points represent the activation curve (that is, the cumulative distribution of input strengths eliciting 50% spiking, n = 39). Dashed lines represent the input strengths recruiting 5% and 95% of the pyramidal cell population. Gray sigmoids indicate the spiking probability of the 39 pyramidal cells making up the activation curve. Their recruitment was staggered along the range of input strengths. Red sigmoid indicates the experiment shown on the left. ( b) Top left, recording configuration. Rec, recording electrode. Top traces represent the responses of two CA1 pyramidal cells simultaneously recorded in loose-patch to threshold stimulation of Schaffer collateral inputs (five superimposed sweeps; successes are shown in black, failures in gray). The pyramidal cell on the left was recruited at weaker stimulus than the pyramidal cell on the right. Bottom traces represent threshold EPSCs (that is, EPSCs evoked at threshold input strength, average of ten traces) recorded in the same two cells voltage clamped at 80 mV. The pyramidal cell on the left necessitated less excitation to reach threshold. Right, summary graph of EPSGTs (black, n = 32, spike threshold determined in loose patch for n = 15 cells and in whole-cell current clamp for n = 17 cells) plotted against input strength at threshold (bin width 0.1). Dashed line represents the average EPSGT for the 00.1 bin. Error bars are s.e.m. (c) Top left, recording configuration. Top traces represent the response of a single CA1 pyramidal cell recorded in loose patch to threshold stimulation of two different Schaffer collateral inputs (stimuli a and b, five superimposed sweeps each). Bottom traces represent threshold EPSCs (average of ten traces) recorded in the same cell voltage clamped at 90 mV. The difference in amplitude of the two threshold EPSCs should be noted. Right, summary graph (n = 19). There was no correlation between EPSGTs evoked by input a and input b (linear regression, R2 = 0.07; spike threshold determined in loose patch for all cells, red data point indicates the experiment shown on the left).

a 100
Spiking probability (%)

Single PC

100 95 Recruited PCs (%)

Population activation curve

100 Spiking probability (%)

50

50

50

30 pA 5 ms 0 0 0.2 0.4 0.6 0.8 Input strength Stimulus Rec 1.0

5 0 0 0.2 0.4 0.6 0.8 Input strength 1.0

Rec

20

PC

PC

Loose patch Weak input Strong input 50 pA

EPSGT (nS)

10

Whole cell

2009 Nature America, Inc. All rights reserved.

500 pA 50 ms

0.1

0.2 0.3 0.4 Input strength EPSGT

0.5

c
Stimulus a
PC

Rec 20 Stimulus b Input b (nS)

Loose patch Input a (0.43) Input b (0.66) 250 pA Whole cell

10

500 pA 50 ms

0 0

R 2 = 0.07 10 Input a (nS) 20

(n = 39) represents the fractional recruitment of the CA1 pyramidal cell population, or activation curve (Fig. 1a). The dynamic range of the pyramidal cell population (that is, the ratio of the input strength necessary to activate 95% versus 5% of the pyramidal cell population) was approximately 34 (Fig. 1a), meaning that the pyramidal cell population can differentially represent a 34-fold increase in the number of active Schaffer collateral inputs before saturating. This is much larger than the dynamic range of an individual pyramidal cell (1.6 0.7, n = 37; Fig. 1a; invariant between pyramidal cells recruited at different input strengths, R2 = 0.034, P = 0.27; Supplementary Fig. 2) and is the result of staggered recruitment of CA1 pyramidal cells over a wide range of stimulus strengths (Fig. 1a). Why are some pyramidal cells recruited at low input strength, whereas others require much stronger stimuli? We compared the excitatory postsynaptic conductance (EPSG) evoked at threshold input strength2 of pyramidal cells recruited over the range of input strengths (EPSGT refers to the EPSG evoked at threshold). Figure 1b illustrates an example of two pyramidal cells, simultaneously recorded in the loose-patch configuration, that required different stimulus strengths to spike. Whole-cell, voltage-clamp recording from the same two cells showed that the EPSGT in the pyramidal cell recruited by the stronger stimulus was much larger than in the pyramidal cell recruited with weaker stimulus (Fig. 1b). Over all of the experiments, we observed a steep increase in EPSGTs with increasing input strength (0.3-nS increase per percentile input strength, n = 32, P = 0.0024; Fig. 1b). The increase in EPSGT with input strength was not unique to pyramidal cells recruited by single-shock stimulation of the Schaffer collateral. Even when pairs of pyramidal cells were recruited by repetitive high-frequency stimulation (26 stimuli at 0.20.5 kHz), mimicking bursting activity in CA3, as recorded in vivo14, EPSGTs were
1578

significantly larger in the cell recruited with higher input strength (1.7 0.2fold larger, n = 12 pairs, P = 0.012; Supplementary Fig. 3). This held true for even higher stimulus frequencies (46 stimuli at 1 kHz, 1.6 0.2fold larger, n = 6 pairs, P = 0.045; Supplementary Fig. 3). Thus, pyramidal cells recruited at higher input strengths need larger EPSGs to reach spike threshold. Are differences in EPSGT amplitudes the results of variability in intrinsic pyramidal cell properties? Input resistance, membrane time constant, resting potential and threshold potential did not significantly differ between pyramidal cells recruited at different input strengths (Supplementary Fig. 4). To further rule out the influence of intrinsic variability between pyramidal cells, we compared EPSGTs between two independent Schaffer collateral inputs converging onto a single pyramidal cell (Fig. 1c). EPSGTs were uncorrelated between the two inputs (Fig. 1c). Furthermore, in an individual pyramidal cell, the EPSGT evoked by the stronger input was invariably larger than the EPSGT evoked by the weaker input (1.5 0.2fold larger; P = 0.002, n = 19; Fig. 1c). Finally, there was no significant difference in the rise and decay kinetics of EPSGTs evoked by the weak and strong inputs (1090% rise time: strong stimulus, 2.8 0.2 ms; weak stimulus, 2.6 0.3 ms; P = 0.67, n = 19; decay time constant: strong stimulus, 7.0 0.4 ms; weak stimulus, 6.8 0.4 ms; P = 0.74, n = 19), ruling out differences resulting from the distribution of the excitatory inputs along the somatodendritic axis. Thus, even in an individual pyramidal cell, the EPSGT varied depending on the activated input, indicating that the same pyramidal cell can be recruited at both the low or high end of the stimulus range. The increase in EPSGT implies that the contribution of each individual afferent in firing the neuron decreases with increasing input strength. By how much does this decrease? Over the range of
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
input strengths from 0 to 0.5, the amplitude of the EPSGT increased approximately linearly (Fig. 1b; see model below) such that gabazine eliminated the increase in EPSGT with input strength (nonsignificant increase of 0.07 nS per percentile input strength, n = 30, P = 0.22; Fig. 2b), demonstrating a crucial role of GABAA receptors. To impose a fixed amount of inhibition, irrespective of input strength (Fig. 2b), we first inhibited GABA release with the -opioid receptor agonist DAMGO (0.51 M, 77.9 7.3% reduction, n = 4; Supplementary Fig. 6)18 and then produced a tonic activation of GABAA receptors by perfusing the selective agonist muscimol (1 M; average hyperpolarization, 2.4 0.5 mV; average conductance increase, 4.7 0.3 nS; n = 4). In the presence of tonic inhibition, EPSGT no longer varied with input strength (nonsignificant decrease of 0.003 nS per percentile input strength, n = 14, P = 0.4; Fig. 2b). Thus, the progressive increase in GABAA receptor activation accounts for the increase in EPSGT. This is supported by the fact that at high input strengths (>0.5), when the amplitude of FFI no longer increased (Fig. 2a), EPSGT remained constant (nonsignificant decrease of 0.02 nS per percentile input strength, P = 0.4; Supplementary Fig. 7). By how much does the dynamic EPSGT increase the range of inputs that the pyramidal cell population responds to? We compared the activation curve of the CA1 pyramidal cell population under control conditions and in the presence of gabazine, where the EPSGT is fixed (Fig. 2c). The number of Schaffer collaterals necessary to recruit the lowest fractions of the pyramidal cell population was comparable in both conditions (for example, 5% recruitment: control, 0.028 input strength; gabazine, 0.03 input strength; Fig. 2c). The situation was, however, radically different when larger numbers of Schaffer collaterals were activated. In the presence of gabazine, an approximately eightfold increase in the number of activated Schaffer collaterals readily led to the saturation of the pyramidal cell population (95% recruitment with 0.26 input strength), whereas the same increase in stimulated

EPSGTN = Nk + EPSGT0

where N is the number of active afferents, EPSGTN is the EPSGT when N afferents are active, EPSGT0 is the EPSG necessary to reach threshold at minimal input strength (under our condition, it was ~6 nS; Fig. 1b) and k is the proportionality factor. Given g, the synaptic conductance produced by an individual afferent, the relative contribution of each afferent toward firing a cell, , is . EPSGTN (Nk + EPSGT0 ) Thus, the relative contribution of individual afferents in firing a cell is normalized by the number of active afferents. FFI expands populations dynamic range What determines the amplitude of the EPSGT and why does it vary with input strength? Stimulation of Schaffer collaterals triggers powerful FFI in CA1 pyramidal cells through the recruitment of GABAergic interneurons1517. There was a strong correlation between the amplitude of the EPSGT and the amplitude of the concomitantly triggered feedforward inhibitory postsynaptic conductance (IPSG; Fig. 2a; see Online Methods and Supplementary Fig. 5). Furthermore, consistent with the correlation between EPSGT and input strength (Fig. 1b), FFI increased with input strength, before saturating at input values above ~0.5 (Fig. 2a). These data suggest that EPSGT may vary with input strength because of a parallel increase of FFI. We directly tested this possibility by either abolishing GABAergic transmission or by imposing a fixed amount of inhibition (Fig. 2b). Abolishing FFI with the GABAA receptor antagonist
Figure 2 Feedforward inhibition expands the dynamic range of the pyramidal cell population. (a) Top traces represent whole-cell current-clamp recording from two CA1 pyramidal cells recruited at threshold by weak (left) or strong (right) Schaffer collateral stimulation (five superimposed sweeps; black indicates successes and gray indicates failures to trigger a spike). Bottom traces, represent threshold EPSC (black, average of five traces recorded in the voltage clamp, 88 and 92 mV for left and right, respectively) and concomitantly evoked feedforward IPSC (blue, recorded at 52 and 59 mV for left and right, respectively, and isolated by subtraction from average of ten sweeps). Insets represent expanded timescale of the sweeps. The size of the two insets has been scaled to match EPSC amplitudes. Bottom left, threshold feedforward IPSG (IPSGT) plotted against EPSGT (bin width of 2.5 nS, n = 30, spike threshold determined in loose patch for n = 19 cells and in whole-cell current clamp for n = 11 cells, dotted line is the linear regression fit of the binned data, R2 = 0.61, slope of 0.82). Bottom right, feedforward IPSG plotted against input strength (bin width of 0.1, n = 50, continuous blue line is a Boltzmann fit of the binned data). Error bars are s.e.m. (b) Summary graph of EPSGTs plotted against input strength in the presence of gabazine (6 M, n = 30, spike threshold determined in loose patch for n = 20 cells and in whole-cell current clamp for n = 10 cells) or under tonic inhibition (1 M muscimol and 0.51 M DAMGO, n = 14, spike threshold determined in loose-patch for n = 11 cells and in whole-cell current clamp for n = 3 cells). Dotted and dashed horizontal lines represent the average EPSGT during tonic inhibition or gabazine treatment, respectively. In contrast with control conditions (black line from Fig. 1b), the EPSGT recorded in gabazine or tonic inhibition changed little with increasing input strength. For all input strengths, the EPSG T during tonic inhibition was larger than during gabazine treatment. (c) Activation curves (cumulative distribution of input strengths eliciting 50% spiking) in control conditions (black symbols from Fig. 1a) and after gabazine treatment (n = 28, spike threshold determined in loose patch for all cells). Dashed lines indicate input strengths recruiting 5% and 95% of the pyramidal cell population. Error bars are s.e.m.

2009 Nature America, Inc. All rights reserved.

Threshold input strength: 0.03 Current-clamp

Threshold input strength: 0.7 Current-clamp 50 mV

Voltage-clamp

Voltage-clamp 6 ms 250 pA 500 pA 20 ms

30

30

IPSGT (nS)

IPSG (nS) R = 0.61


2

20

20

10

10

0 0 10 20 EPSGT (nS) 30

0 0 0.50 Input strength Activation curves 1.00

b
EPSGT (nS)

20 Gabazine Tonic inhibition Control 10

c 100
95 Recruited PCs (%)

50

5 0 0.1 0.2 0.3 0.4 Input strength 0.5 0 0.2

Control Gabazine 0.4 0.6 0.8 Input strength 1.0

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1579

a r t ic l e s
Figure 3 Pyramidal cells spike after the onset of feedforward inhibition. Stimulus 15 (a) Top traces are loose patch recordings from a CA1 pyramidal cell Rec Rec in response to threshold stimulation of Schaffer collaterals (five Loose patch 2 superimposed sweeps). Bottom traces are voltage-clamp recordings from PC 3 PC the same neuron; the EPSC (black line, average of five traces) was 10 40 pA recorded at 85 mV and the feedforward IPSC (blue line, isolated t Whole cell 2 by subtraction from average of ten traces) was recorded at 60 mV. The 1 vertical dashed lines mark the onset of the IPSC and the average timing 5 1 nA of the spike in the pyramidal cell. The IPSC onset occurred before the 1 1 ms spiking of the pyramidal cell. Bottom, summary of 30 similar experiments (spike threshold determined in loose patch for n = 18 cells and in whole0 0 0 00.24 0.25 0 2 4 6 8 cell current clamp for n = 12 cells). The open symbol represents the Spiking Non0.50 t (ms) average. (b) The net threshold charge (EPSC minus IPSC) entering the cell spiking Input strength from the onset of the EPSC to the time of the spike and the EPSG T are shown for two different ranges of threshold input strengths. The threshold charge did not increase significantly with increasing input strength (P = 0.3). In contrast, the peak conductance of the EPSG T recorded in the same cells was significantly larger for larger input strengths ( P = 0.03). Error bars represent s.e.m. (c) Simultaneous recording from two neighboring pyramidal cells in which Schaffer collaterals stimulation was sufficiently strong to reach threshold in one cell (black), but not in the other (gray, n = 15 pairs). The net threshold charge entering in the spiking pyramidal cells was significantly larger than the net charge entering the nonspiking pyramidal cells ( P = 0.004). Circles represent individual experiments and horizontal lines represent averages.
Threshold charge (pC) EPSGT (nS)

Charge (pC)

2009 Nature America, Inc. All rights reserved.

Schaffer collaterals in control conditions recruited only 19.9% of the population (Fig. 2c). Thus, gabazine increased the slope of the activation curve without producing major changes in the offset (Fig. 2c). Hence, a dynamic EPSGT leads to a fourfold expansion of the range of inputs that the CA1 pyramidal cell population can respond to. FFI arrives before spike By what mechanism does the feedforward inhibitory postsynaptic current (IPSC) control the size of the EPSGT? We compared the timing of the spike elicited in pyramidal cells by Schaffer collateral stimulation with the onset of the feedforward IPSC. When stimulated at threshold for spike generation, the spike occurred 5.0 0.4 ms after the onset of the EPSC and 3.3 0.4 ms after the onset of the feedforward IPSC (n = 30; Fig. 3a). The latency between the onset of the EPSC and of the IPSC was 1.65 0.08 ms (n = 30), consistent with previous data15, and did not change with stimulus strength (R2 = 0.007, P = 0.4, n = 30). Thus, in response to threshold Schaffer collateral stimulation, FFI reached pyramidal cells before the membrane potential of the neuron reached threshold for spike generation. Over the period preceding the spike, synaptic inhibition overlapped with the EPSC, thereby reducing the excitatory charge entering the cell by 28.1 4.3% (n = 30). Specifically, although the integral of the EPSC from its onset to the time of the spike (excitatory charge) averaged 2.2 0.2 pC (n = 30), the net synaptic charge (excitatory-inhibitory charge, see Online Methods) entering pyramidal cells was 1.4 0.1 pC (n = 30). In contrast to the EPSGT, this net threshold charge was constant and independent of input strength (threshold charge, 1.1 0.1 pC, (n = 4) at 00.25 input strength versus 1.4 0.1 pC, (n = 14) at 0.250.5, P = 0.3; EPSGT, 7.1 0.9 nS (n = 4) at 00.25 input strength versus 13.1 1.3 nS (n = 14) at 0.250.5, P = 0.03; Fig. 3b). Furthermore, at any given input strength, the threshold charge was significantly larger in pyramidal cells that reached threshold for spike generation as compared with the charge entering over the same time interval in simultaneously recorded cells that did not reach threshold (0.8 0.1 pC, P = 0.004, n = 15; Fig. 3c). In the cells that did not spike, the threshold charge (that is, ~1.5 pC) would have been reached 5.0 0.3 ms after the onset of the EPSC if inhibition had not been present. Thus, by overlapping with excitation before spike occurrence, FFI controls the amplitude of the EPSC necessary to reach spike threshold. Heterogeneous excitation and homogeneous inhibition What determines which pyramidal cells in the population are recruited in response to Schaffer collateral stimulation? We recorded from
1580

two neighboring pyramidal cells simultaneously (somata separated by 50 m) and increased the number of activated Schaffer collaterals until one of the two cells spiked (Fig. 4a). We then compared the EPSGs and feedforward IPSGs in the two cells. Although the EPSG was, on average, 1.6 0.1fold larger in the cell that spiked (P = 0.001, n = 15; Fig. 4b), the IPSG was, on average, not significantly different between the two neurons (1.1 0.1fold difference, P = 0.3, n = 15; Fig. 4b). Furthermore, the latency of the feedforward IPSC (with respect to the onset of the EPSC) did not differ significantly between spiking (1.75 0.09 ms) and nonspiking neurons (1.59 0.06 ms, P = 0.09, n = 15). Thus, differences in the amplitude of synaptic excitation, rather than in the amplitude or timing of inhibition, govern which neuron will spike in response to Schaffer collateral stimulation. To determine whether inhibition is more homogeneously distributed across pyramidal cells as compared with excitation, we computed the spread, that is, the absolute difference in amplitude of simultaneously recorded EPSGs or IPSGs normalized by the average of the amplitudes and divided by two. Although the spread of EPSGs between two simultaneously recorded pyramidal cells was 21 3% (n = 15, same paired values as above; Fig. 4b), the spread of the concomitant IPSGs was only 11 2% (P = 0.03, n = 15). We also calculated how well the amplitude of inhibition in one cell correlated with the amplitude of inhibition in its neighbor and did the same for excitation. For this, we used the same paired values as described above (Fig. 4b), but we randomly allocated the spiking cell to either one of the two axes (Fig. 4b). This randomization removes the correlation bias caused by systematically having the larger amplitude on the same axis. The correlation between IPSGs (RIn = 0.79) was significantly larger than the correlation between EPSGs (REx = 0.30, P < 0.02, see Online Methods; Fig. 4b). Thus, inhibition is more homogeneous than excitation across the pyramidal cell population. To test whether the relative homogeneity of inhibition with respect to excitation also holds true for individual synaptic events, we compared trial-by-trial fluctuations of the amplitude of EPSC and feedforward IPSC between two simultaneously recorded pyramidal cells (Fig. 4c). Using a cesium-based internal solution, we isolated feedforward IPSCs and monosynaptic EPSCs by voltage clamping the cell at the EPSC or IPSC reversal potential, respectively (Fig. 4c). The amplitude of the feedforward IPSC covaried between the two recorded neurons (average correlation, R2 = 0.26 0.06, n = 5; Fig. 4c). This correlation was significantly less pronounced for monosynaptic EPSCs (average correlation, R2 = 0.06 0.005, n = 5, P = 0.033; Fig. 4c). These results indicate that, although FFI is
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 4 Homogeneous inhibition and heterogeneous excitation control the recruitment of pyramidal cells. (a) Top, recording configuration. Left traces represent simultaneous loose-patch recording from two neighboring CA1 pyramidal cells. Schaffer collaterals stimulation was sufficiently strong to reach threshold in one cell (black traces, five superimposed sweeps), but not in the other (gray traces). Right traces represent wholecell voltage-clamp recording from the same two neurons (top, feedforward IPSCs recorded at 60 mV and isolated by subtraction from average of ten traces; bottom, EPSCs recorded at 80 mV, average of ten traces). The amplitude of the feedforward IPSC was similar in both cells, whereas the EPSC was larger in the cell that spiked. (b) Left, summary graph of 15 similar experiments in which the EPSG in the nonspiking cell is plotted against the EPSG in the spiking cell. The majority of the data points are below the unity line (red data point indicates the experiment shown on top). Right, summary graph of the same 15 experiments in which the feedforward IPSG in the nonspiking cell is plotted against the feedforward IPSG in the spiking cell. In contrast with the EPSG, all of the data points are scattered around the unity line (spike threshold determined in loose patch for n = 8 pairs and in whole-cell current clamp for n = 7 pairs; red data point indicates experiment shown in a; same set of experiments illustrated in Fig. 3c). Error bars are s.e.m. Insets have the same data points as are shown in the main graphs, but the spiking cell is randomly allocated to either one of the two axes. Note the larger spread of EPSGs as compared with IPSGs. (c) Trial-by-trial fluctuation of EPSGs (left) and IPSGs (right) simultaneously recorded in two pyramidal cells (PC 1 and PC 2) voltage clamped at the reversal potential of IPSCs (left) or EPSCs (right, same Schaffer collateral stimulation intensity for both holding potentials, cesium internal). Left, single-trial EPSGs recorded in PC 1 are plotted against the EPSGs recorded simultaneously in PC 2. Upper traces are five example EPSCs recorded in PC 1 ordered according to amplitude. Lower traces are the corresponding five EPSCs recorded simultaneously in PC 2. Right, single-trial IPSGs recorded in PC 1 are plotted against the IPSGs recorded simultaneously in PC 2. Upper traces are five example IPSCs recorded in PC 1 ordered according to amplitude. Lower traces are the corresponding five IPSCs recorded simultaneously in PC 2. Note the marked covariation in amplitude of IPSGs recorded in the two pyramidal cells as compared with EPSGs.

a
Loose patch Spiking PC Nonspiking PC

Rec

Stimulus

Rec
PC

PC

Whole cell IPSC

50 pA 5 ms

EPSC

500 pA 25 ms

b
Nonspiking PC (nS)

EPSGT

40 30 20 10 0

IPSGT

10

Nonspiking PC (nS)

20

2009 Nature America, Inc. All rights reserved.

10 20 Spiking PC (nS) EPSG 250 pA 5 ms

10 20 30 Spiking PC (nS) IPSG 1 nA 10 ms

40

c 10
PC 1 (nS)

25 20
PC 1 (nS)

15 10 5

500pA 0 0 5 PC 2 (nS) 10 0 0 5 10 15 PC 2 (nS) 20 25

relatively homogenous across the pyramidal cell population, and thus sets a global threshold for pyramidal cell recruitment by Schaffer collaterals, heterogeneities in excitation determine which pyramidal cells in the population overcome this threshold. Basket cells expand dynamic range of pyramidal cell population Several types of hippocampal inhibitory interneurons are activated in a feedforward manner1922. What type of interneuron controls the amplitude of the EPSGT in pyramidal cells? Because the onset of FFI occurs before the spiking of pyramidal cells (see above; Fig. 3a), these interneurons must spike before pyramidal cells in response to afferent stimulation. We compared the spike timing of different types of interneurons in response to Schaffer collateral stimulation with the timing of the peak of the population spike (recorded in the stratum pyramidale; Fig. 5a,b). Pyramidal cells fired simultaneously with the population spike (0.2 0.2 ms, P = 0.52, n = 21; Fig. 5a,b), whereas regular-spiking interneurons19,20 were recruited after the population spike (0.7 0.3 ms, P = 0.04, n = 34). In contrast, fast-spiking inhibitory interneurons19,20 fired 1 0.2 ms before the population spike (P = 0.0003, n = 18; Fig. 5a,b), consistent with the early onset of FFI. We computed the activation curve of fast-spiking and regularspiking interneurons (Fig. 5c), as we did for pyramidal cells (Fig. 1a). The activation curve of fast-spiking interneurons was much steeper that the one for regular-spiking interneurons and pyramidal cells (Fig. 5c), consistent with the strong and fast excitation received by fast-spiking interneurons21,23,24 (half maximal recruitment of fastspiking interneurons occurred at an input strength of 0.11 compared with 0.37 and 0.4 for regular-spiking interneurons and pyramidal cells,
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

respectively). Notably, the activation curve of fast-spiking interneurons (Fig. 5c) provided a good match for the increase in the amplitude of FFI over the same range of stimuli (half maximal recruitment of FFI occurred at an input strength of 0.13; Fig. 2a). Anatomical identification of these fast-spiking interneurons revealed that 63% were basket cells (for axonal and dendritic distribution, see Supplementary Fig. 8) and the rest were composed of dendrite targeting interneurons exhibiting axonal arborization consistent with bi-stratified and tri-laminar cells19,20 (n = 11 fast-spiking cells; Fig. 5a). Thus, these data indicate that the dynamic EPSGT in CA1 pyramidal cells is primarily enforced by fast-spiking interneurons, the majority of which are basket cells. Model Is the observed change in EPSGT sufficient to account for the expansion of the range of inputs the pyramidal cell population responds to? We created a simple quantitative model of Schaffer collateral excitation onto a population of CA1 pyramidal cells (Fig. 6; see Online Methods). Schaffer collateral inputs contacted pyramidal cells with a probability of 0.06 (ref. 25). We computed the fraction of recruited pyramidal cells as a function of the number of stimulated Schaffer collateral and compared the resulting activation curves under two conditions: with either a fixed or a dynamic EPSGT (Fig. 6a). The threshold to recruit a pyramidal cell was set at 100 Schaffer collateral inputs 2 and remained constant in the fixed EPSGT condition. In the dynamic EPSGT condition, the number of Schaffer collaterals necessary to recruit a pyramidal cell increased linearly with increasing number of stimulated Schaffer collaterals, up to approximately threefold,
1581

a r t ic l e s
Figure 5 Fast-spiking interneurons enforce dynamic EPSGT. (a) Top, recording configuration. Left, simultaneous field (from pyramidal cell layer) and loose-patch recording from three types of neurons in response to Schaffer collateral stimulation (three different experiments): a pyramidal cell (top, five superimposed sweeps), a fast-spiking interneuron (FS, middle, five superimposed sweeps) and a regular-spiking interneuron (RS, bottom, five superimposed sweeps). Although the spike in the pyramidal cell was concomitant with the peak of the population spike recorded with the field electrode, the action potential in the fast-spiking cell preceded, and in the regular-spiking cell followed, the population spike. Right, camera lucida reconstruction of the three neurons on the left (dashed lines mark the margins of the pyramidal cell layer, stratum radiatum is above the dashed lines; d, dendrite; ax, axon). Insets, spiking pattern from same neurons recorded in whole-cell current-clamp configuration in response to depolarizing and hyperpolarizing current steps. ( b) Relative timing of spikes elicited in response to Schaffer collateral stimulation in pyramidal cells (black, n = 21), fast-spiking interneurons (blue, n = 18) and regularspiking interneurons (gray, n = 34) with respect to the peak of population spike (spike threshold determined in loose patch for all cells). On average, fast-spiking interneurons fired before and regular-spiking interneurons fired after the population spike. (c) Activation curves (cumulative distribution of threshold input strengths) for fast-spiking (blue, n = 18), regular-spiking (gray, n = 34) and pyramidal cells (black, n = 39, from Fig. 1a). The input strength that recruited 50% of fast-spiking interneurons elicited spikes in only 13% of pyramidal cells (spike threshold determined in loose patch for all cells). The continuous light blue line is the fit of the feedforward IPSG as a function of input strength (right ordinate from Fig. 2a, asymptote scaled to 100%). There was a good match with the activation curve of fast-spiking neurons.

a
Field PC

Rec Stimulus
IN PC

Rec

Rec

PC layer 40 mV 500 ms

Loose patch

3 mV 20 pA
PC layer

ax

FS

3 mV 200 pA

RS

3 mV 40 pA 5 ms

200 m

b
PC FS RS

c
Recruited neurons (%)

100

Activation curves

20 IPSG (nS)

2009 Nature America, Inc. All rights reserved.

50

4 3 2 1 0 1 2 3 4 Delay relative to population spike (ms)

13 0 0 0.5 Input strength

FS RS PC 1.0

10

and then remained constant to simulate experimental observation (Fig. 6b). The exact increment of the modeled dynamic EPSGT was chosen to yield an activation curve that best approximated the experimentally observed activation curve (Fig. 6c). With a fixed EPSGT, the activation curve was steep and had a narrow dynamic range (Fig. 6c). With a dynamic EPSGT, on the other hand, the activation curve had an onset similar to the fixed threshold activation curve, but rose much less steeply, resulting in a wider dynamic range (Fig. 6c; the sensitivity of the slope of the activation curve to the rate of increase of the dynamic EPSGT is illustrated in Supplementary Fig. 9). The dynamic EPSGT activation curve (Fig. 6c) is the synthesis of a family of fixed EPSGT activation curves, each having progressively larger EPSGT (a subset are illustrated in Fig. 6c). An intersection

occurs at the specific input strength at which dynamic and fixed EPSGT activation curves have equal EPSGT. This simple model captures the basic experimental finding, namely that the dynamic EPSGT expands the range of inputs that the CA1 pyramidal cell population responds to by maintaining sensitivity to weak inputs and preventing saturation to stronger stimuli. It should be noted, however, that the modeled fixed and dynamic EPSGT activation curves both fail to fully account for the experimentally observed activation curves, where more than ~80% of pyramidal cells are active (see Discussion). Dynamic EPSGT in somatosensory cortex Feedforward inhibitory circuits involving fast-spiking interneurons have been described along several cortical projections24,2629. Is the dynamic EPSGT a general mechanism by which the cortex expands the range of inputs it can respond to? We tested this hypothesis at one of the main projections in the neocortical canonical circuit, the
Figure 6 Model of activation curve with dynamic EPSGT. (a) Modeled distribution of EPSG amplitudes in the population of pyramidal cells when N (top row) and 1.5N (bottom) afferent fibers are active. The area above spike threshold T, under the curve (black shaded), is the fraction of pyramidal cells recruited with either fixed (left column) or dynamic (right) EPSGT. For dynamic EPSGT, a smaller fraction of pyramidal cells was recruited when increasing the number of active afferent fibers from N to 1.5N. (b) EPSGT as a function of input strength used in the model; dynamic (continuous line), fixed (dashed line) and experimentally measured dynamic EPSGT (gray columns, same data as in Fig. 1, but for the entire range of stimulus strengths) are shown. ( c) Modeled pyramidal cell activation curves with fixed or dynamic EPSGT. Experimentally measured activation curve in gabazine and control conditions are superimposed. Dotted gray lines represent a set of fixed EPSG T activation curves (for each activation curve, the threshold is given in multiples of T, the threshold at minimal input strength). The dynamic EPSG T activation curve intersects each of the fixed EPSGT activation curves at the specific input strength at which the threshold of the two curves matches. The dynamic EPSGT activation curve thus results from the synthesis of a family of fixed EPSGT activation curves.

a
16 8 PC number (%) 0 16 8 0 0

Fixed EPSGT

above T: 3% T 0 1.5N 33% 0 T 0 EPSG amplitude T

Model EPSGT (au)

Dynamic EPSGT Input strength: Area N 3%

4T 24 3T 2T T 0 Fixed ESPGT Dynamic ESPGT 18 12 6 0 Experimental EPSGT (nS)

8% 1.2T 0 0.2 0.4 0.6 0.8 1.0

c 100
Recruited PCs (%) 80 60 40 20 0 0 1.5 N 1T

1.25T 1.75T 2.2T 2.65T 3.1T Model Fixed ESPGT Dynamic ESPGT Experiment Gabazine Control 0.2 0.4 0.6 Input strength 0.8 1.0

1582

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 7 The stronger the photostimulation of L2/3 pyramidal cells, the larger the excitation necessary to recruit L5 pyramidal cells. ( a) Left, overlay of low-magnification bright-field and red and green fluorescent images of a representative cortical brain slice from a 27-day-old mouse electroporated while in utero with ChR2-Venus (green). The red cells are a pair of L5 pyramids filled with Alexa Fluor 568. Note the green band of ChR2-Venusexpressing cells in L2/3 and the fainter band in L5 representing the axonal arborization of the L2/3-L5 projection. Scale bar represents 200 m. Right, two-photon stack of the same slice at higher magnification. Scale bar represents 50 m. Nearly all of the ChR2-Venusexpressing neurons were in L2/3. Individual axons from L2/3 cells can be seen crossing L4 and arborizing in L5. ( b) Top, simultaneous current-clamp recording from two layer 5 pyramidal cells excited by a 5-ms square light pulse. The pyramidal cell on the left was recruited at threshold by lower-intensity photostimulation as compared with the pyramidal cell on the right. Successes in triggering a spike are shown in black and failures in gray (34 superimposed traces). The blue traces indicate the duration and relative amplitudes of the photostimulations. Bottom traces, threshold EPSCs (average of 510 traces) recorded in the same two cells voltage clamped at 70 mV. The pyramidal cell on the left necessitated less excitation to reach threshold. A summary graph of EPSGTs recorded in cells recruited by the higher-intensity photostimulus plotted against the EPSG T simultaneously recorded in cells recruited by the lower-intensity photostimulus (n = 15) is shown. Most of the data points lie above the unity line (red data point indicates the experiment shown on top). ( c) The same experimental configuration as in b was used, but photostimulation is a 10-ms light ramp. The scatter plot shows the threshold excitatory charge (threshold EPSC time integral) recorded in cells recruited by the higher-intensity photostimulus plotted against the threshold excitatory charge simultaneously recorded in cells recruited by the lower-intensity photostimulus (n = 14). As in b, most of the data points lie above the unity line (red data point indicates the experiment shown on top). ( d) Top, plot of spike latencies of L5 pyramidal cells to stimulation of L2/3 with 5-ms pulses of blue light (black symbols). Bottom, EPSC (black trace, recorded at IPSC reversal potential) and IPSC (blue trace, recorded at EPSC reversal potential) in a layer 5 pyramidal cell with a cesium-based internal solution. The onset of disynaptic inhibition (5% of peak, vertical dotted line) preceded the average spike latency (open symbol). Error bars are s.e.m. (e) Scatter plot of the spread of EPSCs plotted against the spread of IPSCs recorded in pairs of L5 neurons. Most data-point are below the unity line (n = 13 pairs).

L2/3 L4 L5 L6

Light pulse

Light ramp

b
40 mV Current-clamp Voltage-clamp 250 pA 5 ms

c
40 mV Current-clamp Voltage-clamp 500 pA 10 ms

30 High intensity (nS)

EPSGT High intensity (pC) 20

Threshold excitatory charge

2009 Nature America, Inc. All rights reserved.

20

10

10

0 0 10 20 Low intensity (nS) 1 Time (ms) 2 3 4 30

0 0 10 20 Low intensity (pC)

e
IPSC spread (%)

80 60 40 20 0 0 20 40 60 EPSC spread (%) 80

2 nA Voltageclamp

layer 2/3 to layer 5 excitatory projection of the somatosensory cortex1 (Fig. 7). Because cortical architecture does not permit selective stimulation of the axons of layer 2/3 pyramidal cells with an extracellular stimulation electrode, we electroporated mice in utero with channelrhodopsin-2 (ChR2) to target layer 2/3 pyramidal cell progenitors30,31 (Fig. 7a; see Online Methods). Photoactivation of ChR2-expressing layer 2/3 pyramidal cells in slices of juvenile brains triggered both direct excitation and FFI in layer 5 pyramidal cells ( Supplementary Fig. 10). We recorded from two postsynaptic layer 5 pyramidal cells simultaneously and increased the intensity of the photostimulation of the layer 2/3 axons until one of the two recorded layer 5 pyramidal cells reached spike threshold (low-intensity stimulation). We then further increased the intensity of the photostimulation until the second layer 5 pyramidal cell reached threshold (high-intensity stimulation). Similar to the hippocampus, layer 5 pyramidal cells recruited at high intensity required larger EPSGTs as compared with pyramidal cell recruited at low intensity (1.3 0.1fold more, n = 15, P = 0.019; Fig. 7b). This difference in EPSGT was not a result of differences in membrane properties of layer 5 pyramidal cells (input resistance: low-intensity cells, 92 14 M; high-intensity cells, 94 14 M; P = 0.90, n = 15; membrane time constant: low-intensity cells, 19 2 ms; high-intensity cells, 21 2 ms; P = 0.41, n = 15). Does the EPSGT also vary when afferent activity is distributed in time, similar to what occurs under more physiological conditions7?
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

To desynchronize the activation of the input population, we progressively ramped up the intensity of the light stimulus over a period of 10 ms (Fig. 7c; see Online Methods). When recording from two layer 5 pyramidal cells simultaneously, as described above, the excitatory charge necessary to recruit the layer 5 pyramidal cell with the highintensity stimulus was 1.8 0.4fold larger than for pyramidal cells recruited at low intensity (n = 14, P = 0.021; Fig. 7c). This indicates that the EPSGT also varies under conditions in which the activity of presynaptic neurons is desynchronized. Thus, the EPSGT is dynamic for layer 5 cortical pyramidal cells, such that a pyramidal cell recruited when many presynaptic layer 2/3 pyramidal cells are active necessitates significantly larger EPSGs than a pyramidal cell recruited when few layer 2/3 pyramidal cells are active. Is the dynamic EPSGT in the somatosensory cortex based on the same mechanism as the one identified in the hippocampus? As in the hippocampus, the onset of FFI preceded the spiking of layer 5 pyramidal cells (by 2.0 0.4 ms, n = 24; Fig. 7d) and increased with stimulus strength (Supplementary Fig. 10). The overlap between EPSC and IPSC before spike generation was even more pronounced in response to desynchronized activation of the inputs. In fact, although the latency between EPSC and feedforward IPSC was similar to what we observed in response to a synchronized stimulus (1.5 0.1 ms, n = 3), the spiking occurred only 6.2 0.1 ms (n = 26) after the onset of the EPSC, resulting in an almost 5-ms overlap. Finally, similar to the hippocampus,
1583

a r t ic l e s
FFI appeared to be more homogenously distributed across layer 5 pyramidal cells as compared with excitation. Although the spread of the EPSCs simultaneously recorded in two pyramidal cells was 36 5% (n = 13), the spread of the concomitant IPSCs was 23 4% (P = 0.007, n = 13; Fig. 7e). DISCUSSION Our data suggest that the EPSC amplitude necessary to reach spike threshold in hippocampal and neocortical pyramidal cells is dynamic and increases when the number of active neurons in the presynaptic layer increases. Accordingly, the fractional contribution of an individual afferent input in firing a neuron is not fixed, but instead is continuously normalized by the total number of active afferents. Through this simple mechanism, the pyramidal cell population can smoothly operate over a wide range of stimulus strengths. This mechanism is probably important for enabling cortical structures such as the hippocampus to be responsive to weak stimuli, but remain sparsely active even when confronted with stronger inputs. Although FFI sets a global threshold for recruitment of pyramidal cells, local differences in afferent excitation determine which pyramidal cell is recruited. The fraction of neurons active at any given moment in cortical areas strongly fluctuates as a result of either varying sensory stimuli or ongoing intrinsic activity. The probability of spiking in layer 2/3 pyramidal cells in the somatosensory cortex, for example, rapidly fluctuates in response to naturalistic stimuli applied to the whiskers7. Similarly, activity levels in the hippocampus can vary from sparse activity in the exploring animal8 to synchronous activation of a large fraction of neurons during ripples9. As a consequence, downstream targets of these neuronal populations experiences substantial fluctuations in the fraction of afferents that are active at any given time point. The connectivity patterns of cortical excitatory projections, however, are ill-suited to allow postsynaptic populations of neurons to operate over a wide range of afferent activity5; afferent axons typically form very divergent projections to contact a large number of postsynaptic targets through relatively weak contacts, such that the simultaneous activity of several afferents is necessary to recruit a target neuron2,32. Because of this divergence, gradual increases in the number of active afferents produce very steep, or explosive, increases in the fraction of recruited targets5, resulting in a limited range of input strengths that can be differentially represented by the postsynaptic population. Our data indicate that paleo- and neocortical circuits expand the range of afferent input strengths that the cells can respond to by ensuring that, when the input is strong, pyramidal cells necessitate larger EPSCs to reach thresholds. The dynamic range of the population is several-fold wider than the dynamic range of an individual pyramidal cell (34 versus 2). Neither the dynamic range nor the gain (slope of the sigmoidal fit, R2 = 0.046, P = 0.2) of individual pyramidal cells varied between pyramidal cells recruited along the entire input range. Thus, although GABAA receptormediated conductances can regulate the gain of individual neurons1012, our data suggest that pyramidal cell populations can function over a wide range of afferent intensities without requiring gain changes in individual neurons. Individual excitatory afferent inputs to cortical areas diverge to contact fast-spiking basket cells and principal neurons. The match between the increase in FFI and the activation curve of fast-spiking cells (Fig. 5c) suggests that the increase in FFI results from the increased fraction of recruited fast-spiking cells. Because fast-spiking basket cells receive larger and faster EPSPs21,24,26, they were recruited before pyramidal cells in response to afferent activity (Fig. 3)33. This led to a substantial temporal overlap between EPSCs and FFI before spike generation in pyramidal cells (Fig. 3). Only those pyramidal
1584

cells that received large enough EPSCs to overcome the concomitantly occurring inhibition reached spike threshold. With increasing stimulus strength, the amplitude of FFI increased ( Fig. 2a) and larger EPSCs became necessary for pyramidal cells to reach spike threshold. The onset of inhibition before spike generation in pyramidal cells also means that this early phase of inhibition is unlikely to be of feedback origin, as feedback inhibition is a consequence of pyramidal cell spiking. The control of the amplitude of the EPSC necessary to reach threshold by FFI is likely to be even more marked in response to repetitive, burst-like14 or asynchronous afferent activity, as a result of the large temporal overlap between afferent excitation and FFI generated by the previous stimulus. Whether the activation curve of fastspiking cells (Fig. 5c) is also controlled by inhibitory inputs remains to be established. If so, reciprocal inhibition of fast-spiking cells may influence the dynamic range of pyramidal cell populations. Our model captures the initial 80% of the activation curve; that is, until the EPSGT plateaus above input strength of ~0.5. At these greater input strengths, the EPSGT is fixed and the model predicts that the activation curve behaves accordingly. However, the top 20% of the experimentally determined activation curve extends beyond this prediction. It is possible that the observed onset of the plateau is inaccurate because of an error in measurement (for example, the lack of proper voltage clamp) and that the real EPSGT continues to grow with increasing input strength. Alternatively, a small portion of pyramidal cells may receive Schaffer collateral inputs with very low probability as compared with the rest of the population (resulting from heterogeneity in the population or damage to their dendrites) such that they necessitate a much larger stimulus strength to be recruited. Our model also illustrates the fact that the activation curve is sensitive to how steeply the EPSGT varies with input strength. Because the increase in EPSGT is, at least in part, determined by the increase in FFI, any parameter that controls the excitability of GABAergic interneurons, such as neuromodulators, will probably affect the slope of the activation curve, and thus the range of input strength that can be represented by the postsynaptic pyramidal cell population. What determines the specific pattern of pyramidal cells recruited by a stimulus? We found distinct roles for excitation and FFI; a relatively homogeneous inhibition across pyramidal cells sets a global threshold34 that is proportional to stimulus strength and heterogeneities in excitation determine the precise pattern of pyramidal cells recruited in that population. The situation in vivo may create additional biases; through the local action of neuromodulators, such as acetylcholine, some pyramidal cells may be more depolarized than others and reach spike threshold even if they receive less excitation than their neighbors. The local action of presynaptic inhibitors of GABA release, such as opiates, cannabinoids or GABA itself, may create spatial heterogeneities in inhibition that were not present in the slice. Several synaptic and connectivity properties may contribute to homogeneous distribution of FFI, including the strong divergence of individual fast-spiking cells onto the pyramidal cell population21,3537, the large number of contacts made by individual fast-spiking cells onto each pyramidal cell38 and the relatively large number of recruited fast-spiking cells even at low stimulus strength (Fig. 5c). In summary, because the EPSG T is controlled in a feedforward manner, the sensitivity of pyramidal cells is virtually instantaneously adjusted to match the strength of the afferent stimulus. This instantaneous adjustment differs from adaptation because it does not rely on the previous history of the network through a negative feedback mechanism, such as feedback inhibition, spike adaptation, synaptic depression or presynaptic inhibition. The presence of feedforward inhibitory circuits along several major excitatory pathways
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

2009 Nature America, Inc. All rights reserved.

a r t ic l e s
in the brain15,16,29,39,40 suggests that the expansion of the dynamic range by instantaneously varying the amplitude of the EPSC necessary to reach threshold may not be unique to hippocampus and somatosensory cortex. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank P. Abelkop for anatomical reconstructions of biocytin filled neurons, F. Frhlich for developing the initial versions of model, M. Carandini and J. Isaacson for comments and suggestions during the entire course of the project, C. Poo and F. Bertaso for inputs on the manuscript, and all of the members of the Scanziani laboratory for their input on the project and the manuscript. M.S. thanks C. Staub for the original discussions leading to the project. This work was funded in part by the US National Institutes of Health (MH71401 to M.S. and NS061521 to B.V.A.). H.A. is a fellow of the Helen Hay Whithney Foundation. M.S. is an investigator of the Howard Hughes Medical Institute.
14. Tropp Sneider, J., Chrobak, J.J., Quirk, M.C., Oler, J.A. & Markus, E.J. Differential behavioral state-dependence in the burst properties of CA3 and CA1 neurons. Neuroscience 141, 16651677 (2006). 15. Pouille, F. & Scanziani, M. Enforcement of temporal fidelity in pyramidal cells by somatic feedforward inhibition. Science 293, 11591163 (2001). 16. Buzski, G. Feed-forward inhibition in the hippocampal formation. Prog. Neurobiol. 22, 131153 (1984). 17. Alger, B.E. & Nicoll, R.A. Feed-forward dendritic inhibition in rat hippocampal pyramidal cells studied in vitro. J. Physiol. (Lond.) 328, 105123 (1982). 18. Nicoll, R.A., Alger, B.E. & Jahr, C.E. Enkephalin blocks inhibitory pathways in the vertebrate CNS. Nature 287, 2225 (1980). 19. Somogyi, P. & Klausberger, T. Defined types of cortical interneurone structure space and spike timing in the hippocampus. J. Physiol. (Lond.) 562, 926 (2005). 20. Freund, T.F. & Buzski, G. Interneurons of the hippocampus. Hippocampus 6, 347470 (1996). 21. Glickfeld, L.L. & Scanziani, M. Distinct timing in the activity of cannabinoid-sensitive and cannabinoid-insensitive basket cells. Nat. Neurosci. 9, 807815 (2006). 22. Maccaferri, G. & Dingledine, R. Control of feedforward dendritic inhibition by NMDA receptordependent spike timing in hippocampal interneurons. J. Neurosci. 22, 54625472 (2002). 23. Geiger, J.R., Lubke, J., Roth, A., Frotscher, M. & Jonas, P. Submillisecond AMPA receptormediated signaling at a principal neuron-interneuron synapse. Neuron 18, 10091023 (1997). 24. Cruikshank, S.J., Lewis, T.J. & Connors, B.W. Synaptic basis for intense thalamocortical activation of feedforward inhibitory cells in neocortex. Nat. Neurosci. 10, 462468 (2007). 25. Sayer, R.J., Friedlander, M.J. & Redman, S.J. The time course and amplitude of EPSPs evoked at synapses between pairs of CA3/CA1 neurons in the hippocampal slice. J. Neurosci. 10, 826836 (1990). 26. Gabernet, L., Jadhav, S.P., Feldman, D.E., Carandini, M. & Scanziani, M. Somatosensory integration controlled by dynamic thalamocortical feed-forward inhibition. Neuron 48, 315327 (2005). 27. Helmstaedter, M., Staiger, J.F., Sakmann, B. & Feldmeyer, D. Efficient recruitment of layer 2/3 interneurons by layer 4 input in single columns of rat somatosensory cortex. J. Neurosci. 28, 82738284 (2008). 28. Daw, M.I., Ashby, M.C. & Isaac, J.T. Coordinated developmental recruitment of latent fast spiking interneurons in layer IV barrel cortex. Nat. Neurosci. 10, 453461 (2007). 29. Mittmann, W., Koch, U. & Hausser, M. Feed-forward inhibition shapes the spike output of cerebellar Purkinje cells. J. Physiol. (Lond.) 563, 369378 (2005). 30. Saito, T. & Nakatsuji, N. Efficient gene transfer into the embryonic mouse brain using in vivo electroporation. Dev. Biol. 240, 237246 (2001). 31. Petreanu, L., Huber, D., Sobczyk, A. & Svoboda, K. Channelrhodopsin-2assisted circuit mapping of long-range callosal projections. Nat. Neurosci. 10, 663668 (2007). 32. Bruno, R.M. & Sakmann, B. Cortex is driven by weak, but synchronously active, thalamocortical synapses. Science 312, 16221627 (2006). 33. Porter, J.T., Johnson, C.K. & Agmon, A. Diverse types of interneurons generate thalamus-evoked feedforward inhibition in the mouse barrel cortex. J. Neurosci. 21, 26992710 (2001). 34. Poo, C. & Isaacson, J.S. Odor representations in olfactory cortex: sparse coding, global inhibition and oscillations. Neuron 62, 850861 (2009). 35. Holmgren, C., Harkany, T., Svennenfors, B. & Zilberter, Y. Pyramidal cell communication within local networks in layer 2/3 of rat neocortex. J. Physiol. (Lond.) 551, 139153 (2003). 36. Beierlein, M., Gibson, J.R. & Connors, B.W. Two dynamically distinct inhibitory networks in layer 4 of the neocortex. J. Neurophysiol. 90, 29873000 (2003). 37. Thomson, A.M., West, D.C., Wang, Y. & Bannister, A.P. Synaptic connections and small circuits involving excitatory and inhibitory neurons in layers 25 of adult rat and cat neocortex: triple intracellular recordings and biocytin labeling in vitro. Cereb. Cortex 12, 936953 (2002). 38. Buhl, E.H., Halasy, K. & Somogyi, P. Diverse sources of hippocampal unitary inhibitory postsynaptic potentials and the number of synaptic release sites. Nature 368, 823828 (1994). 39. Blitz, D.M. & Regehr, W.G. Timing and specificity of feed-forward inhibition within the LGN. Neuron 45, 917928 (2005). 40. Agmon, A. & Connors, B.W. Thalamocortical responses of mouse somatosensory (barrel) cortex in vitro. Neuroscience 41, 365379 (1991).

2009 Nature America, Inc. All rights reserved.

AUTHOR CONTRIBUTIONS F.P. and A.M.-B. conducted the experiments in the hippocampus; H.A. conducted the experiments in the somatosensory cortex; B.V.A. made the model; and M.S. supervised the project 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. Lefort, S., Tomm, C., Floyd Sarria, J.C. & Petersen, C.C. The excitatory neuronal network of the C2 barrel column in mouse primary somatosensory cortex. Neuron 61, 301316 (2009). 2. Otmakhov, N., Shirke, A.M. & Malinow, R. Measuring the impact of probabilistic transmission on neuronal output. Neuron 10, 11011111 (1993). 3. Marr, D. A theory of cerebellar cortex. J. Physiol. (Lond.) 202, 437470 (1969). 4. Vogels, T.P. & Abbott, L.F. Signal propagation and logic gating in networks of integrate-and-fire neurons. J. Neurosci. 25, 1078610795 (2005). 5. Shadlen, M.N. & Newsome, W.T. The variable discharge of cortical neurons: implications for connectivity, computation and information coding. J. Neurosci. 18, 38703896 (1998). 6. Diesmann, M., Gewaltig, M.O. & Aertsen, A. Stable propagation of synchronous spiking in cortical neural networks. Nature 402, 529533 (1999). 7. Arabzadeh, E., Zorzin, E. & Diamond, M.E. Neuronal encoding of texture in the whisker sensory pathway. PLoS Biol. 3, e17 (2005). 8. Wilson, M.A. & McNaughton, B.L. Dynamics of the hippocampal ensemble code for space. Science 261, 10551058 (1993). 9. Csicsvari, J., Hirase, H., Mamiya, A. & Buzsaki, G. Ensemble patterns of hippocampal CA3CA1 neurons during sharp waveassociated population events. Neuron 28, 585594 (2000). 10. Shu, Y., Hasenstaub, A., Badoual, M., Bal, T. & McCormick, D.A. Barrages of synaptic activity control the gain and sensitivity of cortical neurons. J. Neurosci. 23, 1038810401 (2003). 11. Mitchell, S.J. & Silver, R.A. Shunting inhibition modulates neuronal gain during synaptic excitation. Neuron 38, 433445 (2003). 12. Chance, F.S., Abbott, L.F. & Reyes, A.D. Gain modulation from background synaptic input. Neuron 35, 773782 (2002). 13. Carvalho, T.P. & Buonomano, D.V. Differential effects of excitatory and inhibitory plasticity on synaptically driven neuronal input-output functions. Neuron 61, 774785 (2009).

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1585

ONLINE METHODS
All experiments were conducted in accordance with the animal use guidelines set out by the Institutional Animal Care and Use Committee of the University of California, San Diego. Slice preparation. Acute hippocampal slices (400 m) were prepared from 45-week-old male Wistar rats and incubated for 45 min in an interface chamber at 34 C in normal artificial cerebrospinal fluid (ACSF) containing 119 mM NaCl, 2.5 mM KCl, 1.3 mM NaH2PO4, 1.3 mM MgCl2, 2.5 mM CaCl2, 26 mM NaHCO3 and 11 mM glucose (equilibrated with 95% O2 and 5% CO2). The slices were kept at room temperature (2025 C) for 0 to 6 h before being placed in a submerged chamber for recording at 3133 C. Coronal slices (400 m) from somatosensory cortex were prepared from postnatal day 1525 ICR white mice in modified ACSF containing 83 mM NaCl, 2.5 mM KCl, 1.0 mM NaH2PO4, 3.3 mM MgSO4, 0.5 mM CaCl2, 26.2 mM NaHCO3, 72 mM sucrose and 22 mM glucose (equilibrated with 95% O2 and 5% CO2). The slices were incubated for 45 min in a submerged chamber at 34 C containing the modified ACSF and kept in the same chamber at room temperature (2025 C) for 0 to 6 h before being placed in a recording chamber with normal ACSF at 3133 C.

2009 Nature America, Inc. All rights reserved.

Accordingly, all experiments in which the input strength is reported were performed in the presence of two field-potential recording electrodes, one placed in the stratum radiatum of CA1 (for fEPSPs) and one in the stratum pyramidale (for population spikes). The fEPSP was calibrated with respect to the simultaneously recorded population spike. Specifically, the slope of the fEPSP elicited at any given stimulus intensity was normalized by the slope of the fEPSP evoked at a stimulus intensity that evoked a population spike of 95% of its maximal amplitude (Supplementary Fig. 1). This normalization was done because even if the same number of Schaffer collaterals were recruited in two different slices, the absolute value of the slope of the fEPSP may vary (for example, as a result of a different positioning or depth of the recording electrode). The value of such normalized fEPSP is the input strength. Thus, the input strength is 1 when the number of stimulated Schaffer collateral triggers a population spike of 95% of its maximal amplitude and is 0.1 when the number of stimulated Schaffer collaterals is a tenth of the number it takes to trigger 95% population spike. For each slice, the input strength was always determined in control conditions. Thus, the number of Schaffer collaterals stimulated in control conditions or in gabazine is the same for a given input strength. In a subset of experiments, fEPSP was recorded in the stratum pyramidale. Experiments were only started once the amplitude of the population spike remained stable over a period of at least 10 min. Threshold stimulation, activation curves and the measure of EPSCs and IPSCs peak amplitude, conductance and charge. Neurons were recorded in the loosepatch configuration and the Schaffer collaterals stimulated at increasing intensities until about 50% of the trials elicited a spike in the recorded cells. Neurons recorded in the loose-patch configuration were subsequently re-patched with a pipette containing standard internal solution (see above) to gain whole-cell access. In a subset of the recording reported in Figures 1 and 2 (see legends), we determined threshold input strength using whole-cell recordings; the distribution of EPSGT with input strength did not differ significantly from the one obtained when threshold input strengths was determined in the loose-patch configuration (linear regressions, ANCOVA test on slopes, P = 0.13; intercepts, P = 0.1). The data were thus pooled. The activation curve of a population of neurons (Figs. 1a and 5c) is the cumulative distribution of the input strengths that elicit 50% spiking probability in the neurons of that population. The 50% spiking probability of individual neurons was interpolated by fitting their spiking probability plotted against stimulus 100 , where x0 is the 1 + 10 p(x0 x ) input strength at 50% spiking probability and p is the slope at x0. Schaffer collaterals stimulation evoked an EPSC-IPSC sequence in pyramidal cells voltage clamped at about 60mV (Supplementary Fig. 5). To ensure that the recorded IPSCs were a result of the synaptic recruitment of GABAergic interneurons rather than of direct stimulation of inhibitory axons, we only considered experiments in which IPSCs could be abolished by the glutamate receptor antagonist NBQX15 (at least 70% reduction, average of 84.4 1.2%, n = 50; Supplementary Fig. 5). EPSCs were isolated by voltage clamping the pyramidal cell at the reversal potential of the IPSC (88.9 0.8 mV, n = 43). To estimate the amplitude of the EPSCs recorded at 60 mV (that is, well above the reversal potential of the IPSC), we used the amplitude of the EPSC recorded at the reversal potential of the IPSC after scaling its initial slope to the initial slope of the EPSC recorded at 60 mV (Supplementary Fig. 5). The feedforward IPSC was isolated after subtracting the isolated and scaled EPSC from the EPSC-IPSC sequence. When present, the direct IPSC recorded in NBQX (never larger than 30% of the total IPSC amplitude, average 15.6 1.2%, n = 50, see above) was also subtracted from EPSC-IPSC sequence. Peak synaptic excitatory and inhibitory conductances were computed as the peak of the EPSC or IPSC divided by the driving force at which the synaptic currents were recorded. We can estimate the error on our EPSC measurement if the holding potential of the cell is 10% off with respect to the actual IPSC reversal (say off by 8.5 mV if the IPSC reversal was 85 mV). The ratio between the peak of the EPSGT and the peak of the concomitant feedforward IPSG was approximately 1:0.8 (Fig. 2). Because of the 1.65-ms delay of the feedforward IPSC with respect to the onset of EPSC, at the peak of the EPSC the IPSC is only 55% of its maximal amplitude. Thus, with a driving force of 85 mV for the EPSC and a driving force of 8.5 mV strength (Fig. 1a) with the sigmoid function Y =

Electrophysiology and analysis. Recorded neurons were visually identified using infrared differential interference contrast videomicroscopy. Unless stated otherwise, all whole-cell recordings were performed with patch pipettes (24 M) filled with 150 mM potassium gluconate, 1.5 mM MgCl2, 5 mM HEPES buffer, 1.1 mM EGTA and 10 mM phosphocreatine (pH = 7.25, 280290 mOsm); biocytin (0.2% wt/vol) and 2 mM Mg-ATP were added in the recording solution for interneurons. When recording with a cesium-based intracellular solution, the composition was 115 mM cesium methanesulphonate, 8 mM NaCl, 10 mM HEPES, 0.3 mM Na3-GTP, 4 mM Mg-ATP, 0.3 mM EGTA, 5 mM QX-314-Cl and 10 mM BAPTA tetracesium (pH = 7.4, 290 mOsm). Series resistance was not compensated but was monitored continuously with negative voltage steps, and recordings with series resistances larger than 12 M were not included in the estimation of EPSGT in control conditions or after gabazine treatment. None of the additional recordings in the study had series resistances larger than 20 M. Voltage measurements were not corrected for the experimentally determined junction potential (11.7 1.0 mV, n = 3). Experiments in the hippocampus were performed in the presence of the GABAB receptor antagonist CGP54626 (12 M) and the NMDA receptor antagonist R-()-3-(2-carboxypiperazine-4-yl)propyl-1-phosphonic acid (RS-CPP, 2550 M). The presence of the GABAB and NMDA receptor antagonists had no significant effect on the probability of eliciting a spike in pyramidal cells in response to threshold stimulation of Schaffer collaterals (CPP, 110.1 11.9%; CGP54626, 119.8 50.2%; P = 0.84 and 0.43, respectively). Schaffer collaterals were stimulated (100 s) with constant current (range of 10900 A) using a steel monopolar electrodes placed in the stratum radiatum of CA1. CA1 was isolated from CA3 and the subiculum by two radial cuts to prevent propagation of epileptiform activity. ChR2-expressing layer 2/3 pyramidal cells were activated by a full-field, 5-ms square light-pulse or 10-ms light-ramp from a 5-W luxeon blue LED coupled to the epifluorescence pathway of the Olympus BX51. Loose-patch recordings were performed with ACSF-filled patch pipettes (810 M). Field recordings were performed using patch pipettes (24 M) filled with 3 M NaCl. Data were recorded with Axopatch 200A, Axopatch 200B or Multiclamp 700A amplifiers (Axon Instruments); acquisition (510-kHz digitization) and analysis was performed with pCLAMP 9.2 software (Molecular Devices). Average values are expressed as means s.e.m. Students t test was used for statistical comparisons. We used 2,3-dihydroxy-6-nitro-7-sulfamoylbenzo[f]quinoxaline-2,3-dione (NBQX), SR95531 (gabazine), CGP54626, RSCPP, DAMGO, CTAP and muscimol (Tocris Cookson). Calibrating input strength. The input strength is proportional to the number of activated Schaffer collaterals and varies from 0 to 1. It is a measure that allows comparison of stimulus intensities across slices. To estimate input strength, we used two experimentally determined parameters: the slope of the field EPSP (fEPSP), which is proportional to the number of activated Schaffer collaterals4143, and the amplitude of the population spike (population spike; Supplementary Fig. 1).

nature NEUROSCIENCE

doi:10.1038/nn.2441

for the IPSC, the EPSC amplitude will be under- or overestimated by (0.8 0.55) 0.1 = 0.045. Thus, the error is below 5%. The increase in feedforward inhibition with input strength (Fig. 2a) was fitted with the Boltzmann equation Y = A 1+ e A
( x x0 ) p

similar to the temporal distribution of spikes recorded in the somatosensory cortex in vivo in response to a whisker stimulus7. Computational model. We sought to design a simple model constrained by experimentally determined parameters (anatomical connectivity and average number of inputs required to reach threshold) to establish whether the experimentally measured increase in EPSGT can, in principle, account for the observed dynamic range of the activation curve. Our model predicts the fraction of pyramidal cells activated as a function of the number of active Schaffer collaterals, given a connectivity p of Schaffer collaterals onto pyramidal cells. We assume that all Schaffer collateral-pyramidal cell synapses have the same strength. Thus, the distribution of inputs strengths onto the pyramidal cell population is given by the binomial probability distribution. Accordingly the fraction A of active pyramidal cells given N active Schaffer collateral is
T N N i AN = 1 pi (1 p) i i =0

, where A is the asymptote, x0

is the input strength at half maximal feedforward IPSG and p is the slope at x0. Correlation analysis of EPSC and IPSC amplitudes. The homogeneity of excitation (and inhibition) across the population of recorded neurons was accessed by correlating the average amplitude of EPSCs (or IPSCs) simultaneously recorded in cell pairs (Fig. 4b). These paired data can yield a range of different correlation values depending on whether each pair of amplitudes, A and B, is plotted as (A,B) (that is, A on the x axis and B on the y axis) or (B,A). To account for this, we repeatedly calculated the correlation where the dataset was the same, but each pair of amplitudes was randomly assigned to be (A,B) or (B,A). The mean correlation values from this process were R = 0.3 for EPSC pairs and R = 0.8 for IPSC pairs. The correlation between IPSCs was significantly larger (P < 0.02), assessed using one-tail test of correlation for dependent variables44.

2009 Nature America, Inc. All rights reserved.

Anatomy. Slices containing neurons filled with biocytin were processed as described previously21. Neurons soma, dendrites and axons were reconstructed on a Zeiss light microscope (40) using Neurolucida (MicroBrightField). In utero electroporation. Timed pregnant ICR white mice (Charles River) at embryonic days 15 and 16 were operated on as described previously30. Each embryo was injected with 1 g of pCAGGS-ChR2-Venus31 mixed with 0.5 g of pCAG-EGFP. Light-ramp versus light-pulse photostimulation. Light rampevoked EPSCs recorded in layer 5 pyramidal cells had a slower rise time as compared with light pulseevoked EPSCs, consistent with the asynchronous recruitment of layer 2/3 pyramidal cells (1.3 01 ms versus 2.4 0.4 ms, P = 0.03, n = 6; Fig. 6c). To estimate the temporal distribution of the recruitment of layer 2/3 pyramidal cells, we deconvolved the asynchronous EPSC recorded in layer 5 pyramidal cells with the waveform of spontaneous EPSCs and computed the integral of the deconvolution. With the light ramp, the initial 50% of the activity generated in layer 2/3 occurred over a period of 7.4 0.6 ms (n = 6),

where N is the number of Schaffer collaterals, and T is threshold of excitation expressed in terms of the number of active presynaptic inputs i. p is taken as 0.06 (ref. 25), and the fixed EPSGT activation curve was simulated using a constant threshold, T = 100 (ref. 2). To simulate the dynamic EPSGT activation curve, the threshold was 100 Schaffer collaterals initially, but changed as a function of input strength, T(N), optimized to minimize the distance between the simulated and experiment activation curves. This optimization was constrained by our experimental finding such that it increased by ~threefold, reaching a plateau when ~65% of the pyramidal cells were active (Matlab, Mathworks).

41. Vaillend, C., Mason, S.E., Cuttle, M.F. & Alger, B.E. Mechanisms of neuronal hyperexcitability caused by partial inhibition of Na+-K+-ATPases in the rat CA1 hippocampal region. J. Neurophysiol. 88, 29632978 (2002). 42. Nakamura, M., Sekino, Y. & Manabe, T. GABAergic interneurons facilitate mossy fiber excitability in the developing hippocampus. J. Neurosci. 27, 13651373 (2007). 43. Winegar, B.D. & MacIver, M.B. Isoflurane depresses hippocampal CA1 glutamate nerve terminals without inhibiting fiber volleys. BMC Neurosci. 7, 5 (2006). 44. Steiger, J.H. Tests for comparing elements of a correlation matrix. Psychol. Bull. 87, 245251 (1980).

doi:10.1038/nn.2441

nature NEUROSCIENCE

a r t ic l e s

Microcircuitry coordination of cortical motor information in self-initiation of voluntary movements


Yoshikazu Isomura1, Rie Harukuni1, Takashi Takekawa1, Hidenori Aizawa2 & Tomoki Fukai1,3,4
Motor cortex neurons are activated at different times during self-initiated voluntary movement. However, the manner in which excitatory and inhibitory neurons in distinct cortical layers help to organize voluntary movement is poorly understood. We carried out juxtacellular and multiunit recordings from actively behaving rats and found temporally and functionally distinct activations of excitatory pyramidal cells and inhibitory fast-spiking interneurons. Across cortical layers, pyramidal cells were activated diversely for sequential motor phases (for example, preparation, initiation and execution). In contrast, fast-spiking interneurons, including parvalbumin-positive basket cells, were recruited predominantly for motor execution, with pyramidal cells producing a commandlike activity. Thus, fast-spiking interneurons may underlie command shaping by balanced inhibition or recurrent inhibition, rather than command gating by temporally alternating excitation and inhibition. Furthermore, initiation-associated pyramidal cells excited similar and different functional classes of neurons through putative monosynaptic connections. This suggests that these cells may temporally integrate information to initiate and coordinate voluntary movement. The primary motor cortex changes its neural activity, as seen via electroencephalography, much earlier than the onset of spontaneous voluntary movement1,2. In primates3,4 and rodents57, motor cor- tex neurons are activated or inactivated both long before and during movement expression. The temporally differential activation of these neurons might correspond to sequential phases of single voluntary movement, such as motor preparation, initiation, execution and termination/switch. Antidromic identification techniques can char- acterize the discharge activity of corticofugal neurons long-projecting to the spinal cord3,4,810, striatum8,9 and red nucleus10 during vol- untary movements. However, very few attempts have been made to elucidate the intracortical mechanism underlying motor-command generation for voluntary movement because direct morphological identification of recorded neurons has been technically difficult in actively behaving animals. In the forelimb area of the rat motor cortex1113, layer 2/3 pyrami- dal cells principally project to other cortical areas, layer 5 pyramidal cells project to subcortical structures such as the spinal cord and the striatum, and layer 6 pyramidal cells project to thalamic nuclei. These excitatory pyramidal cells, along with the star-like pyramidal cells in layer 4 of the forelimb area, also innervate local cortico-spinal pyramidal cells in layer 5 via axon collaterals14,15. An in vitro analysis of excitatory laminar connectivity predicted that motor information primarily flows from layer 2/3 to layer 5 circuit of the motor cortex16. On the other hand, a major population of nonpyramidal, inhibitory GABAergic interneurons consists of fast-spiking interneurons (mostly basket cells and chandelier cells), which fire narrow action potentials at a fast rate and usually express parvalbumin17. These interneurons might gate spike outputs from the pyramidal cells with powerful inhibitory synapses on somata or axons18. A single-unit study reported that putative layer 5 pyramidal cells and putative fast-spiking interneurons of the rabbit motor cortex discharge in an antiphasic manner during locomotion cycles19. Locomotion, however, consists of continuous and complex limb movements and is driven by subcorti- cal pattern generator(s) under cortical controls. Therefore, the man- ner in which excitatory pyramidal cells and inhibitory interneurons in the different layers of the motor cortex process sequential motor information along the temporal flow of single voluntary movement remains a mystery. To address these issues, we carried out juxtacellular and multiunit recordings from the motor cortex in head-restrained rats that were trained to spontaneously repeat forelimb movements. The juxta cellular recording technique provides accurate spike events and mor- phological features for a cortical2022 or subcortical neuron23,24. The multiunit recording technique is useful for obtaining spike events of many neurons simultaneously2527 and for exploring their synaptic connectivity2831 in a blind and unbiased manner. Our experimental approach uncovered the functional diversity of pyramidal cells and uniformity of fast-spiking interneurons across all cortical layers in the expression of voluntary movement. Furthermore, we found a pattern of excitatory synaptic interactions among neighboring neurons that have different roles in self-initiated movement. RESULTS Functional activations in motor cortex neurons To obtain a sufficient number of task-trained rats for juxta cellular visualization, we developed a multi-rat task-training system to simultaneously train up to six adult rats on an operant

2009 Nature America, Inc. All rights reserved.

1Neural Circuit Theory and 2Developmental Gene Regulation, RIKEN Brain Science Institute, Wako, Saitama, Japan. 3Complexity Science and Engineering, University of Tokyo, Kashiwa, Chiba, Japan. 4Brain Research Center, Tamagawa University, Machida, Tokyo, Japan. Correspondence should be addressed to Y.I. (isomura@brain.riken.jp).

Received 27 March; accepted 25 September; published online 8 November 2009; doi:10.1038/nn.2431

1586

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 1 Efficient operant learning of the voluntary forelimb-movement task. (a) Schematic diagram of juxtacellular and multiunit recordings from the forelimb area of head-restraint rats that were trained to perform a voluntary (self paced) forelimb-movement task using our multi-rat task-training system. (b) Increase in the total number of successful trials during the 2-week training period (n = 74 rats, mean and s.d.). (c) The convergence of lever-holding time toward 1 s (lower) with less s.d. (upper) over time. R, recording day; T, room transfer day. ( d) Left, lever trajectory and electromyogram (EMG) activity in the right forelimb during task performance. Right, averaged EMG power aligned to the onset of pull or push movements (vertical line). (e) Three representative neurons with distinct firing activity in relation to pull movement. These neurons were recorded simultaneously through the same multiunit electrode. Top, spike waveform for each channel. Middle, lever trajectories. Bottom, task-related firing activity aligned at the onset (0 s) of pull movement (20-ms bins).

a
+

Skull surgery

Automated operant task training (for 2 weeks)

Juxtacellular and multiunit recordings (2 rats per week) Release

Voluntary forelimb movement task Pull Lever (Lock) Push Reward

Hold

b
Total trials (>1 s, 2 h)

Hold > 1 s n = 74 rats 1,000 500 0

c 40
Holding time (s)

20 0 20 10 0

(s.d.)

voluntary forelimb-movement task (Fig. 1a). After primary surgery (Supplementary Fig. 1), the trainee rats did not struggle under a head-restraint condition, instead grasping a lever with their right forelimb in a properly relaxed posture and quickly learning the causal relation between lever movement and reward water in the operant trials (Supplementary Fig. 2 and Supplementary Video 1). All 74 rats learned to perform the operant motor task in 2 weeks (8 d of training, 825 302 trials, holding time of 1.42 0.64 s for 2 h on the eighth day; Fig. 1b,c, Supplementary Fig. 3 and Supplementary Video 2). The rats were then transferred to a recording room where they performed the same motor task during juxtacellular and multiunit recordings from the left forelimb area (879 282 trials, holding time of 1.18 0.34 s; Supplementary Fig. 4). In each trial, the forelimb muscles (for example, biceps brachii) became active just before the onset of pull movement (Fig. 1d and Supplementary Fig. 4). Consistent with previous studies6,7, our multiunit recordings revealed several distinct patterns of neuronal firing at an electrode site in relation to the forelimb-movement task (Fig. 1e). It is unlikely that the different firing patterns merely represented different muscular movements, as they were recorded simultaneously in the same somatotopic position of motor cortex1113; in addition, forelimb movements occurred much later and more rapidly than the slowly increasing activity of these neurons (Fig. 1d,e). We further investigated the variety of motor taskrelated neuronal firing by juxtacellular recording (Supplementary Fig. 5). We classi- fied firing into five temporally and functionally distinct patterns of task-related activity in pyramidal cells identified in the forelimb area: hold-related, pre-movement, movement, movement-off and postmovement activity. Hold-related activity, which might be involved in motor preparation or stillness, showed a gradually increasing or decreasing activation during the immovable (that is, lever holding) state (Fig. 2a). Pre-movement activity, which might be involved in motor initiation, was phasic and preceded the onset of pull or push movement; the activity was then rapidly decreased during movement expression (Fig. 2b). Two well-visualized, deep-layer pyramidal cells showing pre-movement activity extended their axons into subcortical structures (Fig. 2; see also Supplementary Fig. 6 and Supplementary Video 3), presumably to recruit cortico-subcortical loops, such as the basal ganglia, for motor initiation. Movement activity showed a typical command-like activation during the movement expression (Fig.2c). This activity might contribute to motor execution and/or somatosen- sory feedback. Movement-off activity showed a sudden depression of tonic and constant firing during the movement expression (Fig. 2d). Pyramidal cells displaying movement-off activity included a spiny, star-like pyramidal cell in layer 4 of the forelimb area14,15 (Fig. 2d,g); this suggested that the activity might be related to input gating or
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1 2 3 4 5 6 7 8 TR Training (d)

1 2 3 4 5 6 7 8 TR Training (d) Pull Push

1s

2009 Nature America, Inc. All rights reserved.

d
Lever EMG 2 s, 0.2 mV

Lever EMG power

0.5 s, 0.01 mV

e
Lever 10 Unit 1 5 0 1 0 1 0 Time (s) 1 0 Pull 1 ms, 0.1 mV

modulation. Post-movement activity showed a brief activation at the end of the movement (Fig. 2e), possibly engaging in motor termina- tion or switch. We found only one intrinsically bursting pyramidal cell in layer 3 (Supplementary Fig. 6) and the other pyramidal cells had the electrophysiological properties of regular-spiking neurons. Whether the forelimb area of motor cortex (usually agranular) and that of somatosensory cortex (granular) are located very close to13 or overlapping with11,14,15 each other is not known. Our juxtacellular analysis suggests that they overlap, as identified pyramidal cells in the forelimb area containing layer 4 (that is, granular cortex) were acti- vated in relation to motor preparation or initiation lacking sensory feedback (Fig. 2a,b and Supplementary Figs. 4 and 6). We examined two representative interneurons with high-frequency movement activity in superficial and deep cortical layers, which we classified as parvalbumin-positive fast-spiking basket cells using mor- phological and electrophysiological methods (Fig. 3). Most of the fast-spiking interneurons obtained with juxtacellular and multiunit recordings were activated from the baseline level primarily during the expression of movement and showed lower selectivity to pull versus push movement than pyramidal cells (Figs. 2c and 3b). We found no other subtypes of task-related interneurons juxtacellularly. This may be because these interneurons represent a smaller popula- tion and have lower firing activity or have less correlation with the task behavior. Functional diversity in pyramidal cells and interneurons We cleanly isolated 25,26 a total of 166 neurons from multiunit recordings in deep layers of the forelimb area. These neurons were
1587

Firing rate (Hz)

Unit 2

Unit 3

a r t ic l e s a
3 4 5A + 5B Evoked 6 100 m Movement 1 ms, 0.25 mV 0 100 Hold-related Pull 10 5 0 1 0s 10 5 0 0 50 Hz 1s 1.9 Hz Push

Lever Firing rate (Hz)

b
1 2 3 4 5A 5B

Pre-movement

Lever Firing rate (Hz) 20 10 0 1

Pull 20 10 0s 0 0

Push

1s 6.1 Hz

Spontaneous

40 Hz

0 100 ms Push

0 100 Movement-off Pull 20 10 0 1 0s 20 10 0 0 50 Hz

0 Push

100 ms

c
1 2 3 4 5A 5B

Lever 15 10 5 0 Firing rate (Hz)

Pull 15 10 5 0

d
1 2 3

Lever Firing rate (Hz)

0s

0 20 Hz

1s 2.8 Hz 4 5A

1s 8.8 Hz

2009 Nature America, Inc. All rights reserved.

0 100

5B 0 100 ms

0 100

0 100 ms

e
3 4 5A 5B 6

Post-movement

Lever Firing rate (Hz) 40 20 0 1

Pull

Pull (end) 40 20 0s 1 0s 30 Hz 0 0

Push

f
WM

1s FL Str

5.0 Hz 50 m

0 100

Str

10 m

0 100 ms

Figure 2 Diverse functional activation of identified pyramidal cells. ( a) Layer 5B pyramidal cell showing a gradual increase in firing activity during lever holding and a decrease during pull and push movements (hold-related activity). Left, the morphology of the recorded neuron (black, soma and dendrites; red, axons). Top, task-related firing activity aligned at the onset of pull or push movement. Bottom, spike waveforms (left), spontaneous and evoked (I) spiking traces (middle), and auto-correlation histogram (right, 1-ms bins; gray, ongoing firing rate). +, positive. ( b) Layer 5B pyramidal cell showing transiently increased activity just before the onset of pull movement (pre-movement activity). ( c) Layer 5A pyramidal cell showing increased activity with pull movement (movement activity). Arrow indicates an activity reduction in the nonpreferred direction. ( d) Layer 4, spiny, star-like pyramidal cell showing an abrupt decrease in tonic and constant firing during pull movement (movement-off activity). Note that layer 4 exists in the rat forelimb motor area14,15. (e) Layer 6 pyramidal cell showing a peak activity at the end of, rather than during, pull movement (post-movement activity). (f) Axons (triangles, reconstructed) running through the white matter (WM) and striatum (Str) from the neuron shown in b. Inset, location of the soma (asterisk). FL, forelimb area. (g) Spine-rich dendrites of the neuron shown in d.

divided into regular-spiking neurons (n = 150), which should mainly comprise excitatory pyramidal cells, and fast-spiking neurons ( n = 16) on the basis of spike duration 30,31 (n = 22 rats; Fig.4a). The regular-spiking population yielded a diverse range of task-related activity: 41 hold-related, 10 pre-movement, 51 movement, 14 movement-off, 3 post-movement and 31 nontaskrelated neurons (Fig. 4a). In contrast, the majority of fast-spiking neurons exhibited movement activity (13 movement, 1 pre-move- ment, 1 post-movement and 1 nontask-related neurons; 2 test, P < 0.001 between regular-spiking and fast-spiking for movement). The regular-spiking population, however, might include various nonfast-spiking interneurons17,18, which cannot be distinguished from pyramidal cells with similar spike waveforms. Thus, nonfastspiking interneurons may have contributed to the functional diver- sity of regular-spiking population. Juxtacellular recordings provided stronger evidence for the dif- ference between excitatory and inhibitory neuronal activity. Of 87 recorded neurons (n = 69 rats), 68 were injected with biocytin/ Neurobiotin for morphological identification and 38 of these (56%) were visualized successfully. We identified these neurons as 27 taskrelated and 2 nontask-related pyramidal cells and 9 task-related
1588

(including 8 parvalbumin positive) fast-spiking interneurons. The remaining neurons were further classified as putative pyramidal cells (n = 41) and putative fast-spiking interneurons (n = 8, 1 nontask related) on the basis of the distribution of ongoing activity and spike width in identified neurons (Fig. 4b). Identified and putative pyramidal cells (hereafter, pyramidal cells) exhibited a wide range of task-related activity in relation to cortical position (12 (including 3 identified) hold-related, 7 (including 4 identified) pre-movement, 38(including 14 identified) movement, 6 (including 3 identified) movement-off, 4 (including 3 identified) post-movement and 3 (including 2 identified) nontask-related neurons; Fig. 4b), sug- gesting that there is a sequence of concurrent multi-layer integration (Fig. 1e), rather than a layer-by-layer conversion for sequential motor phases. In contrast with the diverse pyramidal-cell activity, all 16 (9 identified) task-related fast-spiking interneurons found in layers 26 exhibited only movement activity (2 test, P < 0.001 between pyramidal cells and interneurons; Fig. 4b). Pyramidal cells and fastspiking interneurons with movement activity were identified in all of layers 2/3, 4, 5 and 6. Thus, juxtacellular recordings revealed the multilayer activation and the different functional diversity of pyramidal cells and fast-spiking interneurons.
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 3 Movement-associated activation in identified fast-spiking interneurons. (a,b) Fast-spiking interneurons in layer 3 (a) and layer 5A (b) showing increased activity with pull movement (movement activity). Note that these biocytin (BC)- or Neurobiotin (NB)-visualized neurons (arrowheads) expressed parvalbumin (PV), a fast-spiking interneuron specific marker, but not calretinin (CR), a marker of nonfast-spiking interneurons. They exhibited a higher evoked firing rate with a narrower spike width than pyramidal cells, which is characteristic of fastspiking interneurons.

Movement 1 2 3 100 m BC PV CR 25 m 1 ms, 0.25 mV Lever 40 Firing rate 20 (Hz) 0

Pull 40 20 1 0s 0 0

Push

1s 50 Hz 8.5 Hz

Spontaneous Evoked

Activity properties of pyramidal cells and interneurons The juxtacellularly recorded pyramidal cells had significantly lower rates of ongoing activity in superficial cortical layers (depth of <800 m) than in deeper layers (superficial: 5.1 6.6 Hz, n = 32; deep: 8.9 7.3 Hz, n = 38; P < 0.03; identified layer 2/3: 2.2 2.6 Hz, n = 6; identified layers 56: 9.4 8.6 Hz, n = 21; P < 0.003; t test; Fig. 5 and Supplementary Figs. 6 and 7). Similarly, the fast-spiking interneurons fired at lower rates in superficial layers than in deep layers, although the difference was not substantial (Supplementary Fig. 7). Moreover, baseline activity in the lever-hold period was much lower in regular-spiking neurons (multiunit; Supplementary Fig. 7) and pyramidal cells (juxtacellular; Supplementary Fig. 7) than in fastspiking interneurons (except for one pre-movement and one postmovement neuron in multiunit analysis). Notably, baseline activity was much higher in movement-off pyramidal cells than in movement pyramidal cells, indicating that the movement-off activity was not inactivation of the movement activity in a nonpreferred direction (multiunit: movement, 3.2 3.4 Hz; movement-off, 9.6 7.4 Hz; t test, P < 0.001; juxtacellular: movement, 5.1 6.5 Hz; movement-off, 20.1 10.0 Hz; P < 0.001; identified movement, 7.5 8.0 Hz; identi- fied movement-off, 22.3 11.5 Hz; P < 0.02). The temporal profile of movement activity in fast-spiking inter neurons (multiunit, n = 11; juxtacellular, n = 16) was parallel with, or slightly slower than, that of regular-spiking neurons (multiunit, n = 25) or pyramidal cells (juxtacellular, n = 28) (Fig. 5a). The peak latency of fast-spiking interneurons was delayed relative to pyramidal cells (fast-spiking, 96.3 40.1 ms; pyramidal, 39.3 50.6 ms; t test, P < 0.001; identified fast-spiking, 113.3 43.6 ms, n = 9; identified pyramidal, 16.4 39.8 ms, n = 11; t test, P < 0.001; Fig. 5b). Invitro experiments have suggested that motor information flows from a superficial-layer loop to a deep-layer loop in cortical microcircuits16. Pyramidal cell and fast-spiking interneuron latencies, however, did
Ongoing activity (Hz)

0 100 Push 50 25 0

0 100 ms

Movement 3 4 5A 5B NB PV CR Lever 50 Firing rate 25 (Hz) 0

Pull

0s

0 50 Hz

1s 16.1 Hz

2009 Nature America, Inc. All rights reserved.

0 100

0 100 ms

Multiunit 60 40 20 RS FS

RS Hold-related Pre-movement Movement FS Movement-off Post-movement Nontask-related PC FS

not differ on the basis of layer in our self-paced movement condition (superficial versus deep; P > 0.3 and P > 0.2, respectively). There were differences in the direction coding of pyramidal cells and fast-spiking interneurons, although pull and push movements were not strictly equivalent in our simple task condition. Consistent with previous studies27,32, movement activity was reduced from base- line levels (in lever-hold period) during the nonpreferred movement (push or pull) in 16 of 51 regular-spiking neurons and 15 of 38 pyram- idal cells (Figs. 2c and 5c and Supplementary Fig. 6). However, only 2 of 13 (multiunit) and 0 of 16 (juxtacellular) fast-spiking inter neurons with movement activity exhibited such antagonistic inacti- vation (Figs. 3 and 5c). Thus, the direction specificity of fast-spiking interneurons was substantial, but was significantly weaker than that of regular-spiking neurons or pyramidal cells (multiunit: regularspiking, 0.50 0.29 for direction specificity; fast-spiking, 0.21 0.16; t test, P < 0.001; juxtacellular: pyramidal cells, 0.48 0.29; inter neurons 0.31 0.19; P < 0.025; identified pyramidal cells, 0.57 0.35; identified interneurons, 0.35 0.22; P = 0.11; Fig. 5ce). Furthermore, multiunit recordings at identical electrode sites revealed that 71% of regular spikingregular spiking neuron pairs with movement activity had similar direction specificity (relative direction specificity, 0.18 0.55, n = 118 pairs, t test, P < 0.001; Fig. 5d), whereas the direction specificity of fast-spiking interneurons seemed to be independent of nearby regular-spiking neurons (0.06 0.59, n = 20, P > 0.6). This
Figure 4 Contrasting functional diversity between pyramidal cells and fast-spiking interneurons. (a) Functionally different groups of regularspiking (RS) and fast-spiking (FS) neurons in the multiunit analysis. Left, classification of isolated units as regular-spiking (gray) and fast-spiking (black) by spike duration (red bar). Filled symbols represent movement activity, open symbols represent other task-related activity and pluses represent nontask-related activity. Right, population ratios of functionally different neuron groups for regular-spiking and fast-spiking neurons. (b) Functionally different groups of pyramidal cells (PCs) and fast-spiking interneurons in the juxtacellular analysis. Left, distribution of identified pyramidal cells (colored triangles) and fast-spiking interneurons (colored circles) and the classification of unidentified neurons as putative pyramidal cells (gray triangles) and putative fast-spiking interneurons (gray circles) by spike width (red bar) and ongoing firing rate. Right, the cortical distribution of functionally different neuron groups. Blue symbols represent layer 2/3, green represents layer 4, orange represents layer 5A, red represents layer 5B and purple represents layer 6.

0 0.2 0.4 0.6 0.8 1.0 1.2 Spike duration (ms) Juxtacellular 50 25 Depth (m) FS 0 400 800 1,200 1,600

b
Ongoing activity (Hz)

PC

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

H Pr old e- -re m la ov te d M em M ove ent o Po ve me st me nt -m n ov t-o em ff en t M ov em en t

0 0.1 0.2 0.3 0.4 0.5 Spike width (ms)

1589

a r t ic l e s
Multiunit (Pull) Figure 5 Time course and specificity in Latency (movement) Activity (movement) 15 RS pyramidal cells and fast-spiking interneurons (Pull) 0 0 PC and FS PC FS with movement activity. (a) Population 40 FS (averaged) firing activity for functionally 400 0 different neuron groups (black line indicates Juxtacellular hold-related activity, gray line indicates 800 30 PC pre-movement, red line indicates movement 0 1,200 and the gray dotted line indicates movement-off) 60 FS aligned with pull onset (0 s) in multiunit and 1,600 0 juxtacellular recordings. (b) Left, cortical 1.0 0.5 0 0.4 0.2 0 0.2 50 75 0 50 0.5 0 25 100 position and onset (left edge) to peak latency Time (s) Time (s) Firing rate (Hz) (symbol) of movement activity for individual Multiunit Multiunit Juxtacellular neurons. Symbols are as described in Figure 4. 100 100 RS-RS Direction RS 1 Open and filled arrowheads indicate mean FS (P PN )/(P + PN ) 0 peak latency in pyramidal cells and fast-spiking 10 10 1 interneurons, respectively. Right, cortical 118 1 Size position and baseline (left edge) to peak activity RS-FS (L S)/(L + S) 1 1 1 (symbol) of movement activity for individual 0 neurons. (c) Normalized changes in movement 0.1 0.1 1 Velocity 0.1 1 10 100 0.1 1 10 100 20 1 activity in preferred and nonpreferred directions (F S )/(F + S ) Nonpreferred direction Neuron pairs (pull/push; normalized with baseline firing rate) 1 0 1 Normalized firing activity (baseline = 1) Specificity in multiunit and juxtacellular analyses. (d) Left, specificities of movement activity for direction Juxtacellular PC n = 50 0 Direction (upper), size (middle) and velocity (lower) in 30 Size Velocity FS n = 12 regular-spiking (n = 51, 25 and 25 available for 400 20 analysis, respectively) and fast-spiking (n = 13, 13 and 11) neurons from multiunit recordings. 800 10 Filled symbols denote significance (P < 0.05) 1,200 0 and open symbols are nonsignificant (P > 0.05) 0 10 20 30 in d and e. Right, coherence of direction 1,600 During current injections 1 0 1 1 0 1 0.5 0 0.5 specificity in local regular spikingregular Mean lever movement (mm s1) Specificity Specificity Specificity spiking and regular spikingfast spiking neuron pairs with movement activity. Positive and negative values indicate the same and opposite preferred directions in individual neuron pairs, respectively. See the Online Methods for details. (e) The cortical position and specificities of movement activity for direction (left), size (middle) and velocity (right) in pyramidal cells ( n = 33, 25 and 27, respectively) and fast-spiking interneurons ( n = 16, 14 and 14) from juxtacellular recordings. ( f) There were no effects of juxtacelluar current injection into single pyramidal cells (n = 50) or interneurons (n = 12) on lever-movement performance. Population activity (Hz) During injection intervals

2009 Nature America, Inc. All rights reserved.

suggested that fast-spiking interneurons might not modulate selective movement activity of regular-spiking neurons. Motor cortex neurons principally encode the force33 and direc- tion32 of movement, but their activity is often modulated by the size and velocity of movement34,35. Some pyramidal cells and fast-spiking interneurons with movement activity conveyed modulatory informa- tion on the size or velocity of movement in our constant force condi- tion (Fig. 5d,e). The size (multiunit, P > 0.06; juxtacellular, P > 0.3) and velocity (multiunit, P > 0.5; juxtacellular, P > 0.5) specificities of fast-spiking interneurons did not differ significantly from those of regular-spiking neurons or pyramidal cells. Furthermore, we found no significant differences in the above specificities between superficial and deep layers (size, P > 0.8; velocity, P > 0.9). Unlike in the whisker motor cortex36, juxtacellular current injection into single neurons did not affect the performance of forelimb movements (Fig. 5f). Synaptic interactions among functionally different neurons To examine the network machinery bridging the sequential motor phases, we searched for excitatory synaptic interactions between neuron pairs2831 in the multiunit (n = 166) and juxtacellular (n = 30) data pool (total of 872 pairs in 22 rats). We examined putatively monosynaptic excitatory connectivity from a hold-related to a movement regular-spiking neuron and from a pre-movement regular-spiking neuron to a movement fast-spiking neuron, respec- tively, along the flow of motor information (Fig. 6a,b). We created cross-correlation histograms and found clear, several-fold, single peaks around 1.5 ms after presynaptic neuron firing. Unexpectedly,
1590

Depth (m)

a pre-movement regular-spiking neuron excited not only a movement fast-spiking interneuron, but also a hold-related regular-spiking neu- ron against the motor information flow (Fig. 6c). This result suggests that the network machinery for voluntary movement expression is not a simple feedforward network for processing sequential motor phases. The single peak observed in the cross-correlation histogram (asymmetric, short latency with small jitter, and statistically large amplitude) most likely represented monosynaptic excitation between two distinct neurons (see refs. 28, 30 and 31), as the spike waveform and amplitude ratio in raw and averaged traces were completely different between the two neurons, the peak latency (typically, 12 ms) was too fast to be burst-spiking of one regular-spiking neuron or a polysynaptic response and too late to be gap junctionmediated synchronization, and the peak position was too tightly asymmetric to be common-input response. In addition, spike clusters including burst activity from one neuron had been concatenated into a single spike cluster in advance and the single peak was preserved after all overlapping outlier spikes were excluded in a two-dimensional plane out of 17 spike-feature parameters (Supplementary Fig. 8). The sin- gle peak disappeared after shuffling trials in the triggering neuron (Supplementary Fig. 8). However, we cannot exclude the possibility that one presynaptic neuron commonly activated two postsynaptic neurons with different synaptic delays. We obtained 19 putatively monosynaptic excitatory connections out of the 872 possible neuron pairs. We did not examine inhibitory connections and suspected common-input responses with no delay, as spikes can decrease or increase over the 0-ms bin in a cross-correlogram
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

Relative direction specificity

Depth (m)

Preferred direction

a r t ic l e s a
Spike ac 10 5 0 1 Time (s) 0 Spike feature Pull RS (hold-related) 4 Hz 2 0 0 25 ms 2 0 1 ms, 0.1 mV Firing rate (Hz) 1 Time (s) 0 10 5 0 1 0 Time (s) 500 RS (pre-movement) cc Spike feature cc RS (movement)

b
Spike

RS (pre-movement)

100 50

cc

FS (movement)

a.c.

25 80 40 0 1 0 Time (s) FS (movement)

Firing rate (Hz)

2009 Nature America, Inc. All rights reserved.

250 Figure 6 Excitatory synaptic interactions between functionally different neurons. (a) Putatively 0 Spike 0 25 monosynaptic excitatory connection from a ac regular-spiking neuron with hold-related activity (blue) 160 to a regular-spiking neuron with movement activity (red). 80 Spike For individual neurons, the top panel depicts spike 2 feature Firing 0 waveforms in each channel, the middle panel shows the RS (hold-related) 1 0 rate 1 auto-correlation histogram (ac, 1-ms bins, 50 ms) and Time (s) (Hz) 0 the bottom illustrates task-related firing activity aligned 1 0 at the onset of pull movement (0 s). Between these Time (s) 100 cc neurons, the top panels shows the cross-correlation histogram (cc, 1-ms bins, 25 ms) triggered by the former 20 50 neuron and the bottom shows their spike distribution in a 10 two-dimensional plane of spike features (total 17 dimensions). 0 0 25 The gap at 0 ms (center) in the cross-correlation histogram is a 0 1 0 result of a lack of spike-detection within 0.5 ms around triggering Time (s) 30,31 . (b) Excitatory connection from a regular-spiking neuron spikes with pre-movement activity (green) to a fast-spiking interneuron with movement activity (gray). ( c) Excitatory connections from a regular-spiking neuron with pre-movement activity (green) to a fast-spiking interneuron with movement activity (gray) and to a regular-spiking neuron with hold-related activity (blue).

by other unknown mechanisms. Notably, we frequently found syn- aptic interactions between presynaptic regular-spiking neurons and postsynaptic neurons engaging in various functional activities, irrespective of the temporal order expected from their functional activities (for example, pre-movement to hold-related). In particu- lar, pre-movement regular-spiking neurons, which might participate in motor initiation, excited similar pre-movement regular-spiking neurons as well as regular-spiking and fast-spiking neurons with dif- ferent functional activities (three pre-movement, four hold-related and one movement regular-spiking neurons, three movement fastspiking neurons; Supplementary Fig. 8). Although more putative monosynaptic connections must be examined for rigorous quanti fication, many of the excitatory connections originated from the small population of pre-movement neurons (see Fig. 4). In at least 12 of the 19 pairs, the synaptic excitation between functionally different neurons was effective during the task performance (Supplementary Fig. 8). The excitatory synaptic communications between functionally different groups of cortical neurons may integrate motor information relevant to self-initiated voluntary movement. DISCUSSION Multilayer activation for sequential motor phases Our behavioral and electrophysiological approach revealed that exci- tatory pyramidal cells in multiple layers of the motor cortex have tem- porally different activations, which probably correspond to sequential motor phases, that most of the inhibitory fast-spiking interneurons were activated during movement expression, and that synchronous spike responses were evoked between functionally different neurons in self-initiation of voluntary forelimb movement. Several in vitro studies have predicted a directional signaling path- way through particular layers of motor cortex1416. An in vivo study reported that putative pyramidal cells in layer 5, but not in layers
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

2/3or 6, were activated on a specific phase of the locomotion cycle19. However, our findings of simultaneous multilayer activation in juxta cellular visualization (Figs. 4 and 5) exclude the possibility that motor information may be converted for sequential motor phases layer by layer. Some striatal neurons had temporally different activations that were similar to those of cortical neurons during the same fore- limb movement (data not shown). Thus, the multilayer activation of pyramidal cells might result from global and concurrent interactions with other cortical areas, as well as subcortical structures for each motor phase. How the information for each motor phase flows from one layer to another and how nonfast-spiking interneurons that were not characterized here may be recruited for coordinating this flow remain unknown. Command-shaping function of fast-spiking interneurons Inhibitory fast-spiking interneurons, typically morphological basket cells or chandelier cells making axo-somatic or axo-axonic synapses, may effectively gate spike outputs from the pyramidal cells18. In fact, putative fast-spiking interneurons of the rabbit motor cortex are activated on the opposite phase to activation of putative pyramidal cells in every cycle of locomotion, which consists of continuous and complex limb movements under subcortical pattern controls19. In contrast, our results from the juxtacellular analysis indicate that the majority of identified fast-spiking interneurons probably participate in an on-going modulation of command-like activity of pyramidal cells during the execution of a single voluntary movement (Figs. 4 and 5). Notably, the absence of fast-spiking interneurons with movement-off activity excludes the possibility of intracortical gating of motor command. Several recent studies have suggested that excitatory and inhibitory inputs should be balanced to regulate spontaneous or sensory-evoked neuronal activity in sensory cortex neurons under anesthesia 37,38.
1591

a r t ic l e s
There is a balance between inhibitory and excitatory synaptic inputs in auditory37 and somatosensory38 cortex neurons, and inhibitory synaptic inputs always succeed excitatory synaptic inputs, making output timing more precise. Fast-spiking interneurons of the motor cortex were also activated slightly after command-like activation of pyramidal cells in behaving rats (Fig. 5a,b), suggesting that balanced and delayed inhibition may achieve temporal sharpening of motor command. Moreover, fast-spiking interneurons with movement activ- ity might suppress other functional activities such as hold-related activity during motor execution. The activity of fast-spiking interneurons was less specific to move- ment direction than that of regular-spiking neurons and was inde- pendent of the preferred direction of neighboring regular-spiking neurons in the rat motor cortex (Fig. 5ce). Thus, the balanced inhi- bition might work effectively at the surround as well as the center of intended movement, presumably to sharpen the direction specificity of pyramidal cells through the so-called iceberg effect37. In fact, a pharmacological blockade of GABAergic inhibition in the primate motor cortex increases the phasic discharge activity in a preferred direction and decreases its directionality39. Alternatively, fast-spiking interneurons activated by movement pyramidal cells might accom- plish recurrent inhibition of other pyramidal cell activity to suppress competing (unnecessary) movements. In the primate motor cortex, where putative fast-spiking interneurons show directional tuning properties27, the cortico-spinal pyramidal cells elicit disynaptic inhibitory responses in neighboring neurons40,41. However, putative fast-spiking interneurons and nearby putative pyramidal cells in the primate prefrontal cortex have increased activation in the same or in the opposite preferred direction, depending on task conditions29,42,43. Although fast-spiking interneurons probably underlie temporal or spatial shaping of motor command with a subgroup of pyramidal cells across all layers of motor cortex, an intracellular measurement of excitatory and inhibitory synaptic inputs37,38 will be needed to conclude whether the fast-spiking interneurons are engaged in bal- anced inhibition, recurrent inhibition or both. Synaptic communications in cortical microcircuits How do functionally different neurons integrate motor information to express self-initiated voluntary movement? The spiking of a single motor-cortex neuron evokes an excitatory or inhibitory postsynap- tic potential in neighboring neurons in awake monkeys44. Our mul- tiunit recording also revealed putatively monosynaptic connectivity between pairs of regular-spiking neurons and between regular-spiking and fast-spiking neurons in the deep layers of the rat motor cortex (Fig. 6). Notably, monosynaptic excitation was detected in regular spikingregular spiking and regular spikingfast spiking neuron pairs with different functional activities, unlike in the primate prefrontal cortex, where the phasic activation of neuron pairs showing excitatory communications does not have a large temporal difference during a working-memory task29. As the direction of synaptic excitation did not always coincide with the temporal order of functional activity (for example, pre-movement to hold-related in Fig. 6c), serial motor phases are unlikely to be relayed solely by the intramicrocircuit syn- aptic interaction. It is possible that specific cortico-cortical path- ways and cortico-subcortical loops also participate in this relay. We found that pre-movement regular-spiking neurons excited excitatory and inhibitory neurons with similar or different functional activity (Fig.6b,c and Supplementary Fig. 8). A recurrent network formed by similar pre-movement regular-spiking neurons may accomplish temporal integration45 of motor signals to decide the timing of selfinitiated movement. Our juxtacellular recordings revealed that some
1592

of the pre-movement regular-spiking neurons sent axonal projections to subcortical structures (Fig. 2b,f and Supplementary Fig. 6). It is possible that these pre-movement regular-spiking neurons are impor- tant for intracortical and cortico-subcortical integration of motor information to coordinate sequential motor phases. Alternatively, the short-latency peak that appeared in our crosscorrelogram may arise from a common excitatory drive on the two recorded neurons with different delays. In this case, spike correlations do not represent direct synaptic interactions between functionally different neurons, but may represent the recruitment of functionally heterogeneous neurons by a presynaptic neuron through highly reli- able synaptic excitation. This possibility cannot be excluded by the present cross-correlation analysis. In either of the above cases, the microcircuit of motor cortex is suggested to process motor informa- tion with a specifically designed neuronal wiring between different functional cell assemblies. Technical advantages and functional implications Juxtacellular recording enabled us to examine accurate spike events and morphological features of a single neuron in arbitrary layer of the cerebral cortex2022 and in many subcortical structures 23,24 (Supplementary Fig. 6) in vivo. The access to deep brain structures is a technical advantage over calcium-imaging with multi-photon laser-scanning microscopy. Although juxtacellular recording has been conducted in anesthetized2023 or sleeping/waking24 animals, the juxtacellular identification of naturally spiking neurons has not been attempted in actively task-performing animals, to the best of our knowledge (see ref. 46 for stimulation). For reliable visual discrimi- nation, we could obtain only one (or a few) juxtacellularly recorded neuron(s) in each rat, which had to be killed within 24 h20. Therefore, juxtacellular experiments focusing on cognitive or motor functions require a substantial number of head-restraint animals performing an operant behavioral task. However, it is usually difficult and time con- suming to train head-restraint rodents, unlike primates47, to learn an operant forelimb-movement task. Our multi-rat task-training system markedly improved the efficiency of preparing task-performing rats for final experiments (Supplementary Figs. 13 and Supplementary Videos 1 and 2). This method should be also useful for other physi- ological measurements, such as whole-cell patch-clamp recordings36, voltage-sensitive dye imaging48 and calcium-imaging with multiphoton laser-scanning microscopy49 in head-restraint rodents. Using the experimental approach, these results shed light on the intracortical mechanism of motor-command generation in the expression of voluntary movements via circuit level analysis. Our results led to the hypothesis of multilayer simultaneous processing for sequential motor phases. In this hypothesis, a group of pyramidal cells in superficial and deep layers may become gradually active to prepare for an intended movement, whereas fast-spiking interneurons do not affect the preparatory activity during the stationary behavior. Once additional pyramidal cells begin to discharge, they self-amplify the transient signal by exciting functionally similar pyramidal cells to initiate the movement. The reverberating excitation may decide the timing as well as the direction of movement. Subsequently, the major population of pyramidal cells across all layers generates an output command to execute the movement and fast-spiking interneurons are coactivated to shape the motor command via balanced inhibition or recurrent inhibition. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

2009 Nature America, Inc. All rights reserved.

a r t ic l e s
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank G. Buzski and J. Tanji for helpful comments and discussion, M. Fujii, K. Ishii, M. Kobayashi, R. Nakatomi and S. Tanaka for technical assistance, and K. Ohara for developing the multi-rat task-training system. This work was supported by the institutional research grants from RIKEN (T.F.) and by Grants-in-Aid for Scientific Research from Ministry of Education, Culture, Sports, Science and Technology (18019041 and 18700386 to Y.I. and 40218871 to T.F.). AUTHOR CONTRIBUTIONS Y.I. designed the experiments. Y.I., R.H. and H.A. performed the experiments. Y.I. and T.T. analyzed the data. Y.I. and T.F. wrote the paper.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
1. Deecke, L., Scheid, P. & Kornhuber, H.H. Distribution of readiness potential, premotion positivity, and motor potential of the human cerebral cortex preceding voluntary finger movements. Exp. Brain Res. 7, 158168 (1969). 2. Hashimoto, S., Gemba, H. & Sasaki, K. Premovement slow cortical potentials and required muscle force in self-paced hand movements in the monkey. Brain Res. 197, 415423 (1980). 3. Tanji, J. & Evarts, E.V. Anticipatory activity of motor cortex neurons in relation to direction of an intended movement. J. Neurophysiol. 39, 10621068 (1976). 4. 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, 155166 (1987). 5. Hyland, B. Neural activity related to reaching and grasping in rostral and caudal regions of rat motor cortex. Behav. Brain Res. 94, 255269 (1998). 6. Chapin, J.K., Moxon, K.A., Markowitz, R.S. & Nicolelis, M.A.L. Real-time control of a robot arm using simultaneously recorded neurons in the motor cortex. Nat. Neurosci. 2, 664670 (1999). 7. Laubach, M., Wessberg, J. & Nicolelis, M.A.L. Cortical ensemble activity increasingly predicts behaviour outcomes during learning of a motor task. Nature 405, 567571 (2000). 8. Bauswein, E., Fromm, C. & Preuss, A. Corticostriatal cells in comparison with pyramidal tract neurons: contrasting properties in the behaving monkey. Brain Res. 493, 198203 (1989). 9. Turner, R.S. & DeLong, M.R. Corticostriatal activity in primary motor cortex of the macaque. J. Neurosci. 20, 70967108 (2000). 10. Fromm, C. Contrasting properties of pyramidal tract neurons located in the precentral or postcentral areas and of corticorubral neurons in the behaving monkey. Adv. Neurol. 39, 329345 (1983). 11. Donoghue, J.P. & Wise, S.P. The motor cortex of the rat: cytoarchitecture and microstimulation mapping. J. Comp. Neurol. 212, 7688 (1982). 12. Rouiller, E.M., Moret, V. & Liang, F. Comparison of the connectional properties of the two forelimb areas of the rat sensorimotor cortex: support for the presence of a premotor or supplementary motor cortical area. Somatosens. Mot. Res. 10, 269289 (1993). 13. Brecht, M. et al. Organization of rat vibrissa motor cortex and adjacent areas according to cytoarchitectonics, microstimulation and intracellular stimulation of identified cells. J. Comp. Neurol. 479, 360373 (2004). 14. Cho, R.-H., Segawa, S., Mizuno, A. & Kaneko, T. Intracellularly labeled pyramidal neurons in the cortical areas projecting to the spinal cord. I. Electrophysiological properties of pyramidal neurons. Neurosci. Res. 50, 381394 (2004). 15. Cho, R.-H. et al. Intracellularly labeled pyramidal neurons in the cortical areas projecting to the spinal cord. II. Intra- and juxta-columnar projection of pyramidal neurons to corticospinal neurons. Neurosci. Res. 50, 395410 (2004). 16. Weiler, N. et al. Top-down laminar organization of the excitatory network in motor cortex. Nat. Neurosci. 11, 360366 (2008). 17. Cauli, B. et al. Molecular and physiological diversity of cortical nonpyramidal cells. J. Neurosci. 17, 38943906 (1997). 18. Markram, H. et al. Interneurons of the neocortical inhibitory system. Nat. Rev. Neurosci. 5, 793807 (2004). 19. Beloozerova, I.N., Sirota, M.G. & Swadlow, H.A. Activity of different classes of neurons of the motor cortex during locomotion. J. Neurosci. 23, 10871097 (2003). 20. Pinault, D. A novel single-cell staining procedure performed in vivo under electrophysiological control: morpho-functional features of juxtacellularly labeled thalamic cells and other central neurons with biocytin or Neurobiotin. J. Neurosci. Methods 65, 113136 (1996). 21. Klausberger, T. et al. Brain state and cell typespecific firing of hippocampal interneurons in vivo. Nature 421, 844848 (2003). 22. de Kock, C.P.J., Bruno, R.M., Spors, H. & Sakmann, B. Layer- and cell typespecific suprathreshold stimulus representation in rat primary somatosensory cortex. J. Physiol. (Lond.) 581, 139154 (2007). 23. Mallet, N., Ballion, B., Le Moine, C. & Gonon, F. Cortical inputs and GABA interneurons imbalance projection neurons in the striatum of parkinsonian rats. J. Neurosci. 26, 38753884 (2006). 24. Lee, M.G., Manns, I.D., Alonso, A. & Jones, B.E. Sleep-wake related discharge properties of basal forebrain neurons recorded with micropipettes in head-fixed rats. J. Neurophysiol. 92, 11821198 (2004). 25. Harris, K.D. et al. Accuracy of tetrode spike separation as determined by simultaneous intracellular and extracellular measurements. J. Neurophysiol. 84, 401414 (2000). 26. Isomura, Y. et al. Integration and segregation of activity in entorhinal-hippocampal subregions by neocortical slow oscillations. Neuron 52, 871882 (2006). 27. Merchant, H., Naselaris, T. & Georgopoulos, A.P. Dynamic sculpting of directional tuning in the primate motor cortex during three-dimensional reaching. J. Neurosci. 28, 91649172 (2008). 28. Csicsvari, J., Hirase, H., Czurko, A. & Buzski, G. Reliability and state dependence of pyramidal cell-interneuron synapses in the hippocampus: an ensemble approach in the behaving rat. Neuron 21, 179189 (1998). 29. Constantinidis, C., Williams, G.V. & Goldman-Rakic, P.S. A role for inhibition in shaping the temporal flow of information in prefrontal cortex. Nat. Neurosci. 5, 175180 (2002). 30. Barth, P. et al. Characterization of neocortical principal cells and interneurons by network interactions and extracellular features. J. Neurophysiol. 92, 600608 (2004). 31. Sirota, A. et al. Entrainment of neocortical neurons and gamma oscillations by the hippocampal theta rhythm. Neuron 60, 683697 (2008). 32. Georgopoulos, A.P., Kalaska, J.F., Caminiti, R. & Massey, J.T. On the relations between the direction of two-dimensional arm movements and cell discharge in primate motor cortex. J. Neurosci. 2, 15271537 (1982). 33. Evarts, E.V. Relation of pyramidal tract activity to force exerted during voluntary movement. J. Neurophysiol. 31, 1427 (1968). 34. Humphrey, D.R., Schmidt, E.M. & Thompson, W.D. Predicting measures of motor performance from multiple cortical spike trains. Science 170, 758762 (1970). 35. Fromm, C. & Evarts, E.V. Relation of size and activity of motor cortex pyramidal tract neurons during skilled movements in the monkey. J. Neurosci. 1, 453460 (1981). 36. Brecht, M., Schneider, M., Sakmann, B. & Margrie, T.W. Whisker movements evoked by stimulation of single pyramidal cells in rat motor cortex. Nature 427, 704710 (2004). 37. Wehr, M. & Zador, A.M. Balanced inhibition underlies tuning and sharpens spike timing in auditory cortex. Nature 426, 442446 (2003). 38. Okun, M. & Lampl, I. Instantaneous correlation of excitation and inhibition during ongoing and sensory-evoked activities. Nat. Neurosci. 11, 535537 (2008). 39. Matsumura, M., Sawaguchi, T. & Kubota, K. GABAergic inhibition of neuronal activity in the primate motor and premotor cortex during voluntary movement. J. Neurophysiol. 68, 692702 (1992). 40. Stefanis, C. & Jasper, H. Recurrent collateral inhibition in pyramidal tract neurons. J. Neurophysiol. 27, 855877 (1964). 41. Georgopoulos, A.P. & Stefanis, C.N. Local shaping of function in the motor cortex: motor contrast, directional tuning. Brain Res. Rev. 55, 383389 (2007). 42. Wilson, F.A.W., Scalaidhe, .S.P. & Goldman-Rakic, P.S. Functional synergism between putative -aminobutyratecontaining neurons and pyramidal neurons in prefrontal cortex. Proc. Natl. Acad. Sci. USA 91, 40094013 (1994). 43. Rao, S.G., Williams, G.V. & Goldman-Rakic, P.S. Isodirectional tuning of adjacent interneurons and pyramidal cells during working memory: evidence for microcolumnar organization in PFC. J. Neurophysiol. 81, 19031916 (1999). 44. Matsumura, M. et al. Synaptic interactions between primate precentral cortex neurons revealed by spike-triggered averaging of intracellular membrane potentials in vivo. J. Neurosci. 16, 77577767 (1996). 45. Okamoto, H. & Fukai, T. Recurrent network models for perfect temporal integration of fluctuating correlated inputs. PLoS Comput. Biol. 5, e1000404 (2009). 46. Houweling, A.R. & Brecht, M. Behavioural report of single neuron stimulation in somatosensory cortex. Nature 451, 6568 (2008). 47. Isomura, Y. et al. Neural coding of attention for action and response selection in primate anterior cingulate cortex. J. Neurosci. 23, 80028012 (2003). 48. Ferezou, I. et al. Spatiotemporal dynamics of cortical sensorimotor integration in behaving mice. Neuron 56, 907923 (2007). 49. Dombeck, D.A. et al. Imaging large-scale neural activity with cellular resolution in awake, mobile mice. Neuron 56, 4357 (2007).

2009 Nature America, Inc. All rights reserved.

nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

1593

ONLINE METHODS
Animal preparation. All experiments were carried out in accordance with the Animal Experiment Plan approved by the Animal Experiment Committee of RIKEN. Adult Long-Evans rats (150250 g, male, Japan SLC) were handled briefly and adapted to a stainless steel cylinder in their home cage (Fig. 1a and Supplementary Fig. 1). A lightweight, sliding head attachment (custom-made by Narishige Co.) was surgically attached to the skull and reference electrodes were implanted above the cerebellum under 2% isoflurane (vol/vol) anesthesia. After recovering from surgery, the rats were deprived of drinking water in their home cages, although food was available ad libitum. Sufficient water was pro- vided as a reward for task performance in the laboratory. If necessary, the rats were provided an agar block (containing 15 ml of water) to maintain over 80% of their body weight. Behavioral training. To efficiently train head-restrained rats to repeat regular movements, we developed a multi-rat task-training system (now available from Ohara & Co.), which consisted of six training boxes that were each controlled by a computer (Supplementary Fig. 2). Each animal was placed in a separate train- ing box and trained to perform the voluntary (self-paced) forelimb-movement task while being head-restrained for 22.5 h a day (8 training days in 2 weeks; Supplementary Fig. 3). At the endpoint of this task learning, the rat spontaneously started each trial by pushing a constant-torque lever and holding the lever for more than 1 s with the right forelimb. If the rat pulled the lever for more than 60% of a full lever-shift (12 mm), then the rat received a drop of 0.1% saccharin water (wt/vol, 0.010.02 ml) in their mouth from a spout connected to a syringe pump after a 0.2-s delay. The lever position was continuously monitored by the task-control computer with an 8-bit encoder and forelimb motion was monitored with an infrared video camera. A high-tone sound (11 kHz) was accompanied by the reward delivery and a low-tone sound (3 kHz) was emitted after a task failure. During the early training stage, the rats were rewarded whenever they moved the lever, regardless of the amount of movement. Lever shift and hold-time requirements for reward acquisition progressively increased during the first week. Initially, rats had to sup- port their body by seizing a side handrail with their left forelimb; however, they were permitted to use an optional armrest during the second week. To start task and recording experiments simultaneously, rats experienced a waiting period of 1 h with the lever locked before every task learning session in the second week. Once the rats completed the operant task learning, they were transferred to the recording room to practice performing the same task in a novel environment for the final behavioral and electrophysiological experiment (Supplementary Fig. 4). The following day, the rats were subjected to a second surgery under isoflurane anesthesia and a tiny hole in the skull and dura matter were made above the left forelimb area (1.0 mm anterior, 2.53.0 mm lateral of bregma). The hole was covered with silicon sealant until the recording experiment commenced. Electrophysiological recordings. We obtained juxtacellular recordings20 from single neuron(s) in the left forelimb area of individual animals behaving in the task-control system 1 d after the second surgery. The glass electrode (BF150-75-10, Sutter Instrument) was prepared by a laser puller (P-2000, Sutter Instrument) and a blunt tip was created (Supplementary Fig. 5). The electrode was filled with 2% biocytin (wt/vol; Sigma-Aldrich) in 0.5 M NaCl or 2% Neurobiotin (wt/vol; Vector Laboratories) in 0.5 M KCl (1224 M). The electrode was inserted ver- tically into the forelimb area with a microdrive (LSS-8000 Inchworm, Burleigh Instrument) that was installed on a fine micromanipulator (1760-61, David Kopf Instruments) on a stereotaxic frame (SR-8N, Narishige). Juxtacellular signals were amplified with an intracellular amplifier (IR-283, Cygnus Tech) and sampled at 20 kHz (final gain of 1,000; band-pass filter, 300 Hz to 10 kHz) with a hard-disc recorder (DataMax II, R.C. Electronics). Subsequently, biocytin or Neurobiotin was electroporated into single recorded neurons with a positive current pulse (220 nA, 0.5-s duration, every 1 s, 1045 min). The electrode depth was used to estimate the cortical position of identified and unidentified neurons (r = 0.85, P <0.001; see Supplementary Fig. 7). Single neuron recordings were combined with extracellular multiunit recordings25,26 from the same area through a wire tet- rode or 16-channel silicon probe placed 1,200 m deep (sampling, 20 kHz; final gain, 2,000; original band-pass filter, 0.5 Hz to 10 kHz). In preliminary experi- ments, EMG recordings47 were obtained during task performance from the right upper forelimb and hindlimb through wire electrodes (Fig. 1d). To determine the coordinates of the forelimb area, we applied intracortical microstimulation

(about 50 A, 20 pulses at 500 Hz) to evoke selective EMG activity in the con- tralateral forelimb47 under urethane anesthesia (Supplementary Fig. 4). Histological staining. Under deep anesthesia, rats were killed by intracardial perfusion with cold saline followed by 4% paraformaldehyde (wt/vol) in 0.1 M phosphate buffer. Postfixed brains were sliced coronally into 50-m-thick serial sections26. Biotin/Neurobiotin-loaded neurons were visualized with streptavidin-AlexaFluor488 (Molecular Probes) in combination with immunostaining for parvalbumin and calretinin (antibody to parvalbumin (MAB1572, Chemicon International) and antibody to calretinin (AB5054, Chemicon) followed by AlexaFluor350- and AlexaFluor594-conjugated secondary antibodies (Molecular Probes), respectively). After fluorescence microscopy, labeled neurons were further examined using the avidinbiotinhorseradish peroxidase complex (Vectstain Elite ABC, Vector) with diaminobenzidine and nickel26. Visualized neurons were reconstructed with camera lucida following counterstaining with Neutral Red. Data analysis. Behavioral scores were compiled from trial-log files produced automatically by the task-control systems. Multiunit (and juxtacellular) record- ing data were processed to isolate spike events by the automatic spike-sorting program KlustaKwik using principal component analysis (17 feature dimensions for 4 channels; high-pass filter, 300 Hz; time resolution, 20 kHz; spike-detection inter- val, >0.5 ms). Next, the sorted spike clusters were combined, divided or discarded manually to refine single-neuron clusters by Klusters and NeuroScope25,26,50. The relationship of spike activity with behavioral performance was analyzed by MATLAB (MathWorks). The behavior-related spike activity was categorized into five temporally (functionally) different groups (Fig. 5a). Hold-related activity was a unimodal (increasing, decreasing or increasing to decreasing) activation during the lever-holding period. Pre-movement was a phasic activation start- ing less than 500 ms before the movement onset and falling down below the half peak at the movement onset. Movement was a phasic activation during the movement. Movement-off activity was an abrupt drop-off during the movement after constant tonic spiking in the intertrial-interval and lever-holding periods. Post-movement was a phasic activation showing a larger peak in the movement endaligned histogram than the movement onsetaligned histogram. The phasic activation/inactivation contained at least three consecutive bins (20 ms) 3 s.d. from the baseline activity (0.251 s before the movement onset); furthermore, these categorized activities were checked by a visual inspection. Ongoing activity was defined as the average firing rate (before juxtacellular current injection). The spike duration was the time from spike onset to the first positive peak and the spike width referred to the elapsed time above the half amplitude of the positive spike waveform. The onset of movement activity was defined as the first of three consecutive bins that exceeded 1 s.d. of the baseline activity. Peak activity was the average firing rate during five 20-ms bins centered at the peak; the peak activity of the nonpreferred movement was acquired from the time window corresponding to the peak bins of preferred movement. P NP The direction specificity was defined as , where P is the firing activity in P + NP the preferred direction (pull/push) and NP is firing activity in the nonpreferred direction. Likewise, the size specificity was defined as L S , where S is the firing L+S activity for small pull movement (<90% of full lever shift) and L is the firing activity for large pull movement (>90%). The velocity specificity was defined F S as , where S is the firing activity for the slower half of pull movements and F +S F is the firing activity for the faster half of pull movements. The significance of the direction, size or velocity specificity was judged in individual neurons using a Mann-Whitney U test. The relative direction specificity was the direction spe- cificity of one neuron, where a standard (positive) direction was the preferred direction of another neuron. The putative monosynaptic excitatory connectivity in a neuron pair was judged by a single asymmetric peak, its short latency (<2.5 ms) with small jitter (<2 ms) and larger amplitude (>3 s.d.; mean, 13.9 s.d.) than the baseline level (5 to 25 ms) in cross-correlation histogram (as verified in Supplementary Fig. 8)2731.
50. Hazan, L., Zugaro, M. & Buzski, G. Klusters, NeuroScope, NDManager: A free software suite for neurophysiological data processing and visualization. J. Neurosci. Methods 155, 207216 (2006).

2009 Nature America, Inc. All rights reserved.

nature NEUROSCIENCE

doi:10.1038/nn.2431

a r t ic l e s

Attention improves performance primarily by reducing interneuronal correlations


Marlene R Cohen & John H R Maunsell
Visual attention can improve behavioral performance by allowing observers to focus on the important information in a complex scene. Attention also typically increases the firing rates of cortical sensory neurons. Rate increases improve the signal-to-noise ratio of individual neurons, and this improvement has been assumed to underlie attention-related improvements in behavior. We recorded dozens of neurons simultaneously in visual area V4 and found that changes in single neurons accounted for only a small fraction of the improvement in the sensitivity of the population. Instead, over 80% of the attentional improvement in the population signal was caused by decreases in the correlations between the trial-to-trial fluctuations in the responses of pairs of neurons. These results suggest that the representation of sensory information in populations of neurons and the way attention affects the sensitivity of the population may only be understood by considering the interactions between neurons. The responses of sensory neurons are variable, and laboratory studies typically deal with this variability by averaging responses to many stimulus presentations. In the real world, however, people and animals must respond to individual stimulus events, and the brain is thought to compensate for neuronal variability by encoding sensory information in the responses of large populations of neurons. To understand the way sensory information guides behavior in everyday life, we need to understand the way information is encoded in populations of neurons. One way to identify the important aspects of a population code is to look at the differences between the neuronal representation of a sensory stimulus when it is used to guide behavior and when it is behaviorally irrelevant. Tasks that control attention provide a powerful way to manipulate behavioral relevance. Attention allows observers to select the most important stimuli and greatly improves perception of the attended location or feature. Attention modulates the firing rates of sensory neurons, typically increasing responses to attended stimuli14. This increased rate of firing acts to improve the signal-to-noise ratio of individual neurons5,6 and a recent study found that attention can cause a small additional reduction in the meannormalized variance (Fano factor) of the responses of some neurons in visual area V4 (ref. 7). However, the net effect of attention on the signal-to-noise ratio of single neurons is modest, suggesting that attention causes large improvements in psychophysical performance by affecting population responses in ways that cannot be measured in single neurons. Attention could also alter the reliability of neuronal representations by affecting the amount of noise that is shared across a population of neurons. Variability in a population partly depends on the variability of single neurons, but can depend greatly on the extent to which variability is shared across the population. The effect of correlated variability on population sensitivity depends on the way in which the population is read out8,9, but its effect can be far greater than the effect of independent variability of single neurons. If the noise in individual neurons is independent, averaging the responses of many neurons will lead to a very accurate estimate of the mean, no matter how noisy the individual neurons are. If, however, there are positive correlations in the trial-to-trial fluctuations of the responses of pairs of neurons, then the shared variability can never be averaged out, leading to a more variable (and less accurate) estimate of the mean activity in the population1012. We found that attention adaptively decreased correlated variability in a population of neurons in visual area V4 in a change-detection task. Furthermore, we found that this decrease accounted for the vast majority of the attentional improvement in the amount of sensory information encoded by the population and is probably the major contributor to the improved psychophysical performance. These results indicate that studies that focus on a single neuron, which necessarily ignore interactions between neurons, miss the most critical aspect of the way sensory information is encoded in populations of neurons. RESULTS We investigated the effect of attention on both single neuron responses and correlated variability by recording from populations of neurons in V4 using chronically implanted microelectrode arrays in two rhesus monkeys (Macaca mulatta). Each monkey had two arrays, allowing us to monitor populations of neurons in both hemispheres simultaneously. The diameter of V4 receptive fields is approximately equal to eccentricity13,14, so the receptive fields the neurons recorded in a hemisphere typically overlapped at least partially (Fig. 1a). We recorded from 376 single units and 2,746 multiunit clusters during 41 d of recording (including 66,578 simultaneously recorded pairs in the same hemisphere and 59,990 pairs in opposite hemispheres).

2009 Nature America, Inc. All rights reserved.

Howard Hughes Medical Institute and Harvard Medical School Department of Neurobiology, Boston, Massachusetts, USA. Correspondence should be addressed to M.R.C. (marlene_cohen@hms.harvard.edu). Received 9 July; accepted 30 September; published online 15 November 2009; doi:10.1038/nn.2439

1594

VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
Figure 1 Methods and behavior. (a) Center of 100 visual receptive fields for the multiunit signals Attended Receptive field centers 200 ms from one monkey. (b) Orientation change 200400 ms detection task. Two Gabor stimuli synchronously 0 Cued flashed on for 200 ms and off for a randomized location 200400-ms period. At an unsignaled and randomized time, the orientation of one of the Analyzed Unattended presentation 5 stimuli changed, and the monkey was rewarded for making a saccade to the stimulus that Attention changed. Attention was cued in blocks, and (behavioral shift) Orientation change the cue was valid on 80% of trials, meaning 0 1 10 100 5 0 5 that on an attend-left block of trials (depicted Orientation change Azimuth here), 80% of the orientation changes were to the left stimulus. The monkey was rewarded for correctly detecting any change, even on the unattended side. Unless otherwise stated, all analyses were performed on responses to the stimulus before the orientation change (black outlined panel). (c) Psychometric performance from a typical example experiment. The percent correct as a function of orientation change in degrees for trials in which the change occurred at the attended (black points) or unattended (gray point) location is shown. Unattended changes occurred only at the middle difficulty level (11). Attentional improvement in behavior was quantified as the lateral shift between the percent correct on unattended trials and the fitted psychometric curve for attended trials.

We did not find any important differences between single and multiunits or between the two monkeys, and our population analyses required large neural populations, so we combined single and multiunits (see Supplementary Results). However, the statistics that we used for single units are based on a subset of 187 single units that we were confident are unique (if there was a single unit on a given electrode on multiple days, it was only counted once). The monkeys performed an orientation changedetection task in which spatial attention was manipulated (Fig. 1b). Two Gabor stimuli flashed on and off and the monkeys task was to detect a change in the orientation of either stimulus. We manipulated attention in blocks by cueing the monkey as to which stimulus was more likely to change (see Online Methods). Each day, the location, size, orientation and spatial frequency of the Gabors were optimized for a selected single unit in each hemisphere. The two stimuli were therefore different, so directing attention to one of the two stimuli probably modulated both feature-based and spatial attention. Because we recorded from neurons with a wide range of receptive field locations and tuning, most neurons were not well driven by the stimulus (mean driven rate was 8.2 spikes per s for single units and 21.5 spikes per s for multiunits, compared with mean spontaneous rate 5.8 spikes per s for single units and 14.1 spikes per s for multiunits). Attention greatly improved behavioral performance in this task. To motivate the monkeys to attend to the cued location, we changed the stimulus at the attended location on 80% of trials (trials in which the attentional cue was valid). On the remaining 20% of trials (invalid trials), we tested performance at the unattended location using only a single orientation change (11), which allowed us obtain reliable estimates of behavior and neural responses despite there being relatively few invalid trials. In an example of a typical recording session, the proportion of trials in which the monkey successfully detected an 11 orientation change was substantially greater on trials when the attended, rather than the unattended, stimulus changed orientation (Fig. 1c). To compute the effects of attention on neural responses during the period in which the monkeys attentional state was most likely to affect its behavioral performance, we focused our analyses on the stimulus presentation directly preceding the orientation change (Fig. 1b). On a given day, the stimuli immediately before the orientation change were identical, regardless of the attentional condition, validity of the attentional cue or size of the orientation change. Invalid trials were randomly interleaved with valid trials, so the neuronal effects of attention were indistinguishable on valid and invalid trials. We observed
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

some adaptation of V4 responses between the first and the second stimulus presentation on each trial, but the average responses to the second through tenth stimuli were statistically indistinguishable (t test, P > 0.5). Because the orientation change occurred no sooner than the third stimulus presentation, the responses to the stimulus directly before the change were unaffected by the length of the trial. Consistent with previous studies3,1518, we found that attention increased V4 firing rates (Fig. 2a). To quantify the increase, we calcu lated a standard modulation index (MIrates), which was the difference between the average firing rates on trials in which the attended stimulus was inside of or outside of the neurons receptive field trials divided by the sum (see Supplementary Results). The mean MIrates was 0.049 for single units and 0.042 for multiunits, both of which were significantly greater than zero (t test, P < 106 for single units, P < 1020 for multiunits). We also found that attention reduces the trial-to-trial variability of individual neurons over a similar time course to its effects on firing rate. As is common to stimulus responses in many cortical areas19, we observed a drop in the Fano factor (the ratio of the variance of the firing rates to the mean) following stimulus onset (Fig. 2b). Following the drop associated with the response transient, the Fano factor remained at a significantly lower level in the attended than in the unattended condition (mean MIFF during the sustained response was 0.011 for single units and 0.017 for multiunits, P < 0.05 and P < 103, respectively). Because the Fano factor (Fig. 2b) was calculated using subdistributions of neurons such that the mean firing rates were the same for each time point and attentional condition19,20, the time course and attentional dependence of the Fano factor were independent of changes in firing rate (see Online Methods). Attention improved the signal-to-noise ratio of individual V4 neurons (Fig. 2a,b), but we found that the effect of attention on the correlated variability in pairs of neurons was even more important. For each pair of simultaneously recorded neurons and each attentional condition, we calculated the correlation coefficient between spike count responses to the stimulus preceding the orientation change. This metric, termed noise correlation, measures the correlation in trialto-trial fluctuations in responses and therefore has a very different timescale than the millisecond timescale synchrony that has been shown to increase with attention21. We did not focus on synchrony here because no more pairs exhibited significant synchrony than would be expected by chance (out of 66,578 pairs, 3,609 had significant synchrony in the attended condition and 3,634 had significant synchrony in the unattended condition, 5.4% and 5.5%, respectively;
1595

2009 Nature America, Inc. All rights reserved.

Percent correct

m Ti

Elevation

a r t ic l e s
Figure 2 Attentional modulation of firing rate, Fano factor and noise correlation. (a) Attention increased firing rates. A peristimulus time histogram of firing rates for all 3,498 single neurons and multiunit clusters on trials in which the stimulus in the same hemifield as the neurons receptive field was attended (black line) or unattended (gray line) is shown. Line width represents the s.e.m. Ripples reflect the 85-Hz frame rate of the video display. (b) Attention decreased mean-matched Fano factor. Plotting conventions are as described in a. (c) Attention decreased noise correlation. Spike count noise correlation (for responses over the period from 60 to 260 ms following stimulus onset) is plotted as a function of the mean stimulus modulation for the pair of neurons (firing rate during the stimulusfiring rate during the interstimulus blank period). For pairs of neurons in the same hemisphere, correlation was lower when the stimulus in the neurons receptive field was attended (black line) than when it was unattended (gray line). Pairs of neurons in opposite hemispheres (dashed lines) had correlations that were close to zero. Error bars represent s.e.m. (d) Raw noise correlation, but not attentional modulation, depended on signal correlation. Mean noise correlation is plotted as a function of signal correlation, which can be thought of as the similarity in spatial and feature tuning of the two neurons (see Online Methods). As has been previously reported, noise correlation increases with signal correlation. However, the difference in correlation between the attended (black line) and unattended (gray line) conditions did not depend on signal correlation. Error bars represent s.e.m.

Firing rate (spikes per s)

30

Attended Unattended

Mean-matched Fano factor

1.2

1.1

1.0 0 200 400 Time from stimulus onset (ms)

0 200 400 Time from stimulus onset (ms) Attended Unattended Opposite hemispheres

c
Noise correlation 0.2

d
Noise correlation

0.1

0.1

0 0 20 40 Evoked response (spikes per s) 0 1 0 Signal correlation 1

2009 Nature America, Inc. All rights reserved.

P < 0.05, bootstrap test described in Online Methods) and synchrony in the attended and unattended conditions was not different (paired t test, P = 0.46). Many spikes are needed to detect statistically significant synchrony and even more are needed to detect modulation of synchrony by processes such as attention. Synchrony has therefore been observed in some studies of visual cortex (for example, see refs. 21,22), but not in others10,23,24 (see ref. 21 for a discussion of the statistical power needed to detect synchrony). The absence of synchrony in our study is probably a result of a combination of the low firing rates of many of our cells caused by stimuli that were suboptimal for most cells, the fact that we calculated synchrony using pairs of spiking neurons rather than correlating spike times with local field potentials21 and the fact that most neuron pairs were separated by millimeters in the cortex. The correlations that we observed were fluctuations on a longer timescale than millisecond-level synchrony. One possibility is that the same mechanisms that cause low-frequency oscillations in electroencephalography and local field potentials (which have been shown to desynchronize with attention2528) caused the correlations we measured. To obtain accurate estimates of noise correlation, we did not calculate a time course of correlation as we did for rate and Fano factor because, over short periods, the distributions of spike counts became non-Gaussian (because spike counts can never be negative) and discrete. Skewed, discrete distributions pose a problem for second-order statistics such as correlation, causing noise correlations to approach zero as the mean number of spikes decreases22,29,30. We therefore calculated noise correlation over the entire 200-ms interval (Fig. 2c). Because the stimuli produced a wide range of responses across the population of neurons, we binned the neuron pairs by their mean evoked response across both attentional conditions (driven rate baseline; Fig. 2c). Noise correlations were highest for pairs of neurons in the same hemisphere that both responded strongly to the stimulus. This result can be explained by the fact that correlations tend to increase with firing rate30 and the observation that noise correlations are highest for neurons with similar tuning10,22,24,29. Our dataset included neurons with a broad range of preferences for orientation and other stimulus properties and different receptive field locations, so two neurons that were both strongly modulated by the stimulus likely had similar tuning.
1596

To test the effect of tuning similarity on noise correlation more directly, we calculated noise correlation as a function of signal correlation (Fig. 2d). In a separate set of trials, we presented Gabor stimuli at a variety of locations and orientations while the monkey performed a change-detection task far outside the neurons receptive fields (see Online Methods). We calculated signal correlation by computing a correlation between the mean responses of each neuron to each stimulus. Consistent with previous results 10,2224, we found that noise correlation is highest for neurons with similar tuning (large, positive signal correlation) and lowest for neurons with opposite tuning (negative signal correlation). Unlike a recent study of noise correlations in V1 using the same electrode arrays that we used here29, we found that noise correlation did not depend on cortical distance. We suspect that the greater retinotopic and tuning organization of V1 compared with V4 accounts for the differences in our results. We found that even for the least responsive neurons (Fig. 2c) and pairs of neurons with dissimilar stimulus preferences (Fig. 2d), correlations within a hemisphere were on average positive, indicating that there is shared variability throughout the population. In contrast, we found that noise correlations for pairs of neurons in opposite hemispheres were close to zero (Fig. 2c), meaning that trial-to-trial fluctuations in the two hemispheres are independent within an attentional condition. The biggest physiological effect of attention in our dataset was a large decrease in the correlations between pairs of neurons in the same hemisphere (Fig. 2c,d). On average, attention reduced noise correlations by about half (mean MIcor = 0.35 for single units and 0.29 for multiunits, P < 105 for single units and P < 109 for multi units). Attentional modulation of correlation depended strongly on how much the neurons were driven by the stimulus; for the most responsive pairs of neurons, noise correlation in the attended condition was roughly one-third the correlation in the unattended condition (Fig. 2c). In contrast, the effect of attention on correlations did not depend on the degree of tuning similarity between the two cells (Fig. 2d). This observed decrease in correlation as a result of attention is the opposite result predicted by the mathematical relationship between firing rate and correlation30. Attention tends to increase firing rates (Fig. 2a), which makes the distributions of spike counts more Gaussian and less discretized, leading to a predicted increase
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

a r t ic l e s
(Figs. 2b and 3b) could arise from a decrease in the independent variability of individual neu0.1 rons, a decrease in shared variability across the 0.4 10 population or a combination of both. Noise correlation measures the degree of shared 0.2 0.8 0 variability. We therefore focused on the other 0 10 20 30 0 10 20 30 0 10 20 30 Evoked response (spikes per s) Evoked response (spikes per s) Evoked response (spikes per s) aspect of variability captured by the Fano factor and asked how much of a decrease in Figure 3 Attention has the biggest effects on the most responsive neurons. ( a) Difference in mean firing rate between trials when the stimulus in the neurons receptive field was attended and independent variability that was large enough unattended as a function of stimulus modulation (rate during stimulus periodinterstimulus to account for the full decrease in Fano factor period) during the full stimulus period (solid line) or the sustained response following the onset would improve population sensitivity. transient (dashed line). Error bars represent s.e.m. ( b) Data are presented as in a for Fano factor. We compared the effects of attentional (c) Data are presented as in a for noise correlation for pairs of neurons in the same hemisphere. modulation of firing rates, independent variability and noise correlation and found that in correlation. Therefore, decreases in correlation cannot be a simple the modulation of correlation had by far the greatest influence on the attentional improvement in population sensitivity ( Fig. 4). We mathematical consequence of increases in firing rate. Previous studies have shown that attention modulates firing rate first quantified the degree to which attention improved the sensitivity more for neurons with the biggest response to the stimulus (whose of the groups of neurons that we recorded (Fig. 4a,c) and then responses may be more informative for the task) 1518. Furthermore, determined the amount of that improvement that was caused by a recent study found that fast-spiking neurons with high firing rates attentional modulation that affected only rate, only independent (putative interneurons, separated from putative excitatory neurons variability or only correlation (Fig. 4d,e). We quantified how much attention improved neuronal sigon the basis of waveform width) had larger differences in Fano factor than regular-spiking neurons7. Our hardware filters prevented nals in our recorded populations (schematized for a hypothetical us from distinguishing these neuron types on the basis of waveform, two-neuron dataset in Fig. 4a and shown for a real 38-neuron dataset in but consistent with this study and studies of attentional modulation Fig. 4c). The monkeys task was to detect a change in the orientation of of firing rate, we found that attention had a bigger effect on the the stimulus, so we defined population sensitivity as the discriminability rates, Fano factors and noise correlations of neurons that responded between the distributions of responses to the original orientastrongly to the stimulus (Fig. 3). Neurons (or pairs of neurons) tion and the changed orientation. For each of the single and multi that were most strongly driven by the stimulus (biggest difference units that we recorded from a given hemisphere on a given day between evoked and baseline firing rate) probably have receptive (mean of 39.5 neurons, range of 1474), we calculated responses field locations and tuning properties that make them well suited for to the stimulus preceding the change from 60 ms to 260 ms this task, and these neurons showed the largest effects of attention following stimulus onset and the changed stimulus starting 60 ms by all three measures. after onset and continuing for either 200 ms or until 60 ms before the Recording from both hemispheres simultaneously allows us to monkeys response, whichever came first. The mean time from the be sure that the correlation changes that we observed were spatially onset of the changed stimulus to the onset of the monkeys response specific effects of attention. The same block of trials that yielded low was 251 ms, and 260 ms fell at least 60 ms before the saccade on correlations in one hemisphere gave high correlations in the other, so 39% of trials. We experimented with other intervals for computing nonspecific factors such as arousal or motivation cannot account for spike counts (including identical periods for the original and changed the changes in correlation that we observed. The fact that trial-to-trial stimuli (200 ms each) and also cutting off the response to the changed variability in the two hemispheres was virtually independent is further stimulus 100 ms or 0 ms before the saccade) and these did not qualievidence that the correlation changes we observed in a hemisphere tatively affect the proportion of the improvement in population senare spatially specific. sitivity accounted for by each of the three factors that we considered. Consistent with many previous studies (for examples, see refs. 21,3134), Using this time period, attentional modulation during the changed we found that attention primarily affected the sustained part of the stimulus was indistinguishable from modulation during the previous response rather than the onset transient (Fig. 2a,b). In our data, stimulus (Supplementary Fig. 1). attentional modulation of firing rate became statistically significant We plotted one point for each stimulus in each trial in an 122 ms after stimulus onset (the first time point at which the 95% n-dimensional space in which each dimension corresponds to the confidence intervals for the means of the two attentional conditions response of one of the n neurons that we recorded in a given hemidid not overlap). In addition to examining the effect of attention on sphere (Fig. 4a). We then calculated the mean response for each rates, Fano factor and correlations during the entire stimulus period, stimulus and projected all responses onto an axis of discrimination we calculated the attentional effects for all three measures during the drawn through the two means. This was done separately for the two sustained response of the response (122260 ms after stimulus onset). attention conditions, producing pairs of one-dimensional distribuAs expected, attentional effects were larger during the sustained period tions of projections for each attention condition (Fig. 4c). by what appeared to be a fairly constant factor (Fig. 3). We measured population sensitivity by calculating d (the difference Attention changed the responses of both single neurons and cor- in the means divided by their root mean square s.d.), which is mono related variability in ways that could allow each to contribute to tonically related to theoretical performance on classifying stimuli, so improvements in population sensitivity (Figs. 2 and 3). A primary attention should increase d to improve behavioral performance. We goal of our study was to determine the relative importance of changes quantified the attentional improvement in population sensitivity as the in firing rates, Fano factor and noise correlations. Because Fano fac- difference in d between the attended and unattended conditions. We tor measures the variability of single neurons without regard to the then normalized the d values to reflect the measured improvement. source of that variability, the decrease in Fano factor that we observed The amount of attentional improvement in our d measure correlated
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009
Noise correlation Fano factor Firing rate

a 20

Full response Sustained response

2009 Nature America, Inc. All rights reserved.

1597

a r t ic l e s
Figure 4 Modulation of noise correlation accounts for the majority of the attentional improvement in population sensitivity. ( a). Procedure for calculating the sensitivity of the population. For each trial and Recorded datasets attentional condition, the firing rate response of the n neurons recorded All three Correlation only simultaneously in a given hemisphere to the stimulus immediately before Rate only Changed Independent variability only the orientation change (open circles) and the changed stimulus (filled stimulus 1.0 circles) is plotted as a point in an n-dimensional space (a fictional Original two-neuron example is plotted here). Each point is projected onto the 0.8 stimulus axis connecting the center of mass of the cloud of points for each 0.6 time period (Xs), leaving a one-dimensional distribution of projected Neuron 1 firing rates values for each time period (dashed and solid curves). The sensitivity 0.4 1.5 of the population to the change in the stimulus is quantified as the 0.2 discriminability of the two distributions in units of d. (b). Population 1.0 d and behavioral improvement were highly correlated. For each 0 hemisphere day, population d is plotted as a function of the behavioral improvement (quantified as the lateral shift between performance at the Simulated data 0.5 unattended location and the fitted psychometric curve for the attended All three Correlation only condition). (c). Procedure for calculating the amount of the observed 2.5 Rate only 0 attentional improvement explained by each factor for a representative Independent variability only example dataset. Histograms of projections onto the axis defined in a are Unattended 2.0 plotted for the real data (left column, for attended and unattended trials) 0.5 0 10 20 and for simulations (right column). We defined the observed attentional 1.5 Behavioral shift (degrees) improvement as the difference between d for the two attentional 1.0 conditions (d = 2.40 for the attended condition and 1.15 for the Real data Simulated data 1.0 Attended unattended condition, giving an improvement of 1.25 in this example). All three 0.5 2.5 The left axis represents d and the right axis represents normalized 1.0 0.5 Correlation 0 proportion of attentional improvement (by definition 1.0 for the attended 0.8 1 10 1001000 2.0 Number of neurons condition and 0.0 for the unattended condition). We calculated the 0.6 0 fraction of the observed attentional improvement explained by each 0.4 1 10 100 1000 Rate 1.5 Population size (neurons) factor(s) by comparing the simulated d (right column of distributions, 0.2 Independent see Online Methods) to the d for the real unattended data. (d). Average 0 Unattended variability 1.0 proportion of actual attentional improvement for all 82 datasets (one dataset for each hemisphere day). Error bars represent s.e.m. (e). Population Original stimulus Changed stimulus sensitivity as a function of the number of neurons involved in the task. Population d was calculated using the method described in a and b, except that data in both the attended and the unattended conditions were simulated. For each population size, we sampled, with replacement, from the entire population of neurons from all datasets combined. Each simulation was run 100 times for 10,000 trials on each run. The inset plots the relative contribution of each factor (which is the ratio of the improvement in d for that factor alone to the improvement in d for all three factors) as a function of population size.

Neuron 2 firing rates

Change in d

2009 Nature America, Inc. All rights reserved.

strongly with the monkeys behavioral improvement resulting from attention. For each hemisphere day, we quantified behavioral improvement as the lateral shift between measured performance in the unattended condition and the fitted psychometric curve in the attended condition (Fig. 1c and Online Methods). Attention shifted the psychometric curve by 7.7, which was typical for our datasets (7.6 0.5, mean s.e.m.; Fig. 1c). Attentional improvement in neuronal d (attended-unattended) for each hemisphere day was highly correlated with behavioral improvement (R = 0.69, P < 1012; Fig. 4c). This strong correlation suggests that our d metric captures the important aspects of the improvements in population sensitivity that lead to improvements in behavior. Each of the physiological changes that we observed in rate, Fano factor and correlation could have contributed to the improvement in population sensitivity. We next compared how much attentional modulation of each factor alone and the three factors together contributed to the actual improvements in d that we calculated. To isolate the contribution of each factor, we simulated the responses of populations of neurons using the same mean rates, noise correlations, Fano factors and number of neurons as the groups of neurons that we recorded (see Online Methods) and compared the calculated d for each simulation to the real data in the unattended condition. In this example dataset, attentional modulation of all three factors together (Fig. 4c) accounted for 95% of the attentional improvement that we observed in the real data. We then calculated the contribution of each factor separately by simulating attentional modulation of the factor
1598

of interest and using the values observed in the unattended condition for the other two factors (Fig. 4c). Correlation alone accounted for 79% of the attentional improvement, rate accounted for 9% and modulation of independent variability accounted for 4%. This example is typical of the 82 datasets (Fig. 4c). On average, attentional modulation of the three factors together accounted for 92% of the attentional improvement that we observed in the actual populations (Fig. 4d). Notably, this result means that population sensitivity is well modeled by accounting only for rate, independent variability and pair-wise noise correlation, and that any other factors (including any higher-order correlations) account for no more than 8% of the observed improvement in population sensitivity. Consistent with this, population responses in the retina are well described by the responses of individual neurons and pair-wise correlations35,36. Overall, modulation of noise correlation was by far the most important factor in explaining the improvement in population sensitivity. Attentional modulation of noise correlation accounted for 81% of the observed improvement, rate accounted for 10% and independent variability accounted for only 0.3% (which was not significantly different than 0.0, t test, P = 0.82). Unsurprisingly, we found that both the observed raw population d and the improvement in d resulting from attention depended on the number of neurons that we recorded. Because there is no a priori way of knowing how many neurons are involved in the task, we examined the dependence of these measures on population size by sampling, with replacement, the firing rates, Fano factors and correlations of all
VOLUME 12 | NUMBER 12 | december 2009 nature NEUROSCIENCE

Population sensitivity (d )

Relative contribution

Population sensitivity (d )

Normalized attentional improvement

Normalized attentional improvement

a r t ic l e s
of the neurons that we recorded over all recording sessions (see Online Methods and refs. 11,24). We calculated population d for simulations in which attention modulated either all three factors, one factor individually or none of the factors (Fig. 4e). In all cases, d increased with population size. Because noise correlations are on average positive for both attentional conditions, d asymptotes for large populations911. Modulation of noise correlation accounted for most of the attentional improvement in sensitivity for population sizes greater than five neurons (Fig. 4e). For very small populations, however, this was necessarily not true (Fig. 4e). If performance depends on a single neuron, there can be no correlation, and the small attentional improvement depends almost entirely on modulation of firing rate. In our simulations, noise correlations became dominant for populations of more than five neurons. If anything, this estimate may be high because we recorded from many neurons with stimulus preferences that were not well matched to the stimuli that we presented, resulting in low firingrate responses and low noise correlations (Fig. 2c). If we had recorded from neurons that were better matched to the stimuli, correlations would probably have been higher (Fig. 2c), shifting the point at which correlation becomes most important to population sizes even lower than five neurons. Many more than five neurons are thought to be involved in virtually every task, so changes in correlation probably dominate attentional improvement in nearly all situations. DISCUSSION Why do changes in shared variability have a bigger effect on population sensitivity than changes in the signal-to-noise ratio of single neurons? One answer is that the changes in correlated variability that we observed were larger than the changes in firing rate or Fano factor. However, we re-ran the simulations (Fig. 4e), assuming that the three factors all had the same modulation index as the changes in rate (see Supplementary Results), and correlation still dominated for population sizes greater than 30 neurons. Instead, the explanation lies in the fact that no matter how noisy individual neurons are, independent variability can be averaged out if the population of neurons is large enough. Correlated variability, however, can never be averaged out by simply adding neurons to the population. Noise correlation can either improve or reduce population sensitivity, depending on the algorithm by which neural responses are read out8,9, and our simulations could, in principle, have revealed that the observed correlation decreases acted to reduce population sensitivity. However, theoretical studies have shown that decreased correlation improves discrimination if the difference between the responses to the stimuli to be discriminated (the original and changed stimuli in our task) are of the same sign for most neurons8,9, which turned out to be the case in our dataset. Most of the neurons that we recorded (92%) responded more strongly to the changed than the previous stimulus, presumably reflecting adaptation to the series of identical stimuli preceding the change. Therefore, the optimal quantity to be read out is similar to a (positively) weighted mean of the responses of the population and the axis of discrimination that we determined (Fig. 4a) was close to the weighted population mean. The attention-related decrease in correlation therefore improved the sensitivity of the population by reducing the amount of shared variability that could not be removed by averaging. In contrast, a recent study found no effect of attention on noise correlations in a situation in which correlations were shown to have no effect on the sensitivity of the population37. There are further situations (such as those in which the optimal readout algorithm is more similar to a subtraction of two populations of cells) in which an increase in correlation would improve the sensitivity of the
nature NEUROSCIENCE VOLUME 12 | NUMBER 12 | december 2009

population8,9. Whether attention would increase correlations in such tasks remains to be determined. It is likely, of course, that the brain uses a different algorithm for extracting stimulus information from the responses of many neurons than the very simple decoding scheme that we used (Fig. 4). However, the observation that attentional modulation of noise correlation explains most of the attentional improvement in population sensitivity is probably true for any sensible decoding algorithm. First, the difference in the amount of attentional improvement explained by pair-wise correlations was very large compared with the amount explained by the changes in the responses of single neurons, suggesting that noise correlations will dominate using any decoding algorithm. Furthermore, correlation was by far the most important factor using any of several linear discriminators that we tried, including the single axis projection described here, Fisher discriminants and support vector machines (data not shown). Higher-order decoders that explicitly read out interneuronal correlations3841 will be even more affected by attentional modulation of correlation than linear discriminators. Finally, any sort of decoding algorithm that incorporates a mean (or weighted mean) of the responses of many neurons will be greatly affected by noise correlations1012. Mathematically, correlation is invariant to the mean response (the correlation coefficient is the ratio of the covariance to the square root of the product of the individual variances, so both the numerator and denominator are proportional to the product of the means), so underlying noise correlations cannot be changed by a simple scaling of neural responses (that is, a gain change). Instead, noise correlations in cortex are thought to arise primarily from common, noisy inputs10,22,23,29. The fact that attention primarily decreased correlations provides clues about the mechanisms by which attention affects populations of sensory neurons. A decrease in correlation combined with an increase in firing rates is consistent with a decrease in the strength of an effectively inhibitory input that is common across the population. One possibility is that attention results in a decrease in the weights or activity of inputs that cause divisive normalization, a mechanism that normalizes responses to many stimuli in a receptive field and has recently been proposed to underlie attention34,42,43. In fact, we found a correlation between the mean attentional modulation of the firing rates of a pair of neurons and modulation of their noise correlation (R = 0.32, P < 104) and also between the average rate and correlation changes in a hemisphere day (R = 0.61, P < 109; see Supplementary Fig. 2), which is consistent with the idea that the two attentional changes may be mediated by the same mechanism. Attention improves perception of the attended location or feature, so studying the effects of attention on populations of sensory neurons reveals the aspects of the population code that are most important for accurately encoding information about a behaviorally relevant stimulus. Here we found that attention improved population sensitivity primarily by changing noise correlations and even the small pair-wise correlations that we observed had a marked effect on the sensitivity of the population. Therefore, understanding the interactions between pairs of neurons is critical for understanding population coding (see also refs. 811,3841,44). Rather than examining mean responses over many trials, the brain makes decisions on the basis of the responses of many neurons over a short period. Our results indicate that studies of average responses of single neurons miss interactions between neurons that have critical effects on behavior. Together, these results suggest that the future of studying population coding will rely on multi-electrode or
1599

2009 Nature America, Inc. All rights reserved.

a r t ic l e s
maging technologies that allow glimpses of population coding on the i timescale of a single behavioral decision. Note added in proof: A recent study in area V4 confirmed that attention reduces noise correlations45. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank D. Ruff for spike sorting and monkey training assistance, M. Churchland for code and advice regarding mean-matching and Fano factor analysis, and M. Histed for many helpful discussions. They, A. Ni and A. Smolyanskaya provided comments on an earlier version of the manuscript. This work was supported by US National Institutes of Health grant R01EY005911 and the Howard Hughes Medical Institute. AUTHOR CONTRIBUTIONS M.R.C. collected the data and performed the analyses. M.R.C. and J.H.R.M. designed the study and wrote the paper.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
16. McAdams, C.J. & Maunsell, J.H. Effects of attention on orientation-tuning functions of single neurons in macaque cortical area V4. J. Neurosci. 19, 431441 (1999). 17. Williford, T. & Maunsell, J.H. Effects of spatial attention on contrast response functions in macaque area V4. J. Neurophysiol. 96, 4054 (2006). 18. Treue, S. & Martinez Trujillo, J.C. Feature-based attention influences motion processing gain in macaque visual cortex. Nature 399, 575579 (1999). 19. Churchland, M.M. et al. Stimulus onset quenches neural variability: a widespread cortical phenomenon. Front. Syst. Neurosci.: Comput. Syst. Neurosci. Abstr. doi:10.3389/conf.neuro.06.2009.03.295 (2009). 20. Churchland, M.M., Yu, B.M., Sahani, M. & Shenoy, K.V. Techniques for extracting single-trial activity patter