nature neuroscience june 2005

www.nature.com/natureneuroscience

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

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

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

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

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

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

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

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

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

i

VOLUME 8 NUMBER 6 JUNE 2005

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

E D I TO R I A L

693 Genetics of pyschiatric disorders

B O O K R E V I E W

695 The Lobotomist: A Maverick Medical Genius and His Tragic Quest to Rid the World of Mental Illnessby Jack El-HaiReviewed by David Healy

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

697 A woman’s prerogativeKevin Staley & Helen Scharfman � see also p 797

699 Subtracting the Math: prominin-positive cerebellar stem cells in white matterAnna Marie Kenney & Rosalind A Segal � see also p 723

701 Blue genes: wiring the brain for depressionStephan Hamann � see also p 828

703 Shaking up sleep researchJoan C Hendricks

705 Where did the time go?Sabrina Ravel & Barry J Richmond

707 Brain’s guard cells show their agilityKalyani Narasimhan � see also p 752

R E V I E W

709 Notch signaling in the mammalian central nervous system: insights from mouse mutantsKeejung Yoon & Nicholas Gaiano

Neural stem cells in postnatal cerebellum

(pp 699 and 723)

Activity-dependent signaling from the synapse to the nucleus is important for synaptic plasticity and neuronal

survival. David Ginty and colleagues generated conditional knockout mice

for serum response factor (SRF), a candidate transcriptional regulator

of activity-induced gene expression, and compared their phenotype with

that of mice lacking CREB family transcription factors. CREB mutants

showed neuronal degeneration, whereas the SRF mutants had synaptic

plasticity deficits. Thus the authors suggest that these factors regulate

distinct gene-expression programs that make differing contributions to survival

and plasticity in mature neurons. (p 759)

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

iii


NATURE NEUROSCIENCE

TRP channels mediate growth cone guidance in vivo

(p 730)

Higher visual areas and 3D perception(p 820)

Microglial dynamics and rapid chemotaxis in vivo

(pp 707 and 752)

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

717 Laminin stimulates and guides axonal outgrowth via growth cone myosin II activityS G Turney & P C Bridgman

720 An fMRI investigation of race-related amygdala activity in African-American and Caucasian-American individualsM D Lieberman, A Hariri, J M Jarcho, N I Eisenberger & S Y Bookheimer

A R T I C L E S

723 Isolation of neural stem cells from the postnatal cerebellumA Lee, J D Kessler, T-A Read, C Kaiser, D Corbeil, W B Huttner, J E Johnson & R J Wechsler-Reya � see also p 699

730 XTRPC1-dependent chemotropic guidance of neuronal growth conesS Shim, E L Goh, S Ge, K Sailor, J P Yuan, H L Roderick, M D Bootman, P F Worley, H Song & Gl Ming

736 RNA editing produces glycine receptor α3P185L, resulting in high agonist potencyJ C Meier, C Henneberger, I Melnick, C Racca, R J Harvey, U Heinemann, V Schmieden & R Grantyn

745 LINGO-1 negatively regulates myelination by oligodendrocytesS Mi, R H Miller, X Lee, M L Scott, S Shulag-Morskaya, Z Shao, J Chang, G Thill, M Levesque, M Zhang, C Hession, D Sah, B Trapp, Z He, V Jung, J M McCoy & R B Pepinsky

752 ATP mediates rapid microglial response to local brain injury in vivoD Davalos, J Grutzendler, G Yang, J V Kim, Y Zuo, S Jung, D R Littman, M L Dustin & W-B Gan � see also p 707

759 SRF mediates activity-induced gene expression and synaptic plasticity but not neuronal viabilityN Ramanan, Y Shen, S Sarsfield, T Lemberger, G Schütz, D J Linden & D D Ginty

768 Subunit interaction with PICK and GRIP controls Ca2+ permeability of AMPARs at cerebellar synapsesS J Liu & S G Cull-Candy

776 Endocannabinoid signaling depends on the spatial pattern of synapse activationP Marcaggi & D Attwell

782 Geometric and functional organization of cortical circuitsG M G Shepherd, A Stepanyants, I Bureau, D Chklovskii & K Svoboda

791 Reversible blockade of experience-dependent plasticity by calcineurin in mouse visual cortexY Yang, Q S Fischer, Y Zhang, K Baumgärtel, I M Mansuy & N W Daw

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

v


NATURE NEUROSCIENCE

Functional correlates of a high-risk gene for depression

(pp 701 and 828)

RNA editing of a glycine receptor(p 736)

797 Ovarian cycle–linked changes in GABAA receptors mediating tonic inhibition alter seizure susceptibility and anxietyJ L Maguire, B M Stell, M Rafizadeh & I Mody � see also p 697

805 Dopaminergic modulation of limbic and cortical drive of nucleus accumbens in goal-directed behaviorY Goto & A A Grace

813 Instructive signals for motor learning from visual cortical area MTM R Carey, J F Medina & S G Lisberger

820 3D shape perception from combined depth cues in human visual cortexA E Welchman, A Deubelius, V Conrad, H H Bülthoff & Z Kourtzi

828 5-HTTLPR polymorphism impacts human cingulate-amygdala interactions: a genetic susceptibility mechanism for depressionL Pezawas, A Meyer-Lindenberg, E M Drabant, B A Verchinski, K E Munoz, B S Kolachana, M F Egan, V S Mattay, A R Hariri & D R Weinberger� see also p 701

835 ERRATA

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

See back pages.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 6 | JUNE 2005 693

E D I TO R I A L

It’s hard to please two masters, as most scientists who do interdis-ciplinary work eventually discover. This problem is particularly acute for researchers who study the neural mechanisms by which

genetic variation influences the risk of psychiatric disorders. A paper in this issue1 illustrates the promise of this approach—along with the biological complexity and scientific uncertainties that make the problem so hard.

Although vulnerability to mental illness tends to run in families, there are no genes for psychiatric disorders in the sense that there are genes for eye color. No known gene is either necessary or sufficient to produce disease. Instead there are many susceptibility genes with small effects, each increasing the risk of illness by 5% or less in an individual. It remains unclear how these genes interact with each other or with the environment, which could increase the system’s complexity enormously2.

There are two basic approaches to finding susceptibility genes for complex disorders. Linkage studies are a hypothesis-neutral search for markers that statistically segregate with a disease, followed by fine mapping to identify the actual gene or genes. Association studies, on the other hand, ask whether particular variants of a candidate gene are more prevalent in patients than would be expected by chance. The definition of chance levels is controversial, and depends—implicitly or explicitly—on the prior probability that the gene is involved in the disease3. This approach may identify effects that are too small to be detected in linkage studies, but it also has considerable potential for false-positive results, so geneticists tend to mistrust these studies.

Many proposed susceptibility genes for psychiatric disorders do not replicate consistently, showing statistically significant effects in some studies but not others. To some extent, this may simply reflect the difficulty of measuring small effects, which can require thousands of subjects to reach significance. However, geneticists attribute some of the variability to population stratification, a confound that may be unfamiliar to many neuroscientists. Differences in the genetic back-ground of experimental and control groups can lead to false-negative or false-positive results. Avoiding such artifacts requires matching groups not only for race, but also for more subtle aspects of ethnicity, such as the proportion of genetic background from northern versus southern Europe. Though the magnitude of the problem is unclear, this is poten-tially an important source of error for studies with small sample sizes, like most neuroscience studies. Researchers can increase sample size by sharing genotype and phenotype data across groups, but this approach is relatively uncommon.

Ultimately, of course, we want to know not only whether a gene is linked to disease vulnerability, but how it exerts its influence. One promising approach is to study intermediate phenotypes, such as a well-defined cognitive symptom or diagnostic subgroup. Genetic polymor-phisms may be more strongly linked to a particular cognitive function than to the heterogeneous set of symptoms that make up a diagnostic

category, as suggested by evidence that some candidate genes are linked to more than one disorder. Exploring specific symptoms may be easier in animal models as well. It is difficult to imagine plausible mouse mod-els of a complete psychiatric disorder, but mutants with hypomorphic alleles that model particular symptoms may be a very useful tool. The hope is that multiple genes involved in a particular disease will be found to converge onto the same biochemical pathway, preferably one that can be manipulated to influence the disease symptoms.

One intriguing possibility is that certain life events may interact with brain vulnerabilities created by particular genes to produce behavioral symptoms. For example, in one prospective study, people homozygous for the short allele of the serotonin transporter gene were more likely to experience depression after stressful life experiences than were het-erozygotes or long allele homozygotes4. This finding—and others like it—have been controversial because the genetic association of this poly-morphism with depression and anxiety is statistically weak and could be artifactual. However, other scientists feel that convergent evidence from biology increases the likelihood that this polymorphism may mediate vulnerability to mood disorders.

In this issue1, Pezawas et al. report neural correlates of these serotonin transporter alleles in people without psychiatric diagnosis. Examining normal subjects allows larger sample sizes, and it avoids potential complications from diagnostic heterogeneity and medication history in patients. Relative to long allele homozygotes, the homozygous and heterozygous carriers of the short allele had reduced amygdala and perigenual anterior cingulate cortex volumes and less correlated activ-ity between these regions during exposure to fearful faces. The authors propose that the short allele affects the development of these brain regions, which may increase vulnerability to depression and anxiety.

Which genetic associations are worthy of such intensive follow-up? That depends on our goals. Successful identification of susceptibility genes seems unlikely to lead to gene therapy for psychiatric illnesses. If the ability to identify people at risk is critical, then we should restrict our efforts to genes with strong evidence for linkage. On the other hand, if our main goal is to develop effective treatments, then it may be more important to determine the biochemical pathways involved in producing particular symptoms, which could be downstream of a group of genes with individu-ally weak effects—and may also have the advantage of affecting a larger proportion of patients than any individual gene variant. Thus it may be worthwhile to pursue some genes with weak linkage data if functional evidence strengthens the case for them. Genes provide a good starting point for the neuroscience of psychiatric disorders—and neuroscientists must take into account the lessons learned from genetics about potential sampling artifacts—but they cannot be an end in themselves.

1. Pezawas, L. et al. Nat. Neurosci. 8, 828–834 (2005).2. Weaver, I.C. et al. Nat. Neurosci. 7, 847–854 (2004).3. Freimer, N. & Sabatti, C. Nat. Genet. 36, 1045–1051 (2004).4. Caspi, A. et al. Science 301, 386–389 (2003).

Genetics of psychiatric disorders

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



A woman’s prerogativeKevin Staley & Helen Scharfman

Hormonal changes during the estrous cycle have profound effects on synaptic transmission, from altering the density of synapses to changing receptor composition. A new paper shows that neurons express different subsets of GABAA receptor subunits during different phases of the estrous cycle, and that this alters tonic inhibition, seizure susceptibility and anxiety in female mice.

A woman’s prerogative is to change her mind. As politically incorrect as that old husbands’ tale may be, during the estrous cycle, profound changes do occur in the neurobiological basis of the mind. From an approximately 30% alteration in excitatory synapse density in the hippocampus1 to changes in receptors for the inhibitory neurotransmitter GABA2 and their modulators3, the physical substrates that sub-serve brain function are radically altered by changing levels of progesterone or estrogen. In this issue, Maguire and colleagues demonstrate yet another profound and unexpected change: neurons express different subsets of GABAA receptor subunits during different phases of the estrous cycle4. This alters the subtle, steady-state dampening of neuronal excit-ability called tonic inhibition. These intriguing new findings may help explain long-standing observations regarding changes in seizure sus-ceptibility and behavior that have been linked to the estrous cycle.

Why might tonic inhibition be important? In the 1850s, after the first transoceanic tele-graph cables were laid, it was discovered that small changes in the quality of insulation on the cable had large cumulative effects on the fidelity of the transmitted signals, due to the inch-by-inch leakage of the electrical impulses along the length of the cables. Similarly, tonic inhibition, by increasing the electrical leaki-

ness, or conductance, of the dendritic mem-brane, substantially and indiscriminately diminishes the size of excitatory signals as they travel from the dendrites to the neuronal soma5. Because of the indiscriminate nature of tonic inhibition, it is unsurprising that agents that selectively enhance tonic inhibition, such as low concentrations of the GABAA receptor agonist THIP and ethanol, tend to have non-specific, depressant actions such as anxiolytic, anticonvulsant and anesthetic effects.

Tonic inhibition is subserved by persis-tent activation of extrasynaptic GABAA ligand-gated receptors. In contrast to subsynaptic GABAA receptors, which are periodically activated by millimolar con-centrations of GABA released from the axon terminals of GABAergic interneurons, extrasynaptic GABA receptors are activated by the stray GABA molecules that escape reuptake by GABA transporters. GABA receptors are built of five subunits; extra-synaptic GABA receptors tend to contain α4 and δ subunits rather than the standard α and γ subunits. The δ subunit, which is found on the dendrites of hippocampal dentate gyrus granule cells, confers important functional characteristics, including a high sensitivity to GABA (which comes in handy given the low concentrations of GABA to which extra-synaptic receptors are exposed), a decreased tendency to desensitize in the sustained pres-ence of GABA, and positive modulation by both neurosteroids and low concentrations of ethanol6,7.

Now, if neurosteroids modulate tonic inhibition, and levels of these neurosteroids change during the estrous cycle, it seems straightforward to suggest that tonic inhi-bition is altered during the estrous cycle by changes in steroid levels. Yet these ‘levels’ refer to assays of serum steroids, not brain or hippocampal levels. Therefore, it is sig-

nificant that the new paper from Maguire et al. demonstrates that during the estrous cycle in mice, neurons in the dentate gyrus alter the membrane expression of the δ sub-units of GABAA receptors that mediate tonic inhibition. During the estrus phase (when serum progesterone levels are relatively low), tonic inhibition was reduced by 50%, with corresponding increases in both seizure susceptibility and anxiety. In late diestrus (Fig. 1a), the authors see an enhanced expres-sion of δ subunit–containing GABAA recep-tors, and this correlates with an increase in tonic inhibition and diminished seizure sus-ceptibility. Removing the δ subunit (either by gene knockout in mice or by antisense tech-niques) prevents the change in excitability during estrus.

The changes in δ subunit expression were correlated with a change in tonic inhibition that would result in roughly a twofold change in membrane conductance for an average granule cell. To get a sense for the impact of such a change, if we treat the dendrite as a transoceanic cable5, a twofold increase in membrane conductance results in a 10% decline in the total current that reaches the soma from the excitatory signals originat-ing in the most proximal dendrites, and an impressive 50% decline in the total current from signals originating in the most distal aspects of the dendrites8.

Given the magnitude of the changes in dendritic signal transmission in the granule cells, it makes sense that Maguire et al. found a large change in seizure susceptibility during different phases of the estrous cycle, particu-larly in light of the dentate’s function as a gate for the propagation of seizure activity through the hippocampus9. This finding may be rel-evant to the human condition of catamenial epilepsy, the increase in seizure susceptibil-ity that occurs in as many as 78% of women

Kevin Staley is at the Departments of Pediatrics and

Neurology, University of Colorado Health Sciences

Center, Denver, Colorado, USA. Helen Scharfman

is at the Departments of Pharmacology and

Neurology, Columbia University, New York, New

York, USA and the Center for Neural Recovery and

Rehabilitation Research (CNRRR), Helen Hayes

Hospital, New York State Department of Health,

West Haverstraw, New York, USA.

e-mail: [email protected]

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

698 VOLUME 8 | NUMBER 6 | JUNE 2005 NATURE NEUROSCIENCE


during progesterone ‘withdrawal’ at the end of the luteal phase10 (Fig. 1). Maguire et al. found dramatic changes in susceptibility to kainate-induced seizures at estrus relative to another point in the cycle (diestrus) when progesterone was relatively high. Although there are caveats to these findings—such as the variability in kainate-induced seizures and the limited information regarding the fluctuations of reproductive hormones and exact timing of distinct phases of the estrous cycle in this particular strain of mice—mice that lack δ receptors had the same seizure severity as wild-type mice when tested during estrus, suggesting that at that time of the cycle, there was very little anticonvulsant protection provided by tonic inhibition (or at least tonic inhibition mediated by δ subunit–containing GABAA receptors).

These findings raise many interesting ques-tions. One major issue is which neurosteroid, of all that change during the estrous cycle, might be responsible for this effect. More experiments will be needed to answer this question, although the correlation of the data with progesterone levels certainly suggest that progesterone is a key. A central role of proges-terone is also suggested by numerous studies showing that GABAergic inhibition and GABA receptors change in response to a reduction in progesterone2, which correlates with increased seizure susceptibility11,12. As the authors point out, progestins have been used to treat epilepsy

in women, although progestins are not nec-essarily effective in women with catamenial epilepsy. This underscores the idea that modu-lation of excitability during the estrous cycle may involve multiple modulators, not all of which are progestin dependent.

One of the difficulties in addressing the role of the estrous cycle in brain function is decid-ing what animal model to use; indeed, as the authors point out, data derived from animal models of progesterone withdrawal2 do not completely agree with the results from the intact mice studied by Maguire et al. However, the prior findings identify changes in the α4 subunit, which also subserves tonic inhibi-tion, and thus they may complement the cur-rent study. Maguire et al. used a mouse with an elongated cycle, which makes it difficult to compare the results of the present study to studies in other rodents, let alone humans (Fig. 1). One critical question is whether the animals that were sampled by Maguire et al. at times when serum progesterone was low had elevated estradiol and estradiol-induced effects. Indeed, they indicate that their ‘estrus’ involves elevated estradiol. Consideration of this issue is important, given the well-described role of estradiol in modulating GABAergic inhibition and excitability in hippocampus2.

A few other issues merit consideration before a clear case for δ subunits in seizure suscepti-bility can be concluded. One is the possibility that compensatory changes in the transgenic

mice confounded this approach, a common concern of transgenic studies. Another is the lack of direct comparison between seizure severity in diestrus and estrus transgenics, and the possibility that recording from elec-trodes in other areas of the brain might have illuminated distinct patterns of seizure activ-ity, compared with recording from only the hippocampus.

Beyond these caveats, however, the find-ings raise some intriguing questions. What is the mechanism of the δ subunit changes—do changes in reproductive steroids induce changes in δ subunit membrane expression directly, or does the increased excitability fol-lowing progesterone withdrawal initiate activ-ity-dependent changes in gene expression? What is happening in other brain areas that do not express such high levels of δ receptors? This may be a particularly interesting ques-tion for brain areas such as the amygdala that are known to have more direct projections to hypothalamus than the hippocampus and thus have potentially more significant responses to variations in steroid concentrations during the estrous cycle.

Also, how relevant are these findings to humans? For example, if the ethanol-sensi-tive δ subunits7 that subserve tonic inhibition change substantially during the estrous cycle, why are the effects of ethanol (and other anes-thetics) not more significantly altered by the estrous cycle13?

Reproductive steroid hormones directly and indirectly modulate peptides, receptors, synapses, transporters, glia and many other factors14 that influence excitability in the dentate gyrus. The enormous change in den-dritic signal transmission owing to just one of these changes, the tonic inhibition medi-ated by GABAA receptor δ subunits, raises this question: how does the brain continue to function (or, why are seizures not more com-mon) in the face of these massive changes in the level of synaptic input? Are there compen-satory changes in GABA levels, GABA release or GABA transport? And what is the function of this type of plasticity across the estrous cycle? Are these alterations really needed for successful mating, nesting and so on, or are we missing a more fundamental reason for these changes in a woman's mind?

1. Woolley, C.S. & McEwen, B.S. J. Neurosci. 12, 2549–2554 (1992).

2. Smith, S.S. & Woolley, C.S. Cleve. Clin. J. Med. 71 (suppl.), S4–S10 (2004).

3. Nakamura, N.H., Rosell, D.R., Akama, K.T. & McEwen, B.S. Neuroendocrinology 80, 308–323 (2004).

4. Maguire, J.L., Stell, B.M., Rafizadeh, M. & Mody, I. Nat. Neurosci. 8, 797–804 (2005).

5. Jack, J.J.B., Noble, D. & Tsien, R.W. Electric Current Flow in Excitable Cells (London, Oxford Univ. Press, 1975).

Figure 1 Comparison of the estrous cycle of the rat and menstrual cycle of the human. (a) The 4-day estrous cycle of the rat (gray bars indicate night, 6 p.m. to 6 a.m.), showing fluctuations in estrogen, progesterone, luteinizing hormone (LH) and follicle-stimulating hormone (FSH). The arrow indicates the approximate time of progesterone withdrawal and potential change in δ subunit expression suggested by the findings of Maguire et al. In the mouse, the estrous cycle is similar, but varies in length (4–7 days) and is less well defined. (b) The human 28-day menstrual cycle. The time of progesterone withdrawal is thought to be at the end of the cycle, approximately at the onset of menses. Seizure susceptibility increases at this time, but in some individuals seizure susceptiblity may also increase at the time of ovulation, or during anovulatory cycles10. Adapted from ref. 15.

Progesterone

LH

FSH

Estrogen

Diestru

s 2

Diestru

s 1

Proes

trus

Estrus

Rodenta b Human

51 1 410

Progesterone

LH

FSH

Estrogen

15 20 25Days

? ?δ δ

Ann Thomson

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



6. Wohlfarth, K.M., Bianchi, M.T. & Macdonald, R.L. J. Neurosci. 22, 1541–1549 (2002).

7. Wallner, M., Hanchar, H.J. & Olsen, R.W. Proc. Natl. Acad. Sci. USA 100, 15218–15223 (2003).

8. Staley, K. & Mody, I. J. Neurophysiol. 68, 197–212 (1992).

9. Ribak, C.E., Gall, C.M. & Mody, I. The Dentate Gyrus

and its Role in Seizures (Springer-Elsevier, New York, 1992)

10. Foldvary-Schaefer, N. & Falcone, T. Neurology 61 (suppl.), S2–S15 (2003).

11. Reddy, D.S., Kim, H.Y. & Rogawski, M.A. Epilepsia 42, 328–336 (2001).

12. Frye, C.A. & Bayon, L.E. Pharmacol. Biochem. Behav.

62, 315–321 (1999).13. Holdstock, L. & de Wit, H. Psychopharmacology (Berl.)

150, 374–382 (2000).14. McEwen, B.S., Alves, S.E., Bulloch, K. & Weiland,

N.G. Psychopharmacol. Bull. 34, 251–259 (1998).15. Knobil, E. & Neill, J.D. The Physiology of Reproduction

Vol. 2 (Raven, New York, 1994).

Subtracting the Math: prominin-positive cerebellar stem cells in white matterAnna Marie Kenney & Rosalind A Segal

A study in this issue describes the first multipotent stem cell identified in the postnatal cerebellum. These cells can generate inhibitory interneurons, astrocytes and oligodendrocytes. They may also be responsible for a class of childhood brain tumors.

Anna Marie Kenney is at the Department of

Pediatric Oncology, Dana Farber Cancer Institute,

44 Binney Street, Boston, Massachusetts 02115,

USA. Rosalind A. Segal is in the Department of

Pediatric Oncology, Dana Farber Cancer Institute,

44 Binney Street, Boston, Massachusetts 02115,

USA and the Department of Neurobiology, Harvard

Medical School, 220 Longwood Avenue, Boston,

Massachusetts 02115, USA.e-mail: [email protected]

A widely accepted model for generation of appropriate numbers and types of CNS cells during development involves progres-sive restriction from a multipotent stem cell to glia- or neuron-specific progenitors. However, multipotent CNS stem cells per-sist after birth and into adulthood. Moreover, there is accumulating evidence for regional specification of stem cells1,2. Thus, just as mature neurons are not homogeneous throughout the CNS, local stem cell popula-tions are unique and are subject to micro-environmental influences that define their fates and regulate their proliferative states throughout development. These findings suggest that the classical model for CNS development requires refinement.

The study from Lee and colleagues in this issue3 identifies a new population of stem cells in the postnatal cerebellum and highlights their potential benefits as well as the risk that these proliferative cells may also give rise to malignant brain tumors. To date, most neural stem cell studies have focused on the subven-tricular zone, olfactory bulb and hippocampal dentate gyrus1. In contrast, the cerebellum has been largely overlooked, despite the detailed description of the timing of proliferation and characterization of developing cell types in the developing cerebellum that has resulted

from the elegant work of Altman and Bayer4. Their studies indicated that the cerebellum is an ideal system for exploring CNS stem cell behavior, given its well-defined cell types and their stereotypical schedules of expansion and differentiation. Add to this the fact that per-turbation of cerebellar development can lead to ataxias or childhood brain tumors, and the study of stem cells in this part of the CNS seems overdue.

The cell types of the mature cerebellum can be readily distinguished and are localized in discrete layers (Fig. 1). Early in development, an additional layer, the external granule cell or external germinal cell layer (EGL), is located on the surface of the developing cerebellum, between the incipient molecular layer and the pia. Based on birth dating studies, Altman and Bayer demonstrated that progenitors of three types of cerebellar neurons (cerebellar granule

Figure 1 Cells of the mature cerebellum have distinctive morphologies and are located in identifiable layers. The outer molecular layer contains the GABAergic stellate (e) and basket cells (b). Below this layer is a single cell layer with Purkinje cells (a). Granule cells (g) are in the internal granule cell layer, near the Bergmann glia (j). Oligodendrocytes (m) are in the white matter tracts in the interior of the cerebellum. From S. Ramon y Cajal, Histologie du Systeme Nerveux de l’Homme et des Vertebres (Maloine, Paris, 1911).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



cells, basket cells and stellate cells) proliferate postnatally, as do the progenitors of Bergmann glia and oligodendrocytes. Proliferation was detected predominantly in the EGL, with smaller numbers of proliferating cells in the white matter below the Purkinje cell layer. However, the lineage relationships between these proliferating populations remained unclear. In agreement with the classical model that CNS stem cells produce distinct neuro-nal and glial precursors, Altman proposed that the EGL precursors give rise to the gran-ule cells, basket cells and stellate cells, whereas the dividing cells in the white matter are glial progenitors.

Subsequent retroviral lineage studies chal-lenged Altman’s model for cerebellar develop-ment. James Goldman’s group demonstrated that proliferating cells in the white matter include precursors for the GABAergic basket and stellate cells as well as glia, whereas the precursors in the EGL give rise exclusively to glutamatergic granule cells5. However, these elegant studies could not address whether either of the postnatal proliferating popula-tions could be explicitly defined as stem cells: that is, if they were capable of self-renewal as well as the generation of differentially fated daughter cells.

Lee et al. now provide new insight as to the nature of the progenitor cells in the white matter and the EGL. The authors identified two populations of dividing cells in the post-natal cerebellum: Math1-positive commit-ted granule cell precursors in the EGL and a newly identified stem cell population in the white matter (Fig. 2). These stem cells express the cell-surface glycoprotein prominin-1 (CD133), which is found in neural and hema-topoetic stem cells. To isolate the stem cells, they used mice with a Math1-green fluores-cent protein (GFP) transgene to remove the numerous granule neuron precursors from the population using flow cytometry. The remaining GFP-negative cells included cells expressing neural, oligodendrocyte or astro-cyte markers, and a small population of prom-inin-1 expressing cells that did not express any of these lineage markers but that frequently expressed nestin. The prominin-1 positive cells were localized to the white matter tracts of the developing cerebellum. These data sug-gested that the proliferating cells identified previously in the white matter could be stem cells, rather than glia-restricted or GABAergic neuron–restricted precursors.

To obtain a purified population of prom-inin-1–positive, lineage-negative cells for

further characterization, Lee and colleagues used flow cytometry to sort away cerebellar cells expressing lineage-indicative markers. In neurosphere assays, the prominin-positive, lineage-negative cells were found to be true, tri-potential stem cells in vitro, capable of self-renewing and also giving rise to neurons, astrocytes and oligodendrocytes. This stem-cell quality was verified in vivo by transplanting the purified, prominin-positive, lineage-neg-ative cells into the developing cerebellum, where they also gave rise to all three lineages. Interestingly, in vitro these stem cells not only generated glial cells and GABAergic basket and stellate cells, as expected for a white mat-ter progenitor, but also produced glutamater-gic granule cells, heretofore thought to derive exclusively from Math1-positive precursors in the EGL. However, in vivo, transplanted prom-inin-positive cells did not generate detectable granule cells. Further studies may define the stimuli that enable the prominin-positive stem cells grown in vitro to generate glutama-tergic granule cells in addition to the other cell types. In the future, isolation and expansion of these cells combined with identification of cues directing their fate may prove to be help-ful in repair strategies for treating cerebellar neurodegenerative disorders.

The study of Lee et al. highlights intriguing distinctions between the proliferating Math1-positive precursor cells of the EGL and the proliferating prominin-1–positive stem cells of the white matter. The Math1-positive cells are committed precursors that give rise exclu-sively to granule cells and do not form neuro-spheres. In agreement with previous studies, Lee et al. find that Sonic hedgehog (Shh) is a potent mitogen for the Math1-positive cells6. However, basic FGF (bFGF), which increases proliferation of cerebellar precursors7, does not function as a mitogen for the Math1-posi-tive granule cell precursors. Indeed, Wechsler-Reya and Scott have shown that bFGF strongly antagonizes Shh-mediated proliferation of EGL-derived granule neuron precursors6. In contrast, the prominin-1 positive cells do not respond to Shh but instead show a clear proliferative response to bFGF. Moreover, Math1-positive EGL cells give rise only to granule neurons, whereas the prominin-posi-tive, white matter–derived cells generate mul-tiple cell types in vitro and in vivo. Thus, the two populations have distinct proliferative and differentiation properties.

In addition to providing insight into pro-cesses regulating cerebellar development, the work of Lee et al. carries powerful implications for understanding the biology of medulloblas-tomas, cerebellar brain tumors. These tumors typically occur in childhood, when cerebellar

Granule neurons

Medulloblastomavariants

(Shh)

(bFGF)

Math1

CD133

Astrocytes Oligodendrocytes

GABAergicinterneurons

EGL ML IGL WM

Figure 2 Basic FGF–responsive stem cells in the cerebellar white matter generate GABAergic interneurons, astrocytes and oligodendrocytes. Folium in the early postnatal cerebellum, sagittal view. The external granule layer (EGL) comprises Math1-positive neural precursor cells. These cells proliferate in response to signaling by Sonic hedgehog (Shh). They then migrate to the internal granule layer (IGL), where they complete terminal differentiation into glutamatergic granule neurons. Medulloblastomas may arise from either of these precursors. WM, white matter.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



development takes place. It has been widely presumed that the cell of origin for medullo-blastomas is the granule cell precursor, which undergoes rapid proliferation during this stage8. Consistent with this idea, mutations that activate the Sonic hedgehog pathway have been identi-fied in human medulloblastomas9. A mouse in which mutation of the Patched receptor leads to constitutive activation of Sonic hedgehog path-way also develops medulloblastoma10, providing additional evidence that this pathway is involved in cerebellar progenitor cell transformation. Many important effectors of hedgehog’s mito-genic function have been identified, including transcription factors such as N-myc11 and Gli family members12. These proteins are upregu-lated in the ‘desmoplastic’ variety of medul-loblastoma in humans13 and are expressed in the medulloblastomas of the Patched mutant mice9,10, providing additional evidence that activation of the Sonic hedgehog pathway in granule cell precursors can cause medulloblas-

toma formation. Small-molecule compounds specifically targeting the hedgehog pathway hold promise for treating medulloblastomas associ-ated with hedgehog pathway activation14,15.

However, as Lee and colleagues discuss, the so-called ‘classic’ medulloblastomas are not associated with hedgehog pathway activation and are genetically and pathologically dis-tinct from the hedgehog-dependent, desmo-plastic medulloblastomas8,13. An intriguing possibility raised by the study of Lee et al. is that the prominin-1 positive cerebellar stem cells constitute an alternative cell of origin for medulloblastoma. If so, the evidence that cerebellar stem cells proliferate in response to bFGF stimulation rather than to stimulation with Shh suggests that compounds blocking the bFGF pathway or its downstream effectors could be therapeutic in classic medulloblasto-mas. Such targeted therapy would be preferable to current medulloblastoma treatments, which do not distinguish the therapies for these two

tumor subtypes, and carry potential physical, developmental and psychological side effects.

1. Alvarez-Buylla, A., Garcia-Verdugo, J.M. & Tramontin, A.D. Nat. Rev. Neurosci. 2, 287–293 (2001).

2. McKay, R. Science 276, 66–71 (1997).3. Lee et al. Nat. Neurosci. 8, 723–729 (2005).4. Altman, J. & Bayer, S.A. Development of the Cerebellar

System in Relation to its Evolution, Structure, and Functions (CRC Press, Boca Raton, Florida, 1997).

5. Zhang, L. & Goldman, J.E. Neuron 16, 47–54 (1996).6. Wechsler-Reya, R.J. & Scott, M.P. Neuron 22, 103–

114 (1999).7. Gao, W.O., Heintz, N. & Hatten, M.E. Neuron 6, 705–

715 (1991).8. Provias, J.P. & Becker, L.E. J. Neurooncol. 29, 35–43

(1996).9. Wetmore, C. Curr. Opin. Genet. Dev. 13, 34–42

(2003).10. Goodrich, L.V., Milenkovic, L., Higgins, K.M. & Scott,

M.P. Science 277, 1109–1113 (1997).11. Kenney, A.M., Cole, M.D. & Rowitch, D.H. Development

130, 15–28 (2003).12. Ruiz i Altaba, A. Trends Genet. 15, 418–425

(1999).13. Pomeroy, S.L. et al. Nature 415, 436–442 (2002).14. Romer, J.T. et al. Cancer Cell 6, 229–240 (2004).15. Berman, D.M. et al. Science 297, 1559–1561

(2002).

Blue genes: wiring the brain for depressionStephan Hamann

How do genes act in the brain to influence susceptibility to mental illness? An imaging study suggests that healthy carriers of a gene variant associated with depression risk have decreased brain volume and neural coupling in affective circuitry involved in depression.

Stephan Hamann is in the Department of

Psychology, Emory University, 532 North Kilgo

Circle, Atlanta, Georgia 30322, USA.


The genetic test result comes back: your baby carries a high-risk gene for depression and anxiety. What does this portend for his future? How will this gene affect his developing brain and temperament, and what can we do to pro-tect him from developing depression? The idea that genes affect risk for mental illness is widely accepted, but exactly how genes influence the brain to increase vulnerability is a mystery that has only recently begun to unfold. A new study1 in this issue uses multiple brain imag-ing techniques to show that carrying a high-risk variant of the serotonin transporter gene profoundly affects both anatomy and function in a key emotion circuit. This work has broad implications for understanding how genetic vulnerability to depression is manifested in the brain’s response to emotional stimuli.

Serotonin is an important modulator of emotional behavior, and considerable evidence links serotonergic dysfunction to depression.

Serotonin also shapes the brain’s development, including that of the limbic system’s affective circuitry2. Given these links, genes regulat-ing serotonin have become prime suspects in tracking down genetic factors in the develop-ment of depression. A variation in the pro-moter region of the serotonin transporter gene (5-HTTLPR) has attracted particular interest. The short (s) allele is associated with reduced serotonin availability compared to the long (l) allele, and individuals who carry at least one s allele (s/s or s/l) have increased anxiety-related traits and risk for depression3.

Depressed patients have decreased brain volume in the subgenual part of the anterior cingulate, together with abnormal activity in a key affective circuit involving the anterior cingulate and amygdala4. However, it has been unclear whether these abnormalities precede the development of depression or are caused by the depressed state. A straightforward approach to this question is to study healthy individuals with respect to their genetic risk for depression to see whether brain anatomy and function in those at higher risk resembles that of the depressed brain, in effect identify-ing neural markers that signal vulnerability to

depression. The new work is one of the larg-est published imaging genomics studies to date, with 94 subjects. In this study, Pezawas et al.1 used a three-pronged imaging approach, hypothesizing that healthy individuals who carry the high-risk 5-HTTLPR s allele would show depression-like changes in brain struc-ture and function in the cingulate-amygdala circuit. They used structural MRI to examine gray matter volume, functional MRI to look at responses during a task with emotional stimuli previously shown to elicit greater amygdala activation in s allele carriers5, and analysis of the functional responses to determine the degree of functional coupling in cingulate-amygdala circuitry.

The primary results strongly supported the conclusion that depression-like changes in anatomy and function are indeed present in healthy carriers of the high-risk s allele, con-stituting neural markers for vulnerability to depression. Carriers of the s allele had more than a 25% reduction in gray matter volume in the perigenual anterior cingulate and an approximately 15% reduction in the amyg-dala (Fig. 1). Notably, the greatest decrease was in the same region where reduced volume

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



has been observed in depression6, the rostral anterior cingulate (rACC), supporting the link to depression. This is the first evidence that healthy s allele carriers show significant vol-ume reductions in brain regions implicated in depression, and the magnitude of these effects is striking. In line with the idea that these regions constitute an affective circuit, the mag-nitude of volume decreases in the cingulate and amygdala were highly correlated within individual subjects.

The authors next examined whether genotype also affected functional coupling between the amygdala and perigenual cin-gulate regions, where s allele carriers showed decreased gray matter volume. Across all sub-jects, analysis of moment-to-moment fluc-tuations in fMRI activity during a task with fearful and angry faces showed that bilateral amygdala activity correlated positively with rACC activity and negatively with caudal anterior cingulate (cACC) activity. Activity in rACC and cACC was positively correlated as well. These correlations suggested a basic affective circuit involving a positive con-nection between the amygdala and rACC, a positive connection between rACC and cACC and, finally, a negative feedback connection from the cACC to the amygdala. Consistent with the authors’ prediction that s allele carri-

ers would show alterations in this functional circuit, the high-risk allele was associated with decreased positive coupling between the amygdala and rACC (Fig. 1) and decreased negative coupling or feedback between the cACC and amygdala, which may lead to elevated amygdala responses to emotional stimuli in s allele carriers5. Subjects’ scores on a personality test assessing anxiety-related traits associated with depression correlated strongly with dynamic coupling between the amygdala and rACC, but not with func-tional and anatomic measures in individual areas, supporting the link between the imag-ing results and depression susceptibility and highlighting the importance of examining dynamic functional interactions.

A recent study7 also looked at 5-HTTLPR genotype and affective processing with fMRI but, intriguingly, found increased rather than decreased functional connectivity between the amygdala and an area near the rACC, the ventromedial prefrontal cortex (vmPFC), (Fig. 1). Pezawas et al. reanalyzed their own data and found qualitatively similar results. In part because the vmPFC may lack signifi-cant direct connections to the amygdala, the authors of the current study propose that the rACC, which has abundant direct con-nections from the amygdala, is part of a pri-

mary affective circuit and that the vmPFC has a more indirect, compensatory role. Statistical techniques such as dynamic causal modeling, which go beyond simple temporal correlations to explore directional, modu-latory and hemispheric asymmetry effects, will help clarify the wiring of this extended affective circuit. Interestingly, Heinz et al.7, using emotional scenes as stimuli, found that positively valenced emotional pictures also elicited amygdala activation, consis-tent with a general role for the amygdala in both positive and negative emotion8, but only responses to negative pictures showed 5-HTTLPR genotype effects. This suggests that the nature of the experimental emo-tional stimuli can influence whether genotype effects will be detected. Future studies using multiple stimuli selected to evoke different types of affective responses may clarify what aspects of emotion are modulated by geno-type and may lead to more sensitive probes of gene effects.

Environmental factors are thought to be critical in determining whether individuals with genetic vulnerability will eventually develop depression. Though somewhat con-troversial, recent studies suggest that carriers of the high-risk serotonin transporter allele may never develop depression unless they are exposed to stressful and traumatic events, particularly in early life9. Even with such exposure, although the serotonin transporter gene clearly influences affective responses, it is only one of an unknown number of genes that may potentially contribute to mental illness susceptibility. Important remaining questions are whether healthy s allele carriers who show stronger anatomical and functional markers will develop depression at a higher rate than those who do not show these markers, and how stressful life experiences factor into this process. The answers will require longitudi-nal, prospective studies following individuals for several years. Another line of questions concerns how early in life the anatomical and functional alterations associated with 5-HTTLPR variation develop. Altered sero-tonin levels can affect early brain develop-ment in animal models, suggesting that by birth the effects of this genetic variation may already be present2. Determining the devel-opmental mechanisms that underlie the structural and functional changes observed here in humans is an important direction for further study.

Surprisingly, most people of European descent carry at least one high-risk allele, and thus along with the baby in our original exam-ple, are at higher risk for depression3. Why should a genetic tendency toward the blues be

Figure 1 Differences in processing of emotional stimuli between s allele carriers (blue arrows) and homozygous l allele carriers (yellow arrows). Negative emotional stimuli are evaluated by the amygdala (red arrow) after preliminary analysis in the ventral visual pathway (not shown). Carriers of the s allele have markedly reduced positive functional coupling between the rostral anterior cingulate (rACC; purple oval) and the amygdala, which results in a net decrease in inhibitory feedback from the caudal anterior cingulate (cACC; green oval), via connections between rACC and cACC (short upward arrows). Brain volume was also substantially reduced in s allele carriers in the rACC and, to a lesser extent, the cACC and amygdala. The consequence of these genotype-based alterations is an emotional hyper-responsivity to negative affective stimuli in s allele carriers (large blue cloud) compared with individuals lacking this allele (small yellow cloud), which may be related to an increased risk of developing depression. As found in a previous study7, functional coupling between the vmPFC (pink circle) and the amygdala was also increased in s allele carriers.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



relatively common? As has often been noted, the function of genes is not to cause disease, and surely, the short 5-HTTLPR allele does not exist for the purpose of causing depression. Instead, it seems to tune and prune affective circuits so as to heighten responses to negative emotional stimuli, a consequence that could easily prove adaptive in threatening environ-ments. Conversely, carriers of the low-risk l variant have attenuated amygdala responses5,7 to negative emotional scenes or faces signaling threat. Always looking on the bright side of

life is fine for warding off depression, but a blunted response to aversive stimuli carries its own risks.

The current study demonstrates the power of imaging genomics to uncover alterations in brain structure and function related to genetic risk for psychiatric disorders. Combining the strengths of functional imaging with genomics holds considerable promise in identifying the neural underpinnings of human individual-ity and understanding how this variability interacts with the environment to produce

both adaptive emotional behavior and dis-ease states.

1. Pezawas, L. et al. Nat. Neurosci. 8, 828–834 (2005).

2. Gross, C. & Hen, R. Nat. Rev. Neurosci. 5, 545–552 (2004).

3. Lesch, K.P. et al. Science 274, 1527–1531 (1996).4. Mayberg, H.S. Br. Med. Bull. 65, 193–207 (2003).5. Hariri, A.R. et al. Science 297, 400–403 (2002).6. Drevets, W.C. Biol. Psychiatry 48, 813–829 (2000).7. Heinz, A. et al. Nat. Neurosci. 8, 20–21 (2005).8. Hamann, S.B., Ely, T.D., Hoffman, J.M. & Kilts, C.D.

Psychol. Sci. 13, 135–141 (2002).9. Caspi, A. et al. Science 301, 386–389 (2003).

Shaking up sleep researchJoan C Hendricks

Sleep deprivation causes all too familiar behavioral impairments and increased need for sleep. A new Drosophila mutant with alterations in the Shaker potassium channel sleeps less than normal but does not show the usual effects of sleep deprivation.

Joan C. Hendricks is at the Department of

Clinical Studies, Philadelphia School of Veterinary

Medicine, University of Pennsylvania, Philadelphia,

Pennsylvania, USA.


Did you sleep well last night? Are you sleepy now? Most of us probably know and envy someone who annoyingly claims to need only a few hours of daily sleep. For many people though, the inability to fall asleep when we want to or the feeling of sleepiness at inopportune times are familiar sensations. In a recent issue of Nature1, Cirelli and colleagues report the iden-tification of a remarkable Drosophila mutant that not only sleeps less but also to some degree is immune to the effects of sleep deprivation (albeit at the price of a shortened life span). This discovery makes an important contribution to our understanding of the remarkably elusive questions of how and why we sleep.

In both mice2 and humans3, potassium channels are implicated in sleep, but studies in Drosophila melanogaster provide the oppor-tunity for a better quality of proof. Sleep-like behavior in Drosophila4,5 was documented just a few months before the fly genome was published. Hopes were immediately raised that flies could help discover the genetic basis for sleep6, despite differences between fly sleep and ‘true’ sleep (most notably, the apparent absence of sleep stages analogous to REM and non-REM). These initial hopes were soon ful-filled by identification of conserved genes7,8 and the demonstration that sleep loss—caused by mutant genes9,10, environmental stimula-

tion10 or drugs11—is lethal for flies. In the cur-rent study, Cirelli and collaborators identified a mutant that slept very little but appeared normal during waking, which they named minisleep (mns). Did these flies need less sleep? Probably not, as the authors also found that mns and other short-sleeping flies with muta-tions in the same gene had shortened lives.

To identify their mutant, the authors took full advantage of the efficiency and low cost of breeding Drosophila, using random muta-genesis, unbiased screening of 9,000 lines of mutant flies, large numbers for statistical analyses and mature genetic tools. The ethyl methane sulfonate (EMS) mutagenesis that led to mns produces point mutations that are notoriously difficult to map. However, the authors noted that the mutants shook when exposed to ether anesthesia. As a per-fect example of chance favoring the prepared mind, the authors recognized this phenotype and focused their attention on the highly con-served Shaker gene, which encodes a voltage-gated potassium channel heavily expressed in the CNS. They soon discovered that mns flies carried a point mutation in the α subunit of the Shaker channel.

However, when the authors examined other existing, independently generated lines of flies with Shaker mutations, they discov-ered that only one null mutant showed short sleep. The authors cleverly considered the possibility that accumulated modifiers might have altered the phenotype and allowed the mutants to show normal sleep. That is, to compensate for a maladaptive mutation, altered forms of genes that counteracted the

mns mutation would have provided a selec-tive advantage to the population over succes-sive generations. To test this possibility, the authors bred the null Shaker mutants that did not show short sleep to unrelated lines of flies that had not been under the same sort of selective pressure (using well-estab-lished Drosophila tricks to preserve the mns mutation). The resulting offspring displayed not only the short-sleep phenotype but also the short-lived phenotype observed in mns. This tendency to accumulate modifiers that prolong both sleep and life further solidifies the evidence that sleep serves an important biological purpose across evolution.

There are several interesting implications of the study for understanding sleepiness and sleep need. The mns flies had reduced baseline sleep but were normally responsive and active. Wild-type flies sleep more deeply after sleep deprivation, which the investiga-tors measured by their ability to sleep during complex mechanical noxious heat stimula-tion. The mns flies responded to deprivation by increasing their total sleep appropriately, but they did not show the changes in sleep intensity. Normal fly sleep is also more con-solidated following deprivation, meaning that they sleep for longer continuous periods. However, even after experimentally imposed all-nighters, mns flies continued to break up their sleep with arousals, fragmenting sleep into relatively short bouts, and they woke up readily when stimulated—they were even hyper-responsive. One might reason that mns flies either have a deficit in deep, consolidated sleep, or that they need to sleep less because

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



they sleep more efficiently. The latter inter-pretation seems highly unlikely, given that the sleep loss phenotype is correlated with a shortened lifespan in several independent lines of Shaker mutations.

We have yet to find a ‘super-sleeper’ muta-tion or manipulation that allows short sleep without negative consequences. In mns flies, the response to sleep deprivation seems to be uncoupled from the need for sleep (Fig. 1a). This suggests that this line of flies might provide a tool to explore the link between longevity and sleep duration. It would be very interesting to pursue studies of the ‘dose-response’ of these flies to sleep deprivation and also to identify whether sleep deprivation hastens the demise of mns flies, as it does in normal flies and in another short-sleeping phe-notype9, cyc0 (ref. 10). Further, does constant darkness (which also reduces sleep9, even in these mutants1) result in a further shortening of the lifespan? The authors used lines of flies

with duplications of the Shaker gene to demon-strate that both the short-lived and the short-sleep phenotypes can be rescued by one or two copies of the gene; it would be relatively easy to discover whether there is any gain-of-function phenotype by overexpressing the gene and/or producing flies with more than two copies. It would be naive to expect that a long-lived, long-sleeping fly might result, but sometimes naivete can lead to interesting findings.

There are several potential explanations for the inability of the mns flies to increase their sleep consolidation and depth after sleep loss. We do not know what the specific sleep-inducing signal is in any species, but most theories are based on the parsimonious assumption that the beneficial function of sleep (for instance, to restore neural energy stores12) is linked to the sleep-inducing signal12. Although this may still be part of the story, the uncoupling of the response to sleep deprivation from deleterious effects on lifespan in mns indicates that the sleep-inducing signal

may be at least partly independent from the life-preserving function of sleep.

Why do mns flies fail to respond normally to the sleep-inducing signal? Thinking about the sleep signal in sleep homeostasis as we think about insulin in appetite regulation, it is pos-sible either that the signal is reduced in mns flies (as in type 1, insulinopenic diabetes) or that the cells are resistant to the signal (as in type 2, insulin-resistant diabetes). This second possibility is consistent with evidence impli-cating Shaker in regulation of general neural excitability. It may be that mns flies are unable to increase their sleep depth due to inability to respond to the sleep signal.

Although it is possible that the mns flies expe-rience a reduction in sleepiness (Fig. 1b, cen-ter), another possibility is that the sleep signal and response are normal but that mns flies are more aroused by sleep deprivation than nor-mal, counter-balancing any sleepiness they feel (Fig. 1b, right). In mammals, including humans,

Figure 1 Shaker channel minisleep mutant flies have altered responses to sleep deprivation. (a) Normally, both a sleep signal and the need for sleep accumulate in parallel as the duration of wakefulness increases. By contrast, minisleep (mns) sleep less and do not sleep more deeply after deprivation, as if they do not perceive the signal for sleep as well as normal flies. However, the short sleep is correlated with a short lifespan, suggesting that the sleep debt accumulates normally. Thus, sleep need is uncoupled from the effect of the sleep signal. (b) Two general mechanisms could cause the reduced response to sleep deprivation in minisleep mutant flies. Sleep loss normally increases sleep depth and duration. This occurs despite the stressful, arousing effects of sleep deprivation, because the effective sleep signal normally effectively overbalances the effect of arousing stimuli so that the homeostatic balance tilts decisively toward sleep. In minisleep mutants, the net effect of sleep deprivation is abnormal, with a reduced depth of sleep. A reduced effective sleep signal would give this result (center), and an increase in the arousing effect of sleep deprivation would also give this result (right). These possibilities are not mutually exclusive—sleepiness could be reduced and arousability increased. The net effect in any of these scenarios would be the same: sleep deprivation would tilt the homeostatic balance less decisively toward sleep.

Sleep need(normal and mns)

Sleep signal(normal)

Sleep signal(mns)

Duration of wakefulness

Sleep loss Sleep lossSleep loss

Sleep Wake Sleep WakeSleep Wake

a b

Ivelisse Robles

Normal Minisleep

Figure 2 Multiple model systems facilitate effective studies. A multidimensional approach using models throughout evolution is proving to be successful in identifying genes involved in sleep. The question or disease condition is identified in humans, and the same behavior is then sought in laboratory mammals and simpler species. Screening and genomic approaches can identify conserved genes in appropriate laboratory animal models. Simple models such as flies are powerful for identifying candidate genes, and laboratory mammals are powerful for demonstrating conserved function and pathophysiology. To improve human health, studies must return to humans, and it must be verified that the genes of interest are involved. Yet another dimension can be added by primary culture of cells from models of disease and by performing in silico studies to identify genes and signaling pathways in all species.

Ann Thomson

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



sleep disorders related to abnormal K+ channel function are accompanied by hyperactivity and restlessness3,13, and even in normal individuals, sleep loss causes stress and arousal, in addition to subjective sleepiness14. Indeed, some have suggested that attention-hyperactivity syn-drome may reflect sleep loss in children15. Even though mns flies are not hyperactive at baseline, the mns mutation may make them so agitated by the sleep deprivation that they cannot get a good, deep night’s sleep.

It will be interesting to identify whether CNS defects are related to the phenotype in mns. A less interesting possibility, at least for understanding sleep mechanisms, would be if the peripheral neuromuscular defect that is responsible for the shaking phenotype leads to hypersensitivity. The apparent hyperarous-ability of mns flies after 24 hours of disturb-ing mechanical stimuli could simply be due to abnormal peripheral excitability. Tissue-selec-tive expression of the normal gene in the CNS will establish whether a normal Shaker channel in the brain can correct the short-sleeping phe-notype, even when the peripheral neurons have the mns mutation. This can be done by rescu-ing only the CNS or only the somatic tissues by using a GAL4-driven UAS-Shaker channel. This tissue-specific rescue will readily demonstrate whether the change in homeostasis is indeed due to mutated Shaker channels in the CNS.

This report provides a general lesson in the utility of a modern protocol using multiple

models to take advantage of their comple-mentary strengths as an approach to the neurogenetics of complex behaviors (Fig. 2). The first step is focusing on an important question for human health—in this case, the wish to understand and control the timing and quality of sleep. Once a conserved phe-notype and candidate gene is established, a coordinated approach is ideal. As perfectly illustrated by the studies of mns, studies in Drosophila can be relatively efficient and convincing in linking a phenotype to a spe-cific gene. For Shaker, as in many cases, the fly often uses a single gene where the mam-malian genome has expanded to multiple genes, so the actual biology itself can also be simpler. Once a gene is known to be impor-tant in sleep, changes in gene expression and cell signaling networks caused by the muta-tion can be investigated across the genome using microarray and in silico approaches readily available in both flies and mammals. Details of mammalian neuroanatomy and neurophysiology are, of course, best under-stood in mammalian models, and studies of the cellular physiology caused by the defect in the Shaker channel will likely be best studied in vitro. Again, in most cases, tissues from mammals are more readily available and more appropriate than fly cells or tis-sues. As hypotheses about sleep mechanisms and function mature, more focused studies and certainly any potential therapies must

be conducted in humans to verify that the results are relevant to human health.

In summary, this study carefully and thor-oughly demonstrates that a point mutation in a conserved K+ channel underlies sleep loss and changes in sleep homeostasis that are correlated with shortened lifespan in Drosophila. Given the urge to control sleep to optimize function in our 24-hour global economy, this strikes a cautionary note: the general rule is that losing sleep is not only generally subjectively unpleas-ant, but deleterious to a long and healthy life.

1. Cirelli, C. et al. Nature 434, 1087–1092 (2005).2. Espinosa, F., Marks, G., Heintz, N. & Joho, R.H. Genes

Brain Behav. 3, 90–100 (2004).3. Josephs, K.A. et al. J. Clin. Neurophysiol. 21, 440–445

(2004).4. Hendricks, J.C. et al. Neuron 25, 129–138 (2000).5. Shaw, P.J., Cirelli, C., Greenspan, R.J. & Tononi, G.

Science 287, 1834–1837 (2000).6. Kilduff, T.S. Neuron 26, 295–298 (2000).7. Hendricks, J.C. et al. Nat. Neurosci. 4, 1108–1115

(2001).8. Graves, L. et al. J. Neurophysiol. 90, 1152–1159

(2003).9. Hendricks, J.C. et al. J. Biol. Rhythms 18, 12–25

(2003).10. Shaw, P.J., Tononi, G., Greenspan, R.J. & Robinson,

D.F. Nature 417, 287–291 (2002).11. Hendricks, J.C., Kirk, D., Panckeri, K., Miller, M.S. &

Pack, A.I. Sleep 26, 139–146 (2003).12. Benington, J.H. & Heller, H.C. Prog. Neurobiol. 45,

347–360 (1995).13. Espinosa, F., Marks, G., Heintz, N. & Joho, R.H. Genes

Brain Behav. 3, 90–100 (2004).14. Horne, J.A. & McGrath, M.J. Biol. Psychol. 18, 165–

184 (1984).15. O’Brien, L.M. & Gozal, D. Minerva Pediatr. 56, 585–

601 (2004).

Where did the time go?Sabrina Ravel & Barry J Richmond

How do we form arbitrary associations, such as ‘stop at red’ or ‘go at green’? A report in Nature suggests that these associations are first formed in the striatum but that activity changes in the prefrontal cortex are more closely related to improved performance.

Sabrina Ravel and Barry J. Richmond are at the

Laboratory of Neuropsychology, National Institutes

of Mental Health, National Institutes of Health,

Department of Health and Human Services,

Building 49, Room 1B80,

Bethesda, Maryland 20892, USA.

e–mail: [email protected]

induce overwhelming desire for the addicting drug1. Learning such associations activates many brain areas, presumably requiring inter-actions among them. Recently, Pasupathy and Miller2 published important and perplexing new information about how the interactions might occur. As monkeys learned new associa-tions, the authors compared the responses of single neurons in two brain areas, the dorsolat-eral prefrontal cortex and the caudate nucleus, which are thought to be important for learn-ing and acting on associations between stimuli and rewarded features. They found that learn-ing-related changes occurred more quickly in the caudate nucleus, suggesting that the basal ganglia first identify these rewarded associa-

Learning and interpreting associations between stimuli and rewarded actions is a fundamental aspect of motivated behavior. These processes are disturbed or distorted in disorders such as depression, where there are feelings that rewards are not worth working toward, or drug abuse, where external stimuli

tions. However, the changes in prefrontal cor-tex were more closely related to improved task performance.

Pasupathy and Miller focused on the pre-frontal cortex and the caudate nucleus for a variety of reasons. First, anatomical studies show that these areas are in one of a set of basal ganglia-cortical processing circuits3, the dor-solateral prefrontal circuit (Fig. 1), in which the connections project from the dorsolateral prefrontal cortex to the head of the caudate, from there to the internal part of the globus pallidus (GPi), and from the GPi back to dor-solateral prefrontal cortex4. These sequential connections have been interpreted to indi-cate that the signals are closely synchronized.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



Second, in behavioral studies, lesions of the dorsolateral prefrontal cortex interfere with delayed response tasks in general5, and lesions in the head of the caudate interfere with learn-ing about reward associations and with eye movements6. Finally, both these areas have been studied in similar delayed response tasks before6,7. These earlier studies had shown that neurons in both areas had activity triggered by a stimulus predicting the direction of the eye movement that would be rewarded: that is, the responses reflected the learned associa-tion between the stimulus and the rewarded eye movement.

However, there is debate about which of these areas causes or drives the signals in the other, and that is what Pasupathy and Miller addressed. They made neuronal recordings in both brain regions using the same behavioral protocol and measured the time when the responses began; that is, the response latency. In their study, monkeys were trained to learn new associations between each of two visual cues and two saccadic eye movements. On each trial, the monkeys were briefly presented with one of two arbitrarily chosen visual stim-uli. After the stimulus disappeared, there was a delay, and then the monkeys were given the opportunity to make a quick eye movement (a saccade) to targets located to the left and right. A movement to the right was rewarded with a drop of juice after one stimulus, and a movement to the left was rewarded after the other stimulus. After training, the mon-

keys became facile at learning the associative rule with a new pair of pictures each day. The authors were then able to record neurons day after day in both areas while the monkeys were learning new associations.

They found that the neuronal activity first developed in the caudate nucleus, within just a few trials (five to ten), with activity predict-ing which eye movement would be rewarded before the animal’s behavior became accurate. The behavior became accurate after many more trials (30–40). The predictive activity in the prefrontal neurons appeared more slowly, more closely paralleling the monkey’s behav-ioral improvement, finally becoming strongest just before the monkey’s behavior became essentially perfect (~20–30 trials). After the monkeys had experience with the cues and the predictive activity had stabilized, the firing of the caudate neurons preceded the saccadic eye movements by more than 200 ms, a very long time for neural processing, whereas the firing of the prefrontal neurons led the saccadic eye movements by a modest amount, approxi-mately 60 ms.

The finding that neurons respond earlier in the caudate nucleus than in the prefron-tal cortex answers the question posed by the authors— the peak predictive activity occurs ~140 ms earlier in caudate neurons than in dorsolateral prefrontal neurons. However, two aspects of these findings cause difficulties for the simplest interpretation, that the activ-ity of the caudate simply drives or directly

elicits the prefrontal activity. First, the cau-date activity is well-established and peaks before the monkeys’ behavior has changed very much, whereas the prefrontal activ-ity closely parallels the improvement in the behavior. If, as the monkeys gain experience with the stimuli, the responses of the caudate neurons accurately predict which saccade to make to get a reward, why does the monkey not make the correct saccade? Second, after the monkeys have learned, the difference between the latencies in the caudate and pre-frontal cortex (about 140 ms) is surprising. There is a general rule of thumb that signals are delayed by 5–10 ms per stage of neural processing8. If, after learning, the processing proceeded in a simple feedforward manner, this simple rule of thumb would lead to the conclusion that there should be several stages of processing. Yet, along the shortest path, it seems as though there are only a few steps (Fig. 1b). The lack of close synchronization makes it unlikely that the processing can be interpreted as a simple, closed feedforward/feedback system.

Because these new results make it unlikely that either one of these two areas drives the other in a simple feedforward fashion, it remains unclear whether or how these brain regions interact. The authors speculate that the caudate might learn first and then ‘train’ the prefrontal cortex, a different causal relation than direct driving of the prefrontal cortex. One implica-tion of this might be that the caudate response would disappear if observed for long enough (perhaps days or weeks). Another possibility is that these regions have different roles and that these roles are independent. Yet another interpretation is that the roles of the caudate nucleus and prefrontal cortex are complemen-tary, rather than causal. The basal ganglia are known to be important for preparing to act, guiding the execution of habits, and monitoring or coordinating programmed motor sequences. The cortex might provide more flexible adapta-tions of behavior, perhaps allowing anticipation of outcomes that had not been experienced pre-viously, so as to modify behavior on a single-trial basis. The cortex in this scenario would be involved in making a conscious choice and triggering the desired action, which would have been prepared by and would be monitored by the basal ganglia. The interdependence of these brain regions can be studied using either abla-tive or pharmacological disconnection of the pathways from the basal ganglia to the frontal cortex, and vice versa.

No matter what the specific roles of these brain regions turn out to be, it seems likely that dopamine, a neurotransmitter known to be involved in learning and maintaining reward-

PFC

CdGPi

Late presaccadicactivityPFC

Cd

Earlypresaccadic

activity

GPi

a b

Ann Thomson

Figure 1 The dorsolateral prefrontal processing loop. (a). Lateral view of a monkey brain with the locations of the regions on the surface (dorsolateral prefrontal cortex) or projected to the surface (basal ganglia separating caudate and globus pallidus). (b). Coronal sections (at positions shown in a) depicting the relative positions and major projections of this circuit. The presaccadic signal in the caudate precedes the saccadic eye movement by more than 200 ms, whereas the signal in prefrontal cortex precedes the movement by only 60 ms after learning is finished. PFC, prefrontal cortex; Cd, caudate nucleus; GPi, globus pallidus internal segment.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce



seeking behavior, is important in the neural plasticity underlying the behavioral learning9. Dopamine is a transmitter in large neurons in substantia nigra pars compacta that respond to reward or to salient cues predicting reward. It is important in learning to associate cues with predicted reward10. Both regions studied in this report are rich in dopamine. After dopamine in the caudate nucleus is disturbed, monkeys have difficulty making saccades toward the contralateral side of the body; thus, dopamine is important in the caudate nucleus for execut-ing and probably for learning this behavior. However, whether dopamine is needed for the learning of the associations seen here in the prefrontal cortex is not yet known.

Because of the timing differences in the activity of these two brain regions, it is

unlikely that either of them simply drives the other in learning or activity. Thus, the new findings require a rethinking of how these brain regions might interact. The miss-ing time makes it appealing to think these two regions complement rather than drive one another. If that were the case, we could imagine that in disorders characterized by failure to control habitual activity, such as drug abuse, in which stimuli become uncon-trollably compelling, the normal balance in strength or timing between the basal ganglia and the cortex may be disrupted. If the cau-date nucleus becomes overexcitable, activa-tion of stimulus-elicited habitual behavior might no longer be controllable by the pre-frontal cortex, which would normally trigger the behavior. Thus, the behavior would occur

without needing activation from the prefron-tal cortex. In any case, these straightforward latency measurements suggest complexity in processing predictive stimuli that had not been guessed previously.

1. Volkow, N.D. Biol. Psychiatry 56, 714–717 (2004).2. Pasupathy, A. & Miller, E.K. Nature 433, 873–876

(2005).3. Alexander, G.E., DeLong, M.R. & Strick, P.L. Annu.

Rev. Neurosci. 9, 357–381 (1986).4. Middleton, F.A. & Strick, P.L. Cereb. Cortex 12,

926–935 (2002).5. Petrides, M. Exp. Brain Res. 133, 44–54 (2000).6. Hikosaka, O., Takikawa, Y. & Kawagoe, R. Physiol. Rev.

80, 953–978 (2000).7. Leon, M.I. & Shadlen, M.N. Neuron 24, 415–425

(1999).8. Rousselet, G.A., Thorpe, S.J. & Fabre-Thorpe, M.

Trends Cogn. Sci. 8, 363–370 (2004).9. Schultz, W. Neuron 36, 241–263 (2002).10. Liu, Z. et al. Proc. Natl. Acad. Sci. USA 101, 12336–

12341 (2004).

Brain’s guard cells show their agilityMicroglia, the principal immune cells of the brain, are thought to be the nervous system’s roaming cleanup crew. When activated by injury or insult (including lesions, stroke, neurodegenerative disorders and tumors), microglia surround dead cells and clear cellular debris from the area. However, most of this work was done in vitro using brain slices, and as the slicing procedure inherently induces some injury, it remained unclear how microglia behave in vivo.

In a technical tour de force, two recent reports by Fritjof Helmchen and colleagues (published online in Science on 14 April) and Wen-Biao Gan (pp 752-758, this issue) describe the imaging of microglia in intact mouse cortex. Both groups took advantage of transgenic mice in which all the microglia were fluorescently labeled and used transcranial two-photon microcsopy to image the behavior of these

cells through the thinned skull. Microglial processes were highly dynamic in the intact brain. Although the somata of microglial cells remained morphologically stable over hours, higher-order branches showed rapid extension and retraction over intervals of seconds to minutes. This high resting mobility may enable the microglia to act as vigilant sentries, constantly screening the surrounding parenchyma.

The microglia also responded rapidly to focal brain injury in both studies. Time-lapse imaging showed that after a small laser ablation, microglia near the site of injury responded within the first minute to extend their processes toward the damaged site. Gan and colleagues report that within 30 minutes after the laser-induced injury, the processes of nearby cells reached the damaged site and appeared to fuse together, forming a spherical containment around it and establishing a potential barrier between the healthy and injured tissue, as shown in the photo. Microglia responded similarly to mechanical injury.

What signals mediate this rapid microglial response? In culture, ATP signaling induces microglial migration. Gan and colleagues extend this work to the in vivo situation, and show that extracellular ATP and activation of P2Y receptors on microglia are necessary for the rapid

microglial response toward the injury site. Simply inserting an electrode containing ATP allowed the authors to mimic—in time, range and kinetics—the rapid response of microglial processes observed following laser ablation. Furthermore, ATP-induced ATP release was essential for this response; when the authors applied apyrase (which degrades endogenous ATP in the extracellular space), and then released non-hydrolyzable ATP from a microelectrode, they observed no such rapid microglial response. Applying connexin channel inhibitors before laser ablation also inhibited the microglial response toward the laser ablation site. Interestingly, baseline motility of microglial processes in the intact brain seems to be modulated by the same ATP signaling mechanisms that mediate injury-induced responses, because apyrase and connexin channel inhibitors also significantly slowed microglial baseline dynamics.

Kalyani Narasimhan

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


R E V I E W

Notch signaling in the mammalian central nervous system: insights from mouse mutantsKeejung Yoon1,2 & Nicholas Gaiano1–4

The Notch pathway, although originally identified in fruit flies, is now among the most heavily studied in mammalian biology. In mice, loss-of-function and gain-of-function work has demonstrated that Notch signaling is essential both during development and in the adult in a multitude of tissues. Prominent among these is the CNS, where Notch has been implicated in processes ranging from neural stem cell regulation to learning and memory. Here we review the role of Notch in the mammalian CNS by focusing specifically on mutations generated in mice. These mutations have provided critical insight into Notch function in the CNS and have led to the identification of promising new directions that are likely to generate important discoveries in the future.

The role of the Notch pathway during animal neural development is best understood in fruit flies, where Notch has been shown to inhibit differentiation by lateral signaling and regulate cell fate through induc-tive interactions1,2. Dissection of these processes in flies has relied heav-ily on loss-of-function genetics, and tremendous progress has been made in understanding both their molecular mechanisms and bio-logical functions. Indeed, to date, the vast majority of known Notch pathway members and molecular interactions were first identified using fly genetics. Similarly, our core understanding of pathway function on a cellular level has emerged almost exclusively from work in flies. In recent years there has been extensive interest in extending our under-standing of the Notch pathway from flies to mammals.

The cloning of Notch pathway members in mice has made it possible to use loss-of-function analyses to examine the role of Notch during mammalian neural development. Over the past decade, many laborato-ries have examined mouse mutants for members of the Notch pathway, including receptors3–11, ligands12–20, modulators21–34 and effectors35–39. In general, these disruptions have resulted in an increase in neuronal differentiation markers and a decrease in progenitor markers, leading to the prevailing view that Notch maintains a progenitor state. While this view of Notch function is consistent with the fly work, it is more limited in scope, suggesting that functions for Notch in the mammalian nervous system remain to be elucidated.

In recent years, our understanding of the Notch signaling cascade and its role in the mammalian CNS have grown more complex. With respect to signaling, for example, although the modulators Numb (encoded by

Numb, also known as Nmb) and Numblike (encoded by Numbl, also known as Nbl) have largely been described as negative regulators of Notch in flies, loss-of-function studies in the mouse have challenged this view21,23,24. In addition, it has become clear that although CBF1 (also called RBP-J or CSL) and the Hes genes40,41 are critical effectors of the pathway, other previously unappreciated effectors are likely to exist42. With respect to pathway function, recent studies have identified previously unknown roles for Notch during glial fate specification2,43, neuronal maturation44–46 and even learning and memory33,47–49. Thus, Notch signaling in the mammalian CNS, already an area of widespread interest, is growing into an even larger topic.

Here, we give a brief overview of the Notch signaling cascade and then review the pathway, with a specific emphasis on mutants gener-ated in the mouse. Loss-of-function analyses are critical for identifying unique and essential functions, and, fortunately, mouse mutants exist for the majority of key Notch pathway components. Although pleio-tropy and redundancy have complicated the analyses of these mutants, the versatility of conditional knockouts has created new opportunity for progress. All told, mouse mutants are an invaluable resource for investigating the role of Notch, both in the CNS and in other tissues.

Overview of the Notch pathwayThe Notch signaling pathway (Fig. 1) is best characterized as mediating cell-cell signaling between adjacent cells1,50. Both the ligands, members of the Delta and Jagged gene families, and the receptors, of which there are four in mammals, are single-pass transmembrane proteins. Upon ligand binding, the intracellular domain of Notch (NICD) is released from the plasma membrane and translocates into the nucleus, where it converts the CBF1 repressor complex into an activator complex. The NICD/CBF1 activator complex, which includes the co-activator Mastermind51, among other proteins52, upregulates targets such as the Hes and Herp (Hes-related protein) genes40,41, which are basic helix-loop-helix (bHLH) transcriptional regulators that antagonize proneural

1Institute for Cell Engineering and Departments of 2Neurology, 3Neuroscience

and 4Oncology, Johns Hopkins University School of Medicine, Baltimore,

Maryland 21205, USA. Correspondence should be addressed to N.G.

([email protected]).

Published online 25 May 2005; doi:10.1038/nn1475

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


R E V I E W

genes like Mash1 (also known as Ascl1) and the neurogenins53. This antagonism blocks early neuronal gene expression and is central to the inhibition of neuronal differentiation by Notch.

In addition to the primary components of the linear Delta/Jagged-Notch-CBF1-Hes/Herp signaling cascade, many other genes play a role in modulating Notch signaling. For example, Notch proteins undergo three proteolytic processing events, the last of which is a ligand-depen-dent intramembraneous cleavage mediated by the presenilin pro-teins (PS1 and PS2; ref. 50). Other pathway components include the phosphotyrosine binding (PTB) domain–containing proteins Numb and Numblike54,55, E3 ubiquitin ligases such as Itch (also known as Suppressor of Deltex (Dx), encoded by Su(dx)) and Fbw7 (also known as Sel-10; refs. 56,57), and glycosyltransferases of the Fringe family58.Although the role of these modulators is only beginning to be under-stood, it is clear that Notch pathway regulation is far more complex than initially had been appreciated.

Beyond the mechanism of signal transduction, another fundamental challenge is identifying the Notch targets relevant in specific contexts. On the basis of homology to the fly gene Suppressor of Hairless (Su(H)), CBF1 was identified as the primary transcriptional effector of Notch in vertebrates59,60. Consistent with such a role, the deletion of Cbf1 (also known as Rbpsuh) has a phenotype similar to the Notch1 knockout (see below) but more severe, as would be expected for an effector that medi-ates signaling through all Notch receptors35,61. Similarly, vertebrate CBF1

targets were identified by screening for genes orthologous to Su(H) targets of the Enhancer of split (E(spl)) complex. Such efforts led to isolation of the Hes genes (related to hairy and E(spl)). Although the identification of Cbf1 and the Hes genes has been critical to mammalian Notch studies, there has been a tendency in the field to focus almost exclusively on these genes when trying to understand pathway output. It is becoming increasingly clear, however, that Notch signaling is mediated by CBF1 targets other than the Hes and Herp genes40,41, and that some aspects of Notch signaling are likely to be CBF1-independent42.

Lessons from the mutantsA conservative estimate of the number of genes in the Notch pathway suggests that there are more than two dozen in mammals, many of which have been mutated in mice (Table 1). Rather than attempting to review them all, we have chosen to focus largely on mutations gen-erated in the core pathway components. The CNS phenotypes of these mutations have been the most carefully studied to date, and they remain an ongoing resource for understanding fundamental aspects of Notch function.

Notch receptor mutationsOne of the first Notch pathway genes to be disrupted by homologous recombination was Notch1, which was mutated by two groups independently3,4. Both studies found that mutant embryos died during early embryo-genesis (around embryonic day 11 (E11)), although little detail was initially provided regarding neural phenotypes. Subsequently,

a detailed analysis of neural development in Notch1–/– mutants was reported61. That study examined the expression both of pathway components such as Hes1, Hes5 and Delta-like 1 (Dll1) and of early differentiation markers such as Math4A (also known as Neurog2), NeuroD (also known as Neurod2) and NSCL-1 (also known as Nhlh1). Consistent with the view that Notch activity is needed for progenitor maintenance, the differentiation markers were found to be upregu-lated in mutants. An interesting though often overlooked finding was that although Hes5 expression was found by northern blot analysis to be reduced in Notch1–/– mutants, Hes1 expression did not seem to be affected. This latter result is puzzling in light of the Hes1–/– mutant phenotype and the extensive literature supporting the notion that Hes1 is a primary Notch/CBF1 target38,40,62,63. Similar results were obtained with CBF1–/– mutants61, suggesting that while Hes1 may well be a bona fide Notch/CBF1 target, it is also likely to be regulated by other signal-ing cascades. This notion is supported by previous findings that Hes1 can be upregulated in PC12 cells cultured in the growth factors NGF, FGF2 or EGF64 and in postnatal cerebellar granule cells cultured in Sonic hedgehodge65.

After the original Notch1 deletion studies, alleles of Notch2 (refs. 5,6), Notch3 (ref. 7) and Notch4 (ref. 8) were also generated, as were floxed alleles of Notch1 (refs. 11,66). Although Notch3 and Notch4 do not seem to have significant phenotypes when deleted, Notch2–/– mutants, similar to Notch1–/– mutants, die around E11 (ref. 5). However, in contrast to

Figure 1 The Notch pathway in mammals. Notch signaling between adjacent cells is a mechanism to generate cellular heterogeneity. Ligand-receptor interactions lead to γ-secretase cleavage of Notch and release of the intracellular domain (NICD). This domain enters the nucleus and together with a complex including CBF1 and Mastermind (MAML) promotes the transcription of targets such as the Hes and Herp genes. These genes encode bHLH proteins that antagonize proneural genes such as Mash1 and the neurogenins. This antagonism blocks neuronal gene expression and consequently inhibits differentiation. Notch signaling may also act through other CBF1 targets and through CBF1-independent cascades, which involve Deltex (Dx) and pathways yet to be identified. The Numb and Numblike (Nbl) proteins have generally been considered negative regulators of Notch. However, recent studies of mice with mutations in these genes have suggested that they may be essential for neural progenitor maintenance and may or may not interact with Notch.

Pen-2

MAML

Notch

γ-secretase

NICD

NICD

PS1/2

NctAph-1

Delta-like/Jagged

Numb/Nbl

Fbw7/Sel-10

Hes/Herp

CBP, ErbB2, others?

Progenitor/stem cellmaintenance

Lgl1

Deltex

?

?

CBF1

Dll1, NeuroD,...

Neuronaldifferentiation

Differentiating neuron Progenitor/stem cell

Proneural genes(e.g., Mash1,neurogenins)

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


R E V I E W

Notch1–/– mutants, Notch2–/– mutants do not display defects in somito-genesis, and they do not show alterations in Hes5 expression in the CNS. Notch2–/– mutants do undergo widespread cell death in the CNS start-ing around E9, but it is unclear whether this phenotype reflects a role for Notch2 in the developing CNS or if it is the indirect consequence of other embryonic perturbations.

To circumvent the early lethality of Notch1 deletion, several studies have addressed the effect of deleting this receptor in specific brain structures. In one case, Cre-loxP–mediated recombination was used to delete Notch1

from the medial cerebellar anlage66. Consistent with the traditional model of Notch function in the nervous system, the authors found that Notch1 deletion resulted in upregulation of proneural genes (such as Mash1 and Math1) and precocious neuronal differentiation. More recently, conditional deletion of Notch1 in the neural progenitor pool (using a floxed allele and the nestin-Cre driver) was also found to result in precocious neuronal differentiation11. Interestingly, deletion of Notch1 in the telen-cephalon (using the foxg1-Cre driver) led to reduced neuronal numbers in vivo later in devel-opment, most likely resulting from precocious neuronal differentiation and earlier progenitor pool depletion9. In support of this contention, the telencephalic deletion of Notch1 led to a reduction in progenitor frequency (assayed as neurospheres) in vitro9. This result is consistent with reduced neurosphere frequencies observed after widespread deletion of Notch1, Cbf1, PS1 and PS2 (ref. 10), or Hes1 and Hes5 (ref. 37). In summary, the conditional deletions of Notch1 support the canonical view that Notch signaling inhibits neuronal differentiation and maintains the neural progenitor pool.

Ligand mutationsIn addition to the receptor mutations, many Notch ligand mutations have been examined in mice. These have mostly been described with respect to their functions during somito-genesis, limb patterning and vascular develop-ment12–20. However, a few studies have exam-ined the effects of deleting Delta-like 1 (Dll1) on neural development19,20. One such study found that Dll1–/– mutant embryos had decreased Hes5 expression, consistent with the expected reduction in Notch activation19. In addition, the study found that Dll1–/– mutants showed a decrease in the radial progenitor marker RC2 and an increase in neuronal mark-ers such as βIII-tubulin and GABA. These find-ings support the traditional view that Notch signaling inhibits neuronal differentiation in the developing CNS. Interestingly, however, on the basis of comparisons of the Dll1–/–, Mash1–/– and other mutants, the authors sug-gest that Notch signaling might also regulate the diversification of the progenitor pool into distinct progenitor subtypes. This function would precede the role of Notch in inhibiting

the differentiation of mature cell types (that is, neurons and oligoden-drocytes) and could convert a homogeneous proliferative pool into a heterogeneous mixture of stem cells, neuroblasts and glioblasts.

Further evidence that Notch signaling may generate progenitor diver-sity was obtained by in vitro analysis of Dll1–/– mutants20. This work suggested that Notch signaling first specifies glial progenitors and then functions in those cells to promote astrocyte versus oligodendrocyte fate. Both this study and the work described above indicate that in mice, Notch influences multiple choice points in the neural progenitor lineage.

Table 1 Brief description of CNS phenotypes of Notch pathway mutants listed together with references.

Gene(s) Ref. CNS phenotype(s)

61 Precocious neuronal differentiation. Hes5↓, Mash1↑, NeuroD↑.

66 Precocious neuronal differentiation. Hes5↓, Mash1↑, Dll1↑.

10 Neurosphere frequency ↓ at E10.5.Notch1

49 Spatial learning and memory deficits in heterozygotes.

11 Neural progenitor deletion: Precocious neuronal differentiation.

9 Telencephalic deletion: neurosphere frequency ↓ at E12.5.

Notch2 5 Widespread neural cell death at E10.

25 Impaired neurogenesis at E14.5, ‘massive neuronal loss’ at E16.5.

27 Loss of Cajal-Retzius cells and cortical dysplasia.

PS1 26 Precocious neuronal differentiation. Hes5↓, Dll1↑.

47 Knockout in postnatal cortex. Spatial memory deficits. No change in Hes1/5, Dll1.

10 Neurosphere frequency ↓ at E14.5.

79 Neural tube disorganization. Hes5 ‘undetectable,’ Dll1↑.

PS1, PS2 10 No neurospheres formed at E14.5.

33 Memory and synaptic plasticity deficits. Neurodegeneration. CBP↓

Adam10 28 Hes5↓ in neural tube, Dll1↑.

19 Hes5↓, preferential production of early born neurons in telencephalon.Dll1

20 ↑ Neurons, ↓ glial cells in differentiated neurosphere cultures

61 Precocious neuronal differentiation. Hes5↓, Dll↑, Mash1↑, NeuroD↑, etc.

Cbf1 10 No neurospheres formed at E8.5.

49 Spatial learning and memory deficits in heterozygotes.

Hes1 38 Precocious neuronal differentiation. Hes5↑.

Hes5 67 30-40% reduction in Müller glial cells in retina.

36 Constitutively active Notch1 unable to inhibit neuronal differentiation.Hes1, Hes5

37 Neurosphere frequency ↓ at E11.5.

Hes1, Hes3,Hes5

68 Severe precocious neuronal differentiation.

21 Precocious neuronal differentiation. MAP2↑, neurofilament↑.

Numb 22 ↓ Neuronal differentiation in hindbrain and spinal cord. No change in Hes1, Hes5.

31 Deficits in granule cell maturation. Reduced Purkinje cell number.

23 Precocious neuronal differentiation along neuraxis.

30 Cortical disorganization and hyperproliferation. Hes1↑, Hes5↑Numb, Numbl

24 Precocious neuronal differentiation. Hes5↓, Mushashi↓ 32 Defects in axonal arborization. Reduced length and branch points.

Lgl1 73 Cortical disorganization and hyperproliferation. Nestin↑, βIII-tubulin↓

Herp1, Herp2 70 Reduced neural tube thickness.

Fbw7 75 Impaired neural tube closure.

Studies describing non-neural aspects of pathway mutants are not listed.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


R E V I E W

While this idea is consistent with what is known about Notch function during fruit fly neural development, it has not received much attention in mammalian studies. However, on the basis of the spatial and temporal concurrence of Notch signaling and progenitor pool diversification in the developing mammalian CNS, it would be surprising if Notch did not influence this process.

Effector mutationsAs discussed above, the Notch signaling cascade is primarily transduced through the transcriptional regulator CBF1, when nuclear translocation of NICD converts CBF1 from a repressor to an activator. Consistent with the Notch1–/– mutant phenotype, Cbf1–/– mutants show altered gene expression suggestive of widespread precocious neuronal differen-tiation (such as decreased Hes5 and increased Dll1 and NeuroD)61. The interpretation of this phenotype is confounded by the fact that these mutants show severe growth retardation by E8.5, a time before neural tube closure. In light of this limitation, and because Cbf1 appears to be a non-redundant bottleneck in the Notch cascade, conditional deletions of Cbf1 in the CNS are likely to be highly informative. Of course, Cbf1 deletion will not uncover purported Cbf1-independent Notch signaling, an aspect of the pathway that will remain difficult to address until the relevant molecular mechanisms are more clearly elucidated.

The most widely accepted Notch/CBF1 targets are the Hes and recently identified Herp gene families40,41. Although there are seven Hes genes, not all are clear Notch targets, and studies in the mam-malian CNS have focused on Hes1 and Hes5. Hes1–/– mutant embryos show severe defects in neural development, including lack of cranial neural tube closure and eventual anencephaly38. However, because these animals die perinatally, it is possible to examine alterations in gene expression much later into development than with Notch1–/– or Cbf1–/– mutants. Consistent with the canonical model, precocious neurogenesis in Hes1–/– mutants was suggested by early expression of markers like Mash1, Nscl and neurofilament.

Based upon both loss-of-function and gain-of-function studies, it seems likely that to some extent Hes1 and Hes5 serve redundant functions in the neural progenitor pool. First, Hes1–/– Hes5–/– double mutants show a far more severe phenotype than Hes1–/– and Hes5–/– sin-gle mutant phenotypes combined36. Second, elevated Hes5 expression was detected in Hes1–/– mutants, suggesting the existence of compensa-tory mechanisms between these Notch targets38. Third, although some precocious neuronal differentiation was evident in Hes5–/– mutants, these animals were largely normal, suggesting that Hes1 is capable of almost completely compensating for lack of Hes5 function. Finally, although a constitutively active form of Notch1 could inhibit neuronal differentiation in either Hes1–/– or Hes5–/– mutant cells, it could not do so in Hes1–/– Hes5–/– double mutants36. As an aside, it is interesting to note that Hes5–/– mutants exhibit a 30–40% decrease in Müller glial cell number67. This loss-of-function data directly supports a role for Notch signaling in promoting glial fate2.

The inability of activated Notch1 to inhibit neuronal differentia-tion in Hes1–/–Hes5–/– double mutants raises the question of whether Hes1 and Hes5 are the only relevant Notch/CBF1 targets in the CNS. This seems unlikely for several reasons. First, recent work has shown that Hes1–/–Hes3–/–Hes5–/– triple mutants show even more precocious neuronal differentiation than Hes1–/–Hes5–/– double mutants68. Second, Herp1 and Herp2 (also called Hey2 and Hey1, respectively), which can form heterodimers with the Hes proteins41, are expressed in the embry-onic neural progenitor pool and could mediate an essential function for Notch signal transduction in that region69. Herp1–/–Herp2–/– dou-ble mutants have recently been described, and although the analysis focused on vascular defects, the authors noted that the neural tube was

substantially thinner in mutants70. Third, several reports have identi-fied ErbB2 as a Notch target that has a role during mammalian neural progenitor maintenance71,72. This finding is intriguing and suggests that other yet-to-be-identified non-canonical targets may exist. Finally, although it is clear that Hes1 and Hes5 are required effectors of Notch during neural cell-fate specification, it is unclear to what extent these genes and/or other targets are needed for later roles of Notch in the CNS. For example, recent data suggests that CREB binding protein (CBP, also known as Crebbp) may be an important Notch/CBF1 target in the postnatal brain (see below)33.

Signaling modulators and the Numb conundrumThe Notch/CBF1 signaling cascade involves a large number of pro-teins that transduce and/or modulate the signal from the cell surface to the nucleus (Fig. 1). These include PS1 and PS2 (ref. 50), nicastrin50, Numb and Numblike55, Lgl1 (ref. 73), Dx74, Itch57, Fbw7 (ref. 75) and Mastermind51, among many others (Fig. 1). As mentioned above, PS1 and PS2 mediate ligand-dependent Notch receptor processing as part of the γ-secretase complex, which also includes nicastrin, Aph1, and Pen-2 (ref. 76). Numb and Numblike are PTB domain–containing proteins generally thought to negatively regulate Notch signaling, possibly by promoting receptor turnover. Numb is of particular interest because its asymmetric localization in dividing neural progenitors, which is depen-dent upon Lgl1, may have a causal role during cell-fate specification in the developing CNS77. Itch and Fbw7 are among a growing family of E3 ubiquitin ligases involved in trafficking and/or turnover of Notch pathway components56,57. Interestingly, although both Itch and Fbw7 promote Notch receptor turnover, the former acts at the plasma mem-brane and the latter acts in the nucleus. Dx also seems to encode an E3 ligase and influence Notch protein localization78. It is worth noting that although the mechanism of Dx action remains unclear, this pathway component stands out as a potential mediator of CBF1-independent Notch signaling42,78.

While there are loss-of-function alleles in mice for many of the genes listed above, PS1, PS2, Numb and Numbl have had the most extensive CNS phenotypes reported. The PS1–/– PS2–/– mutant phenotype dur-ing neural development both in vivo26,79 and in vitro10 is similar to that found during disruption of other positive regulators of Notch and supports a role for the pathway in neural progenitor maintenance. In contrast, the Numb–/–Numbl–/– mutant phenotypes remain the most puzzling aspect of the Notch loss-of-function literature in the mouse CNS, and we will focus on these phenotypes below.

Genetic analysis in flies has clearly demonstrated that Numb can antagonize Notch signaling54. However, although some studies have suggested that mammalian Numb and Numblike can antagonize Notch45,80,81, the mouse mutant analyses do not all support such a function. Two different loss-of-function alleles of Numb were gener-ated by independent groups21,22. In one case, the first coding exon was deleted22, whereas in the other, exons 5 and 6 were deleted to disrupt the PTB domain, which is essential for function in flies21. Unfortunately, both alleles might produce truncated protein products (the exon 1 dele-tion may initiate translation further downstream, whereas the exon 5,6 deletion can be spliced over to generate a mutant protein lacking exactly 72 residues).

The initial characterization of each of the Numb alleles reached dif-ferent conclusions. In one case (exon 1 deletion), the authors concluded that Numb disruption resulted in impaired neuronal differentiation in certain regions of the CNS (such as hindbrain, spinal cord, dorsal root ganglia), but not in others (for example, the forebrain)22. Aspects of their data are consistent with the general view that Numb antago-nizes Notch. The authors highlighted this fact, but they also discussed

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


R E V I E W

the possibility that Numb may act in a Notch-independent manner, as they found no altera-tions in Hes1 or Hes5 expression.

In contrast, a second Numb deletion study (exon 5,6 deletion) found precocious neuronal differentiation in the forebrain, as evidenced by expression of MAP2 and neurofilament21. The authors went on to suggest that these data, together with the apical localization of Numb protein in neocortical progenitors82, indicated that Numb was needed to maintain a progeni-tor state. This result was particularly intriguing because it suggested that if Numb had a role in the Notch signaling cascade, that role was as a positive rather than a negative regulator. It remains unclear how to reconcile these two Numb studies until a more detailed character-ization of the different alleles is obtained. As dis-cussed below, more recent work has suggested that understanding Numb function in the developing CNS poses an ongoing challenge.

To address the possible redundancy between Numb and Numbl, several recent studies have examined the effect of Cre-loxP mediated deletion of Numb in a Numbl–/– mutant back-ground23,24,30 (Numbl–/– mutants are appar-ently normal24). These studies used three different Cre drivers to delete Numb: nestin-Cre23, Emx1-Cre30 and D6-Cre24, which are reported to delete Numb starting at E8.5, E9.5 and E10.5, respectively. Remarkably, although these studies used the same Numb and Numbl alleles, they reached dra-matically different conclusions. One group, using either nestin-Cre or D6-Cre to delete Numb–/– in Numbl–/– mutants, found that the laminar organization of the neocortex was retained, but that there was wide-spread precocious neuronal differentiation23,24. The other group used Emx1-Cre and instead found that the laminar neocortical organization was replaced by rosette-like structures, and that the progenitor pool was expanded at the expense of neuronal differentiation30. This study also found widespread upregulation of the Notch targets Hes1 and Hes5.

One possible explanation of the differences obtained using different Cre drivers is that depending upon the timing of the deletion, different phenotypes might be obtained. However, this explanation is unlikely because deletion at E8.5 and E10.5 had essentially the same phenotype, whereas deletion at E9.5 seemed opposite. Furthermore, Cre-mediated deletions are not instantaneous, and there is likely to be substantial overlap between the timing of all three deletions.

An alternative, more plausible explanation is that something specific to the individual Cre drivers is influencing the experimental outcome. Indeed, one group speculated24 that if the Emx1-Cre deletion were inef-ficient, that study might be characterizing the effects of mosaic and/or hypomorphic Numb function. It is not clear why that would result in the observed phenotype, although it may relate to secondary effects caused by severe tissue disorganization. It is worth noting that a mutation in Lgl1, a gene essential for Numb subcellular localization, has a phenotype remarkably similar to the Emx1-Cre mediated Numb deletion, includ-ing hyperproliferation, tissue disorganization and rosetting73. Because Numb protein is not properly localized in Lgl1–/– mutants, the Lgl1–/– mutant phenotype might mimic a Numb hypomorphic phenotype.

On the basis of the mutant analyses, it seems clear that the traditional view of Numb and Notch interaction is oversimplified in the mammalian

CNS. Many issues remain to be addressed, including the function of the four different Numb isoforms83 and in vitro studies suggesting that Numb or Numblike can indeed antagonize Notch45,80,81. In vivo it remains pos-sible that Numb does not interact with Notch to the extent previously assumed but instead functions in parallel toward the same end. It is also possible that the long-standing view of Notch function in maintaining the progenitor pool is what has been oversimplified and is in need of revision. For example, as described above, before influencing the decision to differentiate into neurons or glia, Notch signaling may promote the generation of distinct progenitor subtypes. It is certainly possible that Numb and/or Numblike inhibit Notch during this process, thereby influ-encing the balance of progenitor subtypes. In any event, ongoing studies of Numb and Numblike will likely precipitate substantial changes in our view of neural progenitor regulation in the mammalian CNS.

Notch beyond cell-fate specificationLiterature is lacking on the effects of Notch loss-of-function on post-mitotic neuronal development and function. Until recently, there was little evidence that Notch signaling played any role in the CNS beyond cell-fate specification. Then several years ago, numerous groups showed that Notch could influence neurite development in vitro44–46. These studies found that Notch activation reduced neurite extension, but pre-sumed signaling blockade (via expression of Numb, Numb or Dx) could promote neurite extension. Subsequent studies have found that Numb deletion disrupts neuronal maturation in the developing cerebellum31, whereas deletion of Numb and Numb disrupts axonal arborization in sensory ganglia in vivo32. The mechanisms behind these phenomena are unclear, although the latter study indicated that, similar to what has been shown recently in fruit flies84, Numb regulates the endocytic trafficking of Notch receptors to promote their degradation. Consistent

Notch Notch

Notch

Notch

MultipotentCNS progenitor

GlioblastAstrocytes

Oligodendrocytes

Adult neural stem cell

Adult neurogenesis

Early neuronal differentiation

Neuronal maturationand function

Neuroblast

Notch

Notch

Notch

Off

Off

Off

Off

Notch

Figure 2 Analyses of mouse mutants have supported many roles for Notch signaling in the developing and adult CNS. Processes that are likely to involve pathway signaling are labeled ‘Notch’ (green), and those that are likely to require downregulation of Notch signaling are labeled ‘Off’ (red). Specific details are provided in the text.

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


R E V I E W

with a function in axonal development, Notch is present in the growth cones of extending axons in fruit flies and has been shown to interact with the axonal tyrosine kinase Abl85.

Beyond the developmental work described above, several recent stud-ies have sought to address the role of Notch signaling in the adult brain in vivo. A gene expression study found that Notch pathway components are expressed in the early postnatal and adult mouse brain, both in germinal zones and in neurons86. Subsequently, several groups have used loss-of-function approaches to examine the role of Notch in the adult brain. One such study found that mice heterozygous for muta-tions in either Notch1 or Cbf1 showed deficits in spatial learning and memory49. While this finding was intriguing, the authors could not rule out the possibility that subtle developmental defects contributed to the observed phenotype. Furthermore, even if there were no such defects, the study did not determine whether the deficits resulted from loss of Notch1/CBF1 signaling in neurons or in postnatal germinal zones such as the subgranular zone (SGZ) of the dentate gyrus. Consistent with a role for Notch in learning and memory, a transgenic antisense strategy which ‘knocked down’ Notch1 levels ~50% found deficits in hippocampal long-term potentiation (LTP)48.

An alternative approach that ostensibly addressed the role of Notch signaling in the adult brain was the postnatal deletion of presenilins 1 and 2 (PS1–/– PS2–/–)33. Because proteolytic processing of the Notch receptors by PS1 and PS2 is essential for Notch signaling, deletion of PS1 and PS2 is an effective means to disrupt the pathway. In a PS2 mutant background, the authors used the αCaMKII-Cre driver to delete PS1 in excitatory neurons beginning 3 weeks after birth. Remarkably, PS1–/– PS2–/– mutant animals showed learning and memory deficits and neuronal dysfunction and underwent gradual neurodegeneration. A strength of this work is that the deletion occured after development was essentially complete. However, the presenilin proteins have numerous substrates50,87, and the Notch pathway itself was not definitively addressed. That said, the authors did identify an optimal CBF1 binding site in the promoter region of CBP, a gene known to function during learning and memory. Since CBP expression was reduced in PS1–/– PS2–/– mutants, this binding site is consistent with a role for Notch in the observed phenotype. This study and those outlined above strongly suggest that Notch functions in the mammalian CNS beyond its role during cell-fate specification (Fig. 2).

Conclusions Although much remains to be learned about Notch function in the mammalian CNS, mouse mutations have been an essential resource. With respect to the vertebrate nervous system, the Notch field has been dominated by studies of cell-fate specification, where the general view is that Notch maintains the neural progenitor state and inhibits dif-ferentiation. Disruption of positive regulators of the pathway has been grossly consistent with this view. However, more detailed analyses have also revealed that Notch is likely to regulate progenitor pool diversifica-tion and neuronal maturation. In addition, emerging data suggests that Notch signaling has a role in neuronal function in the adult brain. It will be of great interest to determine which components of the Notch signal-ing cascade, known or unknown, function in each of these processes. In addition, understanding the perplexing relationship between Notch and Numb function in the mammalian CNS remains a pressing question.

ACKNOWLEDGMENTSThe authors would like to thank M. Starz-Gaiano and H. Mason for critical reading of the manuscript. N.G. is supported by grants from the Burroughs Wellcome fund, the Sidney Kimmel Foundation for Cancer Research and the National Institute of Neurological Disorders and Stroke (NINDS) of the US National Institutes of Health.

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

Published online at http://www.nature.com/natureneuroscience/

1. Artavanis-Tsakonas, S., Rand, M.D. & Lake, R.J. Notch signaling: cell fate control and signal integration in development. Science 284, 770–776 (1999).

2. Gaiano, N. & Fishell, G. The role of notch in promoting glial and neural stem cell fates. Annu. Rev. Neurosci. 25, 471–490 (2002).

3. Swiatek, P.J., Lindsell, C.E., del Amo, F.F., Weinmaster, G. & Gridley, T. Notch1 is essential for postimplantation development in mice. Genes Dev. 8, 707–719 (1994).

4. Conlon, R.A., Reaume, A.G. & Rossant, J. Notch1 is required for the coordinate segmentation of somites. Development 121, 1533–1545 (1995).

5. Hamada, Y. et al. Mutation in ankyrin repeats of the mouse Notch2 gene induces early embryonic lethality. Development 126, 3415–3424 (1999).

6. McCright, B. et al. Defects in development of the kidney, heart and eye vasculature in mice homozygous for a hypomorphic Notch2 mutation. Development 128, 491–502 (2001).

7. Krebs, L.T. et al. Characterization of Notch3-deficient mice: normal embryonic devel-opment and absence of genetic interactions with a Notch1 mutation. Genesis 37, 139–143 (2003).

8. Krebs, L.T. et al. Notch signaling is essential for vascular morphogenesis in mice. Genes Dev. 14, 1343–1352 (2000).

9. Yoon, K. et al. Fibroblast growth factor receptor signaling promotes radial glial identity and interacts with Notch1 signaling in telencephalic progenitors. J. Neurosci. 24, 9497–9506 (2004).

10. Hitoshi, S. et al. Notch pathway molecules are essential for the maintenance, but not the generation, of mammalian neural stem cells. Genes Dev. 16, 846–858 (2002).

11. Yang, X. et al. Notch activation induces apoptosis in neural progenitor cells through a p53-dependent pathway. Dev. Biol. 269, 81–94 (2004).

12. Hrabe de Angelis, M., McIntyre, J., II & Gossler, A. Maintenance of somite borders in mice requires the Delta homologue DII1. Nature 386, 717–721 (1997).

13. Kusumi, K. et al. Dll3 pudgy mutation differentially disrupts dynamic expression of somite genes. Genesis 39, 115–121 (2004).

14. Duarte, A. et al. Dosage-sensitive requirement for mouse Dll4 in artery development. Genes Dev. 18, 2474–2478 (2004).

15. Bulman, M.P. et al. Mutations in the human delta homologue, DLL3, cause axial skeletal defects in spondylocostal dysostosis. Nat. Genet. 24, 438–441 (2000).

16. Jiang, R. et al. Defects in limb, craniofacial, and thymic development in Jagged2 mutant mice. Genes Dev. 12, 1046–1057 (1998).

17. Sidow, A. et al. Serrate2 is disrupted in the mouse limb-development mutant syndac-tylism. Nature 389, 722–725 (1997).

18. Xue, Y. et al. Embryonic lethality and vascular defects in mice lacking the Notch ligand Jagged1. Hum. Mol. Genet. 8, 723–730 (1999).

19. Yun, K. et al. Modulation of the notch signaling by Mash1 and Dlx1/2 regulates sequential specification and differentiation of progenitor cell types in the subcortical telencephalon. Development 129, 5029–5040 (2002).

20. Grandbarbe, L. et al. Delta-Notch signaling controls the generation of neurons/glia from neural stem cells in a stepwise process. Development 130, 1391–1402 (2003).

21. Zhong, W. et al. Mouse numb is an essential gene involved in cortical neurogenesis. Proc. Natl. Acad. Sci. USA 97, 6844–6849 (2000).

22. Zilian, O. et al. Multiple roles of mouse Numb in tuning developmental cell fates. Curr. Biol. 11, 494–501 (2001).

23. Petersen, P.H., Zou, K., Hwang, J.K., Jan, Y.N. & Zhong, W. Progenitor cell mainte-nance requires numb and numblike during mouse neurogenesis. Nature 419, 929–934 (2002).

24. Petersen, P.H., Zou, K., Krauss, S. & Zhong, W. Continuing role for mouse Numb and Numbl in maintaining progenitor cells during cortical neurogenesis. Nat. Neurosci. 7, 803–811 (2004).

25. Shen, J. et al. Skeletal and CNS defects in Presenilin-1-deficient mice. Cell 89, 629–639 (1997).

26. Handler, M., Yang, X. & Shen, J. Presenilin-1 regulates neuronal differentiation during neurogenesis. Development 127, 2593–2606 (2000).

27. Hartmann, D., De Strooper, B. & Saftig, P. Presenilin-1 deficiency leads to loss of Cajal-Retzius neurons and cortical dysplasia similar to human type 2 lissencephaly. Curr. Biol. 9, 719–727 (1999).

28. Hartmann, D. et al. The disintegrin/metalloprotease ADAM 10 is essential for Notch signalling but not for alpha-secretase activity in fibroblasts. Hum. Mol. Genet. 11, 2615–2624 (2002).

29. Hicks, C. et al. Fringe differentially modulates Jagged1 and Delta1 signalling through Notch1 and Notch2. Nat. Cell Biol. 2, 515–520 (2000).

30. Li, H.S. et al. Inactivation of Numb and Numblike in embryonic dorsal forebrain impairs neurogenesis and disrupts cortical morphogenesis. Neuron 40, 1105–1118 (2003).

31. Klein, A.L., Zilian, O., Suter, U. & Taylor, V. Murine numb regulates granule cell maturation in the cerebellum. Dev. Biol. 266, 161–177 (2004).

32. Huang, E.J. et al. Targeted deletion of numb and numblike in sensory neurons reveals their essential functions in axon arborization. Genes Dev. 19, 138–151 (2005).

33. Saura, C.A. et al. Loss of presenilin function causes impairments of memory and synaptic plasticity followed by age-dependent neurodegeneration. Neuron 42, 23–36 (2004).

34. Zhang, N., Norton, C.R. & Gridley, T. Segmentation defects of Notch pathway mutants and absence of a synergistic phenotype in lunatic fringe/radical fringe double mutant mice. Genesis 33, 21–28 (2002).

35. Oka, C. et al. Disruption of the mouse RBP-J kappa gene results in early embryonic death. Development 121, 3291–3301 (1995).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce


R E V I E W

36. Ohtsuka, T. et al. Hes1 and Hes5 as notch effectors in mammalian neuronal differ-entiation. EMBO J. 18, 2196–2207 (1999).

37. Ohtsuka, T., Sakamoto, M., Guillemot, F. & Kageyama, R. Roles of the basic helix-loop-helix genes Hes1 and Hes5 in expansion of neural stem cells of the developing brain. J. Biol. Chem. 276, 30467–30474 (2001).

38. Ishibashi, M. et al. Targeted disruption of mammalian hairy and Enhancer of split homolog-1 (HES-1) leads to up-regulation of neural helix-loop-helix factors, pre-mature neurogenesis, and severe neural tube defects. Genes Dev. 9, 3136–3148 (1995).

39. Zine, A. et al. Hes1 and Hes5 activities are required for the normal development of the hair cells in the mammalian inner ear. J. Neurosci. 21, 4712–4720 (2001).

40. Kageyama, R. & Ohtsuka, T. The Notch-Hes pathway in mammalian neural develop-ment. Cell Res. 9, 179–188 (1999).

41. Iso, T., Kedes, L. & Hamamori, Y. HES and HERP families: multiple effectors of the Notch signaling pathway. J. Cell. Physiol. 194, 237–255 (2003).

42. Martinez Arias, A., Zecchini, V. & Brennan, K. CSL-independent Notch signalling: a checkpoint in cell fate decisions during development? Curr. Opin. Genet. Dev. 12, 524–533 (2002).

43. Wang, S. & Barres, B.A. Up a notch: instructing gliogenesis. Neuron 27, 197–200 (2000).

44. Redmond, L., Oh, S.R., Hicks, C., Weinmaster, G. & Ghosh, A. Nuclear Notch1 signaling and the regulation of dendritic development. Nat. Neurosci. 3, 30–40 (2000).

45. Sestan, N., Artavanis-Tsakonas, S. & Rakic, P. Contact-dependent inhibition of cortical neurite growth mediated by notch signaling. Science 286, 741–746 (1999).

46. Franklin, J.L. et al. Autonomous and non-autonomous regulation of mammalian neurite development by Notch1 and Delta1. Curr. Biol. 9, 1448–1457 (1999).

47. Yu, H. et al. APP processing and synaptic plasticity in presenilin-1 conditional knock-out mice. Neuron 31, 713–726 (2001).

48. Wang, Y. et al. Involvement of Notch signaling in hippocampal synaptic plasticity. Proc. Natl. Acad. Sci. USA 101, 9458–9462 (2004).

49. Costa, R.M., Honjo, T. & Silva, A.J. Learning and memory deficits in Notch mutant mice. Curr. Biol. 13, 1348–1354 (2003).

50. Selkoe, D. & Kopan, R. Notch and presenilin: regulated intramembrane proteolysis links development and degeneration. Annu. Rev. Neurosci. 26, 565–597 (2003).

51. Wu, L. et al. MAML1, a human homologue of Drosophila mastermind, is a transcrip-tional co-activator for NOTCH receptors. Nat. Genet. 26, 484–489 (2000).

52. Fryer, C.J., White, J.B. & Jones, K.A. Mastermind recruits CycC:CDK8 to phosphory-late the Notch ICD and coordinate activation with turnover. Mol. Cell 16, 509–520 (2004).

53. Bertrand, N., Castro, D.S. & Guillemot, F. Proneural genes and the specification of neural cell types. Nat. Rev. Neurosci. 3, 517–530 (2002).

54. Roegiers, F. & Jan, Y.N. Asymmetric cell division. Curr. Opin. Cell Biol. 16, 195–205 (2004).

55. Cayouette, M. & Raff, M. Asymmetric segregation of Numb: a mechanism for neural specification from Drosophila to mammals. Nat. Neurosci. 5, 1265–1269 (2002).

56. Lai, E.C. Protein degradation: four E3s for the notch pathway. Curr. Biol. 12, R74–R78 (2002).

57. Baron, M. An overview of the Notch signalling pathway. Semin. Cell Dev. Biol. 14, 113–119 (2003).

58. Haines, N. & Irvine, K.D. Glycosylation regulates Notch signalling. Nat. Rev. Mol. Cell Biol. 4, 786–797 (2003).

59. Furukawa, T., Maruyama, S., Kawaichi, M. & Honjo, T. The Drosophila homolog of the immunoglobulin recombination signal-binding protein regulates peripheral nervous system development. Cell 69, 1191–1197 (1992).

60. Schweisguth, F. & Posakony, J.W. Suppressor of Hairless, the Drosophila homolog of the mouse recombination signal-binding protein gene, controls sensory organ cell fates. Cell 69, 1199–1212 (1992).

61. de la Pompa, J.L. et al. Conservation of the Notch signalling pathway in mammalian neurogenesis. Development 124, 1139–1148 (1997).

62. Ishibashi, M. et al. Persistent expression of helix-loop-helix factor HES-1 prevents mammalian neural differentiation in the central nervous system. EMBO J. 13, 1799–1805 (1994).

63. Furukawa, T., Mukherjee, S., Bao, Z.Z., Morrow, E.M. & Cepko, C.L. rax, Hes1, and

notch1 promote the formation of Muller glia by postnatal retinal progenitor cells. Neuron 26, 383–394 (2000).

64. Feder, J.N., Jan, L.Y. & Jan, Y.N. A rat gene with sequence homology to the Drosophila gene hairy is rapidly induced by growth factors known to influence neuronal differentia-tion. Mol. Cell. Biol. 13, 105–113 (1993).

65. Solecki, D.J., Liu, X.L., Tomoda, T., Fang, Y. & Hatten, M.E. Activated Notch2 signal-ing inhibits differentiation of cerebellar granule neuron precursors by maintaining proliferation. Neuron 31, 557–568 (2001).

66. Lutolf, S., Radtke, F., Aguet, M., Suter, U. & Taylor, V. Notch1 is required for neuronal and glial differentiation in the cerebellum. Development 129, 373–385 (2002).

67. Hojo, M. et al. Glial cell fate specification modulated by the bHLH gene Hes5 in mouse retina. Development 127, 2515–2522 (2000).

68. Hatakeyama, J. et al. Hes genes regulate size, shape and histogenesis of the nervous system by control of the timing of neural stem cell differentiation. Development 131, 5539–5550 (2004).

69. Sakamoto, M., Hirata, H., Ohtsuka, T., Bessho, Y. & Kageyama, R. The basic helix-loop-helix genes Hesr1/Hey1 and Hesr2/Hey2 regulate maintenance of neural precursor cells in the brain. J. Biol. Chem. 278, 44808–44815 (2003).

70. Fischer, A., Schumacher, N., Maier, M., Sendtner, M. & Gessler, M. The Notch target genes Hey1 and Hey2 are required for embryonic vascular development. Genes Dev. 18, 901–911 (2004).

71. Patten, B.A., Peyrin, J.M., Weinmaster, G. & Corfas, G. Sequential signaling through Notch1 and erbB receptors mediates radial glia differentiation. J. Neurosci. 23, 6132–6140 (2003).

72. Schmid, R.S. et al. Neuregulin 1-erbB2 signaling is required for the establishment of radial glia and their transformation into astrocytes in cerebral cortex. Proc. Natl. Acad. Sci. USA 100, 4251–4256 (2003).

73. Klezovitch, O., Fernandez, T.E., Tapscott, S.J. & Vasioukhin, V. Loss of cell polar-ity causes severe brain dysplasia in Lgl1 knockout mice. Genes Dev. 18, 559–571 (2004).

74. Yamamoto, N. et al. Role of Deltex-1 as a transcriptional regulator downstream of the Notch receptor. J. Biol. Chem. 276, 45031–45040 (2001).

75. Tsunematsu, R. et al. Mouse Fbw7/Sel-10/Cdc4 is required for notch degradation during vascular development. J. Biol. Chem. 279, 9417–9423 (2004).

76. De Strooper, B. Aph-1, Pen-2, and Nicastrin with Presenilin generate an active gamma-Secretase complex. Neuron 38, 9–12 (2003).

77. Zhong, W. Diversifying neural cells through order of birth and asymmetry of division. Neuron 37, 11–14 (2003).

78. Hori, K. et al. Drosophila deltex mediates suppressor of Hairless-independent and late-endosomal activation of Notch signaling. Development 131, 5527–5537 (2004).

79. Donoviel, D.B. et al. Mice lacking both presenilin genes exhibit early embryonic pat-terning defects. Genes Dev. 13, 2801–2810 (1999).

80. McGill, M.A. & McGlade, C.J. Mammalian numb proteins promote Notch1 receptor ubiquitination and degradation of the Notch1 intracellular domain. J. Biol. Chem. 278, 23196–23203 (2003).

81. Sade, H., Krishna, S. & Sarin, A. The anti-apoptotic effect of Notch-1 requires p56lck-dependent, Akt/PKB-mediated signaling in T cells. J. Biol. Chem. 279, 2937–2944 (2004).

82. Zhong, W., Feder, J.N., Jiang, M.M., Jan, L.Y. & Jan, Y.N. Asymmetric localization of a mammalian numb homolog during mouse cortical neurogenesis. Neuron 17, 43–53 (1996).

83. Verdi, J.M. et al. Distinct human NUMB isoforms regulate differentiation vs. pro-liferation in the neuronal lineage. Proc. Natl. Acad. Sci. USA 96, 10472–10476 (1999).

84. Berdnik, D., Torok, T., Gonzalez-Gaitan, M. & Knoblich, J.A. The endocytic protein alpha-Adaptin is required for numb-mediated asymmetric cell division in Drosophila. Dev. Cell 3, 221–231 (2002).

85. Giniger, E. A role for Abl in Notch signaling. Neuron 20, 667–681 (1998).86. Stump, G. et al. Notch1 and its ligands Delta-like and Jagged are expressed and active

in distinct cell populations in the postnatal mouse brain. Mech. Dev. 114, 153–159 (2002).

87. Ni, C.Y., Murphy, M.P., Golde, T.E. & Carpenter, G. gamma -Secretase cleavage and nuclear localization of ErbB-4 receptor tyrosine kinase. Science 294, 2179–2181 (2001).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

E R R ATA

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 10 | OCTOBER 2005 1411

Erratum: Heterogeneity in synaptic transmission along a Drosophila larval motor axonGiovanna Guerrero, Dierk F Rieff, Gautam Agarwal, Robin W Ball, Alexander Borst, Corey S Goodman & Ehud Y IsacoffNat. Neurosci., 8, 1188–1196 (2005)

In the version of this article initially published online, the second author’s name was misspelled. The correct spelling is Dierk F Reiff.

Erratum: Notch signaling in the mammalian central nervous system: insights from mouse mutantsKeejung Yoon & Nicholas GaianoNat. Neurosci., 8, 709 – 715 (2005)

The version of this article that was published contained typographical errors in some gene names. On page 710, in the right column, third paragraph, the fourth sentence should have read as follows: “That study examined the expression both of pathway components such as Hes1, Hes5 and Delta-like 1 (Dll1) and of early differentiation markers such as Math4A (also known as Neurog2), NeuroD and NSCL-1 (also known as Nhlh1).” The last sentence of that paragraph should have read as follows: “This notion is supported by previous findings that Hes1 can be upregulated in PC12 cells cultured in the growth factors NGF, FGF2 or EGF64 and in postnatal cerebellar granule cells cultured in Sonic hedge-hog65.” The fourth and fifth sentences in the second paragraph, right column, on page 713 should have read as follows: “These studies found that Notch activation reduced neurite extension, but presumed signaling blockade (via expression of Numb, Numbl or Dx) could promote neurite extension. Subsequent studies have found that Numb deletion disrupts neuronal maturation in the developing cerebellum31, whereas deletion of Numb and Numbl disrupts axonal arborization in sensory ganglia in vivo32.” In addition, on page 712, in the right column, top line, the authors would like to revise the sentence to read as follows: “Third, several reports have identified ErbB2 as a Notch target that has a role during mammalian radial glial maintenance91,92.”

CO R R I G E N D U M

Corrigendum: Visual field maps and stimulus selectivity in human ventral occipital cortexAlyssa A Brewer, Junjie Liu, Alex R Wade & Brian A WandellNat. Neurosci., 8, 1102-1109 (2005)

The discussion section contains an incorrect citation. In the 3rd paragraph on page 1107, “Tootell et al. 16 (subsequent to Halgren et al.)” should read: “Tootell et al. 16 (subsequent to Hadjikhani et al.)”. The authors regret the error

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

Laminin stimulates and guidesaxonal outgrowth via growth conemyosin II activityStephen G Turney1 & Paul C Bridgman2

Guidance cues and signal transduction mechanisms acting at

the nerve growth cone are fairly well understood, but the

intracellular mechanisms operating to change the direction of

axon outgrowth remain unknown. We now show that growth

cones integrate myosin II–dependent contraction for rapid,

coordinated turning at borders of laminin stripes in response to

signals from laminin-activated integrin receptors; in the absence

of myosin II activity, outgrowth continues across the borders.

Signals derived from integrin and other guidance cue receptors con-verge and act to change vertebrate growth cone motility, coordinatingthe direction and rate of outgrowth1–3. The molecular mechanismsthat link and drive structural rearrangements of the growth coneactin and microtubule cytoskeleton during turning are not fullydefined4–6. Most models of turning emphasize control through thedynamics of peripheral extensions that do not depend on contractilemyosin (myosin II (ref. 7)). To determine if myosin II has a role,we modified a turning assay for growth cones8 and imaged neuritesgrowing at borders of substrate-bound laminin-1 stripes in cultureslacking myosin IIB activity or after pharmacologically blockingmyosin II activity.

We grew peripheral nerve explants and dissociated cells fromembryonic day (E) 13.5 mouse embryos on coverslips that containedalternating stripes of poly-l-ornithine plus laminin-1 (PLO+LN1) andpolyornithine (PLO). Cells from individual embryos were plated

0 0 0 0

2 h 7 h

6 h 21 h

5 h 3.5 h

15 h 13.5 h

a b c

d e f g

Figure 1 Inhibition of myosin II prevents turning at borders between PLO+LN1 and PLO. (a–c) Neurites (green, tyrosinated tubulin immunofluorescence)growing from explants on PLO+LN1 (red, laminin immunofluorescence) to borders (arrowheads) with PLO. (a) Control neurites turn or branch at the border.

(b) When treated with blebbistatin, most control neurites cross the border. (c) Many neurites from a myosin IIB knockout explant cross the border. (d) A time-

lapse series (phase contrast) of a control explant showing the turning response of neurites at the border between PLO+LN1 (red) and PLO (Supplementary

Video 1). (e) Time-lapse series of a control explant treated with blebbistatin. Turning at the border is greatly reduced (Supplementary Video 2). (f) Time-lapse

series of a myosin IIB knockout explant. Turning at the border is intermediate between that seen in d and e (Supplementary Video 3). (g) Time-lapse series

of a myosin IIB knockout explant treated with blebbistatin (no video). Turning at the border is very similar to that seen in e. Elapsed time is indicated at left.

Coverslips were coated with 0.1 mg/ml poly-l-ornithine (M.W. 100–200 kDa), rinsed with sterile water and dried. Silicon masks were made from

polydimethylsiloxane (Sylgard 184, Dow Corning) and then applied to the glass surface. The LN1 mixture (32 mg/ml) was then added for a minimum of 6 h.

After rinsing, the masks were either removed or kept in place until just after the cells or explants (from superior cervical ganglia) were plated. This produced

stripes of laminin without air-drying, which prevented the formation of a visible physical boundary. Blebbistatin was added to cultures 1–2 h after plating from

a 1,000� stock solution in DMSO at a final concentration of either 100 mM (mixed enantiomers; EMD) or 50 mM (– form only; Toronto Research or EMD).

Control cultures received either the + form (EMD) or vehicle only (DMSO). Scale bars: 500 mm (a), 250 mm (b), 80 mm (c) and 40 mm (d–g).


1Department of Ophthalmology and Visual Sciences and 2Department of Anatomy and Neurobiology, Washington University School of Medicine, 660 S. Euclid Avenue,St. Louis, Missouri 63110, USA. Correspondence should be addressed to P.C.B. ([email protected]).

NATURE NEUROSCIENCE VOLUME 8 [ NUMBER 6 [ JUNE 2005 7 1 7

B R I E F CO M M U N I C AT I O N S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

separately to allow comparison between those that were expressingmyosin IIB (wild-type and heterozygous) and myosin IIB knockouts9.Myosin IIB knockout neurons still contain myosin IIA. Blebbistatin canbe used to eliminate myosin IIA activity10.

Neurites of control explants growing on PLO+LN1 changed direc-tions at the border, only rarely crossing onto the PLO (Fig. 1a). Forexplants that were about 500 mm from the border (n ¼ 14), individualneurites could be followed, and 90–100% turned at the border. Neuritesof explants (control, n ¼ 10; myosin IIB knockout, n ¼ 8) that weretreated with blebbistatin rarely changed direction and frequentlycrossed the borders when growth was from PLO+LN1 to PLO(Fig. 1b). Explants from myosin IIB knockout embryos (n ¼ 9)showed an intermediate response to borders. Many neurites crossedthe border without changing direction (Fig. 1c). This suggests thatmyosin II activity is required for a growth cone–mediated turningresponse at LN1 borders. The efficiency and speed of turning alsodepends upon the amount or isoform of myosin II activity. Myosin IIBmay be required for rapid and efficient turning.

We used time-lapse recordings to characterize the turning response(Supplementary Note). To identify the substrate border, LN1 wasconjugated with the fluorochrome Cy3. Control growth cones growingon PLO+LN1 turned or branched upon contacting the border withPLO (Fig. 1d, Supplementary Video 1). After turning, neurites grewalong the border on the side containing LN1. In a representativeexperiment, growth cones responded immediately to the border (thebase of the growth cone never left the PLO+LN1) in one of three ways(Table 1a). The majority extended processes on the PLO+LN1 parallelto the border and continued growing in this direction. Some branchedafter contacting the border and then continued growing along it.Usually one branch retracted. Others sidestepped parallel to the border.

Periodic probing of the border by processes extending onto the PLOwas also observed. The few that crossed the border grew onto the PLOfor a short distance with growth cones spreading and retractingmultiple times before retracting. In blebbistatin-treated cultures, con-trol and myosin IIB knockout neurons grew in irregular curved pathsand rarely detected borders (Fig. 1e,g, Supplementary Video 2,Table 1a). Most neurites continued to grow unimpeded across theborder. All others paused, branched or turned at the border beforeeventually crossing. The behavior of myosin IIB knockout neurites wasin-between: many growth cones crossed the border without turning,but they paused and usually formed a varicosity at the border site.Although some neurites branched or turned at borders, many even-tually crossed at an angle and then sometimes returned to thePLO+LN1 (Fig. 1f, Supplementary Video 3, Table 1a). A consistentprobing of the border was not observed.

The response of the myosin IIB knockout growth cones to the LN1borders could result from decreased levels of myosin II or becauseboth myosin IIA and IIB are needed for turning. To distinguishbetween these possibilities, we expressed green fluorescent protein(GFP)–myosin IIA and GFP–myosin IIB cDNAs in dissociatedmyosin IIB knockout neurons and observed their response at LN1borders. Myosin IIB knockout neurons expressing GFP–myosinIIA turned or branched at LN1 borders at the same frequency asdissociated myosin IIB knockout neurons that were not transfected(Table 1b), indicating that higher levels of myosin IIA do not increaseturning. Their behavior at borders was the same as untransfectedmyosin IIB knockout neurites. Expressing GFP–myosin IIB in myosinIIB knockout neurons significantly increased turning rates com-pared with either GFP–myosin IIA–transfected or untransfectedmyosin IIB knockout neurons (Table 1b). After turning, theyclosely followed the border on the side of the PLO+LN1 with littleor no wandering across the border (Fig. 2, Supplementary Video 4).They also formed periodic probing extensions, similar to controlneurites. Neurites growing from PLO to PLO+LN1 did not turn.This suggests that coordinated turning at LN1 borders does not dependon the level of myosin II, but it requires that myosin IIB or both myosinIIB and IIA be active.

The turning response is reduced or eliminated by myosin II inhibi-tion. Could this be a result of the differences in outgrowth rates on thetwo substrates? To compare the outgrowth rates and how they areinfluenced by myosin II inhibition, we grew dissociated cells on the twosubstrates (with or without blebbistatin) and then fixed at knowntimes. As expected9, myosin IIB knockout outgrowth was slower thanfor control cells on PLO+LN1 (Supplementary Fig. 1). Outgrowth ofboth control and myosin IIB knockout neurons was greatly reducedwhen cells were plated on PLO compared with PLO+LN1. Whentreated with blebbistatin, outgrowth on PLO was significantly faster(P o 0.001 for both; t-test). However, blebbistatin-treated control

a

b

0

5.3 h

10.7 h

Figure 2 Transient expression of GFP–myosin IIB increases turning rates of

myosin IIB knockout neurons. (a) A sequence showing a neurite of a myosin

IIB knockout neuron transfected with GFP–myosin IIB turning and then

growing along the border (arrows) between PLO+LN1 (red) and PLO (phase

and fluorescence images are superimposed; the field is covered with gold

particles from the biolistics; see Supplementary Video 4). Elapsed time is

indicated. (b) GFP–myosin IIB (green) distribution in the same growth cone

after fixation. The growth cone spans the border between the PLO+LN1 (red)and the PLO. The neurite is barely visible (arrowheads). Dissociated neurons

were plated and grown for 5 h and then transfected with pEGFP–myosin IIA

or pEGFP–myosin IIB using biolistics14. The pEGFP (Clontech) was fused to

the N terminus as previously described, except that the cytomegalovirus

promoter was used for both constructs15. Scale bar: 38 mm (a); 9 mm (b).

71 8 VOLUME 8 [ NUMBER 6 [ JUNE 2005 NATURE NEUROSCIENCE

B R I E F C O M M U N I C AT I O N S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

neurons on PLO never reached the high outgrowth rates observed inuntreated control cells on PLO+LN1. This indicates that myosin II actsto inhibit outgrowth on PLO independent of isoform, and myosin IIBactivity is necessary for achieving the maximal outgrowth rates onPLO+LN1.

Is turning a consequence of the abrupt change in outgrowth rates atthe border, or does it depend on differential activation of integrinreceptors? To distinguish between these possibilities, we grew cellson alternating stripes of substrates that differed in their capacitiesfor integrin activation but that produced the same proportionaldifference in outgrowth rates. The frequency that growth cones turnedand the degree to which they remained close to the border did notcorrelate with differences in outgrowth rates (Supplementary Table 1),but it did depend upon substrate-bound laminin, consistent withintegrin activation.

We used immunofluorescence to determine if peripheral neuronsalso express myosin IIC (ref. 11). Growth cones were positive formyosin IIC, indicating that they contain all three myosin II isoforms(Supplementary Fig. 2). Blebbistatin is likely to inhibit all forms ofmyosin II (ref. 12).

To determine if LN1 affects the location of myosin II, we comparedthe distribution of myosin IIA and IIB in cells growing on stripes of thetwo substrates. There was no obvious difference in the distribution ofmyosin IIA in growth cones on the two substrates or at borders.However, myosin IIB consistently changed from a random distributionin growth cones on PLO to the more characteristic concentration in thetransition zone in growth cones on PLO+LN1 (ref. 13). For growthcones that had recently turned or were growing along the LN1 side ofthe border, myosin IIB was often asymmetrically concentrated on theside of the growth cone closest to the border (Supplementary Fig. 2).

These results indicate that integrin activation is linked to myosin II(mainly myosin IIB)-dependent changes in growth cone motility.Depending upon the level and location of activation by integrins,local myosin II contractile activity causes growth cones to turn, branch,retract, stall or advance with precision along a laminin border. Becausethe preference for laminin substrates was strong only when the fullcomplement of myosin II (A, B and C) was present, the growth conemust integrate the activity for rapid, coordinated turning. Thesequence of activity is likely to include integrin activation, signaling,adhesive complex formation, coupling to the actin cytoskeleton,redistribution of actin filaments, local formation of myosin II bipolarfilaments, contraction and development of traction forces. Thus, otherguidance cues modulating the location or degree of integrin activationcan have direct effects on myosin II–dependent growth cone behaviorand can profoundly affect guidance through a contractile mechanism.

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

ACKNOWLEDGMENTSG. Philips and D. Morris provided technical support. We thank J. Cooper forcomments on the manuscript. The US National Institutes of Health providedfinancial support (NS26150 to P.C.B.)


Received 1 April; accepted 21 April 2005


1. Hopker, V.H., Shewan, D., Tessier-Lavigne, M., Poo, M. & Holt, C. Nature 401, 69–73(1999).

2. Stevens, A. & Jacobs, J.R. J. Neurosci. 22, 4448–4455 (2002).3. Nakamoto, T., Kain, K.H. & Ginsberg, M.H. Curr. Biol. 14, R121–R123 (2004).4. Suter, D.M. & Forscher, P. J. Neurobiol. 44, 97–113 (2000).5. Mitchison, T. & Kirschner, M. Neuron 1, 761–772 (1988).6. Schmucker, D. Neuron 40, 4–6 (2003).7. Dent, E.W. & Gertler, F.B. Neuron 40, 209–227 (2003).8. Gundersen, R.W. Dev. Biol. 121, 423–431 (1987).9. Bridgman, P.C., Dave, S., Asnes, C.F., Tullio, A.N. & Adelstein, R.S. J. Neurosci. 21,

6159–6169 (2001).10. Straight, A.F. et al. Science 299, 1743–1747 (2003).11. Golomb, E. et al. J. Biol. Chem. 279, 2800–2808 (2004).12. Limouze, J., Straight, A.F., Mitchison, T. & Sellers, J.R. J. Muscle Res. Cell Motil. 25,

337–341 (2004).13. Rochlin, M.W., Itoh, K., Adelstein, R.S. & Bridgman, P.C. J. Cell Sci. 108, 3661–3670

(1995).14. Bridgman, P.C., Brown, M.E. & Balan, I. Methods Cell Biol. 71, 353–68 (2003).15. Brown, M.E. & Bridgman, P.C. J. Cell Sci. 116, 1087–1094 (2003).

Table 1 Responses to laminin-1 borders

(a) Response of neurites of peripheral neuron explants at laminin-1 borders from

time-lapse recordings.

Initial response at border (%)

Phenotype Blebbistatin treatment Cross Turn Branch Sidestep n

Control � 10 45 27 15 20

Control + 72 11a 17a 0 36

KO � 36 28 36 0 28

KO + 75 6a 19a 0 16

(b) Turning frequencies of dissociated neurons transfected with GFP–myosin IIA or

GFP–myosin IIB compared with untransfected neurons of fixed cultures.

Phenotype Growth direction Turning frequency (%) n

Control PLO+LN1 to PLO 93 89

Control PLO to PLO+LN1 0 30

KO + GFP–myosin IIA PLO+LN1 to PLO 47 93

KO + GFP–myosin IIB PLO+LN1 to PLO 75b 47

KO PLO+LN1 to PLO 48 139

KO PLO to PLO+LN1 0 23

aAfter turning or branching and growing along the border, most of these eventually crossedduring the recording.Direction was PLO+LN1 to PLO in all cases; KO, myosin IIB knockout.bDifference compared with untransfected myosin IIB knockout and myosin IIA–transfectedknockout neurons was significant (w2, P o 0.01 for both).For analysis of turning frequencies in fixed cells, only two categories were used: (i) turned orbranched, and (ii) crossed or undetermined (that is, when a growth cone was just at the border,and we could not determine its future direction). The growth direction was determined byfollowing the individual neurites to their cell bodies. KO, myosin IIB knockout.



2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

An fMRI investigation of race-related amygdala activity inAfrican-American and Caucasian-American individualsMatthew D Lieberman1, Ahmad Hariri2, Johanna M Jarcho1,Naomi I Eisenberger1 & Susan Y Bookheimer3

Functional magnetic resonance imaging (fMRI) was used to

examine the nature of amygdala sensitivity to race. Both

African-American and Caucasian-American individuals showed

greater amygdala activity to African-American targets than to

Caucasian-American targets, suggesting that race-related

amygdala activity may result from cultural learning rather than

from the novelty of other races. Additionally, verbal encoding of

African-American targets produced significantly less amygdala

activity than perceptual encoding of African-American targets.

The amygdala has long been known to play a role in responding to theemotionality of a stimulus, activating to images containing threatening,novel or highly arousing features1. Given the emotionality associatedwith racial interactions between Caucasian-American and African-American individuals, it is perhaps not surprising that recent studieshave also shown that the amygdala is associated with race-relatedprocessing and that the amount of amygdala activity correlates withrace-related prejudice2–4. Previous studies suggest that the amygdalaplays a role in race-related processes, but the nature of that role and theconditions under which it is instantiated remain unclear.

The current study was conducted to investigate two aspects of race-related processing in the amygdala. The first goal was to examinethe differential responses in the amygdala of African-American andCaucasian-American participants to African-American and Caucasian-American faces. Most previous studies have not directly addressed orhave lacked the statistical power to effectively address the issue ofwhether African-American and Caucasian-American individuals pro-duce similar or different amygdala responses to African-American andCaucasian-American faces.

The second goal of this investigation was to examine whether themanner of encoding race-related stimuli affects the amygdala’s responseto target race. Previous paradigms have examined only the perceptualencoding of target race, examining neural responses to images of same-or other-race faces. One possibility is that verbal encoding of the race ofAfrican-American targets should result in greater amygdala activity thanperceptual encoding because perceptual encoding of African-American

targets allows attention and thought to be focused on any number oftarget characteristics such as gender or age, whereas verbal encodingfocuses attention and thought primarily on race5. Alternatively, a secondpossibility is that verbal processing of the race of African-Americantargets should result in less amygdala activity than perceptual processingbecause of the general role of language and resource-limited cognitiveabilities, known as controlled processes, in correcting and overridingautomatic impulses, such as those generated by the amygdala6–8.

The current study examined, for both African-American andCaucasian-American participants, the consequences of both perceptualand verbal processing of race on the amygdala. On perceptual encodingtrials, participants chose the face (from a pair of faces at the bottom ofthe screen) that was of the same race as the target face at the top of thescreen (see Fig. 1a and Supplementary Note for methodologicaldetails). On verbal encoding trials, participants chose the race label(from a pair of labels at the bottom of the screen) that indicated the raceof the target at the top of the screen. Half of the verbal and perceptualencoding trial blocks had predominantly African-American targets andhalf had predominantly Caucasian-American targets.

Perceptual encoding

Right amygdala

1.21.00.80.60.40.20.0

–0.2–0.4–0.6–0.8–1.0

Am

ygda

la a

ctiv

ity

AA participants CA participants

AA targetsCA targets

Caucasian African-American

Verbal encoding Control taska

b c

Figure 1 Task and amygdala responses. (a) Sample stimuli from theperceptual encoding task, the verbal encoding task and the control task. From

the pair of stimuli at the bottom of the screen, participants always chose the

stimulus that best matched or described the target at the top of the screen.

For half of the trial blocks, most targets were African-American faces, and for

the other half of the trial blocks, most targets were Caucasian-American faces.

(b) The region of right amygdala (26, 2, �14) that was more active during the

presentation of African-American faces than during the presentation of

Caucasian-American faces, across both participant races and encoding tasks.

(c) Amygdala activity as a function of participant race and target race during

perceptual encoding of African-American and Caucasian-American targets.

Results are in terms of parameter estimates of signal intensity, relative to the

control task, in ROI analyses of the right amygdala. Error bars represent s.e.m.

Informed written consent was obtained from all participants.

Published online 8 May 2005; doi: 10.1038/nn1465

1Department of Psychology, Franz Hall, University of California Los Angeles, Los Angeles, California 90095-1563, USA. 2Department of Psychiatry, University of PittsburghSchool of Medicine, Western Psychiatric Institute and Clinic, 3811 O’Hara Street, Room E-729, Pittsburgh, Pennsylvania 15213-2593, USA. 3Brain Mapping Center,University of California Los Angeles, School of Medicine, 660 Charles Young Drive, Los Angeles, California 90095, USA. Correspondence should be addressed to M.D.L.([email protected]).



2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Results indicated that both African-American and Caucasian-American partici-pants produced a greater response in theright amygdala to African-American targetsthan to Caucasian-American targets. How-ever, this effect was modulated by encodingtask such that increased amygdala activity toAfrican-American targets was present onlyduring perceptual encoding. Additionally, ageneral pattern emerged such that brainregions involved in affect and motivationalprocessing were more active to African-Amer-ican than to Caucasian-American facesregardless of participant race, whereas brainregions involved in social perceptual processesthat are less affective9 were more responsive toin-group faces than out-group faces for eachparticipant group.

When examining neural activation to Afri-can-American target faces relative to Cauca-sian-American target faces across allparticipants and tasks, region of interest(ROI) analyses of the amygdala were notsignificant in either hemisphere (P 4 0.3);however, whole-brain analyses indicatedgreater right amygdala activity to African-American than to Caucasian-American targetfaces (Fig. 1b; Talairach coordinates: 26, 2, �14; t18 ¼ 3.26, P o 0.005;Supplementary Note). Participant race did not interact with target racein ROI (t18 ¼ 0.22, P 4 0.3) or whole-brain analyses of the amygdala,indicating that the amygdala of African-American and Caucasian-American participants did not differ significantly in their responsesto African-American and Caucasian-American target faces (Fig. 1c).

Whole-brain analyses were conducted to identify other brain regionsthat showed more activity during viewing of African-American thanduring viewing of Caucasian-American target faces. In addition tothe amygdala, ventromedial prefrontal cortex (18, 26, �12; t18 ¼ 3.36,P o 0.005), hippocampus (38, �22, �16; t18 ¼ 4.02, P o 0.005) andthe midbrain in the area of substantia nigra (10, �22, �20; t18 ¼ 4.39,P o 0.005) were all more active in response to African-Americanthan in response to Caucasian-American targets. No brain regionswere more active to Caucasian-American targets than to African-American targets.

Most analogous to previous studies, perceptual encoding yieldedgreater right amygdala activity in response to African-American than toCaucasian-American targets across all participants in ROI (t18 ¼ 2.11,Po 0.05) and whole-brain analyses (14, 0,�18; t18 ¼ 3.16, P o 0.005).Unlike perceptual encoding, however, verbal encoding of African-American versus Caucasian-American targets did not produceincreased amygdala activity in ROI (t18 ¼ 1.62, P 4 0.1, corrected)or whole-brain analyses. When comparing verbal to perceptual encod-ing, ROI analyses yielded marginally less amygdala activity duringverbal encoding, relative to perceptual encoding, of the African-Amer-ican targets (t18 ¼ 1.76, P o 0.1, corrected) but no difference forCaucasian-American targets (t18 ¼ 0.08, P 4 0.4). A whole-brainanalysis (Fig. 2a, left) demonstrated an interaction in the amygdalabetween the encoding task and target race (16, �9, �20, t18 ¼ 2.84,P ¼ 0.005) such that there was less amygdala activity during verbalencoding, relative to perceptual encoding, of African-American targets(16, �9, �20; t18 ¼ 3.01, P o 0.005) but no effect of encoding forCaucasian-American targets (16, �9, �20, t18 ¼ 0.96, P 4 0.15).

Previous investigations have observed activity in right ventrolateralprefrontal cortex (RVLPFC) during verbal encoding, relative to per-ceptual encoding, of emotionally evocative stimuli and have found thisactivity to be inversely correlated with amygdala activity, indicating itspossible role in dampening amygdala reactivity6 (SupplementaryNote). Consequently, we conducted a whole-brain analysis to deter-mine whether RVLPFC was performing a similar role during the verbalencoding of target race. A whole-brain analysis (Fig. 2a, right) showedan interaction in RVLPFC between the encoding task and target race(52, 20, 8, t18 ¼ 3.60, P o 0.005) such that greater RVLPFC activity wasobserved during verbal, relative to perceptual, encoding of African-American targets (52, 22, 8; t18 ¼ 3.35, P o 0.005) but not for theencoding of Caucasian-American targets (52, 20, 8, t18 ¼ �1.83, P 40.01). Additionally, activations in the amygdala and RVLPFC (50, 26, 8;r ¼�0.68, Po 0.005; Fig. 2b) were negatively correlated during verbal,relative to perceptual, encoding of African-American faces. No sig-nificant correlations were observed between the amygdala and RVLPFCduring the encoding of Caucasian-American targets.

Additionally, there were no three-way interactions between partici-pant race, target race and mode of encoding on amygdala activity in ROI(P 4 0.5) or whole-brain analyses of the amygdala, suggesting that theamygdala of Caucasian-American and African-American participantsdid not respond differently to target races in either perceptual or verbalencoding tasks. Thus, during perceptual encoding, African-Americanparticipants produced greater amygdala activity to African-Americantargets than to Caucasian-American targets in an ROI analysis (Fig. 1c;t7 ¼ 2.63, P o 0.05). Similarly, during perceptual encoding, Caucasian-American participants produced greater amygdala activity to African-American targets than to Caucasian-American targets in a whole-brainanalysis (16, �6, �16; t10 ¼ 5.20, P o 0.005), although the related ROIanalysis of Caucasian-American participants was not significant (t10 ¼1.46, P o 0.2 corrected). The amygdala was not significantly activatedduring verbal encoding of African-American targets, relative toCaucasian-American targets, for either African-American or Caucasian-

Perceptual

Right ventrolateralprefrontal cortex Right ventrolateral prefrontal activity

Am

ygda

la a

ctiv

ity

Am

ygda

la a

ctiv

ity

Amygdala

African-American participants

Caucasian-American participants

Caucasian-American

Target race

African-American Caucasian-American

Target race

African-American

1.0

0.5

0.0

–0.5

–1.0

–1.5

–1.5

–1.5

1.5

1.5

RV

LPF

C a

ctiv

ity

1.0

1.0

1.0

0.5

0.5

0.5

0.0

0.00.0

–0.5

–0.5

–0.5

–1.0

–1.0

–1.0

VerbalPerceptual

RVLPFC

Verbal

a

b

Figure 2 Amygdala and RVLPFC responses to target race and encoding task. (a) Interaction of target race

and encoding task effects in the amygdala (left) and RVLPFC (right). (b) Correlational plot of RVLPFC

(50, 26, 8) and right amygdala (16, �9, �20) during verbal encoding, relative to perceptual encoding,

of African-American targets. Each point represents one participant’s activity in the two neural regions.

The dashed line is the best fit for all of the participants (r ¼ �0.68). The gray line anchored by circles

is the best fit for the African-American participants (r ¼ �0.61). The black line anchored by diamonds is

the best fit for the Caucasian-American participants (r ¼ �0.80).



2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

American participants in ROI or whole-brain analyses (all P values 40.2). Finally, correlations between amygdala (16, 9, �20) and RVLPFC(50, 26, 8) activity during verbal encoding, relative to perceptualencoding, was similar for African-American participants (r ¼ �0.61,Po 0.1) and Caucasian-American participants (r ¼�0.80, P o 0.005).

Whole-brain analyses were conducted to identify brain regions thatwere more active for participants when viewing targets of their in-grouprace relative to targets of their out-group race. Between-group analysesindicated that in-group race effects were present in the lateral fusiformgyrus region, typically associated with face processing (42,�52,�12; t17

¼ 3.86, P o 0.005), in the superior temporal sulcus (56, �32, 4; t17 ¼3.40, P o 0.005) and bilaterally in the temporal poles (left: 38, 18, �34;t17 ¼ 4.17, P o 0.005; right: 32, 20, �32; t17 ¼ 4.40, P o 0.005). Thefusiform activation replicates previously observed in-group race effects2.The only brain region demonstrating out-group race effects, such thateach group of participants produced greater activity to out-group racethan in-group race targets, was the insula, bilaterally (left: �34, �16,�6: t17 ¼ 3.73, P o 0.005; right: 34, 24, �4; t17 ¼ 3.28, P o 0.005).

The current finding that both African-American and Caucasian-American participants demonstrated greater amygdala activity toAfrican-American faces than to Caucasian-American faces may providesome insight into the function of the amygdala in race-related proces-sing. Because the amygdala is involved in responding both to threat andto novelty1, it has remained unclear as to whether previous findings ofrace-related amygdala activity in Caucasian-American participantsreflects culturally learned messages that African-American individualsare potentially threatening or whether it reflects the novelty of African-American faces to most Caucasian-American participants. With theinclusion of African-American participants, these two explanationsfor amygdala activity can begin to be disentangled: among African-American participants, novelty effects should be associated with moreamygdala activity to Caucasian-American than African-Americanfaces, whereas negative cultural associations would be associatedwith more amygdala activity to African-American than Caucasian-American faces. Although no single study can conclusively address thisissue (see Supplementary Note for discussion of alternative explana-tions), the present study suggests that the amygdala activity typicallyassociated with race-related processing may be a reflection of culturallylearned negative associations regarding African-American individuals.

The second major finding from this investigation is that the mode ofencoding race-related information can lead to different patterns ofamygdala activation. Unlike previous studies that have focused pri-marily on amygdala responses to visual images of African-Americanand Caucasian-American faces, this study has examined neuralresponses to images as well as to verbal labels of race, and hasdemonstrated that the verbal encoding of African-American targetsresults in less amygdala activity than the perceptual encoding ofAfrican-American targets. Additionally, consistent with other studiesexamining the verbal encoding of affective stimuli4, verbal labeling ofAfrican-American targets recruited RVLPFC, and to the extent thatRVLPFC was active in this condition, the amygdala was less active. Thissuggests that RVLPFC may have been functionally inhibiting theamygdala, possibly by activating inhibitory interneurons in the baso-lateral nucleus of the amygdala10,11, although correlational analyses donot establish a direction of causality. This account is also consistentwith a number of studies linking RVLPFC to general inhibitoryprocesses12. It should be noted that although the current design didnot encourage individuals to regulate their affective responses, theintention to regulate one’s affect may produce results different fromthose observed here13. For a discussion addressing potential limitationsof the current study, see the Supplementary Note.

An issue related to the verbal encoding task that needs furtherinvestigation is whether verbal encoding, strictly speaking, is respon-sible for disrupting amygdala activity, or whether symbolic processes orcontrolled processes more generally cause this effect. Although thecurrent data do not speak directly to this issue, it is possible that verbalprocessing may be one of a number of symbolic processes than can havethis disruptive effect. Regardless of the various forms this symbolicprocess may take, it seems that the content of this process must befocused on the affective or evaluative nature of the stimulus forRVLPFC to be activated and for the disruption process to occur.Multiple studies have found that affective evaluation and labeling,particularly negative evaluation and labeling, selectively activatesRVLPFC relative to non-affective controlled processing14.

The verbal encoding results may seem surprising in light of previousbehavioral work that has used race-related words as primes to increasethe cognitive accessibility of affectively-congruent words15. Thesestudies seem to suggest that race-related words increase negative affect,whereas our results suggest that the controlled processes recruited byrace-related words inhibit the affect-related activity of the amygdala. Itmay be the case that controlled processing of race-related wordsactivates linguistic representations of affect while simultaneously inhi-biting the affect itself.

One intriguing possibility suggested by these results is that verbalprocessing of race diminishes the experience of threat and thus mightbe reinforcing. In other words, the controlled processes invoked byverbal stereotyping may provide some degree of emotional relief fromthe threat associated with the presence of negatively stereotyped groupmembers, and thus may have promoted the further development ofverbal stereotypes. Such a dynamic may have been useful during ourevolution to allow controlled processing responses to threat to overrideautomatic responses, but may now contribute to the ubiquitousdevelopment of intergroup stereotyping.


ACKNOWLEDGMENTSThis work was supported by grants from National Science Foundation (BCS-0074562) and National Institute of Mental Health (MH66709) to M.D.L. Theauthors also wish to thank the Brain Mapping Medical Research Organization,Brain Mapping Support Foundation, Pierson-Lovelace Foundation, theAhmanson Foundation, Tamkin Foundation, Jennifer Jones-Simon Foundation,Capital Group Companies Charitable Foundation, Robson Family, William M.and Linda R. Dietel Philanthropic Fund at the Northern Piedmont CommunityFoundation, Northstar Fund and the National Center for Research Resources(grants RR12169, RR13642 and RR08655) for their support.


Received 5 January; accepted 20 April 2005


1. Zald, D.H. Brain Res. Brain Res. Rev. 41, 88–123 (2003).2. Eberhardt, J.L. Am. Psychol. 60, 181–190 (2005).3. Phelps, E.A. et al. J. Cogn. Neurosci. 12, 729–738 (2000).4. Cunningham, W.A. et al. Psychol. Sci. 15, 806–813 (2004).5. Macrae, C.N. & Bodenhausen, G.V. Annu. Rev. Psychol. 51, 93–120 (2000).6. Hariri, A.R., Bookheimer, S.Y. & Mazziotta, J.C. Neuroreport 11, 43–48 (2000).7. Wilson, T.D. & Schooler, J.W. J. Pers. Soc. Psychol. 60, 181–192 (1991).8. Taylor, S.F., Phan, K.L., Decker, L.R. & Liberzon, I. Neuroimage 18, 650–659 (2003).9. Haxby, J.V., Hoffman, E.A. & Gobbini, M.I. Trends Cogn. Sci. 4, 223–233 (2000).10. Ghashghaei, H.T. & Barbas, H. Neuroscience 115, 1261–1279 (2002).11. Rosenkranz, J.A. & Grace, A.A. J. Neurosci. 22, 324–337 (2002).12. Aron, A.R., Robbins, T.W. & Poldrack, R.A. Trends Cogn. Sci. 8, 170–177 (2004).13. Ochsner, K.N. et al. Neuroimage 23, 483–499 (2004).14. Cunningham, W.A., Johnson, M.K., Gatenby, J.C., Gore, J.C. & Banaji, M.R. J. Pers. Soc.

Psychol. 85, 639–649 (2003).15. Perdue, C.W. et al. J. Pers. Soc. Psychol. 59, 475–486 (1990).



2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Isolation of neural stem cells from the postnatalcerebellum

Audra Lee1, Jessica D Kessler1, Tracy-Ann Read1, Constanze Kaiser1, Denis Corbeil2, Wieland B Huttner2,Jane E Johnson3 & Robert J Wechsler-Reya1

The cerebellum is critical for motor coordination and cognitive function and is the target of transformation in medulloblastoma,

the most common malignant brain tumor in children. Although the development of granule cells, the most abundant neurons in

the cerebellum, has been studied in detail, the origins of other cerebellar neurons and glia remain poorly understood. Here we

show that the murine postnatal cerebellum contains multipotent neural stem cells (NSCs). These cells can be prospectively

isolated based on their expression of the NSC marker prominin-1 (CD133) and their lack of markers of neuronal and glial lineages

(lin2). Purified prominin1lin2 cells form self-renewing neurospheres and can differentiate into astrocytes, oligodendrocytes and

neurons in vitro. Moreover, they can generate each of these lineages after transplantation into the cerebellum. Identification

of cerebellar stem cells has important implications for the understanding of cerebellar development and the origins

of medulloblastoma.

The cerebellum is required for motor coordination and is crucial forcognitive and affective processing1. These functions depend on inter-actions among at least six types of neurons and two types of glia2.Disruption of cerebellar structure and function is associated withdisorders such as ataxia, autism and schizophrenia3–5. In addition,uncontrolled growth of cerebellar precursors results in medulloblas-toma6,7. Elucidating the mechanisms that control the generation ofcerebellar neurons and glia during normal development is critical forunderstanding the basis of these diseases.

The cerebellum differs from most other brain regions in that itcontains two distinct germinal layers: the ventricular zone (VZ), whichis most active during embryonic development, and the externalgerminal layer (EGL), which contributes to neurogenesis after birth2.There is strong evidence that Purkinje cells, the major output neuronsof the cerebellum, originate from the VZ and that granule cells, themost abundant interneurons, arise from the EGL. However, the originof the other cell types in the cerebellar cortex—including astrocytes,oligodendrocytes and stellate, basket, Lugaro and Golgi interneurons—is much less clear. Tissue grafting and transplantation studies suggestthat many of these cells arise from VZ progenitors that migrate into thecerebellar cortex after birth8,9. But whether each class of neuron andglial cell comes from a distinct progenitor or whether they all comefrom a common, multipotent progenitor is not known.

Here we purify a population of multipotent neural stem cells fromthe postnatal cerebellum. We show that this population can undergoself-renewal in culture and can generate neurons, astrocytes andoligodendrocytes both in vitro and after transplantation. Our findings

suggest that cerebellar neurons and glia could be generated from acommon progenitor. In addition, the approach we have used may beapplicable for isolating NSCs from other parts of the nervous system.

RESULTS

Non–granule cell precursors proliferate in response to bFGF

Because the majority of cells in the postnatal cerebellum are granule cellprecursors (GCPs), it has been difficult to study the precursors of othercell types. To circumvent this problem we used Math1-GFP mice, whichexpress green fluorescent protein in their GCPs10. We isolated cellsfrom the cerebellum of 7-d-old (P7) mice and analyzed them by flowcytometry. Among the cells we isolated, 90% were GFP+ GCPs(Fig. 1a). Approximately 10% of the cells were GFP� and thus likelyrepresented precursors of other lineages.

To study these cells in more detail, we sorted them by FACS andmeasured their responses to growth factors. Consistent with ourprevious findings11, GFP+ GCPs proliferated robustly in the presenceof Sonic hedgehog (Shh, Fig. 1b). Although some studies havesuggested that basic fibroblast growth factor (bFGF) can be mitogenicfor GCPs12, we found that purified GFP+ cells did not proliferate inresponse to bFGF. In contrast, GFP� cells showed little response to Shhbut proliferated extensively in response to bFGF (Fig. 1b). These dataindicate that at least two populations of precursors can be isolated fromthe postnatal cerebellum: Math1-GFP+, Shh-responsive GCPs, andMath1-GFP�, bFGF-responsive non-GCPs.

To identify the GFP� cells, we stained them with antibodies specificfor neuronal and glial markers (Table 1). The GFP� population


1Department of Pharmacology & Cancer Biology, Duke University Medical Center, Durham, North Carolina 27710, USA. 2Max-Planck Institute of Molecular Cell Biology andGenetics, D-01307 Dresden, Germany. 3Center for Basic Neuroscience, University of Texas Southwestern Medical Center, Dallas, Texas 75390-9111, USA. Correspondenceshould be addressed to R.J.W.-R. ([email protected]).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

included cells with markers of neuronal (HNK-1, polysialated (PSA)NCAM, MAP-2; refs. 13–15), oligodendrocyte (O4, NG2; refs. 16,17)and astrocyte lineages (GFAP, TAPA-1, CD44; refs. 18–20). In addition,approximately one-third of GFP� cells expressed markers associatedwith neural progenitors and stem cells, including nestin21, prominin-1(ref. 22), Sox-2 (ref. 23) and Musashi24 (Table 1 and data not shown).These studies suggested that the GFP– population includes neurons,astrocytes, oligodendrocytes and stem cells.

Stem cells can be purified from the postnatal cerebellum

Our detection of cells expressing NSC markers raised the possibilitythat the postnatal cerebellum contains multipotent neural stem cells. Toinvestigate this, we sorted the putative stem cells by FACS usingantibodies specific for prominin-1 (CD133), a surface glycoproteinfound on stem cells in the nervous and hematopoietic systems22,25,26.The prominin+ progenitors we isolated were highly enriched for bFGF-responsive cells (Fig. 2). Because these cells co-isolated with GCPs, we

sought to determine whether they were located in the EGL, where GCPsreside. In situ hybridization with probes specific for Prom1, the geneencoding prominin-1, identified Prom1-expressing cells throughoutthe cerebellar white matter (Fig. 3a–d). We also saw occasional Prom1-expressing cells in the internal granule layer (IGL) and Purkinje layerbut not in the molecular layer or EGL. We observed a similar expressionpattern when neonatal cerebellum was stained with antibodies toprominin (Fig. 3e,f). In addition, when cerebellum was microdissectedinto EGL and non-EGL regions and analyzed by FACS, prominin+ cellswere found to be highly enriched in the non-EGL fraction (data notshown). Together these data suggest that the majority of prominin-expressing cells are located in the white matter and not in the EGL.

Among prominin+ cells, 50–60% expressed markers of neurons and30–40% expressed markers of astrocytes and oligodendrocytes; only10% lacked such markers and were considered ‘lineage-negative.’ Tofurther purify these cells, we used antibodies to the surface markersPSA-NCAM, TAPA-1 and O4 to deplete cells associated with neuronaland glial lineages. The resulting prominin-positive, lineage-negative(prominin+lin�) population represented 1–3% of the Math1-GFP�

cells, or 0.1–0.3% of the cells that could be isolated from the neonatalcerebellum (Fig. 4a and Supplementary Fig. 1 online). Notably, only asubset of these cells (B20%) expressed nestin, suggesting that promi-nin and nestin are overlapping but not equivalent markers of NSCs.

Our identification of a population of cells with a phenotypeassociated with neural stem cells prompted us to determine whetherthese cells showed functional properties associated with neural stemcells as well.

Cerebellar stem cells generate neurospheres in vitro

Stem cells from many parts of the CNS proliferate and form macro-scopic spheres when cultured on non-adhesive substrates in thepresence of growth factors27. To determine whether the cells we isolatedcould generate such neurospheres, we cultured them at clonal densityin bFGF and EGF. Prominin+lin� cells cultured in this manner

GFP negative10%

GFP positive90%

GFP negative GFP positive

6050

4030

2010

0

100 101 102 103 104

GFP

40,000

35,000

30,000

25,000

20,000

15,000

10,000

5,000

0Shh bFGF∅ Shh bFGF∅

Stimulus

[3 H]th

ymid

ine

inco

rpor

atio

n (c

pm)

b

a

70,000

60,000

50,000

[3 H]th

ymid

ine

inco

rpor

atio

n (c

pm)

40,000

30,000

20,000

10,000

0∅ bFGF

Stimulus

Prominin+

Prominin–

Figure 2 Prominin+ cells proliferate in response to bFGF. GFP� cells from

Math1-GFP cerebella were sorted by FACS into prominin+ and prominin�

cells, and each population was cultured either with no stimulus (+) or with

bFGF and then harvested for analysis of thymidine incorporation. Datarepresent triplicates 7 s.e.m.

Table 1 Phenotype of non–granule cell precursors

Antibody Specificity Percentage positive (among GFP� cells)

Prominin-1* Stem cells 30

Nestin Stem cells 29

GFAP Astrocytes and stem cells 23

HNK-1* Neuronal progenitors 73

PSA-NCAM* Neuronal progenitors 47

Map-2 Neuronal progenitors 79

O4* Oligodendrocytes 32

NG2* Oligodendrocytes 16

TAPA-1* Astrocytes 58

CD44* Astrocytes 14

GFP� cells were stained with antibodies to surface antigens (asterisks) and analyzedby flow cytometry, or allowed to adhere to coverslips and then fixed and stained withantibodies specific for intracellular antigens (nestin, GFAP, Map-2). Data represent thepercentage of GFP� cells that express a given marker. For intracellular antigens,percentages represent averages of four fields.

Figure 1 Non–granule cell precursors can be purified from the postnatal

cerebellum. (a) Isolation of non-GCPs. Cells from neonatal Math1-GFP

cerebellum were sorted into GFP-negative and GFP-positive populations

by flow cytometry. (b) Proliferative responses of non-GCPs. The GFP� and

GFP+ populations were cultured with no stimulus (+), Shh or bFGF and then

harvested to measure thymidine incorporation. Data represent

triplicates 7 s.e.m.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

reproducibly generated spheres within 6–10 d (Fig. 4b). These cellscould also generate neurospheres when cultured at a density of one cellper well, confirming that neurospheres resulted from expansion ofindividual clones rather than from aggregation of cells. In contrast toprominin+lin� cells, prominin– cells cultured at clonal density survivedpoorly and never formed macroscopic spheres (Fig. 4c). Prominin+lin+

cells survived, but rarely formed free-floating spheres; instead, theyusually gave rise to adherent colonies that showed signs of neuronaldifferentiation (extension of processes and expression of Map-2). Theefficiency of neurosphere formation in various fractions of cerebellarcells is summarized in Supplementary Figure 1. Together these datademonstrate that prominin+lin� cells are highly enriched in the abilityto form neurospheres.

To determine whether the neurospheres generated fromprominin+lin� cells expressed stem cell markers, we stained themwith antibodies specific for nestin, Sox-2, Musashi and GFAP (amarker often associated with astrocytes but recently shown to beexpressed by radial glia and neural stem cells as well28). These markers

were expressed in all neurospheres examined (Fig. 4d–l). The majorityof cells in each sphere expressed these markers at levels well abovebackground (see controls, Fig. 4m–o), with a subset (3–6%) of cells ineach sphere expressing particularly high levels. Thus, prominin+lin–

cells form neurospheres that maintain expression of stem and pro-genitor cell markers.

Cerebellar stem cells undergo extensive self-renewal

A key feature of stem cells is their ability to undergo self-renewal, thatis, to proliferate and generate more stem cells. We examined the self-renewal of neurospheres generated from prominin+lin� cells by dis-sociating them and testing their ability to re-form new neurospheres.

Control Control

PromininRNA

PromininRNA

Promininprotein

Promininprotein

EGL

WM

EGL

EGL

EGL

WM

WM

a b

c d

e f

Figure 3 Prominin+ cells are located in the cerebellar white matter.

(a–d) Detection of prominin mRNA by in situ hybridization. Cryosections of

P7 cerebellum were hybridized with digoxigenin (DIG)-labeled sense (control;

a,b) or antisense (c,d) probes specific for mouse Prom1 (prominin RNA) and

then stained with alkaline phosphatase–conjugated antibodies to DIG and

with NBT/BCIP substrate to detect bound probe. Sections were photographed

at 10� (a,c) and 20� (b,d) magnification; boxes indicate locations of regions

shown in b,d. Arrows indicate prominin+ cells (dark brown spots) in the whitematter (WM). (e,f) Detection of prominin protein by immunostaining.

Cerebellar sections were stained with rat antibodies to prominin-1 and

TRITC-conjugated secondary antibodies. Low-power (10�) image of

cerebellum from a Math1-GFP mouse shows GFP fluorescence (green) in the

outer EGL and prominin+ cells (red) in the white matter (e). The section is

counterstained with DAPI (blue) to highlight cerebellar structure. High-power

(20�) image of cerebellum from wild-type mouse shows prominin+ cells (red)

in the white matter (f). No counterstain is shown. Red, green and blue images

were photographed separately and merged with Openlab software. EGL,

external germinal layer; WM, white matter.

104

103

102

101

100

100 101 102 103 104

Prominin

Line

age

0.2%

Prominin+Lin– Prominin–

Sox2

Musashi Nestin Merge

GFAP Nestin Merge

Nestin Merge

Control (rabbit) Control (mouse) Control (merge)

a b c

d e f

g h i

j k l

m n o

Figure 4 Prominin+lin� cells generate neurospheres. (a) Isolation of

prominin+lin� cells. Cerebellar cells from Math1-GFP mice were stained with

antibodies to prominin-1 (x axis) and neuronal and glial lineage markers

(PSA-NCAM, TAPA-1 and O4, y axis), and analyzed by flow cytometry.

Prominin+lin� cells represent 0.2% of the total population isolated from the

postnatal cerebellum. (b,c) Neurosphere formation. FACS-sorted

prominin+lin� and prominin� cells were cultured at clonal density in the

presence of bFGF and EGF for 10 d. (c–o) Expression of NSC markers.

Neurospheres from clonal cultures of prominin+lin� cells were stained with

rabbit antibodies specific for Sox-2 (d), Musashi (g) or GFAP (j) and with

mouse antibodies specific for nestin (e,h,k); primary antibodies were detected

with FITC-conjugated anti-rabbit (green) and TRITC-conjugated anti-mouse

(red) antisera. Insets show high-magnification images of regions indicated bywhite boxes; high levels Sox-2, Musashi, GFAP and nestin are expressed by a

subset of cells within each sphere. To control for nonspecific staining, some

spheres from each experiment were stained with normal rabbit IgG (m) and

normal mouse IgG (n) followed by the same secondary antibodies. Red and

green pictures were taken at 40� magnification and merged (f,i,l,o) using

Openlab software.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Cells from dissociated neurospheres gave rise to secondary neuro-spheres that were morphologically and phenotypically identical toprimary neurospheres (Supplementary Fig. 2 online). In addition,these neurospheres could be repeatedly dissociated and re-plated,allowing propagation for at least 10 weeks in culture.

The efficiency of neurosphere formation was maintained acrossmultiple passages (Supplementary Fig. 2). When freshly isolatedprominin+lin� cells were cultured at clonal density, approximately 1neurosphere was generated for every 30 cells plated (a frequency of3%). When primary neurospheres were dissociated and re-plated atclonal density, the frequency of secondary and tertiary neurospheregeneration was also B3%. This is similar to the percentage of cells thatexpressed high levels of stem cell markers, as mentioned previously.Like primary neurospheres, secondary and tertiary spheres retainedexpression of nestin, GFAP, Musashi and Sox-2. These studies indicatethat prominin+lin� cells isolated from the cerebellum are capable ofundergoing extensive self-renewal.

Cerebellar stem cells generate neurons and glia in vitro

Another important characteristic of neural stem cells is multipotency:the ability to differentiate into neurons, astrocytes and oligodendro-cytes. To test whether neurospheres derived from prominin+lin� cellsare multipotent, we withdrew bFGF and EGF and cultured the neuro-

spheres in the presence of factors that support neuronal and glialdifferentiation29–31. Neurospheres cultured in retinoic acid (RA),leukemia inhibitory factor (LIF) or platelet-derived growth factor(PDGF-AA) could each generate astrocytic (S100b+, GFAP+), oligo-dendroglial (NG2+) and neuronal (Map-2+) cells (Fig. 5a–k). Notably,mature oligodendrocytes expressing O4 were generated only in LIF(Fig. 5f) and PDGF-AA (Fig. 5i,k). These data demonstrate that asingle prominin+lin� cell can generate neurons, astrocytes and oligo-dendrocytes in vitro.

To determine the types of the neurons generated, we examineddifferentiated neurospheres for markers associated with differentclasses of cerebellar neurons. Neurons in the cerebellum can bedistinguished on the basis of neurotransmitter secretion: granulecells secrete glutamate, whereas all other cerebellar neurons secreteg-aminobutyric acid (GABA)32,33. In our cultures, some neuronsexpressed GABA (Fig. 5l), whereas others expressed glutamate(Fig. 5m). This heterogeneity was also evident from staining withantibodies to lineage-specific transcription factors: some cells expressedPax-2, a transcription factor found in stellate and basket cell progeni-tors34 (Fig. 5n), whereas others expressed Zic-1 and Math-1, markers ofthe granule cell lineage10,35 (Fig. 5o and data not shown). Togetherthese data indicate that cerebellar stem cells can generate various typesof neurons and glia in vitro.

Stem cells generate neurons and glia after transplantation

To determine whether prominin+lin� cells are also multipotent in vivo,we injected them into the cerebellum of neonatal mice and analyzedtheir ability to undergo differentiation. In one set of experiments(Fig. 6), freshly isolated prominin+lin– cells were injected into hosts.This ensured that the ability of donor-derived stem cells to differentiateinto neurons, astrocytes and oligodendrocytes was not influenced byexposure to growth factors or extensive cell culture. However, injecting

RA

LIF

PDGF

a b c

d e f

g h i

j k l

m n o

NG2/S100β

NG2/S100β

NG2/S100β

GFAP

Glutamate

Map2 O4

Map2 O4

Map2 O4

O4/S100β/Map2 GABA

Pax2 Zic1

Figure 5 Prominin+lin� cells exhibit multipotency in vitro. (a–k) Generation of

neurons, astrocytes and oligodendrocytes. Prominin+lin� cells were cultured

at clonal density for 10 d to generate neurospheres. Neurospheres were

transferred onto PDL-coated coverslips and cultured in the absence of bFGF

and EGF and in the presence of all-trans retinoic acid (RA; a–c), leukemia

inhibitory factor (LIF; d–f) or platelet-derived growth factor (PDGF-AA; g–k).

After another 7 d, cultures were fixed and stained with antibodies specific for

NG2 and S100b (red and blue, respectively; a,d,g), Map-2 (b,e,h), O4 (c,f,i)or GFAP (j). Also shown (k) is a single neurosphere cultured in PDGF-AA and

stained with antibodies to O4 (red), Map-2 (green) and S100b (blue).

(l–o) Generation of granule and non-granule neurons. Neurospheres

differentiated in the presence of PDGF-AA were fixed and stained with

antibodies specific for GABA (l), glutamate (m), and Pax-2 (n) and Zic-1 (o)

followed by rhodamine-conjugated secondary antibodies. All pictures were

taken at 40� magnification and processed using Openlab software.

S100βa b c

d e f

NG2 TuJ1

PCL IGL

ML

Figure 6 Prominin+lin� cells differentiate into neurons and glia after

transplantation into the cerebellum. Freshly isolated prominin+lin� cells were

labeled with CM-DiI (red) and injected into the cerebellum of 3-d-old mice

(10,000 cells per host). After 2–3 weeks, host cerebella were fixed and

stained with antibodies specific for S100b (a), NG2 (b) or TuJ1 (c). Primaryantibodies were detected with FITC-labeled antisera (green). Double-labeled

cells (orange-yellow in a–c) were identified and photographed at 40�magnification; 10� bright-field pictures (d–f) indicate the location of

transplanted cells in the corresponding fluorescent panels. PCL, Purkinje cell

layer; IGL, internal granule layer; ML, molecular layer.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

large numbers of prominin+lin– cells could not prove that all threelineages were derived from a single stem cell. To this end we also carriedout experiments in which a single neurosphere (derived from a singlestem cell) was transplanted into each host (Fig. 7). In each case, the fateof transplanted cells was assessed 2–3 weeks after injection.

Both freshly isolated prominin+lin� cells and neurospheres gave riseto cells of all three lineages. S100b+ cells (astrocytes) were found inthe Purkinje layer near the cell bodies of endogenous Bergmann glia(Fig. 6a,d) or in the white matter (Fig. 7a,d). NG2+ cells (oligodendro-cytes) were observed in the IGL (Fig. 6b,e) or near the surface of thecerebellum (Fig. 7b,e). TuJ1+ cells (neurons) were most commonlyseen in the molecular and Purkinje layers (Figs. 6c,f and 7c,f), wherebasket, stellate and Purkinje neurons normally reside. These studiessuggest that individual cerebellar stem cells can generate neurons,astrocytes and oligodendrocytes in vivo and that these cells canintegrate into the cerebellum after transplantation.

The most abundant donor-derived cells were neurons. To furthercharacterize these cells, we stained sections of host cerebellum withantibodies to GABAA receptor a6 (a neurotransmitter receptor foundspecifically on granule cells), parvalbumin (a marker of mature basket,stellate and Purkinje cells) and calbindin (a marker of Purkinje cells).Of the neurons we observed, none were found in the IGL and noneexpressed GABAA receptor a6 (Fig. 7g,j). Rather, the majority ofdonor-derived neurons were small, parvalbumin+ cells located in themolecular layer (Fig. 7h,k), a phenotype consistent with stellateneurons. In addition, we also observed large calbindin+ cells in thePurkinje cell layer, alongside endogenous Purkinje cells (Fig. 7i,l).Together, these data indicate that cerebellar stem cells can give rise toastrocytes, oligodendrocytes and neurons both in vitro and in vivo.

DISCUSSION

Stem cells have been identified in many parts of the nervous system inembryonic life and in adulthood. However, there has been littleevidence for the presence of stem cells in the cerebellum, and thesecells have not been isolated or studied in detail36,37. We have used flowcytometry to prospectively purify multipotent neural stem cells from

the developing cerebellum. We show that these cells express stem cellmarkers, proliferate in response to bFGF and EGF, generate self-renewing neurospheres, and differentiate into neurons and gliain vitro and after transplantation into neonatal mice. These findingshave important implications for our understanding of normal cere-bellar development and tumorigenesis.

Our identification of cerebellar stem cells originated from studies ofbFGF responsiveness in GCPs. Although previous reports have sug-gested that bFGF is a weak mitogen for GCPs12, we have observed thatit is a potent inhibitor of Shh-induced GCP proliferation11. To explainthis discrepancy, we purified GCPs to homogeneity by FACS sortingGFP+ cells from Math1-GFP mice. Analysis of the resulting cellsshowed that highly purified GCPs did not proliferate to bFGF. Onthe other hand, we observed a significant proliferative response tobFGF in a small population of cells that co-fractionates with GCPsduring the purification procedure. Thus, Math1-GFP mice represent apowerful tool for purifying different classes of neurons and glia fromthe cerebellum.

Among the GFP� cells in the postnatal cerebellum, a sub-set expressed markers associated with neural stem cells, includingprominin-1. Prominin is associated with plasma membrane protru-sions in embryonic and adult epithelial cells, and is also a marker ofhematopoietic stem cells22. In the nervous system, prominin has beenreported to colocalize with nestin in the ventricular zone25. Moreover,antibodies to the human homolog of prominin, CD133, have been usedto isolate multipotent stem cells from fetal brain and from brain tumortissue26,38. Thus, prominin is an important surface marker for neuralstem cells. Expression of prominin has not been described previously inthe postnatal cerebellum, but our studies suggest that it marks multi-potent progenitors in this tissue as well.

We used antibodies to prominin to isolate progenitors from thepostnatal cerebellum and tested their ability to form neurospheres. Wefound that prominin+ cells are highly enriched for neurosphereformation as compared to unfractionated cells from the postnatalcerebellum. Moreover, neurospheres derived from prominin+lin– cellscan undergo self-renewal and differentiate into both neurons and glia.The fact that these cells can be prospectively isolated has also allowed usto examine their potential without subjecting them to prolongedculture in vitro. This is important, because a number of recent studieshave indicated that culturing cells under neurosphere-generating con-ditions can alter the types of cells generated39. The fact that we cangenerate neurons, astrocytes and oligodendrocytes after transplanta-tion of either single neurospheres or freshly isolated prominin+lin– cellssuggests that such de-differentiation has not taken place in our studies.Rather, the tri-lineage potential of our cells seems to reflect an intrinsiccharacteristic of the cells we have isolated.

S100β

GABA-ARα6 Parvalbumin Calbindin

WM

IGL ML

PCL

piaML

NG2 TuJ1a b c

d e f

g h i

j k l

Figure 7 Neurospheres derived from a single stem cell can generate neurons

and glia in vivo. Prominin+lin� cells from actin-CFP transgenic mice were

cultured at clonal density for 10 d to generate neurospheres. Neurospheres

were injected into the cerebellum of 3-d-old mice (one neurosphere per host).

After 2–3 weeks, host cerebella were fixed and stained with antibodies

specific for S100b (a), NG2 (b), TuJ1 (c), GABAA receptor a6 (g),

parvalbumin (h) or calbindin (i). These antibodies were detected with TRITC-

conjugated secondary antibodies (red). Chicken antibodies to GFP and FITC-conjugated anti-chicken antiserum were used to amplify the CFP signal

(green). Double-labeled cells (orange-yellow in a–c and g–i) were identified

and photographed at 40� magnification; 10� bright-field pictures (d–f and

j–l) indicate the location of transplanted cells in the corresponding

fluorescent panels. WM, white matter; pia, pial membrane surrounding

surface of cerebellum; ML, molecular layer; IGL, internal granule layer; PCL,

Purkinje cell layer.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Our demonstration that purified cerebellar stem cells can generateneurons, astrocytes and oligodendrocytes in vitro and after transplanta-tion has important implications for cerebellar development. Althoughthe postnatal cerebellum has long been known to contain precursors ofbasket and stellate neurons, astrocytes and oligodendrocytes2, it has notbeen clear whether these cells arise from distinct precursors or from asingle multipotent progenitor. Several studies have hinted at the latterpossibility. For example, oncogene-immortalized cell lines derived fromthe postnatal cerebellum express neuronal and glial markers in vitro andcan generate neurons and glia after implantation in the cerebellum andother parts of the brain40,41. Although it has been suggested that thesecells represent GCPs whose differentiation potential has been altered byoncogenes40, our studies suggest they may be stem cells that have beenselectively immortalized in culture. Similarly, an elegant series ofretroviral lineage-tracing studies8,42 has shown that the white mattercontains progenitors that give rise to interneurons and glia. In thesestudies, a small proportion of retrovirally marked cells did not stainwith lineage markers, and it was suggested that these might representuncommitted stem cells42. Because we have detected expression ofprominin primarily in white matter, the NSCs we have isolated mayrepresent these uncommitted cells.

In addition to generating GABA- and Pax2-expressing interneurons,prominin+lin� cells also generated glutamate- and Zic-1-expressinggranule cells. Although both granule and non-granule cells are gener-ated in the postnatal cerebellum, several studies have suggested thatthey arise from distinct progenitors8,9,43. One interpretation of ourresults is that cerebellar NSCs have the potential to generate granulecells in vitro, but that in vivo they generate exclusively non-granuleneurons and glia. Consistent with this view, we have not observed anygranule neurons generated after transplantation of cerebellar stem cellsinto the neonatal cerebellum. On the other hand, we cannot rule outthe possibility that some of the granule cells generated during postnataldevelopment come from multipotent progenitors rather than fromrestricted GCPs in the EGL. These cells, if they are rare, would bedifficult to detect using conventional lineage tracing and transplanta-tion studies. More saturating methods of fate mapping44 might shedlight on this issue.

In addition to their importance for normal development, our studiesalso have significant implications for understanding medulloblastoma.The cell of origin for medulloblastoma has been a matter of long-standing debate. A subset of medulloblastomas express markers ofGCPs and have mutations in the Shh pathway45,46, and are thereforelikely to arise from committed GCPs. However, the majority ofmedulloblastomas express distinct markers and show no evidence ofShh pathway activation46,47. In light of recent studies demonstratingexpression of CD133 and other NSC markers in human medulloblas-toma38,48, it is important to consider the possibility that some of thesetumors arise from the stem cells we have described. If so, understandingthe signaling pathways that control growth and differentiation of thesecells may yield new approaches to the diagnosis or therapy of thisdevastating tumor.

METHODSAnimals. Math1-GFP mice10 were generated at UT Southwestern Medical

Center. b-actin–cyan fluorescent protein (actin-CFP) transgenic mice were

provided by B. Capel (Duke University). C57Bl/6 � CBAF1 mice were from the

Jackson Laboratories. Animals were maintained in the animal facility at Duke

University and used in accordance with protocols approved by the Duke

Institutional Animal Care and Use Committee.

Isolation of non-GCPs and stem cells. Cells were isolated from the cerebellum

as described previously11. Briefly, cerebella from postnatal day 7 (P7) mice were

digested with 10 U ml�1 papain (Worthington), 200 mg ml�1 L-cysteine and

250 U ml�1 DNase (Sigma). Tissue was triturated to obtain a single cell

suspension and then centrifuged through a Percoll gradient (Sigma). Cells were

harvested, washed and resuspended in buffer consisting of Dulbecco’s PBS with

5% FCS and 2 mM EDTA.

Non-GCPs were purified by FACS sorting of GFP-negative cells from Math1-

GFP mice using a FACSVantage SE (BD Biosciences). Cerebellar stem cells were

purified by sorting cells that expressed prominin-1 (detected with antibody

13A4; ref. 49) and lacked the lineage markers O4, polysialated (PSA)-NCAM

(Chemicon) and TAPA-1 (eBioscience). Prominin+lin� cells were sorted into

Neurobasal medium with B27 supplement (NB-B27, Invitrogen) containing

10 mg ml�1 BSA (Sigma), washed and resuspended in NB-B27 with appro-

priate growth factors.

Proliferation assays. Cells isolated as described above were suspended in NB-

B27 and transferred to poly-D-lysine (PDL)-coated 96-well plates at a density of

2 � 105 cells per well. Growth factors were added immediately. After 48 h, cells

were pulsed with methyl-[3H]thymidine (Perkin Elmer) and cultured for an

additional 16–18 h. Cells were harvested onto filters using a Mach IIIM Manual

Harvester 96 (Tomtec) and incorporated radioactivity was quantified on a

Wallac MicroBeta scintillation counter (PerkinElmer) by liquid scintillation

spectrophotometry.

Flow cytometry and immunofluorescence. To detect expression of surface

markers, cells were stained for 1 h with primary antibodies and 30 min with

secondary antibodies, and analyzed by FACS. To detect expression of intracel-

lular markers, cells were plated on PDL-coated coverslips and allowed to adhere

for 1–2 h before being fixed with 4% paraformaldehyde (PFA). Cells were

incubated in normal goat serum (NGS) with 0.2% Triton X-100 to block

nonspecific binding and permeabilize membranes, and stained overnight with

primary antibodies and for 2 h with secondary antibodies. Coverslips were

mounted with Fluoromount G (Southern Biotechnology Associates). Immuno-

fluorescence was detected using a Nikon TE200 inverted microscope and

Openlab software (Improvision).

For immunostaining of tissue, cerebella were fixed in 4% PFA, cryoprotected

in 25% sucrose and embedded in O.C.T. (Sakura Finetek). Cryosections

(10–16 mM) were blocked and permeabilized overnight in NGS with Triton

X-100 and stained with antibodies as described above. Sections were counter-

stained with DAPI (Molecular Probes) and mounted using Fluoromount G.

Antibodies used for flow cytometry and immunofluorescence included the

following: antibodies to HNK-1 (CD57), nestin and GFAP (BD Pharmingen);

to TuJ1 and Pax-2 (Covance); to CD44 (Biosource); to Musashi, Sox-2, NG2,

Zic-1 and GFP (Chemicon); and to S100b, Map-2, GABA, glutamate, parval-

bumin and calbindin (Sigma). Normal mouse and rabbit IgG, NGS and

secondary antibodies labeled with fluorescein isothiocyanate (FITC), amino-

methylcoumarin (AMCA), tetramethyl rhodamine (TRITC), phycoerythrin

(PE) and Cy5 were from Jackson ImmunoResearch.

In situ hybridization. P7 pups were perfused with 2.5% PFA and their brains

were dissected, embedded in O.C.T., and cryosectioned at a thickness of 12 mM.

Sections were post-fixed in 4% PFA, acetylated and incubated for 1 h at

room temperature (20–25 1C) in pre-hybridization buffer (50% formamide,

5� SSC, 1� Denhardt’s solution, 250 mg ml�1 yeast tRNA, 500 mg ml�1 herring

sperm DNA). Sections were hybridized overnight at 65 1C in hybridization

buffer (50% deionized formamide, 1� Denhardt’s solution, 300 mM NaCl, 20

mM Tris-HCl, pH 8.0, 5 mM EDTA, 10 mM Na2HPO4, pH 7.4, 10% dextran

sulfate, 0.5 mg ml�1 yeast tRNA) containing digoxigenin (DIG)-UTP–labeled

probes for Prom1. Probes were synthesized using a DIG labeling kit (Roche)

according to the manufacturer’s protocol. After hybridization, sections

were incubated overnight at 4 1C with antibodies to DIG conjugated to

alkaline phosphatase (Roche). Bound probe was visualized by incubating slides

in NBT/BCIP overnight in the dark. Coverslips were mounted with Aqua-

Polymount (Polysciences).

Neurosphere growth and differentiation. To generate primary neurospheres,

prominin+lin– cells were cultured at clonal density50 (1–2 cells per mm2)

on uncoated plates in NB-B27 containing 25 ng ml�1 bFGF (Invitrogen) and

25 ng ml�1 EGF (Peprotech). Cells were cultured for 10 d and neurospheres


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

were harvested using AdvanTip-LT low-retention pipette tips (Hamilton). To

generate secondary neurospheres, primary neurospheres were dissociated with

papain and replated under identical neurosphere growth conditions for an

additional 10 d. For differentiation, free-floating neurospheres were centri-

fuged, supernatant was aspirated and neurospheres were resuspended in NB-

B27 containing 10 ng ml�1 PDGF-AA (Sigma), 1 U ml�1 leukemia inhibitory

factor (ESGRO/LIF, Chemicon) or 100 ng ml�1 all-trans retinoic acid (RA,

Sigma) and plated onto PDL-coated coverslips. After 7 d, coverslips were fixed

and stained as described above.

Neurosphere and stem cell transplantation. To assess in vivo the differentia-

tion of prominin+lin– cells and neurospheres, cells were either labeled with

chloromethylbenzamido-DiI (CM-DiI, Molecular Probes) or isolated from

b-actin–CFP mice. Cells were injected into the cerebellum of P3 pups using

a 1.0-mm capillary needle. Each pup received 1 � 104 prominin+lin� cells or

one neurosphere. After 17–26 d, cerebella were fixed, cryosectioned and stained

as described above.


ACKNOWLEDGMENTSThe authors thank M. Cook for flow cytometric analysis; E. Snyder and J.-P. Leefor assistance with intracranial injections; W. Salmon for help with microscopy;and A. Shetty, B. Barres and M. Rao for helpful discussions. R.W.R. is a KimmelFoundation Scholar. This research was supported by a McDonnell Foundation21st Century Award and by grant #1R01MH067916-01 from the US NationalInstitute of Mental Health.


Received 24 March; accepted 26 April 2005


1. Rapoport, M., van Reekum, R. & Mayberg, H. The role of the cerebellum in cognition andbehavior: a selective review. J. Neuropsychiatry Clin. Neurosci. 12, 193–198 (2000).

2. Altman, J. & Bayer, S.A. Development of the Cerebellar System: In Relation to ItsEvolution, Structure and Functions (CRC Press, Boca Raton, Florida, USA, 1997).

3. Rosenberg, R.N. & Grossman, A. Hereditary ataxia. Neurol. Clin. 7, 25–36 (1989).4. Kern, J.K. The possible role of the cerebellum in autism/PDD: disruption of a multi-

sensory feedback loop. Med. Hypotheses 59, 255–260 (2002).5. Martin, P. & Albers, M. Cerebellum and schizophrenia: a selective review. Schizophr.

Bull. 21, 241–250 (1995).6. Wechsler-Reya, R. & Scott, M.P. The developmental biology of brain tumors. Annu. Rev.

Neurosci. 24, 385–428 (2001).7. Wetmore, C. Sonic hedgehog in normal and neoplastic proliferation: insight gained from

human tumors and animal models. Curr. Opin. Genet. Dev. 13, 34–42 (2003).8. Zhang, L. & Goldman, J.E. Generation of cerebellar interneurons from dividing progeni-

tors in white matter. Neuron 16, 47–54 (1996).9. Hallonet, M.E., Teillet, M.A. & Le Douarin, N.M. A new approach to the development of

the cerebellum provided by the quail-chick marker system. Development 108, 19–31(1990).

10. Lumpkin, E.A. et al. Math1-driven GFP expression in the developing nervous system oftransgenic mice. Gene Expr. Patterns 3, 389–395 (2003).

11. Wechsler-Reya, R.J. & Scott, M.P. Control of neuronal precursor proliferation in thecerebellum by Sonic Hedgehog. Neuron 22, 103–114 (1999).

12. Cheng, Y., Tao, Y., Black, I.B. & DiCicco-Bloom, E. A single peripheral injection of basicfibroblast growth factor (bFGF) stimulates granule cell production and increasescerebellar growth in newborn rats. J. Neurobiol. 46, 220–229 (2001).

13. Wernecke, H., Lindner, J. & Schachner, M. Cell type specificity and developmentalexpression of the L2/HNK-1 epitopes in mouse cerebellum. J. Neuroimmunol. 9,115–130 (1985).

14. Theodosis, D.T., Rougon, G. & Poulain, D.A. Retention of embryonic features by an adultneuronal system capable of plasticity: polysialylated neural cell adhesion molecule inthe hypothalamo-neurohypophysial system. Proc. Natl. Acad. Sci. USA 88, 5494–5498(1991).

15. Tucker, R.P., Binder, L.I., Viereck, C., Hemmings, B.A. & Matus, A.I. The sequentialappearance of low- and high-molecular-weight forms of MAP2 in the developingcerebellum. J. Neurosci. 8, 4503–4512 (1988).

16. Sommer, I. & Schachner, M. Monoclonal antibodies (O1 to O4) to oligodendrocytesurfaces: an immunocytological study in the central nervous system. Dev. Biol. 83,311–327 (1981).

17. Dawson, M.R., Polito, A., Levine, J.M. & Reynolds, R. NG2-expressing glial progenitorcells: an abundant and widespread population of cycling cells in the adult rat CNS.Mol. Cell. Neurosci. 24, 476–488 (2003).

18. Bignami, A., Eng, L.F., Dahl, D. & Uyeda, C.T. Localization of the glial fibrillary acidicprotein in astrocytes by immunofluorescence. Brain Res. 43, 429–435 (1972).

19. Geisert, E.E., Jr., Yang, L. & Irwin, M.H. Astrocyte growth, reactivity, and the target of theantiproliferative antibody, TAPA. J. Neurosci. 16, 5478–5487 (1996).

20. Haegel, H., Tolg, C., Hofmann, M. & Ceredig, R. Activated mouse astrocytes and T cellsexpress similar CD44 variants. Role of CD44 in astrocyte/T cell binding. J. Cell Biol.122, 1067–1077 (1993).

21. Lendahl, U., Zimmerman, L.B. & McKay, R.D. CNS stem cells express a new class ofintermediate filament protein. Cell 60, 585–595 (1990).

22. Corbeil, D., Roper, K., Fargeas, C.A., Joester, A. & Huttner, W.B. Prominin: a story ofcholesterol, plasma membrane protrusions and human pathology. Traffic 2, 82–91(2001).

23. Graham, V., Khudyakov, J., Ellis, P. & Pevny, L. SOX2 functions to maintain neuralprogenitor identity. Neuron 39, 749–765 (2003).

24. Sakakibara, S. et al. Mouse-Musashi-1, a neural RNA-binding protein highly enriched inthe mammalian CNS stem cell. Dev. Biol. 176, 230–242 (1996).

25. Sawamoto, K. et al. Generation of dopaminergic neurons in the adult brain frommesencephalic precursor cells labeled with a nestin-GFP transgene. J. Neurosci. 21,3895–3903 (2001).

26. Tamaki, S. et al. Engraftment of sorted/expanded human central nervous system stemcells from fetal brain. J. Neurosci. Res. 69, 976–986 (2002).

27. Reynolds, B.A. & Weiss, S. Generation of neurons and astrocytes from isolated cells ofthe adult mammalian central nervous system. Science 255, 1707–1710 (1992).

28. Imura, T., Kornblum, H.I. & Sofroniew, M.V. The predominant neural stem cell isolatedfrom postnatal and adult forebrain but not early embryonic forebrain expresses GFAP.J. Neurosci. 23, 2824–2832 (2003).

29. Takahashi, J., Palmer, T.D. & Gage, F.H. Retinoic acid and neurotrophins collaborate toregulate neurogenesis in adult-derived neural stem cell cultures. J. Neurobiol. 38,65–81 (1999).

30. Galli, R., Pagano, S.F., Gritti, A. & Vescovi, A.L. Regulation of neuronal differentiation inhuman CNS stem cell progeny by leukemia inhibitory factor. Dev. Neurosci. 22, 86–95(2000).

31. Erlandsson, A., Enarsson, M. & Forsberg-Nilsson, K. Immature neurons from CNS stemcells proliferate in response to platelet-derived growth factor. J. Neurosci. 21, 3483–3491 (2001).

32. Aoki, E., Semba, R. & Kashiwamata, S. New candidates for GABAergic neurons in the ratcerebellum: an immunocytochemical study with anti-GABA antibody. Neurosci. Lett.68, 267–271 (1986).

33. Simmons, M.L. & Dutton, G.R. Neuronal origins of K+-evoked amino acid release fromcerebellar cultures. J. Neurosci. Res. 31, 646–653 (1992).

34. Maricich, S.M. & Herrup, K. Pax-2 expression defines a subset of GABAergic inter-neurons and their precursors in the developing murine cerebellum. J. Neurobiol. 41,281–294 (1999).

35. Aruga, J. et al. A novel zinc finger protein, zic, is involved in neurogenesis, especially inthe cell lineage of cerebellar granule cells. J. Neurochem. 63, 1880–1890 (1994).

36. Shetty, A.K. & Turner, D.A. In vitro survival and differentiation of neurons derived fromepidermal growth factor-responsive postnatal hippocampal stem cells: inducing effectsof brain-derived neurotrophic factor. J. Neurobiol. 35, 395–425 (1998).

37. Laywell, E.D., Rakic, P., Kukekov, V.G., Holland, E.C. & Steindler, D.A. Identification of amultipotent astrocytic stem cell in the immature and adult mouse brain. Proc. Natl.Acad. Sci. USA 97, 13883–13888 (2000).

38. Singh, S.K. et al. Identification of a cancer stem cell in human brain tumors. Cancer Res.63, 5821–5828 (2003).

39. Gabay, L., Lowell, S., Rubin, L.L. & Anderson, D.J. Deregulation of dorsoventralpatterning by FGF confers trilineage differentiation capacity on CNS stem cells invitro. Neuron 40, 485–499 (2003).

40. Gao, W.Q. & Hatten, M.E. Immortalizing oncogenes subvert the establishment of granulecell identity in developing cerebellum. Development 120, 1059–1070 (1994).

41. Snyder, E.Y. et al. Multipotent neural cell lines can engraft and participate in develop-ment of mouse cerebellum. Cell 68, 33–51 (1992).

42. Milosevic, A. & Goldman, J.E. Progenitors in the postnatal cerebellar white matter areantigenically heterogeneous. J. Comp. Neurol. 452, 192–203 (2002).

43. Alder, J., Cho, N.K. & Hatten, M.E. Embryonic precursor cells from the rhombic lip arespecified to a cerebellar granule neuron identity. Neuron 17, 389–399 (1996).

44. Zinyk, D.L., Mercer, E.H., Harris, E., Anderson, D.J. & Joyner, A.L. Fate mapping of themouse midbrain-hindbrain constriction using a site-specific recombination system.Curr. Biol. 8, 665–668 (1998).

45. Pietsch, T. et al. Medulloblastomas of the desmoplastic variant carry mutations of thehuman homologue of Drosophila patched. Cancer Res. 57, 2085–2088 (1997).

46. Pomeroy, S.L. et al. Prediction of central nervous system embryonal tumour outcomebased on gene expression. Nature 415, 436–442 (2002).

47. Katsetos, C.D. et al. Calbindin-D28k in subsets of medulloblastomas and in the humanmedulloblastoma cell line D283 Med. Arch. Pathol. Lab. Med. 119, 734–743 (1995).

48. Hemmati, H.D. et al. Cancerous stem cells can arise from pediatric brain tumors. Proc.Natl. Acad. Sci. USA 100, 15178–15183 (2003).

49. Weigmann, A., Corbeil, D., Hellwig, A. & Huttner, W.B. Prominin, a novel microvilli-specific polytopic membrane protein of the apical surface of epithelial cells, is targetedto plasmalemmal protrusions of non-epithelial cells. Proc. Natl. Acad. Sci. USA 94,12425–12430 (1997).

50. Groszer, M. et al. Negative regulation of neural stem/progenitor cell proliferation by thePten tumor suppressor gene in vivo. Science 294, 2186–2189 (2001).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

XTRPC1-dependent chemotropic guidance of neuronalgrowth cones

Sangwoo Shim1,2, Eyleen L Goh1,2, Shaoyu Ge1,2, Kurt Sailor1,2, Joseph P Yuan3, H Llewelyn Roderick4,Martin D Bootman4, Paul F Worley3, Hongjun Song1,2,3, & Guo-li Ming1,2,3

Calcium arising through release from intracellular stores and from influx across the plasma membrane is essential for signalling

by specific guidance cues and by factors that inhibit axon regeneration. The mediators of calcium influx in these cases are largely

unknown. Transient receptor potential channels (TRPCs) belong to a superfamily of Ca21-permeable, receptor-operated channels

that have important roles in sensing and responding to changes in the local environment. Here we report that XTRPC1, a Xenopus

homolog of mammalian TRPC1, is required for proper growth cone turning responses of Xenopus spinal neurons to microscopic

gradients of netrin-1, brain-derived neurotrophic factor and myelin-associated glycoprotein, but not to semaphorin 3A.

Furthermore, XTRPC1 is required for midline guidance of axons of commissural interneurons in the developing Xenopus spinal

cord. Thus, members of the TRPC family may serve as a key mediator for the Ca21 influx that regulates axon guidance during

development and inhibits axon regeneration in adulthood.

Axons are guided to their targets in the developing nervous system bydiffusible and bound cues that either attract or repel the growing tip ofan axon, the growth cone1–4. Once neuronal circuits are formed, theability of axons to regenerate after injury or disease in the maturemammalian CNS is extremely limited, largely because of the presenceof inhibitory factors5–10. During the last decade, an array of evolutio-narily conserved guidance cues and their receptors have been identi-fied and have been shown to be involved in axon guidance duringdevelopment and in axon regeneration in the adult central nervoussystem1–3,6. How specific guidance signals are transduced from receptoractivation to rearrangement of cytoskeleton within the neuronalgrowth cone is just beginning to be elucidated1–4,11,12. One of theinitial signal transduction mechanisms that triggers the growth coneresponses to a large number of guidance cues and inhibitory factorsassociated with myelin is an increase in cytosolic Ca2+, arising fromboth intracellular store release and influx across the plasma mem-brane5,7,13–17. Notably, different patterns of Ca2+ elevation triggerdifferential attractive and repulsive turning responses of the growthcone13,18. While previous studies have demonstrated the essential roleof Ca2+ signalling in axon guidance and regeneration both in vitro andin vivo5,7,11–13,15–24, it is unknown how initial Ca2+ mobilization fromintracellular stores is coupled to sustained Ca2+ influx from theextracellular environment to regulate growth cone guidance11,12.

Members of the TRPC family (TRPC1–TRPC7 in mammals) arenon-selective cation channels; thus, they are good candidates to mediatesustained Ca2+ influx25,26. For example, TRPC3 and TRPC1 mediateCa2+ influx induced by brain-derived neurotrophic factor (BDNF)27

and by activation of metabotropic glutamate receptor mGluR1 (ref. 28),respectively. Notably, TRPC5 regulates growth cone morphology andneurite extension in cultured hippocampal neurons29,30. XTRPC1, ahomolog of mammalian TRPC1, is the only TRPC family membercloned so far from Xenopus31,32. In this study, with morpholino-mediated gene knockdown, expression of dominant-negative channelsand pharmacological inhibitors, we examined the role of XTRPC1 inXenopus spinal neuron growth cone turning responses to gradients ofspecific guidance cues and inhibitory factors for axon regeneration,including netrin-1, BDNF, myelin-associated glycoprotein (MAG) andsemaphorin 3A (Sema3A). Furthermore, we examined the role ofXTRPC1 in the CNS midline axon guidance of commissural interneur-ons in developing Xenopus spinal cord in vivo, a classic example oflong-range axon guidance by netrin-1 (refs. 33–38). Our in vitro andin vivo studies provide strong evidence that XTRPC1 is required by aselect group of guidance cues for signalling. Two recent independentin vitro studies have also demonstrated roles for TRPC1 and TRPC3 inattractive growth cone responses to netrin-1 and BDNF39,40. Togetherwith recent findings on the roles of TRPs in various biologicalbehaviors25,26,41, these studies support the view that TRP superfamilymembers function as general cellular sensors for extracellular environ-ments, including sensing guidance cues during neuronal pathfinding.

RESULTS

Expression of XTRPC1 at growth cones of spinal neurons

We first examined the expression of XTRPC1 in Xenopus spinalneurons with previously characterized antibodies31. We found that


1Institute for Cell Engineering, 2Department of Neurology, Johns Hopkins University School of Medicine, 733 N. Broadway, Baltimore, Maryland 21205, USA. 3Departmentof Neuroscience, Johns Hopkins University School of Medicine, 725 N. Wolfe Street, Baltimore, Maryland 21205, USA. 4Laboratory of Molecular Signalling, The BabrahamInstitute, Babraham, Cambridge CB2 4AT, UK. Correspondence should be addressed to G.L.M. ([email protected]).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

XTRPC1 was expressed in the majority of Xenopus spinal neurons inthe culture (Fig. 1a). Notably, we observed enriched staining ofXTRPC1 at growth cones of some neurons (Fig. 1a). When injectedinto the early Xenopus embryos, a morpholino specific for XTRPC1,but not a control morpholino, substantially reduced the expressionlevel of XTRPC1 in cultured spinal neurons derived from injectedembryos, as shown by immunocytochemistry (Fig. 1b,c,e), and inwhole embryos, as shown by western blot (Fig. 1d,e; see Methods).

XTRPC1 in netrin-1–induced growth cone attraction

To examine the functional role of XTRPC1 in axon guidance, we used awell-established in vitro growth cone turning assay13–17. In a microscopicgradient of netrin-1 (5 mg/ml in the pipette), Xenopus spinal neuronsdemonstrated chemoattractive turning responses under normal condi-tions (Fig. 1f). When spinal neurons derived from embryos injectedwith the XTRPC1 morpholino were exposed to the same netrin-1gradient, the attraction was completely abolished (Figs. 1g and 2).Instead, we observed significant repulsive turning responses (Figs. 1gand 2b). However, a control morpholino had no effect on netrin-1–

induced attraction (Fig. 2c). As the morpholino treatment did notcompletely eliminate the expression of XTRPC1 (Fig. 1e), these resultssuggest that the expression level of XTRPC1 is critical for the propergrowth cone responses to a netrin-1 gradient. XTRPC1 shows a highdegree of homology to the mammalian TRPC1, especially at theproposed pore region (Fig. 2a). We found that netrin-1–inducedattraction was partially rescued when mRNAs encoding wild-typemammalian TRPC1 were co-injected with the XTRPC1 morpholino(Fig. 2c), further demonstrating the role of TRPC1 in growth coneturning to netrin-1. TRPCs with specific mutations at the pore regionhave been previously shown to block endogenous channel opening andact in a dominant negative fashion28,30. We found that expression of amutant form of mammalian TRPC1, TRPC1(F562A)28, convertednetrin-1–induced attraction into repulsion (Figs. 1h and 2c), verysimilar to the effect of XTRPC1 morpholino treatment. The neuriteextension rate in a netrin-1 gradient was not statistically different underthese conditions (Supplementary Fig. 1). When neurons were culturedin the absence of guidance cues, neurite length was also not significantlydifferent under different conditions (Supplementary Fig. 1). In addition

XTRPC1

XTRPC1

U

U CM (10) XM (10)

CM10

XTRPC1 Tracer Merged

XM102

150

100

Exp

ress

ion

leve

ls (

%)

50

0

ImmunostainingWestern blot

Nor

mal

DN

-TR

PC

1X

TR

PC

1m

orop

holin

oXM

CM 0 min 30 min

0 min 30 min

0 min 30 min

a b f

g

h

cd

e

Figure 1 Requirement of XTRPC1 for growth cone turning responses to a gradient of netrin-1. (a) Sample confocal images of XTRPC1 staining in a cultured

Xenopus spinal neuron (left) and an enlarged view of its growth cone (right). Scale bar: 40 mm (left); 15 mm (right). (b–e) Knockdown expression of XTRPC1 by

a specific morpholino. Sample confocal images show XTRPC1 staining (green) and a tracer (red) in cultured Xenopus spinal neurons derived from embryos

injected with the XTRPC1 morpholino (XM, b) and a control morpholino (CM, c). Scale bar: 40 mm. A sample western blot (d) represents endogenous XTRPC1

in uninjected embryos (U), embryos injected with CM (10 ng per embryo), or embryos injected with XM (2 ng and 10 ng per embryo) at the one-cell stage.Data in e are from western blot of whole-embryo membrane preparations and immunostaining of growth cones of cultured spinal neurons (mean 7 s.d.; n ¼ 4

separated experiments). (f–h) Growth cone turning responses (left, center) in a gradient of netrin-1 of Xenopus spinal neurons derived from normal embryos (f),

embryos injected with the XTRPC1 morpholino (g) or embryos injected with mRNA encoding a dominant-negative (DN) mutant form of mammalian TRPC1

(h) at the start (0 min) and the end (30 min) of exposure to a netrin-1 gradient (5 mg/ml in the pipette). Superimposed trajectories of neurite extension during

the 30-min period for a sample population of ten neurons are shown at right. Arrows indicate direction of gradient. Scale bar: 10 mm.

100

Cum

ulat

ive

dist

ribut

ion

(%)

Mea

n tu

rnin

g an

gle

(deg

)

75

50

25

25

10

0

Nor

mal

Con

trol

mor

ph.

XT

RP

C1

mor

ph.

+ W

T-T

RP

C1

mR

NA

XT

RP

C1

mor

ph.

DN

-TR

PC

1

SK

F96

365

(2 µ

M)

LaC

l 3 (

10 µ

M)

–20

–10

0

–60 –40 –20 0Turning angle (deg)

20 40 60

Normal

DN-TRPC1

XTRPC1morpholino

Repulsion Attraction

** *

*

TRPC1TFHSF I GTCFALFWYIFSLAHVTIFVXTRPC1TFHSFTGTCYALFWYIFSLAHVTLFV

†

a c

b

Figure 2 Summary of growth cone turning responses to a gradient of netrin-1.

(a) TRPCs are highly conserved in the proposed pore region of the channel.

The amino acids in boldface are identical in XTRPC1 and mammalian

TRPC1–TRPC7. (b) Growth cone turning in a gradient of netrin-1 in normal

neurons or in neurons derived from embryos injected with the XTRPC1

morpholino or mRNA encoding DN-TRPC1. Shown are cumulative

distributions of turning angles from individual experiments under various

conditions. The mean turning angles are shown below. (c) Summary of growth

cone turning angles under various conditions (mean 7 s.e.m.; n ¼ 15–29).

* indicates statistically significant difference from normal neurons, and cross

indicates significant difference between two conditions in bracket (P o 0.01,

bootstrap test).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

to genetic manipulation, SKF96365 (2 mM) and LaCl3 (10 mM), twocommonly used inhibitors for TRPCs and other Ca2+ channels, alsoabolished attractive growth cone responses to a gradient of netrin-1(Fig. 2c). Taken together, these results demonstrate that XTRPC1 isrequired for the attractive responses by Xenopus spinal neuron growthcones to a netrin-1 gradient in vitro. Previous studies have shown that anelevation of cytoplasmic Ca2+ owing to intracellular store release andinflux from the extracellular environment are both required for propergrowth cone turning responses to netrin-1 (refs. 13,15). Removal ofextracellular Ca2+ converts growth cone attraction induced by focallaser–induced photolysis of caged Ca2+ to repulsion18. Thus, our resultsare consistent with the notion that XTRPC1 is an essential component ofthe cellular machinery mediating Ca2+ influx in response to netrin-1.

XTRPC1 in formation of commissural interneuron axon tracts

We next examined whether XTRPC1 is also required for CNS midlineguidance of commissural interneuron axons in vivo, a classic exampleof netrin-1–dependent long-range growth cone guidance1,3,37,38. Dis-ruption of netrin-1–dependent signalling has been shown to result insignificant defects in the formation of the contralateral longitudinalcommissural interneuron axon tracts in developing spinal cords inmultiple model systems34,36,37,42,43. To directly compare the axon tractsof neurons derived from the injected and uninjected blastomere in thesame embryo, we injected the XTRPC1 morpholino into one of the twoblastomeres at the two-cell stage, together with a lineage tracer (Fig. 3a;see Methods). In the developing Xenopus spinal cord, commissuralinterneurons send axons ventrally to cross the midline, where they

Figure 3 XTRPC1 is required for formation of

longitudinal axonal tracts by commissural

interneurons in developing Xenopus spinal cord.

(a) Microinjection at the two-cell stage allows

specific manipulation of half of the embryo, with

the uninjected half serving as the internal control.

(b) Diagrams of commissural interneurons in

developing Xenopus spinal cord (adapted fromref. 44). CI: commissural interneuron cell body;

CT: commissural interneuron axon tracts; FP: floor

plate. (c–e) Longitudinal axon tracts of

commissural interneurons in developing Xenopus

spinal cord from stage 28–30 embryos. Confocal

z-stack projections with 3A10 staining of

commissural interneuron axons from an

uninjected embryo (c), embryos injected with the

XTRPC1 morpholino (XM; d, left) or a control

morpholino (CM; e, left), together with fixable

FITC-dextran (green). Dashed lines indicate

midline. Scale bar: 100 mm. Lower images:

optical projections of the coronal section of

ventral spinal cord. Images at right: larger portion

of spinal cords with 3A10 staining of commissural axons (red) and the side of injection (green) at a smaller scale. Scale bar: 250 mm. (f) Comparison of

commissural axons on uninjected and injected sides. Optical projections of coronal sections shown in c–e were used for quantification (see Methods) for

normal embryos or for embryos unilaterally injected with CM, XM, or XM and mRNAs encoding mammalian TRPC1 at the two-cell stage (mean 7 s.e.m.; n ¼ 6

to 9). * indicates statistically significant difference from non-injected control (P o 0.01, bootstrap test).

Morpholino +lineage tracer

Ventral view Coronal view

Two-cell stage

XTRPC1 morpholino (XM) Control morpholino (CM)

Rat

io o

f com

issu

ral a

xons

(uni

njec

ted/

inje

cted

sid

e)

Developingembryo

CI

CT CT

CT

FP

CT

CICI

CI

FP

Injectedside

Uninjectedside

1.25

1.25

0.75

0.50

0.25

0.00

Norm

al CM XM

XM +

TRPC1

*

a

d e f

b c

Midline

Uninjected

CT

CI

UninjectedControl morpholino

injected

XTRPC1 morpholinoinjected 1.5

125Injected sideUninjected side

100

75

50

25

0

1.0

0.5

0.0

Contro

l

mor

pholi

no

Contro

l

mor

pholi

no

XTRPC1

mor

pholi

no

XTRPC1

mor

pholi

no

Rat

io o

f 3A

10+

neu

rons

(inje

cted

/uni

njec

ted

side

)

Per

cent

age

ofcr

osse

d ax

ons

XTRPC1 morpholinoinjected

Spi

nal c

ord

Not

ocor

d

*

a

d e f

b cFigure 4 XTRPC1 is required for attraction of commissural interneuron axons

to the CNS midline. (a) Schematic sagittal view of commissural interneurons

and their axons in the developing Xenopus spinal cord. CT: contralateral

commissural interneuron axon. (b–d) Commissural interneuron projections in

stage 23–25 embryos. Confocal z-stack projections of 3A10 staining of the

uninjected and injected side within the same segment of the spinal cord, in

embryos unilaterally injected with a control morpholino (b) or the XTRPC1

morpholino (d) at the two-cell stage. Scale bar: 40 mm. Cell bodies of

commissural interneurons show 3A10 staining (*). Different phenotypes of

trajectories of commissural interneurons based on three-dimensional

reconstruction of z-stack confocal images: crossed midline (filled arrows);

stalled or turned back from midline (open arrows); joined the ipsilateral (IL)tract (filled triangles); prematurely branched before crossing midline (open

triangles). (e) Comparison of the density of 3A10-positive commissural

interneurons derived from the injected and uninjected side within the same

segment of the developing spinal cord (mean 7 s.d.; n ¼ 9). (f) Summary

of phenotypes of midline axon guidance behavior of 3A10-positive

commissural interneurons in embryos unilaterally injected with the XTRPC1

morpholino or with control morpholino at the two-cell stage (mean 7 s.e.m.;

n ¼ 10–23). * indicates statistically significant difference (P o 0.01,

bootstrap test).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

usually branch to ascend and descend in the marginal zone (Fig. 3b;ref. 44). We used the monoclonal antibody 3A10 to identify axonsof commissural interneurons37,45. Whole-mount staining of stage28–30 Xenopus embryos with 3A10 reveals two longitudinal axontracts of commissural interneurons on each side of the midline as wellas crossing axons within the developing spinal cord (Fig. 3c). Sincethese commissural interneurons have only contralateral projections44,the fibers on the contralateral side of the midline reflect those that hadcrossed the midline (Fig. 3b). We found that contralateral longitudinalaxons of commissural interneurons derived from the XTRPC1 mor-pholino–injected blastomere were significantly less thick than thosefrom the uninjected blastomere in the same embryo in a ventral view(Fig. 3d,f and Supplementary Video 1). We also observed thisreduction in sagittal views of 3A10-positive fibers from the XTRPC1morpholino–injected side, compared with controls, in the same seg-ment of the spinal cord (Supplementary Fig. 2). Injection of a controlmorpholino, however, had no effect (Fig. 3e,f and SupplementaryVideo 2). In addition, injection of mRNA encoding wild-type mam-malian TRPC1 together with the XTRPC1 morpholino largely rescuedthe formation of contralateral longitudinal commissural interneuronaxons from the injected side (Fig. 3f). Thus, XTRPC1 is required forproper formation of contralateral longitudinal commissural axon tractsin vivo. To further examine whether observed defects on commissuralaxon tracts is due to a reduction of the number of commissuralinterneurons, we examined sagittal views of 3A10-positive neurons atboth sides of the spinal cord of embryos injected unilaterally at the two-cell stage (Supplementary Fig. 2). We found that the density ofcommissural interneurons labelled with 3A10 in the injected embryoswas very similar to that previously reported at same developmentalstages using different markers for commissural interneurons46. Moreimportantly, we did not observe any significant difference in thenumber of 3A10-positive neurons between the uninjected side andthe side injected with the XTRPC1 morpholino within the samesegment of spinal cord (Supplementary Fig. 2).

XTRPC1 in attracting commissural axons to the CNS midline

To directly examine whether interfering with XTRPC1 functions leadsto failure to attract commissural axons to the CNS midline in vivo,we examined spinal cords from a sagittal view of stage 23–25embryos when commissural interneuron axons just crossed the midlineand major tracts started to form (Fig. 4a). We found that almostall 3A10-positive commissural axons derived from uninjected blasto-meres or from blastomeres injected with the control morpholinocrossed the midline at this stage (Fig. 4b,d,f). In contrast, a significantpercentage of 3A10-positive commisural axons derived from blasto-meres injected with the XTRPC1 morpholino failed to cross themidline (35 7 3.7%, n ¼ 23 embryos; P o 0.01; Fig. 4f). Out of459 commissural interneurons derived from 23 embryos unilaterallyinjected with the XTRPC1 morpholino (Fig. 4c,d), some of them

branched prematurely (4.8 7 1.2%) or joined the ipsilateral axonaltract (13.9 7 2.2%), and others seemed to stall or even turn back fromthe midline (18.7 7 2.8%). We did not observe a significant differencebetween the injected and uninjected side in the number of 3A10-positive neurons within the same segment of spinal cord (Fig. 4e) or inthe intensity of 3A10 staining of labelled axons (Fig. 4). Taken together,these results suggest that XTRPC1 is essential for midline axonguidance of commissural interneurons, a classic example of netrin-1–dependent axon guidance37,38.

XTRPC1 as a common mediator for a group of guidance cues

To further examine whether XTRPC1 serves as a common mediatorfor multiple guidance factors, we examined the turning responsesof Xenopus spinal neurons to microscopic gradients of MAGand Sema3A, both of which have been implicated in inhibition ofaxon regeneration in the adult CNS3,6,8,10. Xenopus spinal neuronsdemonstrated repulsive turning responses to a gradient of MAG(150 mg/ml in the pipette; Fig. 5a)14,16,47 and to a gradient ofSema3A (50 mg/ml in the pipette; Fig. 5b)16,47. We found thatinterfering with XTRPC1 signalling, either by injection of XTRPC1morpholino or by injection of mRNA encoding DN-TRPC1, comple-tely abolished MAG-induced repulsion (Fig. 5a,c). In contrast,Sema3A-induced repulsion was not significantly affected (P 4 0.05)in neurons injected with the XTRPC1 morpholino (Fig. 5b,c). Wehave previously shown that growth cone responses of Xenopusspinal neurons to a gradient of netrin-1, MAG or BDNF, but notto a gradient of Sema3A, are abolished when extracellular Ca2+ arereduced from 1 mM to 1 mM, indicating a differential requirement forCa2+ influx in the signalling events triggered by these two groups offactors16. Furthermore, we found that attraction to a BDNF gradient(50 mg/ml in the pipette) was also abolished in neurons derived fromXTRPC1 morpholino–injected blastomeres (Fig. 5c). Thus, our resultssuggest that XTRPC1 may be a key mediator of Ca2+ influx for a selectgroup of guidance cues, including factors that inhibit axon regenera-tion in the adult CNS.

MAG gradient Sema3A gradient

Con

trol

XT

RP

C1

mor

phol

ino

0 min 30 min

0 min 30 min

0 min 30 min

0 min 30 min

Normal XTRPC1 morpholino DN-TRPC1

Repulsion Attraction

No gradient

MAG gradient(150 µg/ml)

BDNF gradient(50 µg/ml)

Sema3A gradient(50 µg/ml)

*

*

*

✝✝

✝

**

–30 –20 –10 0

Mean turning angle (deg)

10 20 30

a

c

bFigure 5 XTRPC1 as a general mediator for Ca2+-dependent neuronal growth

cone guidance. (a) Growth cone turning in a gradient of MAG. Shown are

sample images of growth cones at the start (0 min) and the end (30 min)

of exposure to a gradient of MAG (150 mg/ml in the pipette) for a normal

neuron (top) and a neuron derived from embryos injected with the XTRPC1

morpholino (bottom). Scale bar: 20 mm. (b) Growth cone turning responses in

a gradient of Sema3A. Similar to a, except that a Sema3A gradient was used

(50 mg/ml in the pipette). (c) Summary of mean turning angles underdifferent conditions (mean 7 s.e.m.; n ¼ 15–30). * indicates statistically

significant difference from the control (no gradient control), and crosses

indicate statistically significant difference between conditions in brackets

(P o 0.01, bootstrap test).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

DISCUSSION

Our study provides both in vivo and in vitro evidence that XTRPC1 isrequired for guiding neuronal growth cones in response to specificguidance cues. Early studies have shown that patterns of Ca2+ elevation,(for example, initial Ca2+ transients and basal levels of cytoplasmicCa2+) regulate the directionality of growth cone turning13,18. TRPCs arenon-selective cation channels, and they can contribute to the influxof Ca2+ upon stimulation by specific guidance cues at least in twoways (Supplementary Fig. 3). First, TRPCs are Ca2+-permeable chan-nels, thus allowing direct influx of Ca2+. Second, activation ofTRPCs also leads to influx of Na+ and subsequent membrane depolar-ization, which may activate voltage-gated Ca2+ channels and resultin further Ca2+ influx13. Indeed, recent studies have shown that netrin-1can induce TRPC-like currents in neuronal growth cones andsubsequent membrane depolarization of these neurons in culture40.Interfering with XTRPC1 functions would substantially reducethe influx of extracellular Ca2+ after initial release of Ca2+ fromintracellular stores, resulting in a smaller amplitude of Ca2+ transientsand a lower level of global cytoplasmic Ca2+. Consequently, attractivegrowth cone responses to guidance cues such as netrin-1 and BDNF areabolished40 or turned into repulsion (Figs. 2c and 5c) depending on thelevel of XTRPC1 knockdown. Two recent independent studies haveshown that TRPC3 mediates Ca2+ influx in cerebellar granule neuronsand is required for BDNF-induced attraction, and that XTRPC1 isrequired for intracellular Ca2+ elevation and attraction induced bynetrin-1 (ref. 40). In the current study, we have used morpholino-mediated knockdown, expression of dominant-negative mutants andpharmacological inhibitors to interfere with XTRPC1 signalling. Weshow that XTRPC1 is not only involved in attractive signalling tonetrin-1 and BDNF but is also differentially required for signalling oftwo repulsive guidance cues (MAG and Sema3A; Fig. 5c). An emergingview from all of these in vitro findings is that different members of theTRPC family may be widely involved in different guidance behaviors ofneuronal growth cones.

Netrin-1–dependent signalling is essential for midline axon guidanceof commissural interneurons34,36,37,42,43. We provide the first in vivoevidence that XTRPC1 is involved in midline axon guidance ofcommissural interneurons in developing Xenopus spinal cord. Instage 28–30 embryos, we have observed a significant reduction incontralateral longitudinal commissural interneuron axon tracts derivedfrom the side injected with the XTRPC1 morpholino than were derivedfrom the uninjected side within the same spinal cords (Fig. 3 andSupplementary Fig. 2). This defect in the formation of contralateralcommissural interneuron axon tracts is due to a failure of attractingsome of the axons to the CNS midline. In the developing spinal cord ofstage 23–25 embryos, we found that a significant number of commis-sural axons derived from blastomeres injected with the XTRPC1morpholino did not reach the CNS midline (Fig. 4). We did not findany difference in the number of commissural interneurons with theXTRPC1 morpholino injection at the stages we examined (Fig. 3e;Supplementary Fig. 2e). Future mosaic analysis will be needed toprove that the observed defects are stringently cell-autonomous in vivo.As axons derived from the uninjected side can project across themidline normally and form the contralateral commissural axon tracts,a general defect of midline cells owing to the XTRPC1 morpholinotreatment is unlikely. The pathfinding defects of commissural inter-neurons observed following XTRPC1 morpholino treatment, however,are likely to result from a disruption of growth cone responses tomultiple guidance cues involved in midline guidance, including netrin-1. Our previous studies have shown that the turning responses to groupI cues (including netrin-1, MAG and BDNF) are very sensitive to

manipulation of extracellular Ca2+, whereas responses to group II cues(including Sema3A and NT-3) are not affected by the same manipula-tion15,16,47,48. Our current results suggest that XTRPC1 is involved inCa2+ signalling for group I cues. Since multiple inhibitors associatedwith the CNS myelin are likely to belong to group I cues47, inhibition ofthe function of TRPCs may potentially promote axon regeneration inthe injured adult CNS.

How XTRPC1 is activated by netrin-1, BDNF and MAG remains tobe determined. Specific guidance cues can induce tyrosine phosphory-lation of their receptors48,49, providing docking sites for proteins, suchas phospholipase C-g (PLC-g), that contain SH2 domains (Supple-mentary Fig. 3). Together with the requirement of PLC-g in growthcone turning39,48 and activation of TRPCs by PLC25,26, our resultssuggest a model in which guidance cues or inhibitors for axonregeneration activate PLC, which in turn leads to release of intracellularstores of Ca2+ through the IP3 receptor and an influx of extracellularCa2+ through TRPCs (Supplementary Fig. 3). Thus, members of theTRP superfamily may serve as a key component of the cellularmachinery in sensing and responding to changes in the local environ-ment in general, including pathfinding of neuronal growth cones inresponse to specific guidance cues.

METHODSXenopus embryo injection, cell culture, immunocytochemistry and bio-

chemistry. Blastomere injection of morpholino oligonucleotides and mRNAs

encoding mammalian TRPC1 or its mutant forms into early stages of Xenopus

embryos were performed as previously described48,49. Specifically, fertilized

embryos were injected with a mixture of the morpholino (2 or 10 ng/embryo)

or mRNA (2–3 ng/embryo), and a lineage tracer (fixable FITC-dextran or

rhodamine-dextran, 15 mg/ml), at the one- or two-cell stage. A morpholino

oligo specific for XTRPC1 was designed with the following sequence:

5¢-GAGCAGCCATGATGACAGAACTCCC-3¢. A control morpholino was used

with the following sequence: 5¢-CCTCTTACCTCAGTTACAATTTATA-3¢.Injected embryos at stage 22 were used for spinal neuron cultures as previously

described48,49. For immunocytochemistry, cultured cells derived from embryos

unilaterally injected at the two-cell stage were processed as previously

described48,49. Polyclonal antibodies against two different regions of XTRPC1

have been previously characterized31 and were used at a dilution of 1:100 for

immunocytochemistry of cell cultures and at a dilution of 1:1,000 for western

blot. Similar results were obtained with both antibodies. Secondary antibodies

were used at a dilution of 1:250. A majority of neurons (80–90%) were stained

for XTRPC1 in culture. Images of cultured neurons were captured with a Zeiss

510 confocal microscope using the same settings for different conditions, and

fluorescence intensities in a defined box covering only the neuronal growth

cone regions were quantified. Twenty stage 26 uninjected embryos or embryos

injected exclusively at the one-cell stage, with either a control morpholino or

the XTRPC1 morpholino, were used for membrane fractions and western blot

analysis as described31,50. The total protein amount was measured, and an

equivalent amount of proteins was loaded.

Growth cone turning assay. Microscopic gradients of guidance factors were

generated using recombinant netrin-1 (5 mg/ml), MAG (150 mg/ml), BDNF

(50 mg/ml) or Sema3A (50 mg/ml), respectively, in the micropipette as previously

described15,16,47,48. Neurons with the lineage-tracer were identified by fluores-

cence microscopy 14–20 h after plating and were used for the turning assay at

22–25 1C as described48,49. The turning angle was defined by the angle between

the original direction of neurite extension and a straight line connecting the

positions of the center of the growth cone at the onset and the end of the 30

min period. The rates of neurite extension were calculated based on the net

neurite extension during the turning assay (Supplementary Fig. 1). Only

growth cones of isolated neurons with a net neurite extension 45 mm over the

30-min period were included for turning analysis. Neurite lengths of neurons

under different conditions cultured without addition of any guidance factors

were also measured and quantified (Supplementary Fig. 1). Statistical sig-

nificance (P o 0.05) was assessed using bootstrap test as indicated.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Whole-mount immunocytochemistry and confocal imaging. Normal

embryos and embryos unilaterally injected at the two-cell stage were processed

for whole-mount immunocytochemistry when they reached stage 23–25 or

28–30 as previously described20. Monoclonal antibody 3A10 (Developmental

Studies Hybridoma Bank at the University of Iowa), specific for commissural

interneurons37,45, was used at a dilution of 1:100. Secondary antibodies were

used at a dilution of 1:250. For ventral views of spinal cords, individual spinal

cords were dissected out from Xenopus embryos at stage 28–30 and mounted

on coverslips for imaging. Confocal images were taken with a Zeiss LSM 510

META system, and z-series reconstructions were processed with the Zeiss LSM

image acquisition program. Three-dimensional (3D) projections were rotated

with this software to ensure a direct ventral view of the spinal cord as shown in

Fig. 3c–e and sample videos of rotating 3D images were also generated as

shown in supplementary videos. Optical coronal section views of spinal cords

were made from original z-stack confocal images (Fig. 3c–e) and were then used

to measure mean area pixel intensities within a defined box containing either left

or right longitudinal axon tracts for quantification (Fig. 3f). A minimum of

three optical sections were made and measured for each sample, and a minimal

of ten embryos were examined for each condition. For sagittal views of spinal

cord, individual spinal cords were dissected out from stage 23–25 (Fig. 4) or 28–

30 (Supplementary Fig. 2) embryos and were mounted within two layers of

coverslips to allow imaging the same segment of the spinal cord from both sides

for direct comparison. The density of 3A10-positive cells differs along the

developing spinal cord. Thus, we directly compared the number of 3A10-positive

cells in injected and uninjected sides within the same segment of developing

spinal cord for each condition.


ACKNOWLEDGMENTSWe would like to thank A. Kolodkin, C. Montell and T. Dawson for criticalcomments, N. Marsh-Armstrong, L.N. Borodinsky and N.C. Spitzer for theirhelp during this study, and L. Liu for her technical support. This work wassupported by National Institute of Neurological Disorders and Stroke,Charles E. Culpeper Scholarships in Medical Science, Whitehall Foundation,and Basal O’Connor Starter Scholar Research Award Program to G.L.M. S.S.is partially supported by a postdoctoral fellowship from the Korea Scienceand Engineering Foundation. H.L.R and M.D.B. would like to gratefullyacknowledge support from the Biotechnology and Biological Sciences ResearchCouncil and Royal Society.


Received 10 February; accepted 5 April 2005


1. Tessier-Lavigne, M. & Goodman, C.S. The molecular biology of axon guidance. Science274, 1123–1133 (1996).

2. Yu, T.W. & Bargmann, C.I. Dynamic regulation of axon guidance. Nat. Neurosci. 4 suppl:1169–1176 (2001).

3. Huber, A.B., Kolodkin, A.L., Ginty, D.D. & Cloutier, J.F. Signaling at the growth cone:ligand-receptor complexes and the control of axon growth and guidance. Annu. Rev.Neurosci. 26, 509–563 (2003).

4. Song, H. & Poo, M. The cell biology of neuronal navigation. Nat. Cell Biol. 3, E81–E88(2001).

5. Bandtlow, C.E., Schmidt, M.F., Hassinger, T.D., Schwab, M.E. & Kater, S.B. Role ofintracellular calcium in NI-35-evoked collapse of neuronal growth cones. Science 259,80–83 (1993).

6. He, Z. & Koprivica, V. The Nogo signaling pathway for regeneration block. Annu. Rev.Neurosci. 27, 341–368 (2004).

7. Sivasankaran, R. et al. PKC mediates inhibitory effects of myelin and chondroitin sulfateproteoglycans on axonal regeneration. Nat. Neurosci. 7, 261–268 (2004).

8. Fournier, A.E. & Strittmatter, S.M. Repulsive factors and axon regeneration in the CNS.Curr. Opin. Neurobiol. 11, 89–94 (2001).

9. Schwab, M.E. Nogo and axon regeneration. Curr. Opin. Neurobiol. 14, 118–124 (2004).10. Filbin, M.T. Myelin-associated inhibitors of axonal regeneration in the adult mammalian

CNS. Nat. Rev. Neurosci. 4, 703–713 (2003).11. Gomez, T.M. & Spitzer, N.C. Regulation of growth cone behavior by calcium: new

dynamics to earlier perspectives. J. Neurobiol. 44, 174–183 (2000).12. Henley, J. & Poo, M.M. Guiding neuronal growth cones using Ca2+ signals. Trends Cell

Biol. 14, 320–330 (2004).13. Hong, K., Nishiyama, M., Henley, J., Tessier-Lavigne, M. & Poo, M. Calcium signalling in

the guidance of nerve growth by netrin-1. Nature 403, 93–98 (2000).

14. Henley, J.R., Huang, K.H., Wang, D. & Poo, M.M. Calcium mediates bidirectionalgrowth cone turning induced by myelin-associated glycoprotein. Neuron 44, 909–916(2004).

15. Ming, G.L. et al. cAMP-dependent growth cone guidance by netrin-1. Neuron 19, 1225–1235 (1997).

16. Song, H. et al. Conversion of neuronal growth cone responses from repulsion to attractionby cyclic nucleotides. Science 281, 1515–1518 (1998).

17. Song, H.J., Ming, G.L. & Poo, M.M. cAMP-induced switching in turning direction ofnerve growth cones. Nature 388, 275–279 (1997).

18. Zheng, J.Q. Turning of nerve growth cones induced by localized increases in intracellularcalcium ions. Nature 403, 89–93 (2000).

19. Berridge, M.J. Neuronal calcium signaling. Neuron 21, 13–26 (1998).20. Gomez, T.M. & Spitzer, N.C. In vivo regulation of axon extension and pathfinding by

growth-cone calcium transients. Nature 397, 350–355 (1999).21. Wong, S.T. et al. A p75(NTR) and Nogo receptor complex mediates repulsive signaling

by myelin-associated glycoprotein. Nat. Neurosci. 5, 1302–1308 (2002).22. Wen, Z., Guirland, C., Ming, G.L. & Zheng, J.Q.A. CaMKII/calcineurin switch controls the

direction of Ca(2+)-dependent growth cone guidance. Neuron 43, 835–846 (2004).23. Jin, M. et al. Ca2+-dependent regulation of rho GTPases triggers turning of nerve growth

cones. J. Neurosci. 25, 2338–2347 (2005).24. Nishiyama, M. et al. Cyclic AMP/GMP-dependent modulation of Ca2+ channels sets the

polarity of nerve growth-cone turning. Nature 423, 990–995 (2003).25. Montell, C., Birnbaumer, L. & Flockerzi, V. The TRP channels, a remarkably functional

family. Cell 108, 595–598 (2002).26. Moran, M.M., Xu, H. & Clapham, D.E. TRP ion channels in the nervous system. Curr.

Opin. Neurobiol. 14, 362–369 (2004).27. Li, H.S., Xu, X.Z. & Montell, C. Activation of a TRPC3-dependent cation current through

the neurotrophin BDNF. Neuron 24, 261–273 (1999).28. Kim, S.J. et al. Activation of the TRPC1 cation channel by metabotropic glutamate

receptor mGluR1. Nature 426, 285–291 (2003).29. Bezzerides, V.J., Ramsey, I.S., Kotecha, S., Greka, A. & Clapham, D.E. Rapid vesicular

translocation and insertion of TRP channels. Nat. Cell Biol. 6, 709–720 (2004).30. Greka, A., Navarro, B., Oancea, E., Duggan, A. & Clapham, D.E. TRPC5 is a regulator of

hippocampal neurite length and growth cone morphology. Nat. Neurosci. 6, 837–845(2003).

31. Bobanovic, L.K. et al. Molecular cloning and immunolocalization of a novel vertebratetrp homologue from Xenopus. Biochem. J. 340, 593–599 (1999).

32. Brereton, H.M., Harland, M.L., Auld, A.M. & Barritt, G.J. Evidence that the TRP-1protein is unlikely to account for store-operated Ca2+ inflow in Xenopus laevis oocytes.Mol. Cell. Biochem. 214, 63–74 (2000).

33. Hedgecock, E.M., Culotti, J.G. & Hall, D.H. The unc-5, unc-6, and unc-40 genes guidecircumferential migrations of pioneer axons and mesodermal cells on the epidermis inC. elegans. Neuron 4, 61–85 (1990).

34. Harris, R., Sabatelli, L.M. & Seeger, M.A. Guidance cues at the Drosophila CNS midline:identification and characterization of two Drosophila Netrin/UNC-6 homologs. Neuron17, 217–228 (1996).

35. Serafini, T. et al. The netrins define a family of axon outgrowth-promoting proteinshomologous to C. elegans UNC-6. Cell 78, 409–424 (1994).

36. Mitchell, K.J. et al. Genetic analysis of Netrin genes in Drosophila: Netrins guide CNScommissural axons and peripheral motor axons. Neuron 17, 203–215 (1996).

37. Serafini, T. et al. Netrin-1 is required for commissural axon guidance in the developingvertebrate nervous system. Cell 87, 1001–1014 (1996).

38. Kennedy, T.E., Serafini, T., de la Torre, J.R. & Tessier-Lavigne, M. Netrins are diffusiblechemotropic factors for commissural axons in the embryonic spinal cord. Cell 78,425–435 (1994).

39. Li, Y. et al. Essential role of TRPC channels in the guidance of nerve growth cones bybrain-derived neurotrophic factor. Nature 434, 894–898 (2005).

40. Gordon, X.W. & Poo, M. Requirement of TRPC channels in netrin-1-induced chemo-tropic turning of nerve growth cones. Nature 434, 898–904 (2005).

41. Clapham, D.E. TRP channels as cellular sensors. Nature 426, 517–524 (2003).42. Kolodziej, P.A. et al. frazzled encodes a Drosophila member of the DCC immunoglobulin

subfamily and is required for CNS and motor axon guidance. Cell 87, 197–204 (1996).43. Keino-Masu, K. et al. Deleted in colorectal cancer (DCC) encodes a netrin receptor. Cell

87, 175–185 (1996).44. Roberts, A. Early functional organization of spinal neurons in developing lower verte-

brates. Brain Res. Bull. 53, 585–593 (2000).45. Phelps, P.E., Alijani, A. & Tran, T.S. Ventrally located commissural neurons express the

GABAergic phenotype in developing rat spinal cord. J. Comp. Neurol. 409, 285–298(1999).

46. Roberts, A., Dale, N., Ottersen, O.P. & Storm-Mathisen, J. Development and character-ization of commissural interneurones in the spinal cord of Xenopus laevis embryosrevealed by antibodies to glycine. Development 103, 447–461 (1988).

47. Ming, G., Henley, J., Tessier-Lavigne, M., Song, H. & Poo, M. Electrical activitymodulates growth cone guidance by diffusible factors. Neuron 29, 441–452 (2001).

48. Ming, G. et al. Phospholipase C-gamma and phosphoinositide 3-kinase mediatecytoplasmic signaling in nerve growth cone guidance. Neuron 23, 139–148 (1999).

49. Ren, X.R. et al. Focal adhesion kinase in netrin-1 signaling. Nat. Neurosci. 7,1204–1212 (2004).

50. Sayers, L.G. et al. Intracellular targeting and homotetramer formation of a truncatedinositol 1,4,5-trisphosphate receptor-green fluorescent protein chimera in Xenopuslaevis oocytes: evidence for the involvement of the transmembrane spanning domainin endoplasmic reticulum targeting and homotetramer complex formation. Biochem. J.323, 273–280 (1997).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

RNA editing produces glycine receptor a3P185L,resulting in high agonist potency

Jochen C Meier1, Christian Henneberger1, Igor Melnick1,5, Claudia Racca2, Robert J Harvey3, Uwe Heinemann4,Volker Schmieden4 & Rosemarie Grantyn1

The function of supramedullary glycine receptors (GlyRs) is still unclear. Using Wistar rat collicular slices, we demonstrate GlyR-

mediated inhibition of spike discharge elicited by low glycine (10 lM). Searching for the molecular basis of this phenomenon, we

identified a new GlyR isoform. GlyRa3P185L, a result of cytidine 554 deamination, confers high glycine sensitivity (EC50 ~5 lM)

to neurons and thereby promotes the generation of sustained chloride conductances associated with tonic inhibition. The level

of GlyRa3-C554U RNA editing is sensitive to experimentally induced brain lesion, inhibition of cytidine deamination by

zebularine and inhibition of mRNA transcription by actinomycin D, but not to blockade of protein synthesis by cycloheximide.

Conditional regulation of GlyRa3P185L is thus likely to be part of a post-transcriptional adaptive mechanism in neurons with

enhanced excitability.

Appropriate function of neuronal networks in the CNS relies on thebalance between excitation and inhibition. GABA1,2 and glycine3 arethe two major inhibitory neurotransmitters in the mammalian CNS.However, rostrally to the medulla, GlyRs do not, as a rule, participate inphasic synaptic transmission4,5 and are relatively insensitive to both theagonist glycine and the competitive antagonist strychnine. Althoughsupramedullary neurons have been shown to display sustained GlyR-mediated currents6–9, it has remained unclear whether low, ambientglycine (5–10 mM)10–13 can activate these supposedly insensitive GlyRs,or whether more effective GlyR isoforms exist.

Recently, it has become evident that RNA editing–related changes areimportant in the modulation of neuronal information processing.Several neuronal genes undergo RNA editing, among them themRNAs of glutamate14 and serotonin receptors15 as well as potassiumchannels16 where adenosine is edited to inosine (A-to-I). The resultingfunctional consequences are manifested by alterations in Ca2+ perme-ability (GluR6; ref. 17) in the control of G protein activation (5-HT2C;ref. 15) and in the modulation of channel inactivation (hKV1.1; ref. 16).By contrast, mammalian cytidine-to-uracil (C-to-U) editing is rare andis found only in apolipoprotein B18 and neurofibromin19 transcripts.

All single-nucleotide polymorphisms (SNPs) associated with GlyRsto date result in loss-of-function associated with the neurologicaldisorder known as hyperekplexia20. Here we demonstrate the presenceof a gain-of-function GlyRa3 isoform. GlyRa3P185L emerges fromcytidine 554 deamination (C554U) and confers high glycine sensitivityto neurons, thereby promoting the generation of sustained chlorideconductances associated with tonic inhibition. The amount ofGlyRa3-C554U transcripts is conditionally regulated in response to

brain lesion. In embryonic brain slices, slice recovery in artificialcerebrospinal fluid (ACSF) caused an upregulation of GlyRa3-C554U, whereas in adult slices the amount of GlyRa3-C554Udecreased. GlyRa3P185L may represent a future pharmaceutical targetin the treatment of hyperexcitability disorders.

RESULTS

Loose patch recordings from superior colliculus neurons revealed thatstrychnine produced a disinhibition of spontaneous spike dischargerecorded in the continuous presence of 10 mM glycine (Fig. 1a,b). Sinceall recordings were performed in the presence of 10 mM bicuculline, acontribution of GABAA receptor (GABAAR) activity seemed unlikely.In view of the low glycine concentration sufficient to induce a reversibleinhibition of spontaneous spike discharge, we pursued the possibilitythat highly efficient GlyR variants exist in the dorsal midbrain.

The sensitivity of GlyRs mainly depends on the a subunit present20.In mature supramedullary structures, a2- and a3-containing GlyRsdominate over a1-containing GlyRs, and with neuron maturation theexpression of GlyRa3 increases21. We therefore selected GlyRa3 fordetailed investigation. To characterize the regional expression patternof GlyRa3, we performed semiquantitative RT-PCR and immunohis-tochemistry (Supplementary Figure 1). The level of GlyRa3 tran-scripts relative to the glyceraldehyde-3-phosphate dehydrogenase(GAPD, encoded by Gapd) gene expression was high in the adult(4 postnatal day (P) 56) spinal cord and dorsal midbrain (Fig. 1c).Therefore, we used adult dorsal midbrain tissue to isolate GlyRa3cDNA clones. Of the twenty full-length cDNA clones sequenced, tworevealed a variation at position 554 (C554T) as compared with the


1Department of Developmental Physiology, Johannes-Mueller Center of Physiology, Charite University Medicine, 10117 Berlin, Germany. 2School of Biomedical Sciences,University of Leeds, Leeds LS2 9JT, UK. 3Department of Pharmacology, The School of Pharmacy, London WC1N 1AX, UK. 4Department of Cellular Neurophysiology,Johannes-Mueller Center of Physiology, Charite University Medicine, 10117 Berlin, Germany. 5Present address: Bogomoletz Institute of Physiology, Bogomoletz Str. 4, Kiev01024, Ukraine. Correspondence should be addressed to J.C.M. ([email protected]).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

originally described sequence21. This is expected to change theencoded amino acid at position 185 from proline to leucine(a3P185L). P185 is localized between two b-sheet structures thatform a hairpin structure with amino acids 200–204 (Fig. 1d). P185 ishighly conserved in various GlyR subunits (Fig. 1e), but it is notconserved in the entire ligand-gated ion channel (LGIC) superfamily.In fact, GABAARa6 contains, like GlyRa3P185L, a leucine at the samerespective position, and GABAARa4 and d contain functionally equiva-lent amino acids.

To find out whether GlyRa3P185L occurs in situ, we developed anassay for the detection of GlyRa3-C554U mRNA in situ22. This methodinvolves insertion of an artificial HindIII restriction site that will onlybe present in the variant cDNAs (Fig. 2a–c). The 3¢ end of the primercontains part of a HindIII restriction site, which gives rise to thecomplete site by elongation past the C554T change (Fig. 2c). Subse-quent restriction digest by HindIII and BamHI (which cuts the a3cDNA at position 904) followed by cloning into the vector pBluescriptdiscloses the presence of a3C554T. Finally, clones were tested for thepresence of a3 cDNAs by PCR screen (Fig. 2d). a3C554T and a3C554cDNA clones were processed in parallel and served as positive andnegative controls, respectively. The occurrence of GlyRa3-C554Uwithin the various mRNA preparations (Table 1) was determined bythe ratio of samples to positive controls. We found that GlyRa3-C554U

mRNA levels depend on (i) the brain region,(ii) age and (iii) tissue maintenance con-ditions during slice recovery. In dorsal mid-brain, a reciprocal relationship was foundbetween the age-dependent increase ofGlyRa3-C554U and the degree of increaseduring slice recovery. In adult dorsal midbrain,prolonged slice incubation (6 h) was charac-terized by a significant downregulation ofGlyRa3-C554U, whereas in embryonic dorsalmidbrain a significant upregulation was foundwithin 2.5 h (Table 1). Notably, the upregula-tion of GlyRa3-C554U in embryonic dorsalmidbrain slices could be prevented by thecytidine deaminase inhibitor23,24 zebularine(50 mM) or the transcription inhibitor actino-mycin D (1 mg/ml), but not by the proteinsynthesis inhibitor cycloheximide (1 mM).

To characterize the functional properties ofGlyRa3P185L, we analyzed glycine-inducedchloride currents (I(Gly)) in hippocampal neu-rons and oocytes. Hippocampal neurons weretransfected at day 14 in vitro (DIV) to expressenhanced green fluorescence protein (EGFP;insets, Fig. 3a,b) together with either GlyRa3or GlyRa3P185L. Whole-cell patch-clamprecordings were performed 3–6 d after trans-fection. We found that neurons expressingGlyRa3P185L generated chloride currents of asimilar amplitude as neurons transfected withGlyRa3. Likewise, neurons expressing GlyR-a3P185L or a3 exhibited similar maximal cur-rent responses to glycine (Table 2). The latterwere not different from those in EGFP- ornon-transfected neurons. However, there wasa clear difference in the range of effectiveagonist concentrations (Fig. 3a,b). Thepooled concentration-response relations for

glycine-induced currents are shown under four experimental condi-tions (Fig. 3c). The Hill equations fitting individual dose-responsecurves were used to determine mean EC50 values and Hill coefficients(Table 2). The EC50 obtained for GlyRa3P185L (4.7 7 0.6 mM) was 15times lower than that of neurons expressing GlyRa3 (69.6 7 9.2 mM).Oocytes expressing GlyRa3P185L also displayed high agonist sensitivityfor glycine (EC50: 17.1 7 5.2 mM, Hill coefficient: 1.6 7 0.1), taurine(EC50: 116.0 7 16.6 mM, Hill coefficient: 1.6 7 0.2) and GABA (EC50:3.9 7 0.9 mM, Hill coefficient: 1.6 7 0.2). To find out whether thepresence of GlyRa3P185L had an effect on the pharmacological proper-ties of GlyRa3 channels, a3P185L and a3 were co-expressed in oocytes.The resulting dose-response curve was shifted towards the GlyRa3curve, but the slope remained unaltered (data not shown). Thisindicates that GlyRa3P185L co-assembles with proline-containing a3subunits but does not exert functional dominance. Based on theseresults, we conclude that P185L results in higher agonist potency.

Next, we examined the ability of strychnine to suppress I(Gly).Glycine was applied as short (3 s) pulses on the background of variablestrychnine concentrations (Fig. 3d). The latter preceded the glycinepulse by 1 min. The pooled antagonist concentration-response rela-tions are shown for GlyRa3 and a3P185L (Fig. 3e). The mean IC50 valuesobtained from individual dose-response curves were similar (Table 2):123 7 28 nM in neurons expressing a3 and 117 7 39 nM for neurons

m GABAARδr GABAARγ3r GABAARγ2r GABAARγ1r GABAARβ3r GABAARβ2r GABAARβ1r GABAARα6r GABAARα5r GABAARα4r GABAARα3r GABAARα2r GABAARα1

r GlyRβr GlyRα4r GlyRα3 P185Lr GlyRα3r GlyRα2r GlyRα1

h 5HTαh nAChRα7h nAChRα4h nAChRα1

185179

M1 M2 M3 M4

CT204

Y202K200

G160P185L

Y161

NCC

CC

2 s100 pA

Wash

Strych

Control

200

150

100

50

0

Spi

ke fr

eque

ncy

stry

chni

ne +

gly

cine

/

glyc

ine

(%)

(20)**

30

20

10

0

rela

tive

to G

AP

DG

lyR

α3 m

RN

A

> P56 E21 P14 > P56

Dorsal midbrain

(9.6

± 2.

0)

(9.0

± 2.

1)

(12.

6 ± 2

.9)

(0.6

± 0.

1)

(2.6

± 0.

7)

(2.2

± 0.

6)

(29.

9 ± 6

.4)

(1.9

± 0.

3)

Hippoc

ampu

s

Cereb

ellum

Corte

x, fro

ntal

Corte

x, pa

rieta

l

Spinal

cord

a

d e

b c

Figure 1 Inhibition of spontaneous spike discharge by low glycine in P15 dorsal midbrain slices, and

GlyR structure. (a) ‘Loose-patch’ recording of spontaneous spike discharge from a superior colliculus

neuron in an acute horizontal slice. Control: activity in the continuous presence of bicuculline (10 mM)

and glycine (10 mM). Strych: 1 min after the start of strychnine application (3 mM). Wash: 5 min after

the start of control solution. (b) Quantification of the strychnine-triggered disinhibition of spike discharge

in a. (c) Semiquantitative RT-PCR analysis of the amount of GlyRa3 mRNA in selected brain regions at

the indicated age. Mean 7 s.e.m. GlyRa3 mRNA level relative to GAPD (GlyRa3/GAPD) is shown in

parentheses. (d) Membrane topology of the GlyRa3 polypeptide with four transmembrane domains.

Amino acid residues involved in glycine binding (orange dots) are shown together with the P185L

variation (red dot). The two disulfide bridges are shown in yellow. Adapted from ref. 50. (e) Amino acid

sequence alignment of GlyRa3 sequences (position 179 to 190) along with the corresponding sequence

regions of other members representatives of the LGIC superfamily. The position P185 in the rat GlyRa3

polypeptide21 is highlighted. Amino acids in red and orange are conformationally relevant. Aliphatic

amino acids are indicated in blue. r: rat, h: human, m: mouse. **, P o 0.01.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

with a3P185L. A nearly complete block of maximal currents wasobtained with 3 mM strychnine.

To assess possible differences in the kinetics of glycine-inducedcurrents, we determined the time constant of initial desensitization(tdesens) by monoexponential fitting and the steady-state currentpresent 15 s after the start of glycine application (I(Gly)steady15;Fig. 4a,b). The averaged specimen traces and the double logarithmicplot (Fig. 4c) illustrate that the same values of tdesens were obtained bysignificantly different glycine concentrations (Table 2). At tdesens ¼2.5 s, the respective values (derived from the double logarithmic plot,Fig. 4c) were 30 mM for a3P185L and 600 mM for a3. The correlationswere statistically significant at P o 0.001, and the regressions weredifferent from each other. These desensitizing glycine concentrations(20-fold lower in neurons expressing a3P185L) agree well with the 15-fold difference in the agonist potency.

The fraction I(Gly)steady15/I(Gly)max was inversely dependent on theglycine concentration (data not shown), but at similar agonist potencythere was no significant difference in I(Gly)steady15/I(Gly)max betweenneurons expressing GlyRa3 or a3P185L (Fig. 4d). At a Cl� driving forceof 15 mV (holding potential �40 mV) and a glycine concentrationclose to the respective EC50, I(Gly)steady15 had amplitudes between40–120 pA. These results suggest that neurons expressing GlyRa3P185L

are able to produce a tonic conductance in response to low glycine.A tonic conductance requires both the continuous presence of theagonist and the availability of non-desensitized receptors25. Therefore,we first applied standard salt solution to eliminate the access ofendogenous glycine (Fig. 4e) and then applied glycine to test forthe effect of strychnine after 1 min of agonist application. Washingthe stationary bath indeed produced steps in the whole-cellbaseline currents. These were significantly larger in neurons expressing

Figure 2 Detection of GlyRa3-C554U transcripts

in situ. (a,b) GlyRa3 and a3C554T cDNA

sequence (nucleotides 544–564, according to

ref. 21). (c) Primer extension restriction

fragment length polymorphism (RFLP)-PCR.

Oligonucleotide hybridization to the GlyRa3

cDNA is shown. The 3¢ end of the oligonucleotide

contains the partial HindIII restriction site, whichis completed only when the C554T variation is

present. (d) Example of a PCR screen on colonies

using T7 and GlyRa3-specific oligonucleotides

showing the fraction of a3 recombinants for

negative (a3C554 cDNA clone) and positive

(a3C554T cDNA clone) controls, immediately

processed E21 dorsal midbrain slices, E21

dorsal midbrain slices after 2.5 h in oxygenated

ACSF, and in the absence or presence of

zebularine (Zeb, 50 mM). For clarity, only six

colonies are shown for each case. (e) Sequence alignment of apolipoprotein B and GlyRa3 cDNAs. Regulator, spacer and mooring sequence requirements

for apoB editing catalytic subunit 1 (APOBEC1)-dependent RNA editing are shown. The edited cytidine (boldface) and the homologous nucleotides (boldface,

underlined) are indicated.

544 564

544 564

185

185

534 564

Elongation

C554 (Hind III)

C554T (Hind III)

α3C554 α3C554T E210 hE21

2.5 h ACSFE21

2.5 h ACSF Zeb

Apolipoprotein B

GlyRα3

Regulator Spacer Mooring

a c

b

d

e

Table 1 Levels of GlyRa3-C554U transcripts (mean 7 s.e.m., percentage of positive control) in selected rat brain regions at the indicated age

Sample number Tissue Age

Time in

ACSF (h)

ACSF

suppl. with

Levels of

a3C554U transcripts

Significance level compared with

sample number in parentheses

1 Spinal cord 4 P56 0 – 5.4 7 1.5 *(4)

2 Cortex, parietal 4 P56 0 – 4.8 7 2.8 *(4)

3 Cortex, frontal 4 P56 0 – 5.1 7 2.8 *(4)

4 Cerebellum 4 P56 0 – 19.0 7 3.4 *(1,2,3,5)

5 Hippocampus 4 P56 0 – 2.7 7 2.1 *(4)

6 Dorsal midbrain E21 0 – 0.3 7 0.1 ***(7,11,14)

7 E21 2.5 – 4.2 7 0.4 ***(6,8); **(9); *(12)

8 E21 2.5 Zeb 0.2 7 0.1 ***(7)

9 E21 2.5 Act-D 0.4 7 0.1 **(7)

10 E21 2.5 CHX 6.0 7 2.2 –

11 P14 0 – 7.5 7 1.5 ***(6); *(14)

12 P14 2.5 – 9.2 7 1.4 *(7,13)

13 P14 2.5 Zeb 6.8 7 0.6 *(12)

14 4 P56 0 – 14.7 7 0.7 ***(6,16); *(11)

15 4 P56 2.5 – 10.3 7 7.5 –

16 4 P56 6 – 1.4 7 0.5 ***(14)

17 Tail 4 P56 – – ND

ACSF supplemented with 50 mM zebularine (Zeb), 1 mg/ml actinomycin D (Act-D) or 1 mM cycloheximide (CHX). Samples 1–16 represent cDNA probes, and sample 17 is genomicDNA. ND, not detected. Significance levels after an ANOVA are marked with asterisks, *P o 0.05, **P o 0.01, ***P o 0.001.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

GlyRa3P185L than in those expressing a3 (Fig. 4e, Table 2). Further-more, we observed an approximately threefold increase in the base-line noise in a3P185L-expressing neurons. Addition of strychnine incells superfused for 1 min with glycine produced smaller steps(Fig. 4f,g; Table 2), but there was a significant correlation betweenthe amplitudes of the steady currents demonstrated by wash versusstrychnine (Fig. 4h).

To identify neurons best suited for experiments on GlyRa3P185L

function in situ, we analyzed the GlyRa3 distribution in dorsalmidbrain sections from perinatal rats (Fig. 5a,b). We found thatGlyRa3 was particularly abundant in the stratum griseum interme-dium (SGI) of the superior colliculus (Fig. 5a). The colocalization withthe neuronal marker NeuN indicates a preferential neuronal GlyRa3expression. However, GlyRa3 was also present in cells lacking NeuN,which may represent glial labeling (Fig. 5a, arrows). High-power views(Fig. 5c) also indicated that GlyRa3 is found at both non-synaptic(arrows) and presumably postsynaptic loci, colocalizing with thevesicular inhibitory amino acid transporter VIAAT (arrowheads).

We used slices from neonatal rats for the electrophysiologicalcharacterization of GlyRa3P185L in situ, because de novo editing ofGlyRa3 transcripts during slice recovery is stronger in the neonatalbrain (Table 1). As zebularine treatment prevented the de novo editingof GlyR a3 mRNAs (Table 1), we expected an effect of zebularine onthe amplitude of I(Gly), provided that GlyRa3P185L makes a noticeable

contribution to neuronal glycine responses in the dorsal midbrain.Whole-cell recordings of SGI neurons showed that this was the case(Fig. 5d,e). Blockade of C-to-U RNA editing by preincubation of theslices with zebularine resulted in a significant (P o 0.05, unpairedStudent’s t-test) decrease of the glycine-induced steady-state currentdensity (I(Gly)steady in the control: 25.3 7 4.4 pA/pF; I(Gly)steady

after zebularine: 12.5 7 2.8 pA/pF). Thus, the zebularine-sensitiveGlyRa3-C554U upregulation in neonatal dorsal midbrain slices ismatched by a strong increase in the steady-state chloride conductance,at least in a neuron population displaying GlyRa3 at a relativelyhigh level.

DISCUSSION

Using primer extension restriction fragment length polymorphism(RFLP)-PCR to detect SNPs, we found a new GlyRa3 isoform(a3P185L) that occurs in the dorsal tegmentum and spinal cord. Thelack of a corresponding codon (CTT) in the GlyRa3 gene (Glra3) andthe capacity of the cytidine deaminase inhibitor zebularine to block theproduction of GlyRa3-C554U demonstrate the involvement of C-to-URNA editing.

Zebularine may also inhibit DNA methylation26. However, sincezebularine reactivates gene transcription27, and actinomycin D pre-vented the production of GlyRa3-C554U in response to brain slicing,we rejected the possibility that the zebularine effect on GlyRa3-C554U

α3 α3P185L α3P185L

Gly

1 mM

3 mM

10 µM

30 µM

100 µM

300 µM

Gly

100 µM

0.3 µM1 µM3 µM

10 µM

30 µM

Gly

1 nM

Strych10 µM1 µM

0.1 µM

10 nM200

pA

2 s

1.0

0.8

0.6

0.4

0.2

00.0001 0.001 0.01 0.1 1 10

[Glycine] (mM)

I/Imax

1.0

0.8

0.6

0.4

0.2

00.001 0.01 0.1 1.0 10

[Strychnine] (µM)

I/Imax

1 s

200 pA

a b c d

eFigure 3 Concentration-response relations for cultured hippocampal neurons expressing GlyRa3 and a3P185L.

(a,b) Specimen records of current responses to glycine at variable concentrations. Upper trace shows onset of glycine

application. Insets: images of transfected neurons visualized with EGFP. Scale bar, 10 mm. (c) Plot of concentration-

response relations for four different experimental conditions. Filled circles: non-transfected control neurons after

17–20 d in vitro. Filled triangles: neurons expressing EGFP only. Open squares: neurons expressing GlyRa3 and EGFP,

as in a. Filled squares: neurons expressing GlyRa3P185L and EGFP, as in b. Data from seven neurons per group was

pooled. (d) Specimen records of current responses to glycine on the background of variable strychnine concentra-

tions. The latter are given on the left of each trace. Strychnine was applied at least 1 min before the onset of glycine

application. (e) Plot of concentration-response relations for neurons expressing a3 (open circles) or a3P185L (filled

squares). Glycine concentrations were selected to correspond to the minimal concentrations causing IGly saturation:1 mM and 0.1 mM for a3 and a3P185L, respectively. All records at a holding voltage of �40 mV and an ECl of �25 mV.

Table 2 Functional properties of GlyRa3 and a3P185L

Units GlyRa3 GlyRa3P185L Significance level

I(Gly)max pA 928.3 7 34.0 (7) 853.7 7 34.2 (7) NS

EC50 for [Gly] mM 69.6 7 9.2 (7) 4.7 7 0.6 (7) 0.00005

Hill coefficient for Gly 1.87 7 0.12 (7) 1.27 7 0.15 (7) 0.01

IC50 for strychnine against saturating [Gly] nM 123 7 28 (5) 117 7 39 (6) NS

Hill coefficient for strychnine against saturating [Gly] 0.82 7 0.03 (5) 0.86 7 0.06 (6) NS

[Gly] (tdesens ¼ 2.5 s)* mM 600 (48) 30 (32) 0.0001

I(Gly)steady15 / Imax 0.12 (14) 0.16 (19) NS

I(Gly)steady15 (strychnine) pA 4.3 7 3.4 (10) 50.4 7 10.5 (10) 0.0005

Data are presented as mean 7 s.e.m. The number of analyzed neurons is given in parentheses. Saturating [Gly]: 1 mM (a3) and 0.1 mM (a3P185L), *: gives the glycine concentration attdesens ¼ 2.5 s. NS: not significant.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

mRNA levels was due to its methyltransferase inhibiting activity. Inaddition, a period of 2.5 h is insufficient for zebularine integration intogenomic DNA (ref. 28). C-to-U RNA editing has not yet been describedfor a member of the LGIC superfamily and is rare, in general. In case ofglutamate14 and serotonin receptors15 or potassium channels16, A-to-IRNA editing has been shown to play an important role in the regulationof Ca2+ influx and neuron excitability (reviewed in ref. 29). The onlyknown examples of mammalian C-to-U RNA editing have been foundwith apolipoprotein B18 and neurofibromin19. However, since thegenomic sequence flanking the P185 codon is not surrounded by theapoB editing catalytic subunit 1 (APOBEC1) recognition elements30,31

(Fig. 2e), the involvement of another cytidine deaminase is likely.The presence of GlyRa3P185L results in gain-of-function, whereas all

previously described GlyR SNPs have been associated with loss offunction20. GlyRa3P185L responded to glycine, taurine and GABAwithin a physiologically relevant concentration range. Accordingly,the respective EC50 values obtained from both neurons and oocyteswere dramatically reduced in cells expressing GlyRa3P185L. As com-pared with previous data21, where taurine and GABA failed to producea detectable current in oocytes expressing GlyRa3, GlyRa3P185L

responded to both agonists. The IC50 values for strychnine were similar,and the blocking potency of this competitive antagonist decreased

to the same extent as the agonist (glycine) potency increased.This suggests that GlyRa3P185L affects the strychnine-binding siteto the same extent as the agonist-binding site. Based on these results,we propose that P185L confers high ligand sensitivity to supramedul-lary GlyRs.

The GlyRa3 cDNA21 was first cloned in 1990, yet little information isavailable on regional and temporal patterns of Glra3 gene expression.We have performed semiquantitative RT-PCR and immunohisto-chemistry to determine GlyRa3 mRNA and protein levels in a fewselected brain regions. High levels of GlyRa3 transcripts were foundin adult spinal cord and in dorsal midbrain, whereas in cortex,hippocampus and cerebellum expression was significantly lower. How-ever, GlyRa3 protein was found in all investigated brain regions(Supplementary Figure 1). Cultured neurons from the dorsal mid-brain also express the low-efficacy GlyRa2 subunit but lack the high-efficacy GlyRa1 subunit32–34. The occurrence of GlyRa3-C554UmRNA within these brain regions did not match the amount of a3transcripts relative to Gapd gene expression. In the cerebellum, forexample, the level of GlyRa3-C554U mRNA was highest, but theamount of GlyRa3 relative to GAPD mRNA was lowest. A GlyRa3P185L

response value may be predicted by multiplying the GlyRa3 mRNAlevel with the level of GlyRa3-C554U. Among the brain regions tested,

α3 α3P185L

α3

α3P185L

α3

α3

α3P185L

α3P185L

α3 α3P185L

Gly

600

µM

30 µ

M

I(Gly)steady15

I (Gly

)ste

ady1

5 /I m

ax

I(Gly)steady15

100 pA

5 s

100 pA

5 s

τ desens ~ 2.5 s

τ de

sens

(s)

[Glycine] (mM)τ desens ~ 2.5 s

10

2.5

1

0.001 0.01 0.1 1 10

NS

(14)

(19)

0.3

0.2

0.1

0.0

DNQX (100), AP5 (50), Bic (10)

DNQX (100), AP5 (50), Bic (10), Gly (1)

Wash on

DNQX (100), AP5 (50), Bic (10)

Wash on

DNQX (100), AP5 (50), Bic (10), Gly (1)

Strychnine (3)

Strych on

Wash on

Strychnine (3)

300

200

200α3P185L

P < 0.01150

100

50

00 100 200 300 400

100

0

Tota

l shi

ft (p

A)

Shift (wash on) (pA)

Shi

ft (s

tryc

h on

) (p

A)

e g

f h

dcba

Figure 4 Characteristics of current desensitization and tonic glycine-induced currents in cultured hippocampal neurons expressing GlyRa3 or a3P185L.

(a,b) Traces of glycine-induced currents averaged from three neurons per condition at concentrations that produced a current decay with a time constant

of 2.5 s. (c) Dependence of the time constant of desensitization (tdesens) on glycine concentration (data from three to five neurons per data point). Theconcentrations causing desensitization at tdesens ¼ 2.5 s are marked with dotted lines. Open squares: GlyRa3. Filled squares: GlyRa3P185L. (d) Fraction of the

tonic current component recorded 15 s (arrows in a, b) after the maximum glycine-induced current. (e) A stationary bath solution containing the supplements

as indicated (in mM) was maintained for 2 min before powering on the fast superfusion system (‘Wash on’). (f) A continuous superfusion with the same salt

solution as in e, except that glycine (1 mM) was present throughout. The solution also containing strychnine was turned on not earlier than 1 min after the start

of the glycine application. (g,h) Summary of results on the tonic current that was induced by ambient glycine and blocked by wash with a glycine-free salt

solution (‘Wash on’), and on the tonic current induced by addition of 1 mM of glycine and blocked by 3 mM of strychnine (‘Strych on’). All recordings were

made at a holding voltage of �40 mV and ECl of �25 mV, except e–h, where recordings were made at a holding voltage of �70 mV.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

this value was highest in midbrain tegmentum, followed by cortex andcerebellum, and lowest in hippocampus.

As the amount of GlyRa3-C554U is tissue-dependent, one can alsoexpect that GlyRa3P185L fulfills a tissue-specific function. At the level ofthe spinal cord, GlyRa3 has been shown to be specifically involved ininflammatory pain sensitization35. In Glra3�/� animals this process isselectively impaired because the protein kinase A (PKA) phosphoryla-tion site involved in the prostaglandin E2 signal transduction is notpresent in the other predominant adult GlyR isoforms containing theGlyRa1 subunit35. In Glra3�/� retina, however, the overall number ofGlyR clusters is only slightly reduced36, indicating that other GlyRsubunits might have compensated for the loss of GlyRa336. Accord-ingly, the lack of GlyRa3P185L in Glra3�/� animals might not cause anovert phenotype. In this study, GlyRa3 immunostaining of suprame-dullary brain regions showed intensive labeling in neurons of the SGI.

This brain structure controls the execution of orienting eye and headmovements to novel sensory targets37. The degree of inhibition iscritical for the performance of the highly precise ballistic eye move-ments38. Postsynaptic stabilization of GlyRs requires both the GlyRbsubunit39 as well as the presence of permissive gephyrin isoforms4. Insupramedullary structures, C5-cassette–containing gephyrins restrainpostsynaptic enrichment of GlyRs opposite GABA-releasing terminals4,which is in agreement with a recent study of hippocampal neuronsclearly demonstrating the presence of synaptic and non-synaptic GlyRsin the hippocampus40. Our immunohistochemical analysis of the SGIshowed, in addition, that GlyRa3 occurs at both synaptic and non-synaptic loci in the superior colliculus. However, as the antibody doesnot distinguish between edited and non-edited GlyRa3 isoforms, theinterpretation of the immunohistochemical data is ambiguous withrespect to the functional role of GlyRa3P185L. Nevertheless, we proposethat supramedullary GlyRs contribute to tonic rather than phasicinhibition, because we repeatedly could not detect glycinergic post-synaptic currents in superior colliculus slices41,42. While the role ofGABAARs in tonic inhibition of neuron firing has attracted consider-able attention2,25, the potential role of GlyRs has largely been neglected(but see ref. 43). The first evidence for non-synaptically activated GlyRswas provided in 1998 (ref. 7), and this notion was further developed byothers6,8,9. In hippocampal neurons, relatively high (300 mM) glycineconcentrations9 were needed to demonstrate an involvement of GlyRsin tonic inhibition and suppression of hyperexcitability. This is inagreement with reports on the low affinity of GlyRs in hippocampalpyramidal cells (around 270 mM)8 and interneurons (around 270 mM)8

and also the low amount of GlyRa3-C554U mRNA found here in theadult hippocampus.

As GlyRa3 appears to be absent from the majority of hippocampalneurons, we chose these cells, in addition to oocytes, as an expressionsystem for functional tests. We found that a3P185L-GlyRs respond tolow, ambient10,11 glycine with large steady-state currents. The observedhigh agonist potency and the availability of non-desensitized receptorsare a prerequisite for tonic inhibition25. It is therefore very likely thatthe disinhibition demonstrated here on the background of prolonged10 mM glycine application was due to the high-efficacy a3P185L subunit.On the other hand, whether minor amounts of this highly efficientGlyR would suffice to influence neuronal signal processing in vivo wasstill uncertain. However, when considering that we were analyzing a3transcripts from a mixed population of glial and neuronal cells and thatnot all neurons in the regions analyzed may edit a3 mRNA, it is possiblethat some, if not the majority of, neurons may express high levels ofa3P185L. In fact, a greater than 50% decrease in the glycine-inducedsteady state current after zebularine treatment in SGI neurons stronglysupports our assertion that the a3P185L isoform is important in theregulation of neuronal excitability in vivo.

P185L causes the ligand-binding domain of a3P185L to become morehomologous to GABAARa6, which contains a leucine at the equivalentposition. This is notable, as both GABAARa6 and d are known to beinvolved in tonic inhibition25. Taken together with our data fromXenopus oocytes, this suggests that a3P185L-GlyRs, when activated byglycine or taurine and perhaps even by GABA, can add to the tonicinhibition produced via GABAARs.

The occurrence of GlyRa3-C554U in vivo conditionally changes inresponse to brain lesion. In the immature brain, GlyRa3-C554U isupregulated in response to slicing, whereas in the adult brain, theamount of GlyRa3-C554U decreases. Due to the relative directionalflow of chloride ions, which changes during development44, thisobservation suggests a role for the intracellular calcium homeostasisin the regulation of GlyRa3-C554U. This is particularly interesting if

SGS

SGI

SGP

ROI2

ROI1

PAG

SO

DAPIGlyR α3

ROI1 α3 NeuN

DAPI

DAPI

DAPI α3 + peptide NeuN

VIAAT

ROI2 α3 NeuN

α3

N N N N N N

Gly (100 µM)

250 pA

10 s

Zebularine

Control

30 (11)

*(10)20

I (Gly

)ste

ady

dens

ity (

pA/p

F)

10

0Ctrl Zeb

a

b

c

d e

Figure 5 Localization and functional assessment of GlyRa3 in perinatal

(E21–P1) dorsal midbrain slices. (a) Immunohistochemistry of a coronaldorsal midbrain section illustrating the expression of GlyRa3 and NeuN

together with DAPI. High-power views of two regions of interest (ROI) are

provided. Red: GlyRa3. Green: DAPI (left) and NeuN (ROI1 and ROI2, right).

SGS: stratum griseum superficiale; SO: stratum opticum; SGI: stratum

griseum intermedium; SGP: stratum griseum profundum; PAG: periaqeductal

grey. Scale bars: 100 mm. Arrows indicate presumable glial GlyRa3

expression. (b) Negative control experiment, which includes the GlyRa3

peptide (a3 + peptide) used for immunization. (c) High-power view of VIAAT

and GlyRa3 immunoreactivities within the SGI. Note that GlyRa3 is located

at both non-synaptic (arrows) and presumably postsynaptic loci (arrowheads).

The position of nuclei (N) is indicated. Scale bar: 10 mm. (d) Whole-cell

currents of an SGI neuron elicited by bath application of 100 mM glycine in

the presence of AP5 and tetrodotoxin. Currents were fully blocked by 3 mM

strychnine (not illustrated). (e) The steady-state current was calculated as the

difference between averages of 10-s periods before and at the end of glycine

application. The steady-state current (I(Gly)steady) was normalized to the cell

capacitance, which was unaffected by zebularine incubation (data not

shown). Zebularine treatment (Zeb) significantly reduced I(Gly)steady. Thenumber of tested neurons (n) is indicated above the columns. *, P o 0.05.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

one considers that ambient glycine may be elevated (up to 20 mM)during and after brain ischemia45. Therefore, conditional regulation ofGlyRa3-C554U during slice recovery in ACSF is of particular interest,since it suggests that GlyRa3 RNA editing might be a component of theresponse of brain tissue to lesion. GlyRa3P185L may thus represent afuture pharmaceutical target in the treatment of hyperexcitabilitydisorders, and a major challenge will be the identification of themolecular components involved in GlyRa3 RNA editing.

METHODSAnimals and slice preparation. All experiments were in accordance with a

permit from the Office for Health Protection and Technical Safety of the

regional government of Berlin (LaGetSi, T0122/03) that complies with inter-

national and European Union norms.

The preparation and maintenance of Wistar rat dorsal midbrain slices was

similar to previously described procedures46, except that brains were cut in the

coronal plane at the level of the superior colliculi, and slices comprised the

dorsal tegmentum down to the level of the mesencephalic aqueduct. The rats

were investigated at embryonic day (E) 18–21, P1–2, P14–17 and adulthood

(4P56). Briefly, rats were decapitated under ether anaesthesia. Brains were

removed and transferred to ice-cold ACSF, and 400-mm-thick (molecular

biology) or 200-mm–thick (electrophysiology) coronal slices of the dorsal

midbrain were obtained by vibratome sectioning (Campden Instruments).

Slices were stored in oxygenated ACSF at room temperature (23–25 1C) or were

frozen in liquid nitrogen. The ACSF (pH 7.3, 5% CO2/95% O2) contained (in

mM) NaCl (125), KCl (5), glucose (5), NaH2PO4 (1.25), NaHCO3 (25), CaCl2(2), and MgCl2 (1). To inhibit cytidine deaminase, mRNA transcription or

protein synthesis, ACSF was supplemented with the transition state analog in-

hibitor 1-(b-D-ribofuranosyl)-dihydropropyrimidin-2-one (zebularine, 50 mM,

Calbiochem), actinomycin D (1 mg/ml, Sigma) or cycloheximide (1 mM,

Sigma), respectively.

DNA isolation and GlyRa3C554T cloning. Genomic DNA was isolated from

rat tails as previously described47. RNA was isolated from E18, P14 and adult

(4P56) slices using Trizol (Invitrogen). cDNA was obtained by reverse

transcription (GibcoBRL) of RNA using oligo-dT primer. GlyRa3 cDNAs were

amplified using oligonucleotides 5¢-GGATAAGACTGTTTTCAGGATCGG-3¢and 5¢-CTCAATGCTCTGTTTTTGTATGCC-3¢. PCR conditions were 30 cycles

of 45 s at 95 1C, 45 s at 58 1C and 1.5 min at 72 1C. All PCRs were performed

using the High Fidelity Expand PCR system (Roche Applied Science). After

PCR products were cloned into the TA cloning vector pCR2.1 (Invitrogen),

20 clones were sequenced. Nucleotide and amino acid numberings reflect the

mature polypeptide after signal peptide cleavage. The cDNAs encoding

the GlyRa3 isoform with a proline at position 185 (here denoted as a3) and

the new isoform (denoted as a3P185L) lacked the 15–amino acid insert at

position 972. They were inserted into the mammalian expression vector

pEGFP-N1 (Clontech) using SacI and NotI, a combination that leads to

excision of EGFP. The cDNA clones a3C554T and a3C554 are referred to as

positive and negative controls, respectively.

GlyRa3-C554U in situ detection. Based on previous work22, we developed a

new experimental approach for the detection of this variation in situ involving

the insertion of an artificial HindIII restriction site present only in a3C554T

cDNAs. Genomic DNA and cDNA were amplified using the oligonucleotides

5¢-GGCTGAAGGACTCACTAAGC-3¢ (sense; used for amplification of geno-

mic DNA and cDNA) and 5¢-GGGGGATCCATTGTAGTGTTTAGTGCAGTAT-

3¢ (antisense; used for amplification of genomic DNA, end of exon 6; contains a

BamHI site (GGATCC)) or 5¢-TTGACATAGGACACCTTTGG-3¢ (antisense;

used for amplification of cDNA, spans exons 6–8). Positive and negative

controls (a3C554T and a3C554) were always included. The resulting PCR

products were digested at 37 1C for 1 h using 15 units of HindIII and BamHI,

respectively. The digested amplification product was purified and quantified by

intensity comparisons with DNA size markers (New England Biolabs). Equal

amounts (50 ng) of the purified digested DNAs were used for ligation with a

HindIII- and BamHI-digested pBluescript vector. After transformation of

competent bacteria (JM109, 1 � 108 cfu/mg DNA, Promega), the number of

colonies was determined. To rule out vector self-ligation, ten colonies were used

as templates in PCR screens using T7 oligonucleotide together with GlyRa3-

specific primer 5¢-CTGTATTGTAGTGTTTAGTGCAGT-3¢. The fraction of

GlyRa3 recombinant clones was determined and used to adjust the number

of true positive colonies. The amount of a3C554U mRNA relative to a3C554T

(positive control) was calculated in each experimental condition; the mean 7s.e.m. was determined and expressed as a percentage of the positive control.

Semiquantitative RT-PCR. RT-PCR was performed as previously described4.

Variation of template concentration and cycling conditions ensured that PCRs

were carried out in the linear range of amplification. Final parameters were 28

cycles of 45 s at 95 1C, 45 s at 58 1C and 45 s at 72 1C. To amplify GlyRa3

cDNA, the oligonucleotides 5¢-GATCTCAAGAATTTCCCAATGG-3¢ and

5¢-CGTGGTCATCGTAAGTACAG-3¢ were used at 10 mM each. GAPD

was co-amplified in the same PCR reaction tube using oligonucleotides

5¢-CAGTATGACTCTACCCACGG-3¢ and 5¢-CTCAGTGTAGCCCAGGATG-3¢at 1 mM. This served as a reference for quantification of the total amount of

cDNA. Three separate PCR experiments were performed on three separate RNA

preparations. PCR products were visualized by ethidium bromide staining. The

intensity and volume of the analyzed PCR products was quantified using

ImageQuant software (Molecular Dynamics) and expressed as mean pixel

intensity, and then a ratio of the GlyRa3 and GAPD amplification products

was calculated.

Immunohistochemistry. The tissue was dissected and immediately immersed

in 4% paraformaldehyde in 0.1 M phosphate buffer (PBS), pH 7.4, for

10–30 min. Then the tissue was cryoprotected in graded sucrose solutions

(10, 20, and 30%) and embedded in Tissue-Tek OCT compound (Sakura

Finetek). Sections (20 mm) were cut using a cryomicrotome (Jung CM 1800,

Leica), mounted on SuperFrost Plus microscope slides (Carl Roth) and

processed for immunocytochemistry as follows: after three washes in PBS,

sections were washed in PBS containing 0.1% gelatin and reacted overnight

at 4 1C in a humid chamber with primary antibodies diluted in PBS/gelatin,

supplemented with 0.12% Triton X-100. The rabbit polyclonal antibody36

directed against the C-terminal peptide CKILRHEDIHQQQD of GlyRa3 was

used (1:200) in combination either with the mouse monoclonal NeuN antibody

(1:200, Chemicon) or the guinea pig polyclonal VIAAT antibody (1:1,000;

Chemicon). Specificity of the GlyRa3 immunolabeling was checked by applica-

tion of the peptide CKILRHEDIHQQQD (1:200). After washes in PBS/gelatin,

secondary antibodies were applied for 1 h at room temperature (23–25 1C).

Coverslips were mounted on glass slides using Vectashield containing DAPI

(Vector Laboratories). Images were acquired using a 12-bit cooled digital camera

(Photometrics) mounted on an epifluorescence microscope (Axiovert, Zeiss).

Hippocampal cultures. Hippocampal cultures from E18 rats were prepared as

previously described4 and maintained in B27- and 1% FCS-supplemented

Neurobasal medium48. After neuron attachment, the coverslips were transferred

(cell-side down) to dishes containing a confluent layer of hippocampal

glial cells.

Transfection was carried out at DIV 14. Coverslips were transferred to wells

containing the transfection medium (Neurobasal supplemented with 0.25 mM

glutamine and 1% FCS) and were incubated with complexes formed between

5 ml of Effectene (Qiagen) and 300 ng of DNA. The manufacturer’s protocol

was followed, except that the incubation time was reduced to 1 h.

Patch-clamp recording from hippocampal neurons in culture. Patch-clamp

recordings in the whole-cell configuration were performed 3–6 d after transfec-

tion. The growth medium was exchanged by a standard bath solution (pH 7.3)

containing (in mM) NaCl (140), KCl (4), CaCl2 (2), MgCl2 (1), HEPES-NaOH

(20), glucose (10). Transfected neurons were visualized under fluorescence

optics (Axiovert 10, Zeiss) using a 40� objective. Cultures were tested for not

longer than 2 h. Whole-cell patch-clamp recordings were carried out at room

temperature using the patch-clamp amplifier EPC-7 (List Electronics). Patch

electrodes were fabricated from borosilicate glass capillary tubing (WPI) using

a P-87 puller (Sutter Instruments). The pipette solution (pH 7.3) contained (in

mM) potassium gluconate (95), KCl (30), NaCl (5), MgCl2 (2), CaCl2 (0.5),

EGTA (10), HEPES-KOH (20). The pipette-to-bath DC resistance of patch

electrodes ranged from 3 to 5 MO (tip diameter 1–2 mm). Recordings were


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

performed at a holding voltage of �40 mV. The reversal potential of the

Cl� currents was �25 mV. Throughout the whole-cell recordings, series

resistance (RS) compensation was applied as much as possible (at least 70%).

RS was monitored at intervals during experiments. Recordings were rejected if

the initial RS was above 25 MO or changes in RS exceeded 30%.

Glycine and strychnine were locally applied via a fast six-channel super-

fusion/suction system. The 10–90% time of solution exchange was estimated to

be o85 ms using current transients evoked by application of K+-enriched

extracellular solution. An interval of at least 1 min was left between successive

glycine concentrations to allow for recovery from receptor desensitization.

Different concentrations were applied randomly. Glycine concentrations close

to the respective EC50 values (5 mM, a3P185L and 50 mM, a3) were applied at

intervals to ensure that the response had remained stable. Glycine-induced

currents were isolated pharmacologically by blocking glutamatergic (10 mM

DNQX and 50 mM D,L-2-amino-5-phosphonovaleric acid (AP5)) and GABAer-

gic (10 mM bicuculline methiodide) transmission. Signals were sampled at

10 kHz using a 16-bit AD-converter (ITC-16, Instrutech) and the software

WinTida-3.0 (HEKA Elektronik). If not stated otherwise, all drugs were

obtained from Sigma.

Two electrode recordings from Xenopus laevis oocytes experiments were

performed as previously described49.

Patch-clamp recordings from brain slices. Slices were preincubated for 2 h in

ACSF (control), or ACSF supplemented with zebularine, and then placed into a

submersion chamber of an upright microscope (Axioscope 2 FS plus, Zeiss).

During recordings (no longer than 90 min) slices were continuously superfused

(1 ml/min) with oxygenated (95% O2/5% CO2) ACSF. A 63� water immersion

objective (Zeiss) was used to visualize neurons. Neurons were classified on the

basis of their location in the SGI (Fig. 5a).

Spontaneous spike discharge was recorded in the loose (20–50 MO) cell-

attached patch-clamp configuration using ACSF-filled (2–3 MO) pipettes. A

local superfusion system was used to apply the standard extracellular solution

with or without glycine (10 mM) or glycine and strychnine (3 mM), and 10 mM

bicuculline was present during the entire time of recording. The frequency of

spontaneous spike discharge was evaluated offline using the first derivative of

the recording trace for event detection (written in-house by C. Henneberger,

Charite University Medicine, Berlin). Whole-cell patch-clamp recordings were

obtained in ACSF supplemented with AP5 (50 mM) and tetrodotoxin (1 mM).

The composition of the intracellular solution (pH 7.3) was (in mM) KCl (120),

MgCl2 (1), CaCl2 (1), EGTA (10), HEPES (10), glucose (15). Cells were held at

a membrane potential of �70 mV.

Data analysis. The peak currents produced by different glycine concentrations

were plotted and fitted to the Hill equation

I=Imax ¼ ½Gly�n=ð½Gly�n + ECn50Þ

where I is the peak current, Imax the maximum whole cell current amplitude

produced by saturating glycine, [Gly] the glycine concentration, EC50 the

glycine concentration eliciting a current half of Imax, and n the Hill coefficient.

Dose-inhibition curves of strychnine were constructed for glycine-activated

currents obtained in the absence (I) and presence (IStry) of various concentra-

tions of strychnine and fitted to the equation

IStry=I ¼ ICn50=ð½Stry�n + ICn

50Þ

Individually saturating glycine concentrations were applied to cells expressing

a3 (1 mM) and a3P185L (0.1 mM). The IC50 is the strychnine concentration

that inhibited half of the glycine-induced current, and n is the Hill coefficient.

Statistical analysis (unpaired Student’s t-test and ANOVA) was performed

using Origin-4.1 (Microcal). Numerical data is reported as mean 7 s.e.m., with

n being the number of neurons studied. Significance levels are indicated as

*P o 0.05, **P o 0.01, ***P o 0.001.


ACKNOWLEDGMENTSWe thank U. Neumann for excellent technical assistance and S. Kirischuk forcomments on earlier versions of the manuscript. This work was supported by

the Deutsche Forschungsgemeinschaft (grant SFB 515 B2 to R.G.), by ChariteUniversity Medicine funding, and by the Medical Research Council (R.J.H.).




1. Fritschy, J.M. & Brunig, I. Formation and plasticity of GABAergic synapses: physiologicalmechanisms and pathophysiological implications. Pharmacol. Ther. 98, 299–323(2003).

2. Semyanov, A., Walker, M.C., Kullmann, D.M. & Silver, R.A. Tonically active GABAA receptors: modulating gain and maintaining the tone. Trends Neurosci. 27, 262–269(2004).

3. Legendre, P. The glycinergic inhibitory synapse. Cell. Mol. Life Sci. 58, 760–793 (2001).4. Meier, J. & Grantyn, R. A gephyrin-related mechanism restraining glycine receptor

anchoring at GABAergic synapses. J. Neurosci. 24, 1398–1405 (2004).5. Breustedt, J., Schmitz, D., Heinemann, U. & Schmieden, V. Characterization of

the inhibitory glycine receptor on entorhinal cortex neurons. Eur. J. Neurosci. 19,1987–1991 (2004).

6. Mori, M., Gahwiler, B.H. & Gerber, U. Beta-Alanine and taurine as endogenous agonistsat glycine receptors in rat hippocampus in vitro. J. Physiol. (Lond.) 539, 191–200(2002).

7. Flint, A.C., Liu, X. & Kriegstein, A.R. Nonsynaptic glycine receptor activation during earlyneocortical development. Neuron 20, 43–53 (1998).

8. Chattipakorn, S.C. & McMahon, L.L. Pharmacological characterization of glycine-gatedchloride currents recorded in rat hippocampal slices. J. Neurophysiol. 87, 1515–1525(2002).

9. Chattipakorn, S.C. & McMahon, L.L. Strychnine-sensitive glycine receptors depresshyperexcitability in rat dentate gyrus. J. Neurophysiol. 89, 1339–1342 (2003).

10. McGale, E.H., Pye, I.F., Stonier, C., Hutchinson, E.C. & Aber, G.M. Studies of the inter-relationship between cerebrospinal fluid and plasma amino acid concentrations innormal individuals. J. Neurochem. 29, 291–297 (1977).

11. Whitehead, K.J., Manning, J.P., Smith, C.G. & Bowery, N.G. Determination of theextracellular concentration of glycine in the rat spinal cord dorsal horn by quantitativemicrodialysis. Brain Res. 910, 192–194 (2001).

12. Roux, M.J. & Supplisson, S. Neuronal and glial glycine transporters have differentstoichiometries. Neuron 25, 373–383 (2000).

13. Kennedy, R.T., Thompson, J.E. & Vickroy, T.W. In vivo monitoring of amino acids bydirect sampling of brain extracellular fluid at ultralow flow rates and capillary electro-phoresis. J. Neurosci. Methods 114, 39–49 (2002).

14. Higuchi, M. et al. RNA editing of AMPA receptor subunit GluR-B: a base-paired intron-exon structure determines position and efficiency. Cell 75, 1361–1370 (1993).

15. Gurevich, I. et al. Altered editing of serotonin 2C receptor pre-mRNA in the prefrontalcortex of depressed suicide victims. Neuron 34, 349–356 (2002).

16. Bhalla, T., Rosenthal, J.J., Holmgren, M. & Reenan, R. Control of human potassiumchannel inactivation by editing of a small mRNA hairpin. Nat. Struct. Mol. Biol. 11,950–956 (2004).

17. Kohler, M., Burnashev, N., Sakmann, B. & Seeburg, P.H. Determinants of Ca2+permeability in both TM1 and TM2 of high affinity kainate receptor channels: diversityby RNA editing. Neuron 10, 491–500 (1993).

18. Powell, L.M. et al. A novel form of tissue-specific RNA processing produces apolipo-protein-B48 in intestine. Cell 50, 831–840 (1987).

19. Skuse, G.R., Cappione, A.J., Sowden, M., Metheny, L.J. & Smith, H.C. The neuro-fibromatosis type I messenger RNA undergoes base-modification RNA editing. NucleicAcids Res. 24, 478–485 (1996).

20. Breitinger, H.G. & Becker, C.M. The inhibitory glycine receptor-simple views of acomplicated channel. ChemBioChem 3, 1042–1052 (2002).

21. Kuhse, J., Schmieden, V. & Betz, H. Identification and functional expression of anovel ligand binding subunit of the inhibitory glycine receptor. J. Biol. Chem. 265,22317–22320 (1990).

22. Haliassos, A. et al. Detection of minority point mutations by modified PCR technique: anew approach for a sensitive diagnosis of tumor-progression markers. Nucleic Acids Res.17, 8093–8099 (1989).

23. Ireton, G.C., McDermott, G., Black, M.E. & Stoddard, B.L. The structure of Escherichiacoli cytosine deaminase. J. Mol. Biol. 315, 687–697 (2002).

24. Vincenzetti, S. et al. Possible role of two phenylalanine residues in the active site ofhuman cytidine deaminase. Protein Eng. 13, 791–799 (2000).

25. Stell, B.M. & Mody, I. Receptors with different affinities mediate phasic and tonicGABAA conductances in hippocampal neurons. J. Neurosci. 22, RC223 (2002).

26. Zhou, L. et al. Zebularine: a novel DNA methylation inhibitor that forms a covalentcomplex with DNA methyltransferases. J. Mol. Biol. 321, 591–599 (2002).

27. Cheng, J.C. et al. Inhibition of DNA methylation and reactivation of silenced genes byzebularine. J. Natl. Cancer Inst. 95, 399–409 (2003).

28. Yoo, C.B., Cheng, J.C. & Jones, P.A. Zebularine: a new drug for epigenetic therapy.Biochem. Soc. Trans. 32, 910–912 (2004).

29. Seeburg, P.H. & Hartner, J. Regulation of ion channel/neurotransmitter receptor functionby RNA editing. Curr. Opin. Neurobiol. 13, 279–283 (2003).

30. Backus, J.W. & Smith, H.C. Three distinct RNA sequence elements are required forefficient apolipoprotein B (apoB) RNA editing in vitro. Nucleic Acids Res. 20, 6007–6014 (1992).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

31. Wedekind, J.E., Dance, G.S., Sowden, M.P. & Smith, H.C. Messenger RNA editing inmammals: new members of the APOBEC family seeking roles in the family business.Trends Genet. 19, 207–216 (2003).

32. Meier, J., Juttner, R., Kirischuk, S. & Grantyn, R. Synaptic anchoring of glycine receptorsin developing collicular neurons under control of metabotropic glutamate receptoractivity. Mol. Cell. Neurosci. 21, 324–340 (2002).

33. De Saint, J., David-Watine, B., Korn, H. & Bregestovski, P. Activation of human a1 anda2 homomeric glycine receptors by taurine and GABA. J. Physiol. (Lond.) 535, 741–755 (2001).

34. Mangin, J.M. et al. Kinetics properties of the a2 homo-oligomeric glycinereceptor impairs a proper synaptic functioning. J. Physiol. (Lond.) 553, 369–386(2003).

35. Harvey, R.J. et al. GlyR a3: an essential target for spinal PGE2-mediated inflammatorypain sensitization. Science 304, 884–887 (2004).

36. Haverkamp, S. et al. Diversity of glycine receptors in the mouse retina: Localization ofthe a3 subunit. J. Comp. Neurol. 465, 524–539 (2003).

37. Grantyn, A., Moschovakis, A.K. & Kitama, T. Control of orienting movements: role ofmultiple tectal projections to the lower brainstem. Prog. Brain Res. 143, 423–438(2004).

38. Hikosaka, O. & Wurtz, R.H. Effects on eye movements of a GABA agonist and antagonistinjected into monkey superior colliculus. Brain Res. 272, 368–372 (1983).

39. Meyer, G., Kirsch, J., Betz, H. & Langosch, D. Identification of a gephyrin binding motifon the glycine receptor b subunit. Neuron 15, 563–572 (1995).

40. Danglot, L., Rostaing, P., Triller, A. & Bessis, A. Morphologically identified glycinergicsynapses in the hippocampus. Mol. Cell. Neurosci. 27, 394–403 (2004).

41. Juttner, R., Meier, J. & Grantyn, R. Slow IPSC kinetics, low levels of a1 subunitexpression and paired-pulse depression are distinct properties of neonatal inhibitoryGABAergic synaptic connections in the mouse superior colliculus. Eur. J. Neurosci. 13,2088–2098 (2001).

42. Henneberger, C. et al. GluR- and TrkB-mediated maturation of GABA receptor functionduring the period of eye opening. Eur. J. Neurosci. 21, 431–440 (2005).

43. Richerson, G.B. & Wu, Y. Dynamic equilibrium of neurotransmitter transporters: not justfor reuptake anymore. J. Neurophysiol. 90, 1363–1374 (2003).

44. Rivera, C. et al. The K+Cl� co-transporter KCC2 renders GABA hyperpolarizing duringneuronal maturation. Nature 397, 251–255 (1999).

45. Baker, A.J. et al. Changes in extracellular concentrations of glutamate, aspartate,glycine, dopamine, serotonin, and dopamine metabolites after transient global ischemiain the rabbit brain. J. Neurochem. 57, 1370–1379 (1991).

46. Meier, J., Akyeli, J., Kirischuk, S. & Grantyn, R. GABAA receptor activity and PKC controlinhibitory synaptogenesis in CNS tissue slices. Mol. Cell. Neurosci. 23, 600–613(2003).

47. Henneberger, C., Grantyn, R. & Rothe, T. Rapid genotyping of newborn gene mutantmice. J. Neurosci. Methods 100, 123–126 (2000).

48. Brewer, G.J. & Cotman, C.W. Survival and growth of hippocampal neurons in definedmedium at low density: advantages of a sandwich culture technique or low oxygen. BrainRes. 494, 65–74 (1989).

49. Meier, J. & Schmieden, V. Inhibition of a-subunit glycine receptors by quinoxalines.Neuroreport 14, 1507–1510 (2003).

50. Vannier, C. & Triller, A. Biology of the postsynaptic glycine receptor. Int. Rev. Cytol. 176,201–244 (1997).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

LINGO-1 negatively regulates myelination byoligodendrocytes

Sha Mi1, Robert H Miller2, Xinhua Lee1, Martin L Scott1, Svetlane Shulag-Morskaya1, Zhaohui Shao1,Jufang Chang3, Greg Thill1, Melissa Levesque1, Mingdi Zhang1, Cathy Hession1, Dinah Sah1, Bruce Trapp4,Zhigang He3, Vincent Jung1, John M McCoy1 & R Blake Pepinsky1

The control of myelination by oligodendrocytes in the CNS is poorly understood. Here we show that LINGO-1 is an important

negative regulator of this critical process. LINGO-1 is expressed in oligodendrocytes. Attenuation of its function by dominant-

negative LINGO-1, LINGO-1 RNA-mediated interference (RNAi) or soluble human LINGO-1 (LINGO-1-Fc) leads to differentiation

and increased myelination competence. Attenuation of LINGO-1 results in downregulation of RhoA activity, which has been

implicated in oligodendrocyte differentiation. Conversely, overexpression of LINGO-1 leads to activation of RhoA and inhibition of

oligodendrocyte differentiation and myelination. Treatment of oligodendrocyte and neuron cocultures with LINGO-1-Fc resulted

in highly developed myelinated axons that have internodes and well-defined nodes of Ranvier. The contribution of LINGO-1 to

myelination was verified in vivo through the analysis of LINGO-1 knockout mice. The ability to recapitulate CNS myelination

in vitro using LINGO-1 antagonists and the in vivo effects seen in the LINGO-1 knockout indicate that LINGO-1 signaling may be

critical for CNS myelination.

Myelination in the CNS is coordinated by oligodendrocytes and theirinteractions with axons1,2. A detailed understanding of the mechanismsof CNS myelination and of the agents that facilitate oligodendrocytedifferentiation is central to developing therapies for demyelinatingdiseases1, as damage to the myelin sheath disrupts axonal conduction,which results in severe neurological dysfunction3–7. Oligodendrocytesarise from A2B5-positive (A2B5+) progenitor cells that originate fromdistinct locations during late embryonic development. These A2B5+

cells subsequently proliferate, migrate widely throughout the CNS andmature into O4-positive (O4+) pre-myelinating oligodendrocytesbefore differentiating into mature myelin basic protein–positive(MBP+) myelinating oligodendrocytes in white matter. In the CNS,an individual oligodendrocyte can myelinate up to 40 different axonsegments. Myelin, a discontinuous fatty insulation that ensheathsaxons, is critical for the normal functioning of the CNS and peripheralnervous system2,8–10. Several different receptor-ligand pairs are impor-tant for oligodendrocyte differentiation and/or myelination, includingErbB2 and neuregulin11, TrkA and NGF12 and Notch and Jaggedor F3/compactin13–16. Although downregulation of the Rho signalingpathway has been implicated in oligodendrocyte differentiationand myelination17, the connection to the cell surface receptors ispoorly understood.

In this study, we identified a pathway for RhoA regulation byLINGO-1 in oligodendrocytes. LINGO-1 is a transmembrane protein

that contains a leucine-rich repeat (LRR) and an immunoglobulindomain. It is selectively expressed in brain and spinal cord andfunctions as a component of the NgR1/p75 and NgR1/Taj(Troy)signaling complexes that regulate CNS axon growth18–20. In neurons,these complexes activate RhoA in the presence of myelin inhibitors toinhibit neurite outgrowth. Here we show that LINGO-1 is alsoexpressed in oligodendrocytes, where it negatively regulates oligoden-drocyte differentiation and axon myelination. The effects on oligoden-drocytes indicate an even greater role for LINGO-1 in the CNS than wasoriginally anticipated and may provide a new therapeutic target fortreatment of demyelination and dysmyelination disorders.

RESULTS

LINGO-1 is expressed in oligodendrocytes

Analyses of LINGO-1 expression in purified populations of CNS cellsby polymerase chain reaction after reverse transcription (RT-PCR)showed high expression in neurons, lower expression in oligodendro-cytes and a lack of expression in astrocytes (Fig. 1a). Expression ofLINGO-1 in oligodendrocytes was confirmed by in situ hybridizationin sections derived from adult rat optic nerve. Oligodendrocytes thatwere labeled with an antisense LINGO-1 probe were also stained withan antibody to adenomatus polyposis coli (CC1 antibody; Fig. 1b), amarker of oligodendrocytes. No specific labeling was observed using aLINGO-1 sense probe (data not shown). The presence of LINGO-1


1Department of Discovery Biology, Biogen Idec, Inc., 14 Cambridge Center, Cambridge, Massachusetts 02142, USA. 2Department of Neuroscience, Case Western ReserveUniversity School of Medicine, Cleveland, Ohio 44106, USA. 3Division of Neuroscience, Children’s Hospital, Harvard Medical School, 320 Longwood Avenue, Boston,Massachusetts 02115, USA. 4Department of Neuroscience, NC30, Lerner Research Institute, Cleveland Clinic Foundation, 9500 Euclid Avenue, Cleveland, Ohio 44195,USA. Correspondence should be addressed to S.M. ([email protected]).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

protein in oligodendrocytes was confirmed using immunohistochem-istry to show that most CC1+ oligodendrocytes were also labeled byanti–LINGO-1(Fig. 1c). Pre-adsorption of anti–LINGO-1 by additionof competing LINGO-1-Fc completely ablated the signal (data notshown). LINGO-1 mRNA was detected in all stages of the oligoden-drocyte lineage by Taqman RT-PCR.

LINGO-1 antagonists promote oligodendrocyte differentiation

To define LINGO-1 function in oligodendrocytes, endogenous LINGO-1 expression was reduced by infection with LINGO-1 RNAi lentivirus

and was confirmed by RT-PCR (Fig. 2b). Reduction of LINGO-1resulted in more highly differentiated, mature oligodendrocytes ascompared with control infected cells, as was evident by increases inthe length of cell processes and by the presence of abundant myelinsheet structures (Fig. 2a,c). In cells that expressed LINGO-1 RNAi, therewere three times as many mature oligodendrocytes as in control cultures(Fig. 2c). These data indicate that LINGO-1 may negatively regulateoligodendrocyte differentiation. In additional analyses, lentiviralexpression of a dominant-negative LINGO-1 (DN-LINGO-1), whichcontains a truncation of the cytoplasmic signaling domain, was used asan alternative means to block endogenous LINGO-1 function. Oligo-dendrocytes infected with lentivirus that contained the DN-LINGO-1or the control (no insert) were assayed morphologically for differentia-tion (Fig. 2d). Consistent with the RNAi effect, DN-LINGO-1 alsopromoted oligodendrocyte differentiation, producing a fivefoldincrease in the number of mature oligodendrocytes (Fig. 2e). Incontrast, overexpression of full-length LINGO-1 (FL-LINGO-1) hadthe opposite effect and inhibited differentiation (Fig. 2d,e), as wasevident by the presence of less-developed processes and an 80%reduction in the number of mature oligodendrocytes as comparedwith the control (Fig. 2e). Increased MBP is a marker for oligoden-drocyte differentiation and was measured directly by western blotting(Fig. 2f). Cells infected with lentivirus that contained the DN-LINGO-1or FL-LINGO-1 DNA produced fivefold higher and 90% loweramounts of MBP, respectively (Fig. 2f). Expression of FL-LINGO-1and DN-LINGO-1 in infected cells was confirmed by western blotting(Fig. 2f). These data support the notion that endogenous LINGO-1negatively regulates oligodendrocyte differentiation.

Exogenously added LINGO-1-Fc is an antagonist of the NgR1receptor complex signaling pathway18. We therefore examined theeffect of LINGO-1-Fc addition on the differentiation of A2B5+ oligo-dendrocyte progenitors and found that it promoted their differentia-tion into O4+ cells in a concentration-dependent manner (Fig. 2g,h).LINGO-1-Fc treatment also resulted in oligodendrocytes with longer

Merge

Merge

CC1LINGO-1

GAPDH

LINGO-1

CC1Antisense

Ast

rocy

tes

Olig

oden

droc

ytes

CG

neu

rons

a b

c

Figure 1 LINGO-1 is expressed in oligodendrocytes. (a) RT-PCR analysis of

LINGO-1 mRNA expression in P13 CG neuronal, P2 oligodendrocyte and P4

astrocyte cultures. GAPDH expression was analyzed from the same samples

as an internal control. (b) In situ hybridization analysis of LINGO-1 mRNA

expression in adult rat optic nerve sections. Red, probed with LINGO-1

antisense mRNA; green, stained with CC1 antibody; yellow, merge of red andgreen. Scale bar, 100 mm. (c) Immunohistochemistry of LINGO-1 in the

lateral ventricle region of P7 rat cortex tissue sections. Red, stained with

anti–LINGO-1; green, stained with CC1 antibody; blue, DAPI staining; yellow,

merge of red and green. Scale bar, 100 mm. Arrows denote examples of

LINGO-1– and CCI-positive oligodendrocytes.

Protein concentration (µg/ml)301052

LIN

GO

-1-F

c

LIN

GO

-1-F

c

LIN

GO

-1-F

c

LIN

GO

-1-F

c

Con

trol

Fc

Con

trol

Fc

Con

trol

Fc

Con

trol

Fc

LINGO-1 RNAiControl RNAi

β-actin

β-actin

LINGO-1

LINGO-1

RNAi

Contro

l RNAi

Contro

l

**P < 0.001

**P < 0.001

LINGO-1 RNAiControl RNAiNo RNAi0

5

10

15

20

25

Mat

ure

olig

oden

droc

ytes

(per

cent

age

of to

tal)

Mat

ure

olig

oden

droc

ytes

(per

cent

age

of to

toal

)

Mat

ure

olig

oden

droc

ytes

(per

cent

age

of to

tal)

DN–LINGO-1FL–LINGO-1Control

DN-LINGO-1FL-LINGO-1Control

75

50

25

0

*****

* P < 0.05

** P < 0.001

0

10

20

30

50

40

60

70

LINGO-1

MBP

Contro

l

DN–LIN

GO-1

FL–LI

NGO-1

Control Fc

LINGO-1-Fc

a

d e

g

h

b c

f

Figure 2 LINGO-1 antagonists promote oligodendrocyte differentiation. (a) Immunohistochemistry of oligodendrocytes

infected with LINGO-1 RNAi and control RNAi. Red, O4+ cells. Arrow denotes myelin sheet structures. Scale bar, 50 mm.

(b) RT-PCR analysis of LINGO-1 mRNA expression from RNAi-infected oligodendrocytes and an uninfected control. b-actin

was used as an internal standard. (c) Quantification of mature oligodendrocytes in RNAi-treated cultures. (d) Immuno-

histochemistry of oligodendrocytes infected with FL-LINGO-1, DN-LINGO-1 and control lentivirus (day 2). Red, O4+ cells.

Arrows denote myelin sheet structures. Scale bar, 50 mm. (e) Quantification of mature oligodendrocytes. (f) MBP, LINGO-1

and b-actin expression in oligodendrocytes infected with FL-LINGO-1, DN-LINGO-1 and control lentivirus, as detected by

western blotting. (g) Immunohistochemistry of oligodendrocyte differentiation after LINGO-1-Fc treatment (5 mg/ml). Red,

O4+ cells. Scale bar, 50 mm. (h) Dose-dependent differentiation of oligodendrocytes after LINGO-1-Fc treatment. Data in c, e

and h were quantified by counting the number of mature oligodendrocytes (identified by the presence of myelin sheets) as a

percentage of total O4+ oligodendrocytes. For each sample, approximately 800 cells were counted.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

processes and more-abundant myelin sheet structures (Fig. 2g). Thesedata indicate that LINGO-1-Fc, like DN-LINGO-1 and LINGO-1RNAi, inhibits endogenous LINGO-1 function.

LINGO-1 antagonists regulate RhoA and Fyn

A strong candidate signaling pathway that is implicated in the controlof oligodendrocyte differentiation is the Rho family of GTPases. RhoGTPases regulate cellular morphology, and reduced RhoA-GTPamounts are required for oligodendrocyte differentiation17. To deter-mine whether LINGO-1 signals through the RhoA pathway, RhoA-GTP levels in cell lysates of oligodendrocytes treated with LINGO-1-Fcwere compared with levels in the corresponding control. A substantial

70% reduction in RhoA-GTP was seen after LINGO-1-Fc treatment(Fig. 3a), indicating that attenuation of LINGO-1 function may induceoliogodendrocyte differentiation by downregulating RhoA-GTP, with asubsequent increase in MBP expression. Similar reductions in RhoAGTP amounts were seen when oligodendrocytes were treated with DN-LINGO-1 or with LINGO-1 RNAi (data not shown).

The activity of RhoA GTPase is regulated by Fyn kinase17. IncreasedFyn expression and phosphorylation correlate with oligodendrocytedifferentiation17,21. To test if LINGO-1 antagonists affect Fyn function,Fyn expression and phosphorylation were measured directly by westernblotting. DN-LINGO-1 treatment resulted in twofold increases in Fynprotein and in Fyn phosphorylation (Fig. 3b). Conversely, when cellsexpressing FL-LINGO-1 were analyzed, Fyn expression and phosphor-ylation were reduced by 50% (Fig. 3b).

LINGO-1 antagonists promote axonal myelination

The effects of LINGO-1-Fc treatment on oligodendrocyte differentia-tion suggested that LINGO-1-Fc might facilitate CNS myelination. Totest this, we set up oligodendrocyte and neuron cocultures to study theeffects of LINGO-1-Fc on myelination. In cultures of rat primaryoligodendrocytes and dorsal root ganglion (DRG) neurons, low basalamounts of myelination were observed, which are consistent withprevious findings22. Addition of LINGO-1-Fc had a profound effecton the myelination in such cultures. Treatment with LINGO-1-Fc for2 weeks resulted in robust axonal myelination, as was evident by thepresence of MBP+ myelinated axons (Fig. 4a). The number of myeli-nated axons that developed in treated cultures was dependent on theLINGO-1-Fc concentration. This is reflected in the number of clusters

Total RhoA

RhoA-GTP

β-actin

pFyn

FynCon

trol

Contro

l Fc

DN-LIN

GO-1

LINGO-1

-Fc

FL-LI

NGO-1a b

Figure 3 LINGO-1 antagonists regulate RhoA and Fyn. (a) RhoA-GTP

amounts in oligodendrocytes treated with LINGO-1-Fc, as detected by

western blotting. (b) Fyn expression and phosphorylation (pFyn) in

oligodendrocytes infected with lentivirus carrying FL-LINGO-1, DN-LINGO-1

or control plasmids, as detected by western blotting.

β-actin

LINGO-1

MBP

Contro

l

DN-LIN

GO-1

FL-LI

NGO-1

Contro

l

DN-LIN

GO-1

FL-LI

NGO-10

25

50

75

100P < 0.001

**

Mye

linat

ing

MB

P+

cells

/wel

l

ControlDN-LINGO-1FL-LINGO-1Control Fc

Con

trol

Fc

LINGO-1-Fc

LIN

GO

-1-F

c

Node ofRanvier

LINGO-1

-Fc

LINGO-1

-Fc

Contro

l Fc

Contro

l Fc

kDa

252014

βIII-tubulin

βIII-tubulinMBP

MAG

Protein concentration (µg/ml)

301052

LIN

GO

-1-F

c

LIN

GO

-1-F

c

LIN

GO

-1-F

c

LIN

GO

-1-F

c

Con

trol

Fc

Con

trol

Fc

Con

trol

Fc

Con

trol

Fc

**

****

*

* P < 0.05** P < 0.001

60

50

40

30

20

10

0Mye

linat

ing

MB

P+

cel

ls/w

ella

e f g

h i

b c d

Figure 4 LINGO-1 antagonists promote axonal myelination by oligodendrocytes.

(a,b) Immunohistochemistry showing (a) visual and (b) quantitative analysis of

myelination in cocultures that were treated with LINGO-1-Fc (10 mg/ml) for 2 weeks.

Red, anti-MBP. For each sample, ten fields were counted. White arrows denote clusters

of MBP+ myelinated internodes, each of which was derived from a single MBP+

oligodendrocyte. Scale bar, 50 mm. (c,d) Western blots of 4-week cocultures using (c)

anti-MAG and (d) anti-MBP. (e) Confocal microscopy of cocultures that were treated

with LINGO-1-Fc for 4 weeks. Red, MBP staining of myelinated axons. Scale bar,

200 mm. (f) Electron microscopy analysis of cocultures that were treated with

LINGO-1-Fc or control Fc for 4 weeks. Node of Ranvier structure is indicated. Scale bar, 1.5 mm. (g,h) Myelination in cocultures that were infected with

FL-LINGO-1, DN-LINGO-1 and control lentivirus for 2 weeks as shown by (g) immunohistochemical and (h) quantitative analyses. Red, anti-MBP. Scale bar,

50 mm. Arrows denote clusters of myelinated internodes, each of which was derived from a single MBP+ cell. (i) Western blots from cultures infected with

FL-LINGO-1, DN-LINGO-1 and control lentivirus analyzed for MBP and for LINGO-1 using an antibody to the hemagglutinin tag.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

of internodes (myelinated axons) that were derived from single MBP+

cells (Fig. 4b). LINGO-1-Fc–induced myelination was also detectedbiochemically. Western blot analysis for MAG, a protein that isexpressed at the onset of myelination, showed a threefold increase(Fig. 4c), whereas MBP, the major protein component of myelin,showed a tenfold increase in LINGO-1-Fc–treated cultures (Fig. 4d).Like MAG, levels of OMgp, MOG and CNPase (additional myelincomponents) were upregulated by two- to threefold after LINGO-1-Fctreatment (data not shown). Myelination in the presence of LINGO-1-Fc was further confirmed by confocal microscopy, which verified thatMBP had encapsulated the axons (Fig. 4e). Multiple well formedinternodes were observed by electron microscopy in cultures treatedwith LINGO-1-Fc, as well as structures that closely resembled nodes ofRanvier (Fig. 4f). Only occasional myelinated segments and no nodesof Ranvier were detected in the control cultures (Fig. 4f).

The effect of LINGO-1-Fc on axonal myelination was furtherconfirmed using DN-LINGO-1. Expression of DN-LINGO-1 increasedthe total number of myelinating MBP+ cells five- to tenfold whencompared with controls (Fig. 4g,h). In contrast, overexpression of FL-LINGO-1 decreased the number of myelinating MBP+ cells by 50%, ascompared with controls (Fig. 4g,h). Western blot analysis was used toquantify MBP in the cultures. DN-LINGO-1 produced a tenfoldincrease in MBP, whereas FL-LINGO-1 caused a 50% reduction inMBP (Fig. 4i). Expression of FL-LINGO-1 and DN-LINGO-1 proteinsin cultures was confirmed by western blotting (Fig. 4i). These studiesfurther indicate that endogenous LINGO-1 inhibits myelination andthat antagonism of LINGO-1 can reverse the inhibition.

Early-onset CNS myelination in LINGO-1 knockout mice

The role of LINGO-1 in oligodendrocyte differentiation and myelina-tion was further verified using LINGO-1 knockout mice (Fig. 5). Wefirst evaluated cultured oligodendrocytes from LINGO-1 knockoutmice for potential changes in differentiation. Oligodendrocytes thatwere more highly differentiated, and a larger percentage of matureoligodendrocytes were observed, in LINGO-1 knockout than in cul-tures from wild-type littermates (Fig. 5d,e). Because the onset ofmyelination in normal mouse development typically occurs on post-

natal day (P) 5, we next examined myelination in P1 spinal cords fromthe wild-type and knockout mice by electron microscopy. Consistentwith the in vitro cultures, spinal cords from LINGO-1 knockout micecontained more myelinated axon fibers than did their wild-typelittermates (Fig. 5f,g). No obvious changes in peripheral nervoussystem sciatic nerve were detected in the knockout mice, suggestingthat the myelination effects were limited to the CNS (data not shown).

DISCUSSION

This work shows that LINGO-1 is expressed in oligodendrocytes and isa negative regulator of oligodendrocyte differentiation and axonalmyelination. Loss of LINGO-1 function using dominant-negativeLINGO-1, LINGO-1 RNAi or LINGO-1-Fc led to increased processlength, branching, myelin sheet formation and myelination, whereasoverexpression of LINGO-1 led to inhibition of oligodendrocytedifferentiation and myelination. The amount of functional LINGO-1therefore correlates with the extent of oligodendrocyte maturation andmyelination. Studies from the LINGO-1 knockout mice provide in vivosupport for the role of LINGO-1 in oligodendrocyte biology.

To understand the molecular mechanism of LINGO-1, we investi-gated downstream signaling molecules known to be involved inoligodendrocyte differentiation and/or myelination, Fyn and RhoA.Fyn kinase activates p190 RhoAGAP to downregulate RhoA activity,resulting in oligodendrocyte differentiation17,23,24. Fyn disruptionresults in defects in oligodendrocyte differentiation and hypomyelina-tion23,25, whereas activation of Fyn by cross-linking of F3/contactin orintegrin a6b1 (refs. 17,26) results in their differentiation. Consistentwith these observations, LINGO-1 antagonists increased the expressionand phosphorylation status of Fyn and decreased the amount ofactivated RhoA-GTP. The role of LINGO-1 in oligodendrocyte-depen-dent myelination through RhoA correlates with the recent report thatinhibition of Rho kinase (effector of Rho) coordinates Schwann cell–dependent myelination27.

The ability to recapitulate myelination in vitro was particularlyexciting to us. Treatment of oligodendrocyte and neuron cultureswith LINGO-1-Fc resulted in highly developed myelinated axons thatwere separated by specialized regions resembling nodes of Ranvier.

5.7 kb

10.4 kb

HET WTWT KO

β-actin

LINGO-1

NeoGFP

recombinationBacterial

Intron

5′ UTR LINGO-1 coding sequence

WT

WT

KO

KO

WT KOWT KO

**P < 0.001

**P < 0.004

15

10

5

0Mye

linat

ed a

xon

fiber

s/se

ctio

n

0

10

20

30

40

50

Mat

ure

olig

oden

droc

ytes

(per

cent

age

of to

tal)

a b c

d e f g

Figure 5 Oligodendrocyte differentiation and myelination in LINGO-1 knockout mice. (a) Schematic showing strategy used to generate knockout. (b) Southern

blot analysis of Eco RI-digested DNA from heterozygous (HET) and wild-type (WT) mice. Wild-type band, 10.4 kb; knockout band, 5.7 kb. (c) Northern blot

analysis of LINGO-1 mRNA from wild-type and knockout mice. b-actin expression was used as an internal control. (d) Immunohistochemistry of cultured

premyelinating O4+ oligodendrocytes from LINGO-1 knockout mice (KO) and wild-type littermates. Red, O4+ cells. Scale bar, 50 mm. (e) Quantification of

mature oligodendrocytes as a percentage of total O4+ oligodendrocytes. (f,g) Electron microscopy showing (f) visual and (g) quantitative analysis of myelinated

axon fibers in LINGO-1 knockout and wild-type spinal cords from P1 mice. Scale bar, 2 mm. Four spinal cords were analyzed for each; for each sample,

myelinated axon fibers from ten fields were counted. Error bars show s.e.m. Arrows denote myelinated axons.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

In vivo, these nodes are essential for the rapid saltatory conduction ofelectrical impulses8–10. Whereas oligodendrocyte-induced clustering ofsodium channels in cultured rat retinal ganglion cells hasbeen reported28, to our knowledge the development of nodes ofRanvier has not previously been reported in any in vitro CNSmyelination system.

The LINGO-1 knockout mouse allowed us to determine in vivo thenegative effect of LINGO-1 on oligodendrocyte differentiation andmyelination. Early-onset myelination in P1 spinal cord from knockoutmice confirmed our in vitro findings. Increases in oligodendrocytedifferentiation and myelination were also observed by immunohisto-chemical analysis of the corpus callosum and optical nerve tissues fromLINGO-1 knockout mice as compared with those from wild-type mice(data not shown). The knockout mice should serve as a valuable toolfor further elucidating the biology of LINGO-1 in normal developmentand in models of CNS disease.

Several common features of LINGO-1 function are shared betweenneurons and oligodendrocytes. First, LINGO-1 inhibits process elon-gation in both cell types. In neurons, LINGO-1 inhibits axonalgrowth18, and it inhibits process outgrowth and differentiation inoligodendrocyte precursors. Second, LINGO-1-Fc reverses the inhibi-tory properties of endogenous LINGO-1. In neurons, LINGO-1-Fcreverses the inhibition of axonal growth that is caused by myelin18, andin oligodendrocytes, LINGO-1-Fc promotes differentiation fromA2B5+ or premyelinating O4+ to mature MBP+ myelinating oligoden-drocytes. Third, LINGO-1 regulates the amount of RhoA-GTP, aregulator of actin polymerization and of cytoskeletal structure andmorphology. In neurons, LINGO-1-Fc antagonizes the NgR1 receptorsignaling pathway to reduce RhoA-GTP amounts, thereby promotingneurite outgrowth18. In oligodendrocytes, LINGO-1 antagonistsreduce RhoA-GTP to promote oligodendrocyte differentiation andmyelination. We propose that antagonism of LINGO-1 signaling inoligodendrocytes provides a molecular mechanism involving RhoAregulation for coordinating myelination.

In summary, we have provided in vitro and in vivo evidence thatendogenous LINGO-1 is a negative regulator of myelination byoligodendrocytes. The discovery of a pronounced role for LINGO-1in oligodendrocyte biology is an invaluable step for understandingCNS myelination.

METHODSPrimary cell culture. All procedures for animal use were performed in

accordance with US National Institutes of Health guidelines and were approved

by the Biogen Idec Institutional Committee for Animal Care. Embryonic DRG

neurons were grown in vitro as previously described22,28. Briefly, DRGs that

were dissected from embryonic day (E) 14 to E17 Long Evans rats were plated

on coverslips coated with poly-L-lysine (100 mg/ml). They were grown for 2

weeks in Neurobasal medium supplemented with B27 (Invitrogen). To remove

proliferating glial cells, the cultures were pulsed twice with fluorodeoxyuridine

(20 mM) from days 2–6 and from days 8–10. Enriched populations of

oligodendrocytes were grown from female Long Evans P2 rats. Briefly, the

forebrain was dissected and placed in Hank’s buffered salt solution (HBSS;

Invitrogen). The tissue was cut into 1-mm fragments and was incubated at

37 1C for 15 min in 0.01% trypsin and 10 mg/ml DNase. Dissociated cells were

plated on poly-L-lysine-coated T75 tissue culture flasks and were grown at

37 1C for 10 d in DMEM medium with 20% fetal calf serum (Invitrogen).

Oligodendrocyte precursors (A2B5+) were collected by shaking the flask over-

night at 200 rpm at 37 1C, resulting in a 95% pure population. Cultures were

maintained in high-glucose Dulbecco’s modified Eagle’s medium (DMEM)

with FGF PDGF (10 ng/ml; Peprotech) for 1 week. Removal of FGF PDGF

allowed A2B5+ cells to differentiate into O4+ premyelinating oligodendrocytes

after 3–7 d, and to differentiate into O4+ and MBP+ mature oligodendrocytes

after 7–10 d. These differentiation states are readily apparent from changes in

morphology: A2B5+ cells are bipolar in shape, O4+ premyelinating oligoden-

drocytes have longer and more-branched processes and MBP+ mature oligo-

dendrocytes contain myelin sheet structures between processes. For assessing

differentiation, A2B5+ cells were plated in 4-well slide chambers in FGF PDGF–

free growth medium supplemented with 10 ng/ml CNTF and 15 nM triiodo-L-

thyronine and were immediately treated with LINGO-1-Fc, DN-LINGO-1,

FL-LINGO-1 or LINGO-1 RNAi. After 48 h (72 h for RNAi), cultures were

stained with antibody to O4, and the number of total O4+ and mature O4+

oligodendrocytes was quantified. Samples were analyzed in duplicate. For

coculture studies, A2B5+ oligodendrocytes were added to DRG neuron drop

cultures in the presence or absence of 10 mg/ml LINGO-1-Fc. The culture

medium (Neurobasal medium supplemented with B27 and 100 ng/ml NGF)

was changed, and fresh LINGO-1-Fc was added to the cells every 3 d. To

identify changes in myelination, 2-week-old cultures were labeled with anti-

bodies, and 4-week-old cultures were subjected to SDS-PAGE followed by

western blot analyses (see below) and electron microscopy. Myelinated axons in

the 2-week-old cultures were quantified by counting the number of myelinated

internode bundles that were derived from single MBP+ oligodendrocytes.

Samples were analyzed in duplicate. Error bars denote individual determina-

tions. P values in all studies were determined using a one-way analysis

of variance.

Immunohistochemistry. Monoclonal antibodies against O4, MBP and CNPase

were from Sternberger Monoclonals; antibody to APC (clone CC-1; ref. 29) was

from Calbiochem. Other antibodies were to bIII tubulin (Covance), hemag-

glutinin (Roche) LINGO-1 (Biogen Idec), Fyn (Santa Cruz Biotechnology) and

phospho-Fyn (Biosource). To visualize antibody labeling in tissue sections or in

cell cultures grown on chamber slides, samples were prefixed in 4% parafor-

maldehyde and were incubated with the indicated antibodies using standard

protocols18. Tissue sections were incubated in primary antibodies overnight at

4 1C, washed thoroughly and incubated with an appropriate Alexa-labeled

secondary antibody (Molecular Probes) for 2 h. They were then mounted in

VectaMount (Victor) and visualized by fluorescence microscopy. Cultures were

incubated for 2 h in primary antibodies and 1 h in secondary antibodies and

were visualized by fluorescence microscopy.

In situ hybridization. Rat optic nerve sections were prepared and processed as

described18 and were probed with digoxigenin-labeled LINGO-1 antisense and

sense RNA. Sections were stained using Tyramide Signal Amplification (Amer-

sham Biosciences) plus fluorescent anti-digoxigenin conjugated antibody kit

(Perkin Elmer) per manufacturer’s instructions. For combined in situ and

immunofluorescence analyses, cultures were first probed with the digoxigenin-

labeled RNAs and then with anti-CC1.

RT-PCR. mRNA extracted (Ambion kit) from rat brains was subjected to RT-

PCR using forward primer (5¢-AGAGACATGCGATTGGTGA-3¢) and reverse

primer (5¢-AGAGATGTAGACGAGGTCATT-3¢). Taqman RT-PCR was used to

quantify LINGO-1 mRNA extracted from rat brains (Ambion kit), using

forward primer (5¢-CTTTCCCCTTCGACATCAAGAC-3¢) and reverse primer

(5¢-CAGCAGCACCAGGCAGAA-3¢) and 6-carboxyfluorescein-labeled probe

(5¢-ATCGCCACCACCATGGGCTTCAT-3¢). The primers and FAM-labeled

probes were designed using Primer Express v1.0 (Applied Biosystems). Data

were normalized to GAPDH levels.

LINGO-1-Fc protein purification. LINGO-1-Fc (residues 1–532 of human

LINGO-1 fused to the hinge and Fc region of human IgG1) was expressed

in CHO cells and was purified on protein A–Sepharose (Amersham

Biosciences). The purified protein (495% pure) ran on SDS-PAGE with

Mr ¼ 90 kDa under reducing conditions and Mr ¼ 180 kDa under nonredu-

cing conditions.

FL-LINGO-1 and DN-LINGO-1 plasmid construction and cell infections.

Full-length mouse LINGO-1 (FL-LINGO-1; amino acid residues 34–614) DNA

sequence was inserted into the NotI and BamH1 sites of HRST-IRESeGFP

lentivirus vector (gift from R. Mulligan, Harvard University) using oligonu-

cleotide primers 5¢-GAGGATCTCGACGCGGCCGCATGGAGACAGACACA

CTCCTG-3¢ and 5¢-GGGGCGGAATTGGATCCTCACAGATCCTCTTCTGAGA

TGAG-3¢. Dominant-negative LINGO-1 (DN-LINGO-1; amino acid residues


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

34–581) was inserted into the NotI and BamH1 sites of the HRST-IRESeGFP

lentivirus vector using primers 5¢-GAGGATCTCGACGCGGCCGCATGGAGA

CAGACACACTCCTG-3¢ and 5¢-GATACGGATC CTCAGCCTTTGCCCCGG

CTCCATAGAAACAGC-3¢. FL-LINGO-1 and DN-LINGO-1 plasmids were

transfected into 293 cells to produce lentivirus as described30. Oligodendrocytes

or cocultures were infected with lentivirus at a multiplicity of infection of two

per cell.

LINGO-1 RNAi construction and cell infections. Mouse and rat LINGO-1

DNA sequences were compared to identify homologous regions for use as

candidate small hairpin RNAs. CH324, for lentivirus expression of LINGO-1

RNAi30, was constructed by annealing the following oligonucleotides: 5¢-TGATCGTCATCCTGCTAGACTTCAAGAGAGTCTAGCAGGATGACGATCTT

TTTTC-3¢ and 5¢-TCGAGAAAAAAGATCGTCATCCTGCTAGACTCTCTT

GAAGTCTAGCAGGATGACGATCA-3¢ and ligating to HpaI- and XhoI-

digested pLL3.7. Control RNAi was designed from the same oligonucleotide

sequence with the nucleotide changes indicated in lower-case letters: 5¢-TGATCcTCATcCttCTAtACTTCAAGAGAGTgTAGCAGGATGAcGATCTTTTT

TCTCGA-3¢ and 5¢-TCGAGAAAAAAGATCGTCATCCTGCTAGACTCTCTTG

AAGTaTAGaAGGATGACGATCA-3¢. RNAi lentiviruses carrying green fluores-

cent protein (GFP) were generated as described30. Oligodendrocytes were

infected with lentivirus at a multiplicity of infection of two per cell and were

stained for O4 after 3 d. In cultures treated with either control or LINGO-1

RNAi, approximately 80% of the oligodendrocytes were GFP positive. Total cell

number was not altered by the RNAi treatments. To quantify the effects of

RNAi on differentiation, only GFP-expressing oligodendrocytes were counted.

RhoA Assay. Progenitor oligodendrocytes were treated for 2 d with LINGO-1-

Fc or control Fc. The cells were lysed, and RhoA-GTP and total RhoA amounts

were determined as described17.

Western blots. To quantify expression of myelin proteins, cells were grown on

35-mm plates, harvested and lysed in 200 ml lysis buffer (50 mM HEPES,

pH 7.5; 150 mM NaCl; 1.5 mM MgCl2; 1 mM EGTA; 1% Triton X-100 and

10% glycerol). The lysates were clarified by centrifugation, boiled in Laemmli

sample buffer and subjected to SDS-PAGE on a 4–20% gradient gel. They were

then analyzed by western blotting with monoclonal anti-MBP and monoclonal

antibody to MAG (Zymed) and with horseradish peroxidase–conjugated anti–

mouse antibody (Roche) as a detection reagent. Anti–b-tubulin was used as an

internal control. To quantify expression and phosphorylation of Fyn, progeni-

tor oligodendrocytes were treated for 5 d with DN-LINGO-1, FL-LINGO-1 or

control lentivirus and were then lysed and subjected to western blotting. b-actin

was used as an internal control.

Electron microscopy. For ultrastructural analyses, DRG neuron and oligo-

dendrocyte cultures were grown for 4 weeks in the presence of

10 mg/ml LINGO-1-Fc or control Fc. They were then fixed in 2.5% gluter-

aldehyde in 0.1 M cacodylate buffer, pH 7.4, and incubated with 1% osmium

tetroxide/1.5% potassium ferrocyanide and were washed extensively. After

en bloc staining in uranyl acetate, samples were dehydrated and embedded in

TAAB Epon. Sections were collected on 300-mesh hex grids, were stained with

uranyl acetate and lead citrate and were examined on a Jeol 100CX microscope

at 80kV31,32. For the ultrastructural analysis in tissues, mice were perfused with

2.5% gluteraldehyde in 0.1 M cacodylate buffer, pH 7.4. Spinal cords were

excised and then were fixed, stained and processed as described above.

Generation and genotyping of LINGO-1 knockout mice. LINGO-1 knockout

mice (Fig. 5a) were generated with a GFP/neo (neomycin-resistance gene)

replacement vector that targeted the entire single-exon coding sequence of

LINGO-1 (ref. 18). Mouse genomic 129/SvJ DNA was isolated from a lambda

genomic library (Stratagene #946313). A 14.6-kb EcoRV fragment was sub-

cloned into pBSK+ and then was targeted by homologous recombination in

bacteria33 to insert the enhanced GFP (eGFP) reporter gene at the initiating

ATG. The final construct deleted the entire 1–1,841 nucleotides of the single-

exon coding sequence of LINGO-1. This construct was used to target the

LINGO-1 locus in D3 (129/Sv) embryonic stem cells. Correctly targeted cells

were identified by Southern blotting of EcoRI-digested embryonic stem cell

DNA and were injected into C57Bl/6 blastocysts to generate chimeric mice.

Chimeras were crossed to C57Bl/6 mice to generate heterozygous founder mice.

Genotypes were determined by three-primer PCR of tail DNA. The forward

primer, 5¢-CTATCCAAGCACTGCCTGCTC-3¢, and the two reverse primers, 5¢-GAGTTCTAGCTCCTCCAGGTGTG-3¢ and 5¢-GATGCCCTTCAGCTCGAT

GCG-3¢, yielded 275-bp wild-type and 356-bp mutant allele products, respec-

tively, in a 35-cycle reaction (94 1C for 20 s, 65 1C for 30 s, 72 1C for 30 s)18.

Validation of LINGO-1 gene deletion was accomplished by Southern blot, RT-

PCR and northern blot analyses (see Fig. 5b,c for Southern and northern

analyses). Prominent bands were detected in northern blot and RT-PCR

in wild-type mice, but a complete absence of bands was found in the knock-

out mice. Southern blots of the heterozygotes showed both the wild-type

and modified LINGO-1 allele. LINGO-1 knockout mice appeared normal,

with no obvious physical abnormalities or alterations in behavior, locomotion

or fecundity.

ACKNOWLEDGMENTSWe thank J. Mason and other LINGO-1 team members and the researchmanagement team from Biogenidec for discussions.




1. Edgar, J.M. & Garbern, J. The myelinated axon is dependent on the myelinating cellfor support and maintenance: molecules involved. J. Neurosci. Res. 76, 593–598(2004).

2. Garbay, B., Heape, A.M., Sargueil, F. & Cassagne, C. Myelin synthesis in the peripheralnervous system. Prog. Neurobiol. 61, 267–304 (2000).

3. Trapp, B.D., Ransohoff, R. & Rudick, R. Axonal pathology in multiple sclerosis: relation-ship to neurologic disability. Curr. Opin. Neurol. 12, 295–302 (1999).

4. Trapp, B.D., Bo, L., Mork, S. & Chang, A. Pathogenesis of tissue injury in MS lesions.J. Neuroimmunol. 98, 49–56 (1999).

5. Villagra, N.T. et al. PML bodies in reactive sensory ganglion neurons of the Guillain-Barresyndrome. Neurobiol. Dis. 16, 158–168 (2004).

6. Kiewe, P. et al. Progressive multifocal leukoencephalopathy with detection of JC virus ina patient with chronic lymphocytic leukemia parallel to onset of fludarabine therapy.Leuk. Lymphoma 44, 1815–1818 (2003).

7. Kolodny, E.H. Dysmyelinating and demyelinating conditions in infancy. Curr. Opin.Neurol. Neurosurg. 6, 379–386 (1993).

8. Salvati, S., Attorri, L., Avellino, C., Di Biase, A. & Sanchez, M. Diet, lipids and braindevelopment. Dev. Neurosci. 22, 481–487 (2000).

9. Stoffel, W. & Bosio, A. Myelin glycolipids and their functions. Curr. Opin. Neurobiol. 7,654–661 (1997).

10. Friede, R.L. Relation between myelin sheath thickness, internode geometry, and sheathresistance. Exp. Neurol. 92, 234–247 (1986).

11. Michailov, G.V. et al. Axonal neuregulin-1 regulates myelin sheath thickness. Science304, 700–703 (2004).

12. Chan, J.R. et al. NGF controls axonal receptivity to myelination by Schwann cells oroligodendrocytes. Neuron 43, 183–191 (2004).

13. Wang, S. et al. Notch receptor activation inhibits oligodendrocyte differentiation.Neuron 21, 63–75 (1998).

14. Vartanian, T., Goodearl, A., Viehover, A. & Fischbach, G. Axonal neuregulin signals cellsof the oligodendrocyte lineage through activation of HER4 and Schwann cells throughHER2 and HER3. J. Cell Biol. 137, 211–220 (1997).

15. Givogri, M.I. et al. Central nervous system myelination in mice with deficient expressionof Notch1 receptor. J. Neurosci. Res. 67, 309–320 (2002).

16. Garratt, A.N., Britsch, S. & Birchmeier, C. Neuregulin, a factor with many functions inthe life of a Schwann cell. Bioessays 22, 987–996 (2000).

17. Liang, X., Draghi, N.A. & Resh, M.D. Signaling from integrins to Fyn to Rho familyGTPases regulates morphologic differentiation of oligodendrocytes. J. Neurosci. 24,7140–7149 (2004).

18. Mi, S. et al. LINGO-1 is a component of the Nogo-66 receptor/p75 signaling complex.Nat. Neurosci. 7, 221–228 (2004).

19. Shao, Z. et al. Taj/Troy, an orphan TNF receptor family member, interacts with the Nogo-66 receptor and regulates axonal regeneration. Neuron 45, 353–359 (2005).

20. Park, J. et al. A TNF receptor family member TROY is a co-receptor with Nogo receptorin mediating the inhibitory activity of myelin inhibitors. Neuron 45, 345–351(2005).

21. Osterhout, D.J., Wolven, A., Wolf, R.M., Resh, M.D. & Chao, M.V. Morphologicaldifferentiation of oligodendrocytes requires activation of Fyn tyrosine kinase. J. CellBiol. 145, 1209–1218 (1999).

22. Svenningsen, A.F., Shan, W.S., Colman, D.R. & Pedraza, L. Rapid method for culturingembryonic neuron-glial cell cocultures. J. Neurosci. Res. 72, 565–573 (2003).

23. Sperber, B.R. & McMorris, F.A. Fyn tyrosine kinase regulates oligodendroglial celldevelopment but is not required for morphological differentiation of oligodendrocytes.J. Neurosci. Res. 63, 303–312 (2001).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

24. Wolf, R.M., Wilkes, J.J., Chao, M.V. & Resh, M.D. Tyrosine phosphorylation of p190RhoGAP by Fyn regulates oligodendrocyte differentiation. J. Neurobiol. 49, 62–78(2001).

25. Sperber, B.R. et al. A unique role for Fyn in CNS myelination. J. Neurosci. 21, 2039–2047 (2001).

26. Kramer, E.M., Klein, C., Koch, T., Boytinck, M. & Trotter, J. Compartmentation of Fynkinase with glycosylphosphatidylinositol-anchored molecules in oligodendrocytes facil-itates kinase activation during myelination. J. Biol. Chem. 274, 29042–29049 (1999).

27. Melendez-Vasquez, C.V., Einheber, S. & Salzer, J.L. Rho kinase regulates schwann cellmyelination and formation of associated axonal domains. J. Neurosci. 24, 3953–3963(2004).

28. Kaplan, M.R. et al. Induction of sodium channel clustering by oligodendrocytes. Nature386, 724–728 (1997).

29. Bhat, R.V. et al. Expression of the APC tumor suppressor protein in oligodendroglia. Glia17, 169–174 (1996).

30. Rubinson, D.A. et al. A lentivirus-based system to functionally silence genes in primarymammalian cells, stem cells and transgenic mice by RNA interference. Nat. Genet. 33,401–406 (2003).

31. Partadiredja, G., Miller, R. & Oorschot, D.E. The number, size, and type of axons in ratsubcortical white matter on left and right sides: a stereological, ultrastructural study.J. Neurocytol. 32, 1165–1179 (2003).

32. Li, C., Trapp, B., Ludwin, S., Peterson, A. & Roder, J. Myelin associatedglycoprotein modulates glia-axon contact in vivo. J. Neurosci. Res. 51, 210–217(1998).

33. Zhang, Y., Muyrers, J.P., Testa, G. & Stewart, A.F. DNA cloning by homologousrecombination in Escherichia coli. Nat. Biotechnol. 18, 1314–1317 (2000).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

ATP mediates rapid microglial response to local braininjury in vivo

Dimitrios Davalos1, Jaime Grutzendler1,3, Guang Yang1, Jiyun V Kim2, Yi Zuo1, Steffen Jung2, Dan R Littman2,Michael L Dustin2 & Wen-Biao Gan1

Parenchymal microglia are the principal immune cells of the brain. Time-lapse two-photon imaging of GFP-labeled microglia

demonstrates that the fine termini of microglial processes are highly dynamic in the intact mouse cortex. Upon traumatic brain

injury, microglial processes rapidly and autonomously converge on the site of injury without cell body movement, establishing a

potential barrier between the healthy and injured tissue. This rapid chemotactic response can be mimicked by local injection of

ATP and can be inhibited by the ATP-hydrolyzing enzyme apyrase or by blockers of G protein–coupled purinergic receptors and

connexin channels, which are highly expressed in astrocytes. The baseline motility of microglial processes is also reduced

significantly in the presence of apyrase and connexin channel inhibitors. Thus, extracellular ATP regulates microglial branch

dynamics in the intact brain, and its release from the damaged tissue and surrounding astrocytes mediates a rapid microglial

response towards injury.

Microglia in the cerebral cortex have a highly branched morphology,with each cell soma decorated by long processes with fine termini.They share many molecular structures with phagocytic macrophagesand are thus expected to perform tissue surveillance in the nervoussystem1,2. Activated microglia are found in the brain under almostall pathological conditions and are involved in tissue repair, amplifica-tion of inflammatory effects and neuronal degeneration3–5. Microgliahave been shown to phagocytose dead cells and clear the cellular debrisin brain slices6–9 and respond to various substances, including purineand pyrimidine analogs, complement factors, cytokines and chemo-kines in cell culture10–13. However, studies in brain slices are limitedbecause the slicing procedure can inherently activate microgliaand induce their transformation into a morphologically distinct andhighly reactive state, therefore obscuring potentially importantdynamic processes9,14,15. To investigate the dynamic properties ofmicroglia and the mechanisms underlying their activation uponinjury in the intact living brain, we took advantage of transgenicmice in which all microglia are fluorescently labeled after replacingthe Cx3cr1 gene with the gene encoding enhanced green fluore-scent protein (EGFP) by homologous recombination in embryonicstem cells16. Using transcranial two-photon microscopy17, wewere able to image the behavior of GFP-expressing parenchymalmicroglia through the thinned skull of anesthetized hetero-zygous Cx3cr1GFP/+ mice (see Methods). Our results indicate thatmicroglial processes are highly dynamic in the intact brain and respondrapidly towards the site of traumatic injury. Furthermore, extra-

cellular ATP released from astrocytes is essential in mediating boththe rapid baseline dynamics and the injury-induced response ofmicroglial processes.

RESULTS

Microglial processes are highly dynamic in the intact brain

We first examined the normal behavior of microglia in the living mousebrain through a thinned but intact area of the skull. Whereas microglialcell bodies and main branches were morphologically stable during theentire period of observation (hours), higher-order branches of theramified processes underwent rapid extension and retraction overintervals of seconds to minutes, reaching up to several micrometersin length or retracting until they completely disappeared (Fig. 1;Supplementary Video 1; 430 cells in ten animals). Over a period ofminutes to hours, many small processes appeared and disappearedwithout net change in the total number of branches (125 processesfrom five cells in three animals over 15 min; P 4 0.8). This baselinedynamism of microglial processes is in sharp contrast to the stability ofthe surrounding neuronal processes17 and is in line with their proposedfunction to sense and respond to tissue abnormalities.

Microglia respond rapidly to focal brain injury

To directly examine the response of microglia to brain injury, we tookadvantage of the focal properties of the two-photon laser and per-formed a small laser ablation, B15 mm in diameter, inside the cortexthrough the thinned skull (Figs. 2,3; see Methods). This type of injury


1Molecular Neurobiology Program, Department of Physiology and Neuroscience and 2Molecular Pathogenesis Program, Skirball Institute, New York University Schoolof Medicine, 540 First Avenue, New York, New York 10016, USA. 3Present address: Northwestern University Department of Neurology, 303 East Chicago Avenue,Chicago, Illinois 60611, USA. Correspondence should be addressed to W-B.G. ([email protected]).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

has been studied and involves non-linear processes that can be confinedto the laser focus, much like the two-photon fluorescence excitation18.Time-lapse imaging showed that all the cells near the site of injury (cellbodies within B50 mm from ablation) responded within the firstminute post-ablation: the tips of their processes close to the site ofinjury appeared bulbous and slightly enlarged. Within the next fewminutes, these cells extended their processes toward the damaged site atan average rate of 1.25 7 0.06 mm/min (Fig. 2b,c and SupplementaryVideo 2; n 4 20; five quantified in Fig. 3g–i). Approximately 30 min

after the laser-induced injury, the processes of the nearby cells reachedthe damaged site and appeared to fuse together, forming a sphericalcontainment around it (Fig. 2d). During this period, the same cells alsoretracted those processes that previously lay in directions opposite tothe site of injury (Fig. 2d,e). Most of the cellular content of each of theimmediate neighbors was directed towards the damaged site within thefirst 1–3 h (Fig. 2e,f) whereas the cell bodies remained at the samelocation for at least 10 h (Supplementary Video 2). Cells locatedfurther away (cell bodies between 75 and 125 mm from the ablation,82 cells in six animals) also responded in a directional way, sendingtheir processes towards the ablation without ever reaching the already-contained injury site (data not shown). In addition, when the laserablation was performed on two or more sites very close to each other,we observed that the same cells that initially committed a number oftheir processes to the first ablation could still detect and send theirremaining processes to the new ablation 10–20 min later (Supplemen-tary Video 3). These results indicate that the site of laser-induced injuryreleases highly localized signals able to rapidly attract individualmicroglial processes.

To determine whether microglia can also respond to mechanicalinjury in a similar fashion, we induced local injury to a small area of thecortex (B50 mm in diameter) with a glass electrode through a small

b

a

0

2Leng

th (

µm)

4

6

8

10

0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5Time (min)

Process 1Process 2Process 3Process 4

1

2

3

40 s 30 s 60 s 90 s

240 s210 s180 s 150 s

Figure 1 Transcranial two-photon imaging shows rapid dynamics of fine

microglial processes. (a) Time-lapse imaging of the same microglial branches

demonstrated rapid extension and retraction of fine microglial processes

over seconds. Circles and rounded box indicate four representative processes

that change in length and shape over time (Supplementary Video 1).

(b) Length changes of the four processes marked in a as a function of time.

Scale bar, 5 mm.

Before 15 min

XY

2 h 45 min32 min

2 min

1 h 35 min

6 min

30 min

c

d e f

g h i

a b

Figure 2 Microglial processes move rapidly

towards the site of injury induced either by the

two-photon laser, or mechanically with a glass

electrode. (a–f) After we created a localized

ablation inside the cortex (B15 mm in diameter)

with a two-photon laser (b; see Methods), nearby

microglial processes responded immediately

with bulbous termini (b) and extended toward

the ablation until they formed a spherical

containment around it (c–f). At the same time,

the same cells retracted those processes thatlay in directions opposite to the site of injury

(arrows in d and e; Supplementary Video 2).

(g,h) Mechanical injury, induced with a glass

electrode in a small area of the cortex marked

with a hexagon (B50 mm in diameter), led to a

similar response of microglial processes as with

the laser-induced injury (Supplementary Video 4).

(i) To quantify the microglial response toward the

laser-induced injury, we measured the number of

microglial processes entering from the outer

area Y (70 mm in radius) into the inner area X

(35 mm in radius) as a function of time. The

number of white pixels in area X or Y were

measured at each time point (Rx(t) or Ry(t)), and

the microglial response was defined as

R(t) ¼ (Rx(t) – Rx(0))/Ry(0) (see Methods). The

results of such quantification are shown in

Figure 3g. Scale bar, 10 mm.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

craniotomy (see Methods). We observed that microglial processesassumed the bulbous morphology at their termini immediately uponinjury and rapidly moved into the damaged tissue (Fig. 2g,h). Thelocalized nature and the kinetics of the microglial response to mechan-ical damage were very similar to those observed in the laser ablationinjury model (Supplementary Video 4). The rapid and highly direc-tional responses of microglia to the exact location of both laser- andmechanically induced injuries suggest that the same signaling cascade isinvolved in attracting the microglial processes in both models. Owingto the relative ease in controlling the extent of injury with a focusedlaser beam, we performed all the subsequent experiments using laser-induced ablation.

ATP mediates rapid microglial response towards injury

Previous studies of microglia in culture have shown that ATPsignaling through G protein–coupled P2Y receptors induces rapidmicroglial membrane ruffling and whole-cell migration11. To testwhether ATP released as a result of tissue injury is important inmediating microglial response, we applied 100 mM, 1 mM or 10 mMATP in artificial cerebrospinal fluid (ACSF) directly through asmall craniotomy (see Methods), followed by laser ablation. Wereasoned that if ATP released from the injured tissue were res-ponsible for initiating the observed microglial response, such aresponse should be prevented by eliminating the ATP concentrationgradient or by saturating ATP receptors after bathing the tissuewith a high concentration of ATP, comparable to the intracellularATP concentration19. Indeed, we found that applying 1–10 mM (butnot 100 mM) ATP significantly reduced both the extent and speedof the microglial response towards the injury site (Fig. 3b,g–i;the extension rate of processes was 0.66 7 0.03 mm/min and0.29 7 0.04 mm/min in the presence of 1 mM and 10 mM ATP,respectively (n ¼ 3–5 for each concentration), and the extensionrate of the control was 1.25 7 0.06 mm/min; P o 0.001). Similarto ATP application, 1 mM ADP or UTP also significantly reducedthe microglial response, but 1 mM CTP had no effect (Fig. 3i, n ¼ 3–5

for each case). These results suggest that factors such as ATP, ADP andUTP released from the injured tissue are involved in attracting micro-glial processes towards the injury site.

To further test the role of ATP and ADP in the microglial response,we performed the laser ablation in the presence of apyrase, an ATPasethat hydrolyzes both extracellular ATP and ADP19,20. Similarto application of ATP and ADP, application of 300 U/ml apyrasealso substantially reduced the microglial response to laser ablation(Fig. 3c,i; extension rate of processes was 0.5 7 0.03 mm/min, n ¼ 5;P o 0.001), suggesting that the presence of ATP and/or ADP in theextracellular space is necessary for the rapid microglial responsetowards the site of injury. In addition, the rapid baseline motility ofmicroglial processes slowed in the presence of 300 U/ml apyrase: over a15-min interval, the average absolute length change of the fine dynamicprocesses was 1.91 7 0.25 mm after applying apyrase (36 processes,four cells from four animals), significantly smaller than the control withACSF solution (2.73 7 0.29 mm, 35 processes, five cells from threeanimals; P o 0.003), suggesting that extracellular ATP and/or ADP isalso involved in regulating the baseline motility of microglial processes.

P2Y receptor activation is necessary for microglial response

ATP, ADP and UTP are potent agonists for P2X ligand-gated ionchannels and P2Y G protein–coupled receptors21. To investigate theinvolvement of activation of these receptors in mediating the microglialresponse, we applied various P2 receptor inhibitors directly to thecortex through a small craniotomy and recorded their effect onmicroglial dynamics after laser ablation. Two P2Y inhibitors, reactiveblue 2 (100 mM) and PPADS (10 mM), markedly reduced the numberand motility of microglial processes toward the ablations, while up to1 mM suramin (mainly a P2X inhibitor) was not effective in preventingthe laser-induced response (Fig. 3d–f, i; n ¼ 3–4 for each inhibitor).Together, these results suggest that activation of P2Y G protein–coupledreceptors by ATP, ADP or UTP, either on microglia or cells in thesurrounding tissue (see discussion below), is necessary for the rapidmicroglial response to injury.

g h

i

Control ATP 1 mM Apyrase

RB2 PPADS Suramin

–0.050

0.050.100.150.200.250.300.35

0 10 20 30 40 50 60

Time (min)

Res

pons

e

ControlsATP 1 mM

ATP 10 mM

00.20.40.60.81.01.21.4

Control ATP1 mM

ATP10 mM

Ext

ensi

on R

ate

(µm

/min

)

0

20

40

60

80

100

120

Controls ATP1 mM

ATP10 mM

ADP1 mM

UTP1 mM

CTP1 mM

Apyrase300 u/ml

RB2100 µM

PPADS10 mM

Suramin1 mM

Res

pons

e 30

min

afte

r ab

latio

n re

lativ

e to

con

trol

a

d

b

e

c

f

Figure 3 Extracellular ATP and activation of purinergic G protein–coupled receptors are necessary for rapid chemotactic microglial response. (a–f) Microglial

response (30 min after laser ablation) in the presence of various compounds applied through a small craniotomy (see Methods): ACSF (a), 1 mM ATP (b), 300

U/ml apyrase (c), 100 mM reactive blue 2 (d), 10 mM PPADS (e) and 1 mM suramin (f). While many microglial processes reached the site of injury in a and f,

few or none moved towards the ablation in b–e. (g–i) Quantification of microglial response to laser ablation (described in Fig. 2i and Methods) after a 45-min

bath application of various compounds. (g) Detailed kinetics of microglial response over 1 h in the control (ACSF), 1 mM and 10 mM ATP solutions. The

response in control reaches a plateau within B35 min (when the converging processes have reached and contained the damage), whereas in the presence

of 1–10 mM ATP, the response is smaller and slower. (h) Extension rate of microglial processes toward the laser ablation in control (ACSF), 1 mM ATP and

10 mM ATP solutions. (i) Microglial response measured at the 30-min time point after laser ablation in the presence of various compounds (n ¼ 3–5 for

each case). Scale bar, 10 mm.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Local release of ATP induces rapid microglial response

To test whether ATP is sufficient to induce a microglial responsesimilar to the one we observed after laser ablation, we inserteda sharp microelectrode containing ATP (tip diameter o1 mm) intothe cortex and monitored the surrounding microglial behavior(Fig. 4). We found that simply inserting an ACSF-containing electrodeinto the cortex (without moving it laterally to cause tissue damage)induced no significant response from the surrounding microglia(Fig. 4a,e,f; n ¼ 5). However, a similar electrode containing 10 mMATP in an ACSF solution was able to mimic in time, range and kineticsthe rapid response of microglial processes observed following laserablation (Fig. 4b; 15 injections in ten mice, three quantified inFig. 4e,f). Within the first minute after the electrode was inserted,many nearby microglial processes showed enlarged termini and beganextending toward the tip of the electrode (extension rate of processestoward the tip was on average 1.10 7 0.05 mm/min; 30 processes inthree quantified experiments). Within B45 min, a large number ofprocesses converged and tightly surrounded the tip of the electrode(Supplementary Video 5). If the electrode was withdrawn from thecortex 10–15 min after insertion while the microglial responsewas underway, the convergence of the processes towards theprevious location of the tip stopped immediately (data not shown),suggesting that the constant release of ATP from the tip of theelectrode was necessary for the chemotactic response of micro-glial processes. Notably, when electrodes containing a high con-centration of ATP (100 mM) were used, the surroundingmicroglial processes responded rapidly at the beginning but slowedsignificantly and formed a sphere midway around the tip, possiblybecause ATP receptors within the sphere were saturated by the highconcentration of ATP released from the tip of the electrode (n ¼ 4,data not shown).

Similar to the response to ATP, release of 10 mM ADP throughan electrode also induced a rapid response from microglialprocesses (Fig. 4f; n ¼ 4); 10 mM UTP induced a smaller response(Fig. 4f; n ¼ 5), and we observed little or no response with 10 mMCTP (Fig. 4f; n ¼ 4). Since ATP in the extracellular space ishydrolyzed very rapidly by ectonucleases19,20, we further testedwhether ATP alone could attract microglia by injecting non-hydrolyzable ATP analogues (AMPPNP or ATPgS) into the cortex.We found that local application of 10 mM AMPPNP caused achemotactic response similar in range and size to that with ATP

application (Fig. 4c,f; average extension rate towards the tip was1.07 7 0.05 mm/min; 30 processes, n ¼ 4), and 10 mM ATPgS induceda smaller response (n ¼ 3, data not shown).

ATP-induced ATP release is essential for microglial response

The above experiments indicate that ATP released from either thedamaged tissue or an electrode can trigger the rapid response ofmicroglial processes towards the source of ATP. Many lines of evidencehave shown that extracellular ATP released from astrocytes, andsubsequent activation of purinergic receptors, mediate intercellularcommunication among astrocytes as well as communication betweenastrocytes and microglia, as indicated by calcium wave propagation22–27.In addition, ATP can induce ATP release from astrocytes, and such aregenerative ATP release is important for the spreading of Ca2+

waves23,25,28. To test whether ATP-induced ATP release is necessaryfor the rapid microglial response, we applied 300 U/ml apyrase on thecortex through a small craniotomy for 1 h and then tested the effect ofnon-hydrolyzable ATP (either AMPPNP or ATPgS at 10 mM) releasedfrom a microelectrode. Although a non-hydrolyzable ATP-containingelectrode alone can rapidly attract microglial processes (Fig. 4c,f), noresponse was observed in the presence of apyrase (Fig. 4d,f; n ¼ 5).Furthermore, neither inserting a microelectrode containing high con-centration of AMPPNP (up to 100 mM) nor continuously injecting10 mM AMPPNP was able to reverse the inhibitory effect of apyrase onthe microglial response towards the electrode tip (n ¼ 4 for each, datanot shown). As only the endogenous ATP, not the non-hydrolyzableATP, was rapidly degraded by the applied apyrase in the extracellularspace, this suggests that non-hydrolyzable ATP–induced release ofendogenous ATP is required for attracting microglial processes. In addi-tion, ATP released from the electrode appears to function as a trigger forfurther release of ATP and possibly other factors from the surroundingtissue (astrocytes, neurons, etc.) that are important for microglialresponse, because constant release of ATP from the electrode is notable to attract microglial processes directly in the presence of apyrase.

Connexin hemichannels are important for microglial dynamics

Because ATP induces ATP release from astrocytes, and such aninduction serves as an important mechanism for inter-astrocyte com-munication23,25,28, the above experiments suggest that astrocytes andtheir release of ATP are important for the microglial response uponinjury. It has been shown that ATP release from cultured astrocytes and

Figure 4 ATP-induced ATP release is essential

for rapid microglial response. (a–d) Microglial

response to local injection of ATP or AMPPNP

(non-hydrolyzable ATP analogue) 30 min after

insertion of the glass electrode through a

craniotomy. Whereas a control microelectrode

containing ACSF and rhodamine-dextran (a)

caused little or no microglial response, similarelectrodes containing 10 mM ATP (b) or 10 mM

AMPPNP (c) induced rapid extension of microglial

processes towards the tip of the electrode

(Supplementary Video 5). However, a

microelectrode containing 10 mM AMPPNP in the

presence of 300 U/ml apyrase (d; applied 1 h

before the electrode was inserted) did not elicit

a response towards the tip. (e) Detailed kinetics of

microglial response towards control electrode (a)

and towards electrodes containing 10 mM ATP

(b). (f) Microglial response measured as described in Figure 2i, 30 min after the insertion of electrodes containing only ACSF (control) or 10 mM of various

compounds in ACSF (n ¼ 3–5 for each case). For the last bar graph, electrodes containing 10 mM AMPPNP were inserted in the brain after a solution of

300 U/ml apyrase was applied for 1 h. Scale bar, 10 mm.

e

0

0.05

0.10

0.15

0.20

0.25

0.30

0

0.05

0.10

0.15

0.20

0.25

0 10 20 30 40 50 60

Time (min)

Res

pons

e

ATP 10 mMControl

f

Control ATP ADP UTP CTP AMPPNPAMPPNPin apyrase

Res

pons

e 30

min

afte

r lo

cal i

njec

tion

Control ATP

AMPPNP AMPPNP in apyrase

a b

c d


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

glioma cell lines involves connexin hemichannels24,25,29–32. Adultastrocytes express very high levels of connexin proteins33, and theconnexin channel blocker flufenamic acid (FFA)34 inhibits ATP releasefrom astrocytes25,30. To further explore the involvement of ATP andastrocytes in microglial response, we applied connexin channel inhibi-tors such as FFA or carbenoxolone (CBX) before laser ablation. We firstfound that the baseline motility of microglial processes was abolishedwithin 30 min after applying 250 mM CBX or within 10 min afterapplying 1 mM FFA. In the presence of FFA, the average absolute lengthchange of the small processes is 0.19 7 0.04 mm within 15–30 min (33processes from ten cells, three animals), significantly smaller than thatof the control (2.73 7 0.29 mm; P o 0.0001), suggesting that connexinchannels are important for the dynamics of microglial processes intheir native state, possibly by releasing ATP and/or other factors. Inaddition, 30 min after laser ablation, we did not observe any microglialprocesses moving towards the injury site in the presence of either250 mM CBX (n ¼ 3, not shown) or 1 mM FFA (Fig. 5; n ¼ 7). Thedirectional responses of microglia towards an electrode containing10 mM ATP were also completely abolished in the presence of CBX orFFA (n ¼ 4 for each inhibitor; data not shown). Notably, after washingout FFA with ACSF, the microglial processes extended towards the siteof ablation that was performed in the presence of 1 mM FFA B30 minearlier (Fig. 5; n ¼ 7). This result indicates that the inhibitory effect ofFFA is reversible and that the signals attracting microglial processespersist for at least 30 min after laser ablation is performed. Together,these experiments suggest that astrocytes are involved in releasing ATPthrough connexin channels and are important for mediating the rapidmicroglial response to local brain trauma.

DISCUSSION

By taking advantage of transcranial two-photon imaging and GFPknock-in mice, we have been able to observe, for the first time,rapid dynamics of microglial processes in the intact living brain. Inaddition, within the first minutes of laser-induced or mechanicalinjury, microglial processes respond and move directly towards the

site of injury. The directional convergence of microglial processestowards the trauma implies the presence of a gradient of one ormore highly diffusible and abundant molecules that mediate thisphenomenon. For at least 30 min after the laser ablation, the signalsattracting the microglial processes still remain (Fig. 5a,b), suggesting arole of the surrounding tissue in preserving and regenerating the signalsthat fuel the observed response.

ATP is an important signaling molecule mediating interactionsamong various cell types in the brain10,35,36. Massive release of purinesoccurs after metabolic stress and trauma36,37, and high levels of ATPpersist in the peritraumatic zone for many hours after the insult38. ATP,acting through purinergic receptors, can induce membrane rufflingand ramification of cultured microglia11 and can stimulate them torelease various biologically active substances10. Our results demonstratethat extracellular ATP and activation of P2Y receptors are necessary forthe rapid microglial response towards the injury site, suggesting thatextracellular ATP could itself be the chemoattractant responsible for thedirectional extension of the processes by activating P2Y receptors onmicroglia. Alternatively, ATP may activate microglia and/or act as atrigger (through purinergic mechanisms) to induce the release ofunidentified chemoattractant(s) from the surrounding tissue. Ourresults with non-hydrolyzable ATP injection and apyrase applica-tion indicate that (i) release of endogenous ATP triggered by non-hydrolyzable ATP (or ATP) is necessary for the microglial response, and(ii) ATP may not be the endogenous chemoattractant, because non-hydrolyzable ATP from a point source (an electrode) in the presence ofapyrase is not sufficient to attract the microglial processes.

Although neurons, oligodendrocytes and endothelial cells are likelyto release large amounts of ATP upon injury and thereby contribute tomicroglial reaction, two lines of evidence suggest that astrocytes areimportant in mediating the rapid and widespread microglial response.First, our results indicate that this response involves ATP-triggeredATP release from the surrounding tissue; such a regenerative releaseof ATP has been shown to be an important mechanism for long-rangesignaling among astrocytes23,25,28. Second, we have found thatconnexin channel inhibitors block microglial response reversibly.Connexin channels are highly expressed in astrocytes but notmicroglia in the resting state33 and are involved in ATP releasefrom astrocytes25,30. Together, our results suggest that the initialATP release from the damaged tissue triggers further release ofATP and other factors from the surrounding astrocytes, a processrequiring the opening of connexin channels and the activation ofpurinergic receptors. Although factors other than ATP may beimportant for directional chemotaxis of microglial processes,ATP-triggered ATP release from the surrounding astrocytes is keyin mediating such a response.

Furthermore, the baseline motility of microglial processes in theintact brain is also modulated by the same ATP signaling mechanismsmediating injury-induced microglial responses, because baseline dyna-mism slows significantly in the presence of apyrase and connexinchannel inhibitors. The high baseline motility of microglial processesmay reflect the fluctuation of the ATP concentration in the surround-ing tissue. It is worth noting that owing to technical limitations, weimaged microglial behavior only in anesthetized mice. As anestheticscan induce microglial activation in the brain39, the degree to which thebaseline microglial dynamics is altered by anesthetics and/or the state ofanesthesia remains to be determined. Upon injury and release ofintracellular ATP, the immediate chemotactic response of microglialprocesses seems to be specific for microglia, as we did not observeany obvious motile responses of nearby neuronal processeswithin hours of a laser ablation in the cortex of mice with YFP-labeled

FFA 1 mM 30 min After washout 1 h

–40–20

020406080

100120

1 mMFFA

30 min

30 minafter

washout

Control30 min

Res

pons

e af

ter

abla

tion

re

lativ

e to

con

trol

ca b

Figure 5 The microglial response towards the laser ablation is inhibitedafter blocking connexin hemichannels, the conduits of ATP. (a) Very few or

no microglial processes extended towards the site of injury within the first

30 min after laser ablation in the presence of 1 mM flufenamic acid (FFA)

applied through a small craniotomy. (b) Microglial processes extended

toward the same site of injury 30–60 min after FFA was washed out.

(c) Quantification of microglial response in the presence of 1 mM FFA

and 30 min after washout. Scale bar, 10 mm.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

neurons (D.D., J.V.K., M.L.D., W.-B.G., unpublished observations). Itis also important to point out that not all brain pathologies wouldtrigger a rapid microglial response as in the case of traumatic injury. Forexample, the ATP concentration near sites of focal ischemia and othertypes of vascular injury are likely to be low owing to the lack of oxygenin the surrounding cells, and the amounts of ATP released from theinjured tissue does not appear to be sufficient to generate a rapidmicroglial response (J.G., D.D., W.-B.G., unpublished observations).The functional significance of the immediate microglial responsetowards sites of traumatic injury remains to be elucidated. Such aresponse may allow for rapid isolation of damaged areas and could beparticularly important when trauma-associated local bleeding occursinside the brain and immediate containment of the damage is a priorityfor restoring a stable environment for the nearby neurons.

METHODSTransgenic mice. Heterozygous Cx3cr1GFP/+ mice were used for all experi-

ments16. Cx3cr1 (encoding fractalkine receptor) is expressed in microglia, some

monocytes and dendritic cells. No obvious defects of parenchymal microglia

were found in homozygous mutant Cx3cr1GFP/GFP mice16, and the rapid

microglial response to laser-induced injury was also observed in these mice.

These mice were housed and bred in Skirball Institute’s animal facilities, and all

experiments were done in accordance with protocols approved by the institu-

tional animal committee.

In vivo imaging of microglia with two-photon microscopy. GFP-labeled

microglia were imaged by two-photon microscopy, either through a thinned

intact skull as described previously17 or through a small craniotomy. Briefly,

adult transgenic mice were anesthetized intraperitoneally with ketamine

(200 mg/kg body weight) and xylazine (30 mg/kg body weight) in 0.9% NaCl

solution. For transcranial imaging, a region (B1 mm in diameter) over barrel

or motor cortex was first thinned with a high-speed drill under a dissecting

microscope and then scraped with a microsurgical blade to a final thickness of

o20 mm. For imaging through a window in the skull, a thinned region (B1–2

mm in diameter) was opened either with a needle or forceps. A drop (B200 ml)

of artificial mouse cerebrospinal fluid (ACSF) was applied on the exposed

region for the duration of the experiment. The skull surrounding either the

thinned region or the open skull window was attached to a custom-made steel

plate to reduce respiratory-induced movement. The animal was placed under

either a Bio-Rad multi-photon microscope (Radiance 2001) or a custom-made

two-photon microscope described previously17.

The Ti-sapphire laser was tuned to the excitation wavelength for GFP

(890 nm). A stack of image planes with a step size of 0.75–2 mm was acquired

using a water-immersion objective (Olympus 60�, 0.9 N.A.; Nikon 60� 1.0

N.A. or 40�, 0.8 N.A.) at zoom of 1.0–3.0. The maximum imaging depth was

B200 mm from the pial surface. Images were acquired using low laser power

(o30 mW at the sample) and a low-pass emission filter (o700 nm).

Two-photon laser ablation and mechanical injury inside the cortex. A highly

localized injury was achieved by focusing a two-photon laser beam (B1 mm in

size) in the superficial layer of the cortex either through a thinned, intact skull

or with a small piece of the skull removed. The wavelength of the two-photon

laser was set at 780 nm and the laser power was B60–80 mW at the sample.

The beam was parked at the desired position for approximately 1–3 s to create a

small injury site as indicated by a bright autofluorescent sphere (B15 mm in

diameter) around the focal point of the beam (Figs. 2–4). The injury was

confined to the area B15–20 mm in diameter around the laser focal point,

because microglia within this area lost their GFP immediately after laser

ablation, whereas those B30 mm from the injury site still responded to the

ablation. The laser-induced focal ablation is a useful injury model, as the site

and degree of injury are easy to control, and the response of microglia toward

the injury is highly reproducible.

Mechanical injury inside the cortex was generated with a glass microelec-

trode (B1 mm tip diameter). The electrode was loaded with an ACSF solution

containing 3% rhodamine-dextran (10 kDa, Molecular Probes) to make it

visible under a fluorescent microscope. The electrode was inserted into the

exposed cortex through an open skull window using a micromanipulator. The

local mechanical injury was generated by moving the fine control of the

micromanipulator laterally within a region B50 mm in diameter (Fig. 2g,h).

Drug application through an open window in the skull. Various reagents

were dissolved in ACSF, and a small drop of the solution (B200 ml) containing

the compound of interest was applied directly onto the cortex through the open

window. Imaging started 10–45 min after drug application. Because of rapid

diffusion of small molecules within the cortex40, we estimated that the effective

drug concentration in the cortex was B10 times lower than that in the solution

directly applied to the cortex. For the local drug application, the reagent was

added to the ACSF-rhodamine solution and loaded in a sharp glass electrode

(1.0 mm outer diameter; 10–12 MO resistance as measured in ACSF). A

picospritzer (General Valve Incorporation) was used to release a small volume

of the solution at 20–40 psi for 20–50 ms.

Image processing and quantification. All image processing was done using US

National Institutes of Health Image J software. All z-stacks of images were

projected along the z-axis to recreate a two-dimensional (2D) representation of

3D structures. Time-lapse movies were generated by z-projections of stacks of

images taken sequentially over time. In making both the movies and figures, we

made sure that the 3D image stacks at all time points pass above and below the

processes studied.

The laser ablation appeared as a small autofluorescent sphere B15 mm in

diameter (Fig. 2b,i). Following the ablation, the processes of neighboring

microglia invaded the area around the site of injury and eventually reached the

sphere. To quantify the extent and speed of microglial responses to laser-

induced injury, we measured the number of microglial processes entering from

the outer area Y (70 mm in radius) into the inner area X (35 mm in radius)

surrounding the ablation site as a function of time (Fig. 2i). To account for

signal intensity differences among different experiments, we thresholded every

image so that all processes had the maximum value (255), and all background

was set to 0 (Fig. 2i shows the thresholded version of Fig. 2b). We then counted

the number of white pixels in area X over time (Rx(t)) and compared it with the

first picture taken immediately after the ablation (Rx(0)). The number of white

pixels corresponds to the region covered by processes within the area X, and its

increase over time provides a measure of the microglial response. To account

for the variability in the number of microglia located in the outer area Y in

different experiments, we calculated the microglial response relative to the

number of processes in the outer area Y immediately after the ablation (Ry(0)).

The microglial response at any time point (R(t)) is therefore given by R(t) ¼(Rx(t) – Rx(0))/Ry(0).


ACKNOWLEDGMENTSWe thank H. Suk-Woo, G. Shakhar, Y-C. Chen and R. Uglesich for offering usefulcomments and help with experiments. This work is supported by grants from theNational Institute of Health and the Dana Foundation to M. L. D and W-B. G.




1. Thomas, W.E. Brain macrophages: evaluation of microglia and their functions. BrainRes. Brain Res. Rev. 17, 61–74 (1992).

2. Hickey, W.F. Basic principles of immunological surveillance of the normal centralnervous system. Glia 36, 118–124 (2001).

3. Gonzalez-Scarano, F. & Baltuch, G. Microglia as mediators of inflammatory anddegenerative diseases. Annu. Rev. Neurosci. 22, 219–240 (1999).

4. Aschner, M., Allen, J.W., Kimelberg, H.K., LoPachin, R.M. & Streit, W.J. Glial cells inneurotoxicity development. Annu. Rev. Pharmacol. Toxicol. 39, 151–173 (1999).

5. Kreutzberg, G.W. Microglia: a sensor for pathological events in the CNS. TrendsNeurosci. 19, 312–318 (1996).

6. Smith, S.J., Cooper, M.W. & Waxman, A. Laser microscopy of subcellular structure inliving neocortex: can one see dendritic spines twitch? in The Biology of Memory, XXIIISymposium Medicum Hoechst (eds. Squire, L.R., Lindenlaub, E.) 49–71 (SchattauerVerlag, Stuttgart, 1990).

7. Brockhaus, J., Moller, T. & Kettenmann, H. Phagocytozing ameboid microglial cellsstudied in a mouse corpus callosum slice preparation. Glia 16, 81–90 (1996).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

8. Czapiga, M. & Colton, C.A. Function of microglia in organotypic slice cultures.J. Neurosci. Res. 56, 644–651 (1999).

9. Petersen, M.A. & Dailey, M.E. Diverse microglial motility behaviors during clearance ofdead cells in hippocampal slices. Glia 46, 195–206 (2004).

10. Inoue, K. Microglial activation by purines and pyrimidines. Glia 40, 156–163 (2002).11. Honda, S. et al. Extracellular ATP or ADP induce chemotaxis of cultured microglia

through Gi/o-coupled P2Y receptors. J. Neurosci. 21, 1975–1982 (2001).12. Nolte, C., Moller, T., Walter, T. & Kettenmann, H. Complement 5a controls motility of

murine microglial cells in vitro via activation of an inhibitory G protein and therearrangement of the actin cytoskeleton. Neuroscience 73, 1091–1107 (1996).

13. von Zahn, J., Moller, T., Kettenmann, H. & Nolte, C. Microglial phagocytosis ismodulated by pro- and anti-inflammatory cytokines. Neuroreport 8, 3851–3856(1997).

14. Stence, N., Waite, M. & Dailey, M.E. Dynamics of microglial activation: a confocal time-lapse analysis in hippocampal slices. Glia 33, 256–266 (2001).

15. Koshinaga, M. et al. Rapid and widespread microglial activation induced by traumaticbrain injury in rat brain slices. J. Neurotrauma 17, 185–192 (2000).

16. Jung, S. et al. Analysis of fractalkine receptor CX(3)CR1 function by targeted deletionand green fluorescent protein reporter gene insertion. Mol. Cell. Biol. 20, 4106–4114(2000).

17. Grutzendler, J., Kasthuri, N. & Gan, W.B. Long-term dendritic spine stability in the adultcortex. Nature 420, 812–816 (2002).

18. Galbraith, J.A. & Terasaki, M. Controlled damage in thick specimens by multiphotonexcitation. Mol. Biol. Cell 14, 1808–1817 (2003).

19. Dubyak, G.R. & el-Moatassim, C. Signal transduction via P2-purinergic receptorsfor extracellular ATP and other nucleotides. Am. J. Physiol. 265, C577–C606(1993).

20. Zimmermann, H. & Braun, N. Extracellular metabolism of nucleotides in the nervoussystem. J. Auton. Pharmacol. 16, 397–400 (1996).

21. Illes, P. & Alexandre Ribeiro, J. Molecular physiology of P2 receptors in the centralnervous system. Eur. J. Pharmacol. 483, 5–17 (2004).

22. Cotrina, M.L. et al. Connexins regulate calcium signaling by controlling ATP release.Proc. Natl. Acad. Sci. USA 95, 15735–15740 (1998).

23. Guthrie, P.B. et al. ATP released from astrocytes mediates glial calcium waves.J. Neurosci. 19, 520–528 (1999).

24. Cotrina, M.L., Lin, J.H., Lopez-Garcia, J.C., Naus, C.C. & Nedergaard, M. ATP-mediatedglia signaling. J. Neurosci. 20, 2835–2844 (2000).

25. Anderson, C.M., Bergher, J.P. & Swanson, R.A. ATP-induced ATP release from astro-cytes. J. Neurochem. 88, 246–256 (2004).

26. Schipke, C.G., Boucsein, C., Ohlemeyer, C., Kirchhoff, F. & Kettenmann, H. AstrocyteCa2+ waves trigger responses in microglial cells in brain slices. FASEB J. 16, 255–257(2002).

27. Verderio, C. & Matteoli, M. ATP mediates calcium signaling between astrocytes andmicroglial cells: modulation by IFN-gamma. J. Immunol. 166, 6383–6391 (2001).

28. Wang, Z., Haydon, P.G. & Yeung, E.S. Direct observation of calcium-independentintercellular ATP signaling in astrocytes. Anal. Chem. 72, 2001–2007 (2000).

29. Beyer, E.C. & Steinberg, T.H. Evidence that the gap junction protein connexin-43 isthe ATP-induced pore of mouse macrophages. J. Biol. Chem. 266, 7971–7974 (1991).

30. Stout, C.E., Costantin, J.L., Naus, C.C. & Charles, A.C. Intercellular calcium signalingin astrocytes via ATP release through connexin hemichannels. J. Biol. Chem. 277,10482–10488 (2002).

31. Braet, K., Vandamme, W., Martin, P.E., Evans, W.H. & Leybaert, L. Photoliberatinginositol-1,4,5-trisphosphate triggers ATP release that is blocked by the connexinmimetic peptide gap 26. Cell Calcium 33, 37–48 (2003).

32. Coco, S. et al. Storage and release of ATP from astrocytes in culture. J. Biol. Chem. 278,1354–1362 (2003).

33. Rouach, N. et al. Gap junctions and connexin expression in the normal and pathologicalcentral nervous system. Biol. Cell. 94, 457–475 (2002).

34. Srinivas, M. & Spray, D.C. Closure of gap junction channels by arylaminobenzoates.Mol. Pharmacol. 63, 1389–1397 (2003).

35. Hansson, E. & Ronnback, L. Glial neuronal signaling in the central nervous system.FASEB J. 17, 341–348 (2003).

36. Fields, R.D. & Stevens-Graham, B. New insights into neuron-glia communication.Science 298, 556–562 (2002).

37. Ciccarelli, R. et al. Involvement of astrocytes in purine-mediated reparative processes inthe brain. Int. J. Dev. Neurosci. 19, 395–414 (2001).

38. Wang, X. et al. P2X7 receptor inhibition improves recovery after spinal cord injury.Nat. Med. 10, 821–827 (2004).

39. Nakki, R., Nickolenko, J., Chang, J., Sagar, S.M. & Sharp, F.R. Haloperidol preventsketamine- and phencyclidine-induced HSP70 protein expression but not microglialactivation. Exp. Neurol. 137, 234–241 (1996).

40. Binder, D.K., Papadopoulos, M.C., Haggie, P.M. & Verkman, A.S. In vivo measurementof brain extracellular space diffusion by cortical surface photobleaching. J. Neurosci.24, 8049–8056 (2004).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

SRF mediates activity-induced gene expression andsynaptic plasticity but not neuronal viability

Narendrakumar Ramanan1,2, Ying Shen1, Sarah Sarsfield1,2, Thomas Lemberger3, Gunther Schutz3,David J Linden1 & David D Ginty1,2

Synaptic activity-dependent gene expression is critical for certain forms of neuronal plasticity and survival in the mammalian

nervous system, yet the mechanisms by which coordinated regulation of activity-induced genes supports neuronal function is

unclear. Here, we show that deletion of serum response factor (SRF) in specific neuronal populations in adult mice results in

profound deficits in activity-dependent immediate early gene expression, but components of upstream signaling pathways and

cyclic AMP–response element binding protein (CREB)-dependent transactivation remain intact. Moreover, SRF-deficient CA1

pyramidal neurons show attenuation of long-term synaptic potentiation, a model for neuronal information storage. Furthermore, in

contrast to the massive neurodegeneration seen in adult mice lacking CREB family members, SRF-deficient adult neurons show

normal morphologies and basal excitatory synaptic transmission. These findings indicate that the transcriptional events underlying

neuronal survival and plasticity are dissociable and that SRF plays a prominent role in use-dependent modification of synaptic

strength in the adult brain.

Activity-dependent gene expression is critical for neuronal develop-ment, plasticity and survival1–3, yet the identities of synapse-to-nucleussignaling pathways, as well as the nuclear targets contributing toactivity-dependent gene expression underlying synaptic plasticity,remain poorly understood. Early studies on transcriptional regulationof the archetypical immediate early gene (IEG) c-Fos (also known asFos) identified SRF and CREB as candidate regulators of activity-dependent IEG expression4,5. Much subsequent work has focused onthe neuronal functions of CREB. Indeed, CREB activity is regulated bybursts of synaptic activity, and CREB binding sites are present ingenomic regulatory regions of many IEGs6,7. However, recent find-ings have raised doubt about the requirement of CREB in activity-dependent synaptic plasticity8–11. Furthermore, CREB and its familymembers appear to be critically involved in the maintenance ofneuronal viability6. Given the essential requirement of CREB familymembers in neuronal survival and maintenance, interpretation of theroles of these transcription factors in synaptic plasticity is complicated.Thus, the distinct nuclear events mediating activity-dependent survivaland plasticity of neurons are not known; nor are the transcrip-tional mechanisms that support use-dependent modifications ofsynapse strength in the adult hippocampus and elsewhere in thenervous system.

In contrast to the large body of work addressing CREB function inthe nervous system, very little is known about the role of SRF and itstarget genes in regulating neuronal survival and plasticity. SRF is atranscription factor containing the highly conserved MADS domain

that binds to the serum response element (SRE) found in the promo-ters of several cardiac-specific and stimulus-dependent genes12,13. Thein vivo functions of SRF-dependent gene expression in the brain andelsewhere are poorly understood, owing in part to early embryoniclethality of Srf-null mice due to severe defects in mesoderm forma-tion14. In this study we have generated conditional Srf mutant mice tostudy SRF function in the nervous system of adult mice. Using twodifferent Cre recombinase transgenic lines, we show that hippocampaland cortical neurons lacking SRF show major deficits in activity-induced gene expression. Notably, in contrast to the massive neuronaldegeneration observed in brain of mice lacking CREB-family transcrip-tion factors, mice lacking SRF in the brain show normal gross structuralmorphology and neuronal cytoarchitecture, suggesting that SRF-dependent gene expression is dispensable for neuronal viability. Wealso found that CA1 pyramidal neurons lacking SRF show an attenua-tion of both the early and late phases of long-term synaptic potentia-tion. Thus, SRF is required for activity-induced gene expression andsynaptic plasticity but not for neuronal survival in the adult brain.

RESULTS

Neuron-specific deletion of Srf in adult brain

To ascertain the in vivo functions of SRF and SRF-dependent geneexpression and to ask whether distinct transcriptional events mediateactivity-dependent survival and plasticity of adult neurons, we gener-ated conditional Srf knockout mice, as Srf null mice show earlyembryonic lethality owing to severe defects in mesoderm formation.


1Department of Neuroscience, 2Howard Hughes Medical Institute, 725 North Wolfe Street, Preclinical Teaching Building Room 1015, The Johns Hopkins University Schoolof Medicine, Baltimore, Maryland 21205, USA. 3Molecular Biology of the Cell I, Deutsches Krebsforschungszentrum, Im Neuenheimer Feld 280, 69120 Heidelberg,Germany. Correspondence should be addressed to D.D.G. ([email protected]).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

We targeted 1.4 kb of the promoter along with exon I of the Srf geneusing the Cre-loxP strategy15 (Fig. 1). As expected, crosses between Srf f/f

mice and mice expressing Cre recombinase in the germline did notyield any viable Srf-null mutants (data not shown), consistent with theearlier observation of embryonic lethality resulting from Srf deletion14.In contrast, crosses with either a synapsin1-Cre line16 (Srf f/f;SCre) or anaCaMKII-Cre17 line (Srf f/f;CKCre) to ablate Srf exclusively in neuronsyielded viable conditional mutants in the expected mendelian ratios.We found no observable differences between Srf f/f;SCre and Srf f/f;CKCre

mutant mice and their respective littermates in their general health orgrowth and breeding behavior, except that Srf f/f;SCre females failed tobreed beyond 3 months of age.

Loss of SRF in distinct populations of hippocampal neurons

Brains of Srf f/f;SCre and Srf f/f;CKCre mutant mice were analyzed byimmunohistochemistry using an SRF-specific antibody18 to establishthe extent of Srf deletion (Fig. 2). In Srf f/f;SCre mice, loss of SRF protein

was confined to dentate granule cells and CA3 pyramidal neurons ofthe hippocampus beginning 10 d after birth (Supplementary Fig. 1).This is in contrast to the reported embryonic expression of Crerecombinase and more widespread target allele recombination inmice harboring this synapsin1-Cre allele16. Differences in recombina-tion efficiency may be attributable to positional influences of the loxPsites at the targeted locus19. The observed hippocampus-specificdeletion of SRF in Srf f/f;SCre mice is essentially complete in youngadults (6–8 weeks old; Fig. 2b,f,j), which were used for subsequentexperiments. Furthermore, we assessed Srf mRNA expression by RT-PCR using total RNA isolated from microdissected hippocampal CA3and dentate gyrus of Srf f/f;SCre mutants and control littermates (6–8weeks old, n ¼ 3). Consistent with the immunohistochemical analysis,we did not detect Srf mRNA in the Srf f/f;SCre mutant mice (Fig. 1c).

In contrast to Srf f/f;SCre mice, Srf f/f;CKCre mice show SRF deletion inhippocampal CA1 pyramidal neurons and, to a lesser extent, in CA3pyramidal and neocortical neurons, but not in the dentate gyrus at

Partial Srf genomic structure

Targeting vector

Following FIpE/Cre recombinationPstl

(793)F1

Pstl

Pstl

(793)

Short arm

Long arm (6,590 bp)(955 bp)

(1,359) (2,314) (3,375)

Pro region

Pro regionpGKneo

E1

E1

E2

E2

E2 E3

E3

E3

3,708 4,542

ES cells (Southern) Mouse tail PCR

Srf

Rps29

Per3 pro

Tf pro

+/+ +/+f/+

f/+ f/f

(4,839) (11,429)

BsrGl

BsrGl

BsrGl

Spel/blunt

BamHl

BamHl/blunt

Ssp I

Ssp I

Ssp I

loxPloxP FRT FRT

Spel Pstl

F2R1

a b

cSrf

f/f;S

Cre

Srff/f

Figure 1 Generation of mice lacking SRF in postnatal forebrain neurons. (a) Partial structure of the Srf locus and the region targeted for deletion. loxP sites

flanked the 1.4 kb upstream of the transcriptional start site, which included the regulatory elements critical for SRF transcription (‘Pro region’) and exon I

(840 bp), which encodes the SRF dimerization and DNA binding regions. Schematics of the genomic locus, targeting vector and final targeted alleles

after FlpE and Cre-mediated recombination are depicted. (b) Southern blot of embryonic stem (ES) cells and PCR screening of mutant mouse tail DNA

samples. Mice were genotyped using two sets of PCR primers (shown in a, lower panel): F1+R1 and F2+R1 (F1, 5¢-TGCTTACTGGAAAGCTCATGG-3¢; F2,

5¢-GGCACTGTCTCAGGGTGTCT-3¢; R1, 5¢-TGCTGGTTTGGCATCAACT-3¢). f, floxed allele. +, wild-type allele. (c) RT-PCR of total RNA isolated from

microdissected CA3 and dentate gyrus regions of Srf f/f and Srf f/f;Scre mice (6- to 8-week-old mice, n ¼ 3). The Srf primers amplify a 328-bp fragment

corresponding to exons 3–5. Lack of signal in the lanes corresponding to mutant samples indicates absence of any truncated Srf transcript. The

positive control for amplification was riboprotein S29 (Rps29), and PCR for promoters of Per3 and transferrin (Tf) served as the control for genomicDNA contamination.

DG

Srf f/f Srf f/f;SCre

6 weeks 6 weeks 3 months

Srf f/f;CKCre Srf f/f;CKCre

CA1

CA3

a b c d

e f g h

i j k l

Figure 2 Cre-mediated recombination of Srf

results in loss of SRF in the brains of Srf f/f;SCre

and Srf f/f;CKCre mice as shown by SRF

immunohistochemistry. (a,e,i) Srf f/f control mice

show robust expression of SRF at all ages.

(b,f,j) At 6 weeks of age, Srf f/f;SCre mice show a

loss of SRF in dentate granule cells (b) and CA3

pyramidal neurons (j), but no SRF loss is observed

in CA1 neurons (f). A few isolated cells in dentate

gyrus (DG) showed Srf expression (arrows in b).

(c,g,k) In Srf f/f;CKCre mutants, loss of SRF begins

around 6 weeks in all structures of the

hippocampus. (d,h,l) At 3 months of age,

depletion of SRF was near complete in CA1 and

CA3 neurons (h,l) and in greater than 70% of

dentate granule neurons (d). Scale bar, 100 mm.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

6 weeks of age (n ¼ 6; Fig. 2c,g,k; Supplementary Fig. 1). We did notobserve loss of SRF expression in Srf f/f;CKCre mice within the first twopostnatal weeks (Supplementary Fig. 1). At 3–4 months of age,Srf f/f;CKCre mutants showed near-complete loss of SRF expression inCA1 and CA3 pyramidal neurons, greater than 70% loss in dentategyrus and loss to varying degrees in neurons of the neocortex(Fig. 2d,h,l; Supplementary Fig. 1).

IEG expression and MAPK signaling in SRF mutants

To study the consequence of SRF ablation on neuronal activity-inducedgene expression, Srf f/f;SCre mutants and control littermates were sub-jected to electroconvulsive shock (ECS), a procedure that inducesrobust activation of IEGs in the dentate gyrus20. Expression of IEGs,including c-Fos, FosB, Egr1, Egr2, cJun (also known as Jun), JunB, Arc,Actb and Actg1, was greatly enhanced in the dentate gyrus of controlanimals 30 min after ECS. In contrast, the Srf f/f;SCre mutants showedlittle to no induction of any of these IEGs in dentate gyrus neurons,which lack SRF, although these mice exhibited comparable basalexpression of Actb, Actg1 and Egr1 in SRF-positive hippocampalCA1 pyramidal neurons (n ¼ 6; Fig. 3; Supplementary Fig. 2; datanot shown).

To determine whether dentate granule cells in Srf f/f;SCre mutantsrespond normally to ECS, we analyzed the activation of a signalingpathway that is rapidly induced by ECS and is critical for both SRE- andCRE-mediated transcription. The mitogen-activated protein kinases(MAPKs), including extracellular signal–regulated kinases-1 and -2(Erk1/2), convey synaptic signals to the nucleus21,22 and in particularcontrol the activities of the ternary complex transcription factors thatassociate with SRF23 and CREB6. Phospho-Erk1/2 (pErk1/2) immuno-histochemistry showed that ECS leads torobust activation and nuclear translocationof Erk1/2 in both Srf f/f;SCre mutants andcontrol littermates (Fig. 4a–d). Thus, whereasSrf f/f;SCre dentate granule cells are responsive,and MAPKs become active and translocate toneuronal nuclei following ECS, we did notobserve transcription of many IEGs in SRF-deficient dentate granule cells.

Normal CRE-dependent gene expression in

Srf f/f;SCre mice

We then sought to determine whether SRF-independent transcriptional events are alsoaffected in the Srf f/f;SCre mutants. Activation

of CREB (as indicated by phosphorylation of Ser133) and CRE-dependent gene expression are well-characterized readouts of activ-ity-dependent gene expression in response to many extracellularstimuli, including ECS in neurons6,7. We did not observe differencesin the levels of pCREB immunoreactivity in Srf f/f;SCre mutants and incontrol littermates 5 min after ECS (Fig. 4e–h). We also analyzed theexpression of three activity-regulated CREB target genes, TrkB,Homer1a and Cpg15 (also known as Nrn1 or neuritin). TrkB (thereceptor for brain-derived neurotrophic factor (BDNF)), neurotrophin4/5 (NT-4/5) and Cpg15, an activity-regulated membrane proteininvolved in dendritic growth, are critically dependent on CREB fortranscriptional activation24,25 but do not contain obvious SREs. Simi-larly, Homer1a, the activity-induced splice variant of the Homer1 gene,has three CREs but no obvious SRE in its promoter, and its activation ismediated by MAPK signaling26,27. We observed robust activation ofeach of these genes in both groups of mice (Fig. 4i–p; data not shown).Taken together, these results indicate that the lack of activity-inducedIEG responses observed in the Srf f/f;SCre mutants is due to the absenceof SRF, not to pleiotropic effects on neuronal signaling or activation ofthe CREB-dependent transcriptional machinery.

Novelty-induced IEG expression in Srf f/f;CKCre mutant mice

We next examined the importance of SRF for IEG transcriptionalresponses during natural modes of synaptic stimulation. We found thatablation of Srf in Srf f/f;CKCre mutants results in decreased basal expres-sion of Actb and Egr1 mRNA at 3–4 months of age in hippocampal CA1pyramidal neurons (Supplementary Fig. 3; data not shown). We thenused exploration of a new environment to induce gene expression inSrf f/f;CKCre mice (Fig. 5). Exposure to novelty strongly activates IEG

0 min 30 min


0 min 30 min

c-F

osF

osB

Egr

1E

gr2

Arc

Act

b

a b c d

e f g h

i j k l

m n o p

q r s t

u v w x

Figure 3 SRF is required for expression of many

IEGs after electroconvulsive shock (ECS).

(a–x) Induction of c-Fos (a–d), FosB (e–h),

Egr1 (i–l), Egr2 (m–p), Arc (q–t) and Actb (u–x)

mRNA after ECS as visualized by in situ

hybridization (6-week-old age- and sex-matched

control and Srf f/f;SCre mice; n ¼ 6 for stimulated

and unstimulated). Strong IEG induction was seen

at 30 min in control Srf f/f mice (b,f,j,n,r,v), but

IEG induction was observed in only a few,

scattered neurons in Srf f/f;SCre mutants

(d,h,l,p,t,x). The few isolated neurons showingexpression of c-Fos and Arc mRNA in brain

sections from Srf f/f;SCre mutants (arrowheads in

d,t) are likely explained by the finding that few

neurons remain immunopositive for SRF in these

mice (Fig. 2d). Scale bar, 100 mm.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

responses in place cells of the hippocampus and in neurons in theneocortex, including the somatosensory cortex28,29. Upon explorationof a previously unknown space for 30 min, control littermates showedrobust induction of c-Fos, Egr1 and Arc mRNA in the somatosensoryneocortex (Fig. 5f,j; data not shown) as well as in other neocorticalregions and in the hippocampus (data not shown). In marked contrast,Srf f/f;CKCre mutants showed little to no induction of these IEGs in CA1pyramidal cells, and very few neurons in the dentate gyrus andsomatosensory cortex showed novelty-induced gene expression (Fig.5h,l; data not shown). The few cells that do show induction coincidewell with the extent of SRF deletion in these brain regions (Fig. 5c; datanot shown). Nissl staining revealed normal cell density and layerformation in the neocortex (Fig. 5d), suggesting that the observedloss of IEG expression in cortical neurons is not due to gross structural

alterations or neuronal loss. Taken together, these findings demonstratethat SRF is critical for synaptic activity-induced IEG expression inseveral populations of postmitotic CNS neurons.

Hippocampal neuronal survival in Srf and Creb/Crem mutants

Recent findings have implicated synaptic activity-dependent geneexpression in neuronal survival3,6. Moreover, many IEGs, includingthose shown here to require SRF for their expression, also containbinding sites for the CREB family of transcription factors, which arecritical for cell survival6. Indeed, deletion of CREB30, compounddeletion of CREB with its paralog ATF-1 in the germline31 or com-pound deletion of CREB with its paralog CREM in the forebrain32 leadsto progressive and massive apoptosis. As SRF is essential for mostactivity-dependent IEG expression, and as SRF and CREB family

0 min

sosr

5 min 0 min 5 min

0 min 30 min 0 min 30 min


pErk

1/2

pCR

EB

Trk

BH

omer

1a

a b c d

e f g h

i j k l

m n o p

Figure 4 ECS-induced activation of MAPK,

CREB and transcription of TrkA and Homer1a

is normal in Srf f/f;SCre mice. (a–h) Immuno-

histochemistry for activated pErk1/2 (a–d) and

pCREB (e–h) using Srf f/f and Srf f/f;SCre mice after

ECS (5 min). pErk1/2 immunoreactivity was

largely restricted to mossy fibers (arrows in a,c)

in unstimulated mice, whereas pErk1/2 immuno-reactivity was robust in the dendritic field of

dentate gyrus, stratum oriens (so) and stratum

radiatum (sr) of CA1 and CA3 in both Srf f/f and

Srf f/f;SCre mice (b,d). Strong nuclear staining was

seen only in dentate granule cells from stimulated

mice (compare insets in b,d; insets show high

magnification of dentate gyrus). The intensity

of pErk1/2 immunoreactivity was comparable

in Srf f/f;SynCre mutants and control littermates

(n ¼ 3, induced and uninduced). The robust

nuclear pCREB immunoreactivity observed at

5 min was comparable in both controls and

mutants. (i–p) Induction of CREB-dependent

IEGs TrkB (i–l) and Homer1a (m–p) after ECS

(30 min) as observed by in situ hybridization

(6-week-old age- and sex-matched control and

Srf f/f;SCre mice; n ¼ 6 for stimulated and

unstimulated). Very strong induction of these

CREB-regulated genes was seen in both controland Srf f/f;SCre mice. Scale bar, 100 mm.

I

II/III

IV

V

VI

Srf f/f

α–SRF Nissl α–SRF Nissl

Srf f/f;CKCre

Caged control Novelty exploration (30 min) Caged control Novelty exploration (30 min)

Egr

1A

rc

a b c d

e f g h

i j k l

Figure 5 SRF is required for activity-induced IEGexpression in the somatosensory cortex after

exploration of an enriched environment.

(a–d) Extent of Srf deletion in the somatosensory

cortex in the Srf f/f;CKCre and control littermates

at 3 months (n ¼ 6, age- and sex-matched) as

seen by SRF immunoreactivity (a-SRF; a,c).

Corresponding cresyl violet (Nissl) staining

showed normal cortical architecture in both

mutant and control animals (b,d). (e–l) In situ

hybridization for Egr1 (e–h) and Arc (i–l) mRNA in

home-caged animals and in animals that explored

a novel enriched environment (30 min after

exploration). Strong induction of Egr1 and Arc

mRNA was seen in control littermates in cortical

layers IV and VI after exploration, but IEG

induction was mostly absent in the Srf f/f;CKCre

mutants. Scale bar, 100 mm.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

members may control expression of a common or overlapping set oftarget genes, we sought to compare the consequences of SRF deletionon neuronal survival and maintenance to the consequences of CREBand CREM deletion. Toward this end, we analyzed cresyl violet–,Hoescht- and Golgi-stained coronal brain sections obtained (i) fromadult Srf f/f;CKCre and Srf f/f;SCre mutants, (ii) from mice lacking CREB(encoded by Creb, also known as Creb1) and CREM in forebrain(CrebCamkCre4 Crem�/� mice) and (iii) from control animals (Fig. 6).We found neither changes in cell number nor defects in the structuralintegrity in any of the forebrain regions analyzed in adult mice lackingSRF, including the dentate gyrus and hippocampal CA1 fields (Fig. 6;data not shown). This is in stark contrast to adult mice lacking CREBand CREM in the forebrain which show massive apoptotic cell deathand extensive neuronal degeneration of the hippocampus (Fig. 6).These findings clearly distinguish the in vivo functions of SRF and theCREB family of transcription factors and indicate that SRF andactivity-dependent expression of SRF target genes, including theIEGs described above, are not necessary for maintenance and survivalof postnatal forebrain neurons.

Basal synaptic transmission in Srf f/f;CKCre mutant mice

Although it is well established that IEG expression is critical for long-lasting forms of use-dependent changes in synaptic strength, recent

findings have cast some doubt on the necessity of CREB familymembers for activity-dependent synaptic plasticity in the hippocam-pus8–11. Given that SRF is critical for activity-dependent IEGexpression, and as SRF and CREB functions are dissociable, wewondered whether SRF-mediated IEG expression is critical for synaptictransmission and plasticity. To measure basal excitatory synaptictransmission, we made whole-cell voltage clamp recordings fromCA1 pyramidal cells in hippocampal slices obtained from adultSrf f/f;CKCre mice at a time when SRF deletion is complete (3–4 months;Fig. 2h). We recorded miniature excitatory postsynaptic currents(mEPSCs) by holding cells at a command potential of –90 mVin external saline supplemented with tetrodotoxin (0.5 mM) toblock action potentials and GABAzine (10 mM) to block GABAA

receptors. Following a stabilization period (410 min), we recordedmembrane current for 3–10 min, after which we detected mEPSCsoffline using a sliding template algorithm. Comparison of 17 Srf f/f

control cells with 27 Srf f/f;CKCre cells demonstrated no significantdifference in either mEPSC amplitude (21.4 7 1.2 pA and24.6 7 1.6 pA, respectively; Fig. 7a) or frequency (1.91 7 0.43 Hzand 1.91 7 0.28 Hz, respectively).

To assess the role of glutamate receptor subtypes in excitatorysynaptic transmission, we removed tetrodotoxin, and we activatedSchaffer collateral/commissural fibers with a stimulating electrode

Srf f/f Srf f/f;SCre Srf f/f Srf f/f;CKCre

Srf

f/f;S

Cre

Srf

f/f

Srf

f/f

Srf

f/f;C

KC

re

Creb1Camkcre4 Crem+/– Creb1Camkcre4 Crem–/–

Cre

b1C

amkc

re4

Cre

m+

/–

Cre

b1C

amkc

re4

Cre

m–/

–

Cre

b1C

amkc

re4

Cre

m–/

–

Hip

poca

mpu

sD

GC

A1

DG CA1

3,000

2,500

2,000

1,500

1,000

500

0CA1 DG CA1 DG

Cel

l num

ber/

unit

area

Cre

b1C

amkc

re4

Cre

m+

/–

a b c d e f

g h i j k l

m n o p q

s t u v w

r

Figure 6 SRF, unlike CREB family members, is not necessary for neuronal survival or for structural integrity in the adult hippocampus. (a–r) Structural and

cellular integrity is comparable in Srf f/f control mice, Srf f/f;Scre, Srf f/f;CKCre mutant mice and CrebCamkCre4 Crem+/� control mice but is severely disrupted in

CrebCamkCre4 Crem�/� mutants. Nissl staining done on coronal brain sections from 1-year-old Srf f/f;SCre mutants and controls (a,b,g,h,m,n). High magnification

of the dentate gyrus shows normal cytoarchitecture in both controls and mutants (g,h). Brain sections from 6-month-old Srf f/f;CKCre mice, 6-month-old Srf f/f

control littermates (n ¼ 4; c,d,i,j,o,p) and CrebCamkCre4Crem+/� control mice (n ¼ 2; 7 months old) also showed normal cytoarchitecture (e,k,q). CA1 neurons

appear normal in Srf f/f controls and mutants (m–o), but Nissl staining of CrebCamkCre4Crem�/� mutant mice (n ¼ 2; 7 months old) shows massive

neurodegeneration of the hippocampus (f,l,r). High magnification shows near-complete loss of CA1 neurons and severe reduction of dentate gyrus (l,r).

(s–v) Golgi-stained coronal brain sections showed similar architecture of dentate gyrus cells in Srf f/f (s) and Srf f/f;SCre (t) animals and of CA1 neurons in Srf f/f

(u) and Srf f/f;CKCre mice (v). Scale bar, 100 mm. (w) Comparable CA1 and dentate gyrus cell numbers in Srf f/f, Srf f/f;CKCre and CrebCamkCre4Crem+/�, but

near-complete cell loss was observed in CA1 and dentate gyrus of CrebCamkCre4 Crem�/� mutants. Boxes in e and f: regions in which cell counts were

obtained (described in Methods). Bars: mean 7 s.d.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

placed in stratum radiatum. When we delivered test pulses at lowfrequency (0.05 Hz) and the command potential was set to �80 mV, alarge, rapidly inactivating inward current was evoked, which is typicallymediated by AMPA/kainate receptors (Fig. 7b). Changing the com-mand potential to +40 mV yielded an evoked outward current, whichinactivated more slowly. This current has a rapid component mediatedby AMPA/kainate receptors and a slow tail mediated by NMDAreceptors. Thus, taking the ratio of the peak inward current at�80 mV and the outward tail current at +40 mV provides a goodestimate of the ratio of neurotransmission mediated by AMPA/kainatereceptors versus NMDA receptors33. This ratio was similar when15 Srf f/f control cells were compared with 15 Srf f/f;CKCre cells (8.9 70.8 and 7.7 7 0.6, respectively). Thus, low-frequency basal excitatorysynaptic transmission seems to be normal in CA1 pyramidal neuronslacking SRF.

LTP in Schaffer collateral synapses in Srf f/f;CKCre mice

To measure long-term synaptic potentiation (LTP), we used fieldpotential recordings, as these allow for resolution of a late phase ofLTP, which has been previously reported to require transcription34 butwhich appears to be independent of CREB8–11. We recorded fieldexcitatory postsynaptic potentials (fEPSPs) in the stratum radiatum inresponse to test pulses delivered at 0.05 Hz. After recording stablebaseline responses for 30 min, we delivered theta-burst stimulation,and then test pulses were resumed for an additional 150 min (Fig. 7c).Recordings from Srf f/f control synapses revealed robust LTP witha strong initial phase (n ¼ 6; 266 7 16% of baseline fEPSP slope att ¼ 20 min) and a sustained late phase (178 7 19% at t ¼ 146 min).In contrast, Srf f/f;CKCre synapses showed a significant reduction of boththe early and late phases of LTP (n ¼ 9; 170 7 18% and 117 7 7% fort ¼ 20 and 146 min, respectively). Field potential recordings made

Srf f/f

Srf f/f

Srf f/f

Srf f/f;CKCre

Srf f/f;CKCre

Srf f/f;CKCre

Srf f/f;CKCre

Srf f/f

10 pA

100 pA

200 ms

100 msn = 17 n = 27

Srf f/f Srf f/f;CKCre

n = 17 n = 27

Srf f/f

Srf f/f

Srf f/f;CKCre

Srf f/f;CKCre

n = 15 n = 15

2.5

2.0

1.5

1.0

0.5

0mE

PS

C fr

eque

ncy

(Hz)

mE

PS

C a

mpl

itude

(pA

)

302520151050

+40 mV

+40 mV

–80 mV

–80 mV

10

8

6

4

2

0

AM

PA/N

MD

A r

atio

1

1

1

1

1

3

3

3

3

2

2

3

22

2

10 ms

TBS Tet

Time (min) Time (min)

1 mV

10 ms

1 mV

300

250

200

150

100

50

–20 0 20 40 60 80 100 120 140

Per

cent

age

of b

asel

ine

fEP

SP

slo

pe

Per

cent

age

of b

asel

ine

fEP

SP

slo

pe

–20 –10 0 10 20 30 40 50 60 70 80 90

200

180

160

140

120

100

80

1

2 3

a b

c d

Figure 7 Schaffer collateral/commissural–CA1 pyramidal cell synapses from Srf f/f;CKCre mice show normal basal synaptic transmission but attenuated long-

term synaptic plasticity. (a) mEPSCs were recorded from CA1 pyramidal cells in slices derived from Srf f/f and Srf f/f;CKCre mice (3–4 months old). Sample traces

are consecutive recordings from a single cell. Bar graphs represent mean 7 s.e.m. for frequency and amplitude measures for mEPSCs detected and measuredoffline with a sliding template algorithm implemented in pCLAMP 9.0. (b) EPSCs recorded in CA1 pyramidal cells evoked by stimulation in the stratum

radiatum. Recordings were made at a holding potential of �80 mV and then again at +40 mV. Sample traces are the average of five consecutive responses. The

ratio of the peak at –80 mV to the tail amplitude at +40 mV (measured at the dotted line) was taken as an index of AMPA/NMDA conductance. Bars represent

mean 7 s.e.m. (c) Field potential recordings were made in stratum radiatum of hippocampal area CA1 to measure long-term synaptic potentiation induced by

theta-burst stimulation (TBS) at t ¼ 0 min. Sample traces from representative cells are from the numbered points indicated. Each trace is the average of six

consecutive responses. (d) An LTP experiment similar to that shown in c, except that the conditioning stimulation delivered at t ¼ 0 min consisted of a weak

tetanus (50 Hz, 1 s) rather than theta bursts.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

during the conditioning theta-burst stimulus showed no significantdifference between Srf f/f and Srf f/f;CKCre synapses (data not shown).

Theta-burst stimulation as used here is a strong form of synapticactivation which gives rise to a large and sustained form of LTP. Todetermine if Srf f/f;CKCre synapses would also show impaired LTP inresponse to less robust stimulation, we repeated LTP experiments usinga weak tetanic stimulation (50 Hz, 1 s) designed to elicit LTP of smalleramplitude (Fig. 7d). In this situation, LTP evoked in Srf f/f synapses wasstill significantly larger than that evoked in Srf f/f;CKCre synapses. Thisdifference was apparent soon after tetanic stimulation (Srf f/f: n ¼ 8, 1457 8% of baseline; Srf f/f;CKCre: n ¼ 10, 129 7 3% of baseline; t ¼ 6 min)and continued throughout the recording (Srf f/f: 135 7 8% of baseline;Srf f/f;CKCre: 112 7 3% of baseline; t ¼ 90 min). These experimentsindicate that although SRF is dispensable for basal excitatory synaptictransmission in CA1 neurons, SRF-dependent gene expression iscritical for LTP. This is in contrast to CREB, which seems to bedispensable for hippocampal LTP induced by synaptic activationusing either theta bursts or sustained tetani9.

DISCUSSION

Here, we have characterized the function of SRF in adult brain and havedemonstrated that SRF is critical for activity-induced gene expressionand synaptic plasticity. However, activation of MAPK and CREB, andat least some CREB-dependent transcriptional events, are normal in theabsence of SRF. We further show that neither SRF nor SRF-dependentgene expression is required for survival of adult hippocampal neurons,which is in marked contrast to the massive neurodegeneration observedin brains lacking CREB and CREM. Our electrophysiological datasuggests that SRF is not required for basal synaptic transmission butthat it is required for both early and late phases of LTP. Thus, SRF andthe CREB family of transcription factors control distinct geneticprograms that mediate plasticity and survival.

Early studies addressing the mechanisms of stimulus-dependentgene expression showed that, in addition to the CRE, the SRE in thepromoter region of c-Fos is critical for its activation5,35–38. However,until now, very little has been known about the requirement of SRF foractivity-dependent gene expression in neurons, synaptic plasticity andneuronal survival. Our results show that hippocampal dentate gyrusand cortical neurons that lack SRF do not exhibit activity-dependentinduction of many IEGs, including c-Fos, FosB, cJun, JunB, Egr1, Egr2,Actb and Arc. This impaired IEG response reflects a specific require-ment for SRF for the transcriptional control of IEG expression but notfor signaling from the synapse to the nucleus, as immunostaining afterECS showed comparable activation of nuclear pErk1/2 and CREB inhippocampus of Srf f/f;SCre mutants and control littermates. Further-more, ECS-induced activation of at least some CREB target genes,including TrkB, Homer-1a and Cpg15, is normal in neurons lackingSRF. Thus, SRF ablation does not have pleiotropic effects on the generaltranscriptional machinery, and neurons lacking SRF are capable ofresponding to synaptic activity and demonstrate normal activation ofSRF-independent genes.

Are SRF and activity-dependent gene expression necessary forneuronal survival? Our analyses of cell number and cytoarchitecturein brains of SRF mutants strongly suggest that SRF and its target genesare dispensable for postnatal survival and maintenance of at least someneuronal subtypes. This is in contrast to recent findings that haveshown a requirement for SRF for cell survival during embryonic stemcell differentiation in vitro and during early embryogenesis in vivo39 andfor BDNF-dependent survival of cortical neurons following trophicdeprivation and DNA damage in vitro40. Although SRF may regulateexpression of the anti-apoptotic gene Bcl-2 in embryonic stem cells

during differentiation in vitro, its precise role during survival of corticalneurons is not clear. Furthermore, other recent in vivo findings haveconcluded that SRF per se might not be required for cell survival duringdevelopment in mice and that the embryonic lethality observed inSRF-null mice is likely due to cardiac defects and poor circulation41,42.In the present study, although we did not observe neuronal loss in SRFmutant mice, we cannot rule out the possibility that SRF might beimportant for cell survival during early embryonic neuronal develop-ment. Nevertheless, the lack of neuronal loss and structural abnorm-ality in brains of SRF mutant mice as old as 1 year is in striking contrastto the massive neurodegeneration observed in 6-month-old micelacking CREB and CREM. These findings indicate that differenttranscriptional programs exist to regulate neuronal survival andmaintenance in the adult nervous system.

Previous work has suggested that there is a transcription-dependentlate phase of hippocampal LTP, as drugs that inhibit transcription leavethe early phase of LTP intact but reduce a later phase starting 1–2 h afterinduction34. Moreover, earlier studies that used strategies to interferewith CREB function showed similar deficits in the late phaseof hippocampal LTP and concomitant deficits in hippocampus-dependent memory tasks34. Also, it is very difficult to assess therequirement of CREB-dependent gene expression for hippocampalLTP in the Creb/Crem mutant mice due to extensive cell loss in thehippocampus. Nevertheless, recent findings have shown that CREB-dependent gene expression might be dispensable for the late phase oftetanus-evoked LTP8–11. Here, we find that in Srf f/f;CKCre mice,although basal excitatory synaptic transmission is normal in hippo-campal area CA1, LTP evoked by either weak or strong stimulation issubstantially attenuated. However, this attenuation is not similar tothat previously reported with acute application of transcription inhi-bitor drugs, as it manifests in both the early and the late phases of LTP.At present, it is unclear if the LTP deficits in CA1 pyramidal neurons inwhich SRF is deleted result from an acute lack of SRF during LTPinduction, from a chronic effect or from both. The defects in the earlyphase of LTP in Srf f/f;CKCre mice are consistent with a chronic require-ment for SRF-regulated genes in both the induction and the main-tenance of LTP. One possibility is that SRF affects the expression of oneor more components of a signaling cascade engaged by theta burst orby weak tetanic stimulation to support the early phase of LTP.

The differences in transcription factors required for LTP induced bysynaptic activation, neuronal maintenance and viability indicate thatSRF and CREB regulate distinct genetic programs, which in turnregulate different functions in adult forebrain neurons. Our resultssupport a model in which SRF regulates a broader range of genesrequired for neuronal plasticity than does CREB, whereas CREB-dependent genes, but not SRF-dependent genes, support neuronalmaintenance and viability. Taken together, these findings indicate thattranscriptional events involved in neuronal survival and maintenanceare dissociable from those involved in use-dependent changes insynaptic strength. Future insights into mechanisms underlying use-dependent changes in synapse strength should come from studies thatprobe the control of SRF activity and the identity of key SRF targetgenes in adult neurons as well as from behavioral analysis in Srfconditional mutant mice.

METHODSGeneration of Srf and Creb mutant mice. To generate Srf f/f mice, an 11.5-kb

fragment of the Srf gene (3.5 kb upstream and 8 kb downstream from the

transcription start site) was cloned from a BAC clone into pBluescript

(Stratagene). This subcloned fragment along with a neoFRT/loxP cassette

(modified from a construct provided by K. Takamiya and R. Huganir, Johns


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Hopkins University School of Medicine) was used to generate the targeting

vector using standard DNA cloning techniques. The loxP sites are located

approximately 1,400 bp upstream and 1,100 bp downstream of the transcrip-

tion start site, and Cre recombination results in the deletion of both the

promoter and exon 1. The targeted promoter region includes all the regulatory

elements required for SRF expression43, and exon 1 encompasses the

N terminus and the MADS box domain critical for SRF dimerization and

DNA binding13. After electrotransformation, 129/SV embryonic stem cells

harboring the targeted allele were selected for G418 resistance. Positive clones

identified by southern blot analysis were then injected into C57BL/6 blastocysts

to yield chimeric mice, which were crossed with C57BL/6 mice to generate

heterozygous mice carrying the conditional allele (Srf +/f). Srf +/f mice were

crossed with mice expressing FlpE recombinase or Cre recombinase in the germ

cell lineage44,45 to excise the neo cassette or generate a null allele, respectively.

Mice were genotyped using PCR primers for the presence of the targeted allele

and the Cre recombinase gene. The primers indicated in Figure 1 are F1 (5¢-TGCTTACTGGAAAGCTCATGG-3¢), F2 (5¢-GGCACTGTCTCAGGGTGTCT-3¢)and R1 (5¢-TGCTGGTTTGGCATCAACT-3¢). Crosses with germline Cre

transgenic mice yielded no viable Srf null mutant mice at P0 (eight litters,

n ¼ 64 pups), indicating embryonic lethality of the mutation14.

For neuron-specific deletion of SRF, the Srf +/f mice were crossed with mice

expressing Cre recombinase under the control of either the synapsin-1

promoter16 or the CaMKIIa promoter17. For all experiments, age- and sex-

matched homozygous conditional mice (Srf f/f) and mutants carrying a single

copy of Cre recombinase (Srf f/f;SCre and Srf f/f;CKCre) were used (after three

backcrosses into C57BL/6 background). RT-PCR analysis failed to detect Srf

mRNA, indicating both the complete deletion of SRF and the absence of any

truncated transcript following Cre recombination. Furthermore, the anti-SRF

antibody used in this study recognizes the C terminus of SRF, and immuno-

histochemical analysis showed the complete absence of full-length or truncated

SRF in all SRF-ablated regions.

The Creb/Crem double mutant mice were generated as described32. These

mice have a null mutation in Crem and a conditional deletion of Creb in

forebrain neurons. The animals used in the present study were 7 months old.

All animal procedures used in this study were approved by the Animal Care and

Use Committee of the Johns Hopkins University School of Medicine.

RNA isolation and RT-PCR. Total RNA was isolated from microdissected CA3

and dentate gyrus regions of 6-week-old Srf f/f;SCre mutants and control

littermates using TRIzol reagent (Invitrogen) according to the manufacturer’s

protocol; 1 mg of total RNA was used in first-strand cDNA synthesis using the

first strand synthesis kit (Invitrogen). The following primers were used: Srf,

5¢-ACCAGTGTCTGCTAGTGTCAGC-3¢ and 5¢-CATGGGGACTAGGGTACAT

CAT-3¢; Rps29, 5¢-CCAGCAGCTCTACTGGAGTCA-3¢ and 5¢-AGACTAGCAT

GATCGGTTCCA-3¢; Per3 promoter, 5¢-AGGCACTTTGTGAGACACTGATAG

-3¢ and 5¢-CAGGTAGAATCCAACCGTATTTTC-3¢; Transferrin promoter,

5¢-CAGGATTGAAATGCACAACTTTAG-3¢ and 5¢-CTTCCATGTTGTACTCTT

TGGTTG-3¢.

Electroconvulsive shock (ECS). Electroconvulsive shock (ECS) was adminis-

tered as previously described17, except that electroshock consisted of 0.5-s,

100-Hz, 10-mA stimulus of 0.5-ms square wave pulses delivered using the Ugo

Basile ECT unit, Model 7801. At these parameters, both Srf f/f;SCre mutants and

control littermates showed similar minor convulsions lasting 1–2 s; 30 min after

ECS, animals were decapitated and their brains processed for in situ hybridiza-

tion. For immunohistochemistry analysis, brains were processed 5 min after

ECS, as explained below.

In situ hybridization and immunohistochemistry. In situ hybridization was

performed as previously described46. Fresh-frozen 18-mm coronal brain sections

from mutants and control animals were mounted on the same glass slides to

minimize differences in handling and processing. The cDNA clones for ribop-

robes were either generated or were gifts from P. Worley, Johns Hopkins

University. No hybridization signal was detected with sense strand probes.

For immunohistochemistry, the animals were anesthetized with i.p. injection

of Avertin (2,2,2-tribromoethanol; Sigma) and transcardially perfused with

cold 4% paraformaldehyde (PFA) in phosphate buffered saline (PBS, pH 7.4).

After 2- to 3-h postfixation in 4% PFA, the brains were equilibrated in 30%

sucrose in PBS at 4 1C, frozen and stored at �80 1C until use. Free-floating

coronal sections (50 mm) were incubated overnight at 4 1C in the primary

antibody (C terminus–specific anti-SRF, 1:2,000, a gift from R. Misra,

University of Wisconsin; pCREB, #06-519, Upstate, 1:1,000; pErk1/2, #9106,

Cell Signaling, 1:400) after 1-h incubation in blocking solution at room

temperature (20–22 1C). After washes and incubation with the secondary

antibody (biotinylated goat anti-mouse or anti-rabbit, Vector labs, 1:250) for

1 h at room temperature, sections were processed using the ABC Elite Kit

(Vector labs) according to the manufacturer’s instructions.

Novelty exposure. Mice were moved from their home cages into an enriched

environment and allowed to explore for 30 min, after which they were killed and

brains processed for in situ hybridization as described above. One group of mice

was killed directly from their home cages without being subjected to environ-

ment exploration, and these served as caged controls (‘home-caged’ mice).

Golgi staining. Golgi staining for Srf f/f, SRFf/f;SCre and Srf f/f;CKCre mice (6

month old; n ¼ 3) were carried out using the FD Rapid GolgiStain Kit (FD

NeuroTechnologies) according to the manufacturer’s instructions.

Cell counts. CA1 pyramidal neurons were counted from Nissl-stained 10-mm

coronal brain sections from Srf f/f, Srf f/f;CKCre, CrebCamkcre4Crem+/� and

CrebCamkcre4Crem�/� mice (6–7 months old) at 40� magnification (boxed

regions in Fig. 6). Similarly, dentate gyrus neurons were counted from Srf f/f,

Srf f/f;SCre, CrebCamkcre4Crem+/� and CrebCamkcre4Crem�/� mice (Srf, 1 year old;

Creb/Crem, 6–7 months old). For each animal, 28 defined regions (correspond-

ing to boxes in Fig. 6) were counted and analyzed blind from seven coronal

sections using Openlab imaging software. Mean 7 s.d. from two animals

is represented.

Hippocampal slice preparation. Mice (3–4 months old) were anesthetized

with halothane and decapitated. Hippocampi were rapidly removed and placed

in ice-cold saline containing (in mM) 110 choline chloride, 2.5 KCl,

1.2 NaH2PO4, 25 NaHCO3, 0.5 CaCl2, 7 MgCl2, 2.4 pyruvate, 1.3 ascorbic

acid, 20 glucose, adjusted to pH 7.4 and oxygenated with 95% O2/5% CO2.

Transverse slices (400-mm thick for field potential recording or 250 mm for

whole-cell recording) were prepared with a vibrating tissue slicer (Leica

VT1000S) and kept in a solution containing (in mM) 126 NaCl, 5 KCl,

2 MgCl2, 2 CaCl2, 1.25 NaH2PO4, 26 NaHCO3, 20 glucose, pH 7.4 (ACSF1;

field potential recording) or 125 NaCl, 2.5 KCl, 1.25 NaH2PO4, 25 NaHCO3,

2 CaCl2, 2 MgCl2, 25 glucose, pH 7.4 (ACSF2; whole-cell recording) for at least

1 h at room temperature.

Field potential recording. Slices were transferred to a Haas-style interface

recording chamber perfused with ACSF1, warmed to 32.0 7 0.3 1C at a rate of

2 ml min�1 and allowed to recover for at least 1 h. Field potentials were

recorded with glass micropipettes (1–2 MO) filled with external saline and

placed in CA1 stratum radiatum. Schaffer collateral/commissural fibers were

simulated at 0.05 Hz with brief (100-ms) constant-voltage pulses delivered

by a concentric bipolar electrode. At the beginning of each experiment, an

input/output curve was established and the strength of the test stimuli was

adjusted to produce a response equal to B50% of the maximum fEPSP

slope. Signals were low-pass filtered at 3 kHz and digitized at 10 kHz.

GABAzine (10 mM) was added to block GABAA receptors. To elicit LTP,

two different stimulus protocols were used. In one set of experiments, theta-

burst stimulation was used, consisting of five sets with an interset interval of

20 s. Each set consisted of 15 bursts delivered at 5 Hz, and each burst contained

five pulses delivered at 100 Hz. In another set of experiments, weak tetanic

stimulation consisting of 50 pulses delivered at 50 Hz was used to evoke a

smaller LTP response.

Whole-cell recording. Slices were placed in a submerged chamber and perfused

with ACSF2 and warmed to 32 1C at 2 ml min�1. Whole-cell recordings were

obtained under visual guidance from pyramidal neuron somata located 2–3 cell

diameters below the slice surface. Synaptic transmission was elicited by passing

current through an extracellular glass stimulation electrode placed in stratum

radiatum layer of CA1, B100 mm from the recording electrode. Patch pipettes

had a resistance of 3–5 MO when filled with an internal solution containing


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

(in mM) 120 cesium methanesulfonate, 10 HEPES, 5 CaCl2, 10 Cs4BAPTA,

2 Na2ATP, 0.3 Na3GTP, 5 QX314 (adjusted to pH 7.3 with CsOH). 10 mM

GABAzine was added in all experiments, and 0.5 mM tetrodotoxin was added

for mEPSC experiments. Signals were low-pass filtered at 1 kHz and digitized at

10 kHz. All values were expressed as mean 7 s.e.m. and compared statistically

using unpaired Student’s t-test. QX314 was purchased from Tocris, cesium

BAPTA from Molecular Probes and tetrodotoxin from Calbiochem; all other

chemicals were from Sigma.


ACKNOWLEDGMENTSWe thank J. Baraban, A. Kolodkin, R. Misra and P. Worley for comments on themanuscript; R. Misra and P. Worley for valuable advice and reagents; the JohnsHopkins Transgenic Core facility for help with generating conditional mice;members of the Ginty, Linden, Kolodkin and Ghosh laboratories for helpfuldiscussions and suggestions and K. Takamiya and R. Huganir for constructs andmice. This study was supported by grants from the US National Institutes ofHealth to D.J.L. and D.D.G; N.R. is supported by a fellowship from the DamonRunyon Cancer Research Foundation. D.D.G. is an Investigator of the HowardHughes Medical Institute.




1. Lanahan, A. & Worley, P. Immediate-early genes and synaptic function. Neurobiol. Learn.Mem. 70, 37–43 (1998).

2. West, A.E., Griffith, E.C. & Greenberg, M.E. Regulation of transcription factors byneuronal activity. Nat. Rev. Neurosci. 3, 921–931 (2002).

3. Deisseroth, K., Mermelstein, P.G., Xia, H. & Tsien, R.W. Signaling from synapse tonucleus: the logic behind the mechanisms. Curr. Opin. Neurobiol. 13, 354–365 (2003).

4. Sheng, M. & Greenberg, M.E. The regulation and function of c-fos and other immediateearly genes in the nervous system. Neuron 4, 477–485 (1990).

5. Robertson, L.M. et al. Regulation of c-fos expression in transgenic mice requiresmultiple interdependent transcription control elements. Neuron 14, 241–252 (1995).

6. Lonze, B.E. & Ginty, D.D. Function and regulation of CREB family transcription factors inthe nervous system. Neuron 35, 605–623 (2002).

7. Shaywitz, A.J. & Greenberg, M.E. CREB: a stimulus-induced transcription factoractivated by a diverse array of extracellular signals. Annu. Rev. Biochem. 68, 821–861 (1999).

8. Gass, P. et al. Deficits in memory tasks of mice with CREB mutations depend on genedosage. Learn. Mem. 5, 274–288 (1998).

9. Pittenger, C. et al. Reversible inhibition of CREB/ATF transcription factors in region CA1of the dorsal hippocampus disrupts hippocampus-dependent spatial memory. Neuron34, 447–462 (2002).

10. Balschun, D. et al. Does cAMP response element-binding protein have a pivotal role inhippocampal synaptic plasticity and hippocampus-dependent memory? J. Neurosci. 23,6304–6314 (2003).

11. Rammes, G. et al. Synaptic plasticity in the basolateral amygdala in transgenic miceexpressing dominant-negative cAMP response element-binding protein (CREB) inforebrain. Eur. J. Neurosci. 12, 2534–2546 (2000).

12. Norman, C., Runswick, M., Pollock, R. & Treisman, R. Isolation and properties of cDNAclones encoding SRF, a transcription factor that binds to the c-fos serum responseelement. Cell 55, 989–1003 (1988).

13. Chai, J. & Tarnawski, A.S. Serum response factor: discovery, biochemistry, biological rolesand implications for tissue injury healing. J. Physiol. Pharmacol. 53, 147–157 (2002).

14. Arsenian, S., Weinhold, B., Oelgeschlager, M., Ruther, U. & Nordheim, A. Serumresponse factor is essential for mesoderm formation during mouse embryogenesis.EMBO J. 17, 6289–6299 (1998).

15. Nagy, A. Cre recombinase: the universal reagent for genome tailoring. Genesis 26, 99–109 (2000).

16. Zhu, Y. et al. Ablation of NF1 function in neurons induces abnormal developmentof cerebral cortex and reactive gliosis in the brain. Genes Dev. 15, 859–876 (2001).

17. Tsien, J.Z. et al. Subregion- and cell type-restricted gene knockout in mouse brain.Cell 87, 1317–1326 (1996).

18. Misra, R.P., Rivera, V.M., Wang, J.M., Fan, P.D. & Greenberg, M.E. The serum responsefactor is extensively modified by phosphorylation following its synthesis in serum-stimulated fibroblasts. Mol. Cell. Biol. 11, 4545–4554 (1991).

19. Vooijs, M., Jonkers, J. & Berns, A. A highly efficient ligand-regulated Cre recombinasemouse line shows that LoxP recombination is position dependent. EMBO Rep. 2, 292–297 (2001).

20. Cole, A.J., Saffen, D.W., Baraban, J.M. & Worley, P.F. Rapid increase of an immediateearly gene messenger RNA in hippocampal neurons by synaptic NMDA receptoractivation. Nature 340, 474–476 (1989).

21. Bading, H. & Greenberg, M.E. Stimulation of protein tyrosine phosphorylation by NMDAreceptor activation. Science 253, 912–914 (1991).

22. Adams, J.P. & Sweatt, J.D. Molecular psychology: roles for the ERK MAP kinase cascadein memory. Annu. Rev. Pharmacol. Toxicol. 42, 135–163 (2002).

23. Hill, C.S. et al. Functional analysis of a growth factor-responsive transcription factorcomplex. Cell 73, 395–406 (1993).

24. Deogracias, R., Espliguero, G., Iglesias, T. & Rodriguez-Pena, A. Expression of theneurotrophin receptor trkB is regulated by the cAMP/CREB pathway in neurons. Mol.Cell. Neurosci. 26, 470–480 (2004).

25. Fujino, T., Lee, W.C. & Nedivi, E. Regulation of cpg15 by signaling pathways thatmediate synaptic plasticity. Mol. Cell. Neurosci. 24, 538–554 (2003).

26. Bottai, D. et al. Synaptic activity-induced conversion of intronic to exonic sequence inHomer 1 immediate early gene expression. J. Neurosci. 22, 167–175 (2002).

27. Sato, M., Suzuki, K. & Nakanishi, S. NMDA receptor stimulation and brain-derivedneurotrophic factor upregulate homer 1a mRNA via the mitogen-activated proteinkinase cascade in cultured cerebellar granule cells. J. Neurosci. 21, 3797–3805(2001).

28. Guzowski, J.F., McNaughton, B.L., Barnes, C.A. & Worley, P.F. Environment-specificexpression of the immediate-early gene Arc in hippocampal neuronal ensembles.Nat. Neurosci. 2, 1120–1124 (1999).

29. Wallace, C.S. et al. Correspondence between sites of NGFI-A induction and sites ofmorphological plasticity following exposure to environmental complexity. Brain Res.Mol. Brain Res. 32, 211–220 (1995).

30. Lonze, B.E., Riccio, A., Cohen, S. & Ginty, D.D. Apoptosis, axonal growth defects, anddegeneration of peripheral neurons in mice lacking CREB. Neuron 34, 371–385(2002).

31. Bleckmann, S.C. et al. Activating transcription factor 1 and CREB are important forcell survival during early mouse development. Mol. Cell. Biol. 22, 1919–1925(2002).

32. Mantamadiotis, T. et al. Disruption of CREB function in brain leads to neurodegenera-tion. Nat. Genet. 31, 47–54 (2002).

33. Wu, G., Malinow, R. & Cline, H.T. Maturation of a central glutamatergic synapse. Science274, 972–976 (1996).

34. Lynch, M.A. Long-term potentiation and memory. Physiol. Rev. 84, 87–136 (2004).35. Bading, H., Ginty, D.D. & Greenberg, M.E. Regulation of gene expression in

hippocampal neurons by distinct calcium signaling pathways. Science 260, 181–186(1993).

36. Treisman, R. Transient accumulation of c-fos RNA following serum stimulation requiresa conserved 5¢ element and c-fos 3¢ sequences. Cell 42, 889–902 (1985).

37. Greenberg, M.E., Siegfried, Z. & Ziff, E.B. Mutation of the c-fos gene dyad symmetryelement inhibits serum inducibility of transcription in vivo and the nuclear regulatoryfactor binding in vitro. Mol. Cell. Biol. 7, 1217–1225 (1987).

38. Treisman, R. The SRE: a growth factor responsive transcriptional regulator. Semin.Cancer Biol. 1, 47–58 (1990).

39. Schratt, G. et al. SRF regulates Bcl-2 expression and promotes cell survival duringmurine embryonic development. EMBO J. 23, 1834–1844 (2004).

40. Chang, S.H., Poser, S. & Xia, Z. A novel role for serum response factor in neuronalsurvival. J. Neurosci. 24, 2277–2285 (2004).

41. Parlakian, A. et al. Targeted inactivation of serum response factor in the developing heartresults in myocardial defects and embryonic lethality. Mol. Cell. Biol. 24, 5281–5289(2004).

42. Miano, J.M. et al. Restricted inactivation of serum response factor to the cardiovascularsystem. Proc. Natl. Acad. Sci. USA 101, 17132–17137 (2004).

43. Spencer, J.A. & Misra, R.P. Expression of the serum response factor gene is regulated byserum response factor binding sites. J. Biol. Chem. 271, 16535–16543 (1996).

44. Rodriguez, C.I. et al. High-efficiency deleter mice show that FLPe is an alternative toCre-loxP. Nat. Genet. 25, 139–140 (2000).

45. Schwenk, F., Baron, U. & Rajewsky, K. A cre-transgenic mouse strain for the ubiquitousdeletion of loxP-flanked gene segments including deletion in germ cells. Nucleic AcidsRes. 23, 5080–5081 (1995).

46. Pasterkamp, R.J., De Winter, F., Holtmaat, A.J. & Verhaagen, J. Evidence for a role of thechemorepellent semaphorin III and its receptor neuropilin-1 in the regeneration ofprimary olfactory axons. J. Neurosci. 18, 9962–9976 (1998).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Subunit interaction with PICK and GRIP controls Ca2+

permeability of AMPARs at cerebellar synapses

Siqiong June Liu1,2 and Stuart G Cull-Candy1

At many excitatory central synapses, activity produces a lasting change in the synaptic response by modifying postsynaptic AMPA

receptors (AMPARs). Although much is known about proteins involved in the trafficking of Ca21-impermeable (GluR2-containing)

AMPARs, little is known about protein partners that regulate subunit trafficking and plasticity of Ca21-permeable (GluR2-lacking)

AMPARs. At cerebellar parallel fiber–stellate cell synapses, activity triggers a novel type of plasticity: Ca21 influx through GluR2-

lacking synaptic AMPARs drives incorporation of GluR2-containing AMPARs, generating rapid, lasting changes in excitatory

postsynaptic current properties. Here we examine how glutamate receptor interacting protein (GRIP, also known as AMPAR

binding protein or ABP) and protein interacting with C-kinase-1 (PICK) regulate subunit trafficking and plasticity. We find that

repetitive synaptic activity triggers loss of synaptic GluR2-lacking AMPARs by selectively disrupting their interaction with GRIP

and that PICK drives activity-dependent delivery of GluR2-containing receptors. This dynamic regulation of AMPARs provides a

feedback mechanism for controlling Ca21permeability of synaptic receptors.

Because AMPARs have a central role in synaptic transmission andefficacy, much attention has been focused on the molecular mechan-isms involved in their dynamic control. At a number of excitatorysynapses, repetitive presynaptic activity produces a lasting change intransmission by modifying the postsynaptic response1,2. This modifi-cation can be associated with a rapid change in the total number ofsynaptic AMPARs. Regulation of the density of receptors in thesynaptic membrane is mediated by exocytosis and endocytosismechanisms controlled by various AMPAR-interacting proteinsinvolved both in constitutive trafficking of receptors and in theactivity-induced recruitment of new receptors to the synapse2–5.

Recently, we have identified a previously unknown type of plasticityat the parallel fiber input to cerebellar stellate cells6. During high-frequency synaptic activity, Ca2+ influx through GluR2-lackingAMPARs drives the rapid incorporation of GluR2-containing receptorsto produce a change in excitatory postsynaptic current (EPSC) proper-ties at this synapse6. Although many studies have examined the proteinsinvolved in regulating Ca2+-impermeable AMPARs, the trafficking ofCa2+-permeable (GluR2-lacking) AMPARs is much less well under-stood. GluR2-lacking receptors exhibit a number of characteristicproperties that make them of particular interest. These include elevatedCa2+ permeability7–10, high single-channel conductance11, block byendogenous intracellular polyamines at depolarized potentials12–14 andactivity-dependent facilitation of synaptic currents15. In addition, thedeactivation kinetics of GluR2-lacking receptors can be rapid com-pared with those of GluR2-containing receptors7. Furthermore, tetra-meric assemblies formed by unedited subunits (GluR1, 3 and 4) exitfrom the endoplasmic reticulum more readily than edited varieties16,

suggesting differences in the intracellular molecules involved in traf-ficking of unedited and edited AMPARs. The fact that Ca2+-permeableAMPARs have a pivotal role in various pathological conditions17–22 andcontribute to heterogeneity of transmission6,23–26 provided a strongimpetus to try to understand the regulation of Ca2+-permeableAMPARs at central synapses.

Here, we aimed to determine the intracellular protein partners thatdifferentially modulate the trafficking of GluR2-lacking and GluR2-containing AMPARs and ultimately control the postsynaptic Ca2+ influxin cerebellar stellate cells. The Ca2+-permeable AMPARs in these cells arelikely to be homomeric GluR3 assemblies26–29, whereas the Ca2+-impermeable AMPARs are expected to be mainly heteromeric GluR2/3assemblies6,27–29. As PICK and GRIP occur at high levels in stellatecells30,31, we focused our attention on these proteins. GRIP and PICKare known to bind via their PDZ domains to the C terminus of GluR2and GluR3 subunits. Phosphorylation of the C-terminal serine residueof GluR2 disrupts the GluR2-GRIP interaction without affecting GluR2/3-PICK interaction32,33, whereas GluR2-PICK is targeted along withactivated protein kinase C (PKC) to spines of hippocampal neurons34,and their interaction can be selectively disrupted by N-ethylmaleimide–sensitive fusion protein (NSF)35–41. Thus, the interaction of GluR2/3with GRIP and PICK appears to be critical in AMPAR regulation.

As an influx of Ca2+ through synaptic AMPARs is necessary and suf-ficient to trigger the switch in the EPSC properties and AMPAR subtypein stellate cells6, we examined whether synaptic activity selectively dis-rupts the interaction between GRIP and GluR2-lacking AMPARs, allo-wing their loss, and whether PICK is involved in a concomitant activity-dependent delivery of Ca2+-impermeable receptors at these synapses.


1Department of Pharmacology, University College London, Gower Street, London WC1E 6BT, UK. 2Department of Biology, Penn State University, State College, Pennsylvania16801, USA. Correspondence should be addressed to S.J.L. ([email protected]) or S.G.C.-C. ([email protected]).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

RESULTS

PICK is not required for maintenance of synaptic AMPARs

Our first approach was to interfere with the interaction between PICKand GluR2/3. The peptide inhibitor pep2-SVKI, which corresponds tothe C terminus of GluR2 (NVYGIESVKI), binds both PICK and GRIPand hence inhibits their interaction with GluR2 and GluR3 (refs.30,31,42). When the serine residue in pep2-SVKI is replaced with analanine, the mutant peptide (pep2-AVKI; NVYGIEAVKI) no longerbinds to GRIP but continues to bind PICK. Therefore, this can be usedto selectively block interaction between GluR2/3 and PICK43. As asecond approach, we used the inhibitor peptide pep2m (correspondingto the NSF binding site of GluR2) to inhibit interactions between NSFand GluR2 (refs. 38,39,44) and between clathrin adapter protein 2(AP-2) (ref. 36) and AMPARs.

We tested whether inhibiting the interaction between PICK andGluR2/3 altered the basal level of synaptic transmission at parallelfiber–stellate cell synapses by including 1 mM of the mutant peptide(pep2-AVKI) in the pipette solution. We monitored miniature EPSCs(mEPSCs; at –60 and +40 mV) over a 2-h period in the presence of1 mM tetrodotoxin (TTX). Pep2-AVKI did not affect mEPSC amplitude(Fig. 1a–d) or frequency at positive or negative potentials, suggestingeither that PICK is not required for the maintenance of AMPARs at thissynapse or that it binds tightly to receptors.

To interfere with the interaction between NSF and GluR2 and theinteraction between AP-2 and AMPARs, we included 1 mM pep2m(KRMKVAKNAQ) in the pipette solution. We did not observe achange in the amplitude or frequency of mEPSCs and spontaneousEPSCs (sEPSCs; at –60 or –80 mV) in cells monitored over B2 h(Figure 1e–h). As the effect of pep2m depends on synaptic activity41,45,we examined its time course of action on evoked EPSCs. The currentamplitude remained unaltered at –60 mV (Fig. 1g), and the smallreduction at +40 mV was not significant (P ¼ 0.18, n ¼ 5). This lack ofeffect of pep2m on basal levels of transmission is consistent with the

idea that NSF and AP2 are not involved in the constitutive delivery andremoval of synaptic AMPARs, respectively, in these cells, as indeedmight be expected for GluR2-lacking receptors.

Notably, in these experiments, the mEPSC amplitude was unaffectedby prolonged voltage clamp of cells at resting potential, although itis known that the subunit composition of postsynaptic AMPARsis regulated by the level of spontaneous synaptic activity instellate cells29.

Delivery of synaptic GluR2-containing AMPARs requires PICK

We have previously found that high-frequency stimulation of presy-naptic parallel fiber inputs to the stellate cell increases the EPSCamplitude at +40 mV and changes its current-voltage relationshipfrom inwardly rectifying to linear6. This alteration is consistent with arapid activity-dependent insertion of Ca2+-impermeable (GluR2-con-taining) receptors that are no longer blocked by polyamines at positivepotentials12–14. Is PICK required for this activity-dependent change inAMPAR subtype? We examined whether infusion of the mutantpeptide (pep2-AVKI) into cells blocked the activity-induced increasein the EPSC amplitude at +40 mV and modified its current-voltagerelationship. EPSCs were measured at various potentials before andafter 50-Hz stimulation of parallel fibers.

Typical inwardly rectifying EPSCs (Fig. 2) signified the presence ofCa2+-permeable AMPARs at these synapses. After high-frequency pre-synaptic stimulation, the EPSCs were reduced in amplitude (at +40 mV;Fig. 2a,d,l). Pep2-AVKI therefore prevented the normal activity-dependent increase that occurs at positive potentials6. Furthermore,EPSCs at –60 mV were also reduced after stimulation (Fig. 2a,c,k). Asa consequence, the ratio of EPSC amplitude at +40 mV to amplitudeat –60 mV (R+40/–60mV; the rectification index) remained unaltered(Fig. 2e,m). The decrease in EPSC amplitude at both potentialsoccurred within 10–20 min of stimulation and lasted for at least1 h (Fig. 2b).

40 80 120

–250

–200

–150

–100

–50

0

50

mE

PS

C a

mpl

itude

(pA

) Time (min)

1 2 3

4 5

mE

PS

C a

mpl

itude

(pA

)

–160

–120

0 20 40 60 80 100 120 140

–80

–40

0

Time (min)

0–40 40–800

0.05

0.10

0.15

0.20

0.25

Fre

quen

cy (

no. o

f eve

nts/

s)

Time (min)

Miniature EPSCSpontaneous EPSC

a

–60 mV

+40 mV

10 pA5 ms

2 3

4 5

1

b

0–40 40–800.00

0.02

0.04

0.06

0.08

0.10m

EP

SC

freq

uenc

y (H

z)

Time (min)

d

e 0–60 min

5 ms10 pA

60–135 minf

hg

0 10 15 20 25 30 35 400.0

0.2

0.4

0.6

0.8

1.0

1.2

Time (min)

–60 mV

+40 mV

Nor

mal

ized

evo

ked

EP

SC

am

p

20 40 60 80 1000

–20

–40

–60

–80

Time (min)

EP

SC

am

plitu

de(p

A)

Pep2-AVKI Pep2m

40 80 120

–60

–40

–20

0

20

mE

PS

C a

mpl

itude

(pA

)

c

Time (min)

+40 mV

–60 mV 5

Figure 1 PICK is not involved in the maintenance of synaptic AMPA receptors in stellate cells. (a–d) pep2-AVKI, a peptide that binds to PICK, did not alter

basal synaptic transmission in cerebellar stellate cells when present (1 mM) in the pipette solution. (a) Amplitude of mEPSCs recorded at –60 mV (open circles,

time periods 1, 2 and 3) and +40 mV (filled circles: time periods 4 and 5) during a period of 2 h. (b) Average traces of mEPSCs at –60 mV and +40 mV over

each time period shown in a. (c) The mean amplitude of mEPSCs at –60 and +40 mV (n ¼ 5; mean 7 s.e.m.) remained constant over a time period of 2 h.

(d) The mEPSC frequency in each cell did not change with time. (e–h) pep2m did not alter the amplitude or the frequency of mEPSCs and spontaneous EPSCs

when present (1 mM) in the pipette solution. (e) mEPSCs were measured in the presence of 1 mM TTX over a period of 2 h. (f) Averaged traces of mEPSCs.

(g) Plot of amplitude of averaged evoked EPSCs against time; amplitude was normalized to the value determined during the first 1–10 min (n ¼ 5). The inset

shows the mean amplitude of mEPSCs and sEPSCs (n ¼ 5). (h) Frequency of miniature EPSC and spontaneous EPSC at 0–40 min and 40–80 min.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

To test whether this effect of pep2-AVKI was due to the selectivebinding of pep2-AVKI to PICK, we examined the effect of the inactivecontrol peptide, pep2-SVKE (NVYGIESVKE), on EPSCs. Pep2-SVKEcontains a glutamate residue at the C terminus instead of isoleucine,rendering it ineffective at binding either to PICK or GRIP46. Asexpected, with pep2-SVKE included in the pipette, high-frequencysynaptic stimulation triggered a normal increase in EPSC amplitude at+40 mV and a decrease at –60 mV (Fig. 2f–i). The ratio increased from0.28 7 0.02 to 0.45 7 0.06 (n ¼ 7; P o 0.01; Fig. 2j,m). This isconsistent with our earlier observations made in the absence of addedintracellular peptides6.

As expected, the ratio of EPSC amplitudes (R+40/–60mV) measuredbefore stimulation was similar regardless of whether the peptideincluded in the intracellular solution was active or the control (pep2-AVKI or pep2-SVKE; Fig. 2m). Hence, the fact that pep2-AVKI clearlyinhibited the activity-induced increase in the ratio of EPSC amplitudes(R+40/–60mV) suggested that it blocked the change in subunit composi-tion. Its ability to block the increase in EPSC amplitude at +40 mV(mediated entirely by GluR2-containing receptors) therefore is likely toreflect specific binding to PICK.

We next investigated whether interferingwith GluR2-NSF interaction also blocked theactivity-dependent increase in GluR2-contain-ing receptors at this synapse. With 1 mMpep2m in the pipette, high frequency stimula-tion did not alter the current-voltage relation-ship or amplitude of EPSCs (Fig. 3a,c).However, in the presence of a scrambled ver-sion of the peptide (pep2s; VRKKNMAKQA)that does not interact with NSF38, parallel fiberstimulation triggered an increase in EPSCamplitude at positive potentials and a decreaseat negative potentials (Fig. 3b,d), as has beenobserved in the absence of added peptides6.

The fact that both peptide inhibitors (pep2-AVKI and pep2m)blocked the increase in synaptic GluR2-containing receptors supportsthe view that GluR2-PICK interaction is responsible for the activity-dependent delivery of GluR2-containing receptors at these synapses.

Is the total number of synaptic AMPARs altered by activity?

Whereas repetitive activation of synaptic Ca2+-permeable AMPARsreduced EPSC amplitude by B30% at –60 mV, the current wasincreased at +40 mV6. As GluR2-lacking receptors are blocked byintracellular spermine at depolarized potentials, this indicates anincreased contribution from GluR2-containing receptors. However,the cause of the decreased EPSC amplitude at –60 mV, when the EPSCis mediated both by GluR2-containing and GluR2-lacking AMPARs, isless clear. It could reflect either a reduction in the total number ofpostsynaptic AMPARs or an alteration in channel conductance due toinclusion of edited subunits in the receptor11.

We have previously found that AMPARs in the soma of stellate cellsundergo an activity-dependent change in subtype6,29. Somatic recep-tors in control cells have an outward or linear current-voltage relation-ship and low Ca2+ permeability, characteristic of GluR2-containing

5 ms

20 pA

+40 mV–60 mV

Beforestimulation

15 min afterstimulation

4 ms

20 pA

+40 mV

–60 mV

Time (min)

Stimulation

Before After0.0

0.1

0.2

0.3

0.4

stim.stim.Before After

10

15

20

25

30

35

stim.stim.

+40 mV

Before After–40

–60

–80

–100

–120 –60 mV

Before After0.0

0.2

0.4

0.6

stim.stim.Before After

10

20

30

40

stim.stim.

+40 mV

Before After–30

–40

–50

–60

–70

–80

–90

–100 –60 mV

stim.stim.

Beforestimulation

15 min afterstimulation

a

c d e

f g

h i j

k l m

Time (min)

–60 mV+40 mV

Stimulation

b

Nor

mal

ized

EP

SC

amp

(pA

)E

PS

C a

mp,

+40

/–60

mV

Nor

mal

ized

EP

SC

amp

(pA

)

EP

SC

am

plitu

de (

pA)

EP

SC

am

plitu

de (

pA)

EP

SC

am

plitu

de (

pA)

EP

SC

am

plitu

de (

pA)

EP

SC

am

p,+

40/–

60m

V

Before stimulationAfter stimulation

SVKE AVKI0

–20

–40

–60

–80

–100 –60 mV

****

EP

SC

am

plitu

de (

pA)

SVKE AVKI0

5

10

15

20

25

30+40 mV

**

**

EP

SC

am

plitu

de (

pA)

SVKE AVKI0.0

0.1

0.2

0.3

0.4

0.5

0.6

***

EP

SC

am

p,+

40/–

60 m

V

stim.stim.

Pep2-AVKI

Pep2-SVKE

1.4

1.2

1.0

0.8

0.6

–20 20 40

0.6

0.8

1.0

1.2

1.4

–20 20 40

–60 mV+40 mV

0

0

Figure 2 Pep2-AVKI inhibited the activity-induced

increase in the amplitude of EPSCs at +40 mV.

Pep2-AVKI (1 mM) (a–e), or the control peptide

pep2-SVKE (f–j), was included in the pipette

solution. (a,f) Averaged EPSCs obtained at +40

and –60 mV before and after high-frequency

stimulation. (b) When pep2-AVKI was included in

the pipette solution, high-frequency presynapticstimulation induced a lasting decrease in the

EPSC amplitude at +40 mV and –60 mV. (g) High-

frequency synaptic stimulation induced an

increase in the EPSC amplitude at +40 mV but a

decrease at –60 mV when a control peptide was

present. (c,h) EPSC amplitude in an individual

cell at –60 mV before and after stimulation.

(d,i) EPSC amplitude in an individual cell at +40

mV before and after stimulation. EPSC amplitude

decreased in the presence of pep2-AVKI; it

increased when control peptide was present.

Different symbols indicate recordings from

different cells. (e,j) Ratio of EPSC amplitudes at

+40 mV versus –60 mV was not altered by high-

frequency stimulation in the presence of pep2-

AVKI but increased when the control peptide was

included. (k,l) Mean EPSC amplitudes at –60

and +40 mV. (m) Rectification index before and

after stimulation. * P o 0.05; ** P o 0.02;*** P o 0.01 by t-test; pep2-AVKI: n ¼ 5 cells;

pep2-SVKI: n ¼ 7 cells.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

receptors. However, in cells pretreated with TTX (to block spontaneousaction potentials), the somatic AMPARs showed a partial inwardlyrectifying current-voltage relationship and high Ca2+ permeability29.The rectification index of these receptors is similar to that of synapticreceptors before high-frequency stimulation.

To measure the single-channel conductance of somatic AMPARs incontrol and TTX-treated cells, we applied a low concentration ofglutamate (10–50 mM, in the presence of the NMDAR antagonistD-AP5 (50 mM )) to outside-out patches. Because of their rectificationproperties and their sensitivity to Joro spider toxin6, we deduced thatAMPAR channels in patches from cells treated with TTX expressedboth GluR2-containing and GluR2-lacking AMPARs. Single-channelcurrents in these patches had larger mean conductances than didcontrol cells (untreated with TTX) that expressed predominantlyGluR2-containing somatic AMPARs (Fig. 4). The weighted meansingle-channel current was B30% higher in TTX-treated cells (7.27 0.3 pA at –60 mV; P o 0.003) than the mean value of 5.5 7 0.2 pA(n ¼ 5) in control cells. This matches the B30% reduction in EPSCamplitude (at –60 mV) measured after high frequency stimulation,suggesting that synaptic activity can alter the receptor subtypewithout necessarily eliciting a marked change in the total number ofsynaptic AMPARs.

GRIP is involved in maintenance of synaptic AMPARs

To test whether GRIP interacts with AMPARs, stabilizing them in thesynaptic membrane of stellate cells, we used pep2-SVKI to inhibit theinteraction of GluR2/3 with both GRIP and PICK31,43,46. When pep2-SVKI was included in the patch pipette (Fig. 5), the mEPSC amplitudedecreased by B30% (P o 0.03) over a period of 2 h, whereas themEPSC frequency remained unaltered (Fig. 5a,b,g). When pep2-SVKIwas replaced with a ‘scrambled’ peptide (pep2Cs), the mEPSC

amplitude remained constant over the same time period (Fig. 5c,d,g).As pep2-SVKI reduced mEPSC amplitude at both –60 mVand +40 mV,the ratio of mEPSC amplitudes remained unchanged (Fig. 5h). To testthe possibility that interaction between pep2-SVKI and other intracel-lular proteins might underlie this reduction in mEPSC amplitude, weused the inactive peptide pep2-SVKE, which does not bind to GRIP orPICK43. This did not alter the amplitude of mEPSCs (Fig. 5e,f,g).

As we described earlier, pep2-AVKI selectively bound to PICKwithout altering mEPSC amplitude (Fig. 1). Collectively, ourdata suggests that GRIP, but not PICK, is involved in synapticAMPAR stabilization.

Activity disrupts the GRIP–GluR2-lacking AMPAR interaction

The activity-dependent change in stellate cell EPSCs predicts thatGluR2-lacking receptors may be released from their anchor proteins.Therefore, we next tested whether activation of Ca2+-permeablesynaptic AMPARs disrupted the interaction between GRIP andGluR2-lacking receptors. We assumed that GRIP molecules that dis-sociate from AMPARs can bind with a competing peptide (pep2-SVKI)and that receptors released from GRIP are subsequently removed from

Pep2m Pep2s

5 ms20 pA

–60

–60

–60 –60 –60 –20–20–20–20 20 60

20

60

–60

20

60

–60

20

60

–60

20

60

20 60 20 2060 60

+60 mV–60 mV

10 pA5 ms

Before stimulation After stimulationBefore stimulation After stimulation

0.2

0.4

0.6

0.8

1.0

1.2

1.4–60 mV+40 mV

Stimulation

–20 –10 0 10 20 30 40 50

0.2

0.4

0.6

0.8

1.0

1.2

1.4

Time (min)–20 –10 0 10 20 30 40 50

Time (min)

Stimulation

Nor

mal

ized

EP

SC

am

plitu

de

Nor

mal

ized

EP

SC

am

plitu

de

a b c dPep2m Pep2s

Voltage (mV)

Cur

rent

(pA

)

Figure 3 Effect of pep2m and control scrambled peptide (pep2s) on the activity-dependent change in EPSC amplitude, and current-voltage relationships.

(a,b) Evoked EPSCs (top) at –60 mV and +60 mV before and after stimulation. Current-voltage relationships of average EPSCs (bottom; n ¼ 5 cells in both

sets). The presence of 1 mM pep2m (a) blocked the activity-induced change in the EPSC amplitude and suppressed the change in current-voltage relationship.

With the control peptide (pep2s) included in the pipette solution (b), high-frequency presynaptic stimulation triggered an increase in EPSC amplitude at

depolarized potentials and a decrease at hyperpolarized potentials. As expected, the current-voltage relationship changed from inwardly rectifying to linear.

(c,d) Plot of EPSC amplitude versus time with pep2m (n ¼ 9) and with control peptide (n ¼ 5).

20 ms

0.5 pA

–1.2 –0.8 –0.4 0.0 0.40

2,000

4,000

6,000

pA–1.2 –0.8 –0.4 0.0 0.4

1,000

2,000

3,000

4,000

Cou

nt /

0.01

pA

Cou

nt /

0.01

pA

pA

2

4

6

8

10

12

14

16

18

TTX-treatedControl

Indi

vidu

al c

ondu

ctan

ce (

pS)

2

4

6

8

TTX-treatedControl

Wei

ghte

d co

nduc

tanc

e (p

S)

Control TTX-treateda

b c

0

Figure 4 Single-channel currents recorded from outside-out patches excised

from the soma of stellate cells. Recordings were made from cells in control

slices and in slices treated for at least 3 h with 1 mM TTX. Patches were held

at –60 mV, and single channels currents were activated by bath application of

10–50 mM glutamate. (a) Representative single-channel currents (top) and

corresponding all-point amplitude histograms (bottom). (b) Individual

estimates, and average values, for single-channel conductances. (c) Weighted

mean conductance of single-channel currents in patches from control and

TTX-treated cells (n ¼ 5 patches, P o 0.003). Each symbol represents a

different conductance state. Large symbols indicate mean values 7 s.e.m.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

the synapse. Since pep2-SVKI binds not only to GRIP but also to PICK,we used either pep2-SVKI or the peptide inhibitor pep2-AVKI, whichbinds selectively with (and inactivates) PICK, leaving GRIP unaffected.The difference in the effect produced by these two peptide inhibitorswas used to dissect the interaction between GRIP and AMPARs.

If synaptic activity selectively disrupts the interaction between GRIPand Ca2+-permeable AMPARs, a difference in the effect produced bypep2-SVKI and pep2-AVKI will be seen only at hyperpolarized poten-tials when a proportion of the EPSC is carried by GluR2-lackingAMPARs. We measured EPSCs at –60 and +40 mV (Fig. 6) beforeand after high-frequency synaptic stimulation, with 1 mM pep2-SVKIincluded in the pipette solution. We then determined the stimulation-induced change in EPSC amplitude (EPSCbefore stim – EPSCafter stim) atboth potentials (Fig. 6a–c). We compared the activity-induced changeobtained with pep2-SVKI, which is expected to block delivery andenhance loss of AMPARs by blocking PICK and GRIP, respectively(experiments described earlier; Fig. 2), with that obtained in thepresence of pep2-AVKI (to block AMPAR delivery by blockingPICK). The difference in EPSCs was expected to reflect the reductionin EPSC amplitude caused by enhanced loss of AMPARs as a result ofactivity-induced dissociation from GRIP.

Inclusion of pep2-SVKI in the patch pipetteinhibited the increase in EPSC amplitude thatwas typically seen at +40 mV after high fre-quency stimulation (Fig. 6a). Instead, theEPSC amplitude was slightly reduced. Thissmall decrease resembled that seen with thePICK selective blocker pep2-AVKI (Fig. 6d).Therefore, the interaction between Ca2+-impermeable AMPARs and GRIP was appar-ently unaffected by synaptic activity. Incontrast, the reduction in EPSC amplitudeproduced by intracellular pep2-SVKI at –60mV was significantly greater than thatobtained with pep2-AVKI (Fig. 6e,f). Thisenhanced reduction when both PICK andGRIP were inhibited suggests that a propor-tion of the synaptic AMPARs dissociate fromGRIP after activity. Furthermore, as the inter-action between Ca2+-impermeable AMPARs

and GRIP was unaffected by synaptic activity (Fig. 6d), our findingssuggest that activity is selectively disrupting the interaction betweenGRIP and GluR2-lacking AMPARs.

The validity of this interpretation requires that before high frequencystimulation, a similar relative proportion of Glu2-lacking synapticAMPAR be present in the pep2-SVKI– and pep2-AVKI–filled cells.Indeed, we found no significant difference between the two populationsof cells in the ratio of EPSC amplitudes at +40 versus –60 mV (Fig. 6f).After high-frequency activity, the ratio R+40/–60mV increased by 35 715% (n ¼ 7) when both GRIP and PICK were inhibited (with pep2-SVKI), whereas little change was observed when we used the PICK-selective peptide (pep2-AVKI). As the EPSC amplitude was slightlyreduced at +40 mV after stimulation (Fig. 6d), the increase inR+40/–60mV that we observed is likely to arise entirely from selectivelydisruption of the interaction between GRIP and Ca2+-permeableAMPARs rather than from any increase in GluR2-containing receptors.

DISCUSSION

By infusing specific peptide inhibitors into stellate cells and monitoringEPSC properties, we have identified interacting protein partnersinvolved in the activity-dependent switch in the Ca2+ permeability of

Pep2-SVKI

mEPSC (0–60 min)

5 ms

10 pA

mEPSC (60–120 min)a

20 pA

5 ms

mEPSC (60–125 min)

Pep2CsmEPSC (0–60 min)

10 pA

5 ms

Pep2-SVKE

0–40 40–80 80–1200.0

0.5

1.0

1.5

Time (min)

Nor

mal

ized

freq

uenc

y

c

e

Pep2CsPep2-SVKE

mE

PS

C a

mpl

itude

(pA

)

Time (min) Time (min)

–60

–50

–40

–30

–80

–70–80

–70

–60

–50

–40

g

Pep2-SVKI Pep2-SVKE0.0

0.1

0.2

0.3

0.4 0–60 min60–120 min

h

d

20 40 60 80 100

60

50

40

30

20

10

0

10

20

Time (min)–60 mV+40 mV

20 40 60 80 100

–60

–50

–40

–30

–20

–10

0

mE

PS

C a

mpl

itude

(pA

)m

EP

SC

am

plitu

de (

pA)

b

–60

20 40 60 80 100

–50

–40

–30

–20

–10

0

10

20

Time (min)

mE

PS

C a

mpl

itude

(pA

)

f

*

*

mEPSC (20–30 min) mEPSC (90–120 min)

Pep2-SVKI

mE

PS

C a

mp,

+40

/–60

mVmE

PS

C am

plitude (pA)

0–40 80–120 0–40 80–120

Time (min)

Figure 5 Pep2-SVKI reduced the amplitude of

synaptic currents. (a,b) Time-dependent change in

the amplitude and frequency of mEPSC recorded

in the presence of intracellular pep2-SVKI

(1 mM): averaged mEPSCs at –60 mV (a; inset

shows frequency of mEPSCs plotted against time)

and plot of mean amplitude of mEPSC versus time

(b; –60 and +40 mV; n ¼ 5; *, P o 0.03compared with the amplitude at 20 and 40 min).

(c,d) Averaged mEPSCs at –60 mV (c) and plot

of mean amplitude of mEPSC versus time

(d; –60 mV; n ¼ 5) in the presence of 1 mM

pep2Cs. (e,f) Averaged mEPSCs at –60 mV

(e) and mean mEPSC amplitude versus time

(f; –60 and +40 mV; n ¼ 5) in the presence of

pep2-SVKE. (g) Comparison of time-dependent

change in mEPSC amplitude in individual cells

infused with the various peptides (at –60 mV).

Values were compared at 0–40 and 80–120 min.

(h) Ratio of mEPSC amplitudes at +40 versus

–60 mV in pep2-SVKI and pep2-SVKE at 0–60

and 60–120 min (n ¼ 3).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

synaptic AMPARs. Repetitive activation of the Ca2+-permeableAMPARs disrupts their interaction with GRIP, resulting in their lossfrom the synapse; PICK drives the activity-dependent delivery of Ca2+-impermeable receptors into the synaptic membrane. This dynamicregulation of edited and unedited AMPARs, triggered by receptor-mediated Ca2+-entry, therefore provides a feedback mechanism forlimiting subsequent postsynaptic Ca2+ influx.

Delivery of AMPAR subtypes at Ca2+-permeable synapses

Several observations show that PICK proteins are required for activity-dependent delivery of the GluR2-containing receptors. First, selectivelyoccluding PICK blocks the increase in EPSC amplitude at depolarizedpotentials. Second, disrupting the NSF-GluR2 interaction and the AP-2–AMPAR interaction blocks the change in EPSC amplitude at positiveand negative potentials. NSF-GluR2 binding is known to stimulate NSFATPase activity35, which disrupts GluR2-PICK interaction and maythereby allow stabilization of GluR2-containing AMPARs in themembrane. Hence, interfering with NSF-GluR2 interaction may pre-vent uncoupling of PICK-GluR2 complexes, hindering delivery andmaintenance of edited receptors.

On the other hand, constitutive delivery of Ca2+-permeableAMPARs does not require GluR2/3-PICK interaction or NSF-GluR2interaction. However, we cannot exclude that NSF and AP-2 areinvolved in constitutive delivery and removal, respectively, of AMPARsand that blocking the NSF-GluR2 interaction with pep2m inhibits bothprocesses. It is notable that in hippocampal neurons, pep2m reducesEPSC amplitude within B30 min36,38–40. Our observations, therefore,clearly suggest differences in the molecules involved in constitutiverecycling and stabilization of unedited AMPARs in stellate cells and ofedited AMPARs in hippocampal cells.

Since Ca2+ influx is required for the switch in receptor subtype instellate cells6,29, our findings suggest a model in which Ca2+ influx asso-ciated with repetitive synaptic activity initiates the targeting and deliveryof the PICK-GluR2 complex (Supplementary Fig. 1). NSF may thencause disruption of the PICK-GluR2 interaction, allowing editedAMPARs to bind with anchor proteins which stabilize them at thesynapse. Our previous experiments have shown that the somaticAMPARs are Ca2+-impermeable (GluR2-containing) in these cells.

Since AMPARs can diffuse laterally between extrasynaptic and synapticlocations47, this raises the intriguing possibility that activity triggersextrasynaptic insertion of edited AMPARs, which diffuse laterally toaccumulate at the synapse.

Activity disrupts GRIP–Ca2+-permeable AMPAR interaction

GRIP is thought to anchor AMPARs at many glutamatergic synapses42.Blocking the GRIP-AMPAR interaction in stellate cells reduces themEPSC amplitude by B30%, consistent with the view that GRIP doesindeed act as an anchor protein in these cells. However, only GluR2-lacking AMPARs seem to be released from their association with GRIP,whereas the level of GluR2-containing receptors remain unaltered. Onepossible explanation for such selective disruption is that local Ca2+

influx through the unedited AMPARs causes their phosphorylation,reducing their binding affinity for GRIP. Once dissociated, they couldthen be replaced by GluR2-containing receptors, delivered by PICK. Itis notable that the interaction between GluR2 and GRIP can bedisrupted by phosphorylation of Ser880 (refs. 32,33) and that acomparable serine residue is present on GluR3. However, whetherthis serine residue can be phosphorylated and whether phosphorylatedGluR3 subunits no longer interact with GRIP remains to be seen.

Involvement of a second anchor protein could explain the activity-dependent reduction in EPSCs when PICK is occluded with pep2-AVKI. Stellate cells express stargazin, which can interact with GluR1, 2and 4 subunits; whether stargazin interacts with GluR3 is not known.

Comparison with other types of plasticity

Although the protein partners involved in the dynamic regulation ofedited and unedited AMPARs in stellate cells also have key roles inAMPAR trafficking and plasticity in hippocampal CA1 neurons andPurkinje cells, the stellate cells show some interesting differences. First,following induction of hippocampal long-term potentiation (LTP) andcerebellar long-term depression (LTD), the number of synapticAMPARs is thought to increase2 and decrease, respectively48. Incontrast, the change in EPSC amplitude that accompanies the switchin Ca2+ permeability in stellate cells can be accounted for by theexpected reduction in mean channel conductance, suggesting thatCa2+-permeable AMPARs are replaced rather than supplemented by

Figure 6 Activity-dependent reduction in EPSC

amplitude at –60 mV was greater with pep2-SVKI

than with pep2-AVKI. Either pep2-SVKI or pep2-

AVKI was included in the pipette solution. EPSCs

were recorded before and after high-frequency

stimulation. (a) Averaged EPSCs recorded at

–60 mV (lower traces) and +40 mV (upper

traces). (b,c) Comparison of EPSC amplitudebefore and after stimulation at –60 mV (b) and

+40 mV (c). Data are from individual stellate

cells. (d) Comparison of averaged EPSC

amplitudes at +40 mV before and after

stimulation in the presence of pep2-AVKI, SVKI

and pep2Cs (pep2-AVKI, P o 0.02, n ¼ 5;

pep2-SVKI, P o 0.03, n ¼ 7). (e) Time-

dependent change in mean EPSC amplitude

(at –60 mV) in the presence of pep2-AVKI and

pep2-SVKI. Data shown were normalized to

the values obtained before high-frequency

stimulation. Note the greater reduction in the

EPSC amplitude after high frequency stimulation

in the presence of pep2-SVKI, when compared with pep2-AVKI (P o 0.05). (f) Comparison of the mean ratio of EPSC amplitude at +40 versus –60 mV

before and after stimulation in the presence of pep2-AVKI and pep2-SVKI. Axis at right illustrates the percentage change in this ratio. Values obtained with

pep2-SVKI were significantly different from those obtained with pep2-AVKI (P o 0.05, t-test).

5 ms20 pA

Beforestimulation

a15 min afterstimulation

Pep2-SVKI

d

–20 0 20 40 600.5

0.6

0.7

0.8

0.9

1.0

e

Stimulation SVKI

At–60 mV

AVKI

Before After

10

20

30

40

SVKI

Pep2Cs

AVKI

At +40 mV

Time (min)

EP

SC

am

plitu

de (

pA)

EP

SC

am

p (I

/I-co

ntro

l)

EP

SC

am

plitu

de (

pA)

EP

SC

am

plitu

de (

pA)

Before After–40

–60

–80

–100

–120 –60 mV

Before After

15

25

35 +40 mV

0.2

0.3

0.4

0.5 Pep2-SVKIPep2-AVKI

–20

–10

0

10

20

30

40

50

60

EP

SC

am

p, +

40/–

60 m

V

Percentage change

f

b c

Before Afterstim.stim. stim.stim.

0.0

0.1

Pep2-SVKIPep2-SVKI

0

Glu

R2

Glu

R2


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

new receptors. Second, in hippocampal and cerebellar LTD, loss ofAMPARs depends on PICK-GluR2 interaction31,46. Furthermore,recent studies have shown that overexpression of PICK1 in CA1 cellsselectively downregulates synaptic GluR2-containing receptors33,34,49

and increases the level of GluR2-lacking receptors49. In stellate cells, wehave found that inhibiting PICK-GluR2/3 interaction suppressesAMPAR delivery rather than loss. Third, in keeping with the ideathat NSF interacts selectively with GluR2, NSF/AP2 is not required forconstitutive delivery of Ca2+-permeable AMPARs in stellate cells, but itis important for AMPAR delivery in hippocampal CA1 neurons40.Fourth, GRIP anchors AMPARs at synaptic and intracellular locationsin hippocampal cells46. In stellate cells, while GRIP appears to play a keyrole in AMPAR anchoring, activity triggers the release by GRIP of theunedited, but not the edited, AMPARs.

A study also identifying protein partners involved in the change ofAMPAR subtype in cerebellar stellate cells appeared while our paperwas in review50.

METHODSSlice preparation. Coronal slices (250 mm) of the cerebellum were cut from

postnatal day 18–21 Sprague-Dawley rats and maintained in slicing solution (in

mM: 125 NaCl, 2.5 KCl, 1 CaCl2, 2 MgCl2, 26 NaHCO3, 1.25 NaH2PO4 and 25

glucose, saturated with 95% O2/5% CO2, pH 7.4), as described previously6.

Experimental procedures were in accordance with the animal welfare guidelines

of University College London.

Electrophysiology. External solution (in mM: 125 NaCl, 2.5 KCl, 2 CaCl2,

1 MgCl2, 26 NaHCO3, 1.25 NaH2PO4, and 25 glucose) contained GABAA

and NMDA receptor blockers (20 mM SR95531, 100 mM picrotoxin, 50 mM

D-AP5). Patch pipettes (4–8 MO) were filled with a Cs+-based internal solution

(in mM: 95 CsF, 45 CsCl, 10 Cs-HEPES, 10 Cs-EGTA, 2 NaCl, 2 ATP-Mg,

1 QX314, 1 TEA, 1 CaCl2, 0.1 spermine, pH 7.3). Whole-cell patch-clamp and

outside-out patch recordings were made from stellate cells, identified as

described previously6, at room temperature (21–22 1C), using an Axopatch

200A amplifier (Axon instruments).

Synaptic currents were minimally evoked with a patch electrode filled with

the external solution by stimulating parallel fibers in the molecular layer at

0.33 Hz. High-frequency stimulation and recordings and analysis of EPSCs

were carried out as described earlier6. Briefly, parallel fibers were stimulated

by three trains of 100 stimuli at 50 Hz, separated by a 10-s interval, while the

postsynaptic cell was voltage-clamped at –80 mV. The experiment was

terminated if the series resistance changed by more than 20%.

Single-channel recordings were made from outside-out patches excised from

the soma of stellate cells in cerebellar slices maintained in external solution for

2–8 h or incubated with 1 mM TTX for at least 3 h. The current response to

application of 10–50 mM glutamate was recorded.

Recordings were filtered at 2 Hz and digitized at 20 Hz. All-point amplitude

histograms were constructed using Origin 6.0 software and were fitted using a

sum of Gaussian distribution functions. The amplitude distribution curves

obtained from control cells were fit well with a function with three Gaussian

peaks, and a four-peak Gaussian function best fits the data from the TTX-

treated cells. Individual chord conductance was calculated [(I – Ibaseline)/

holding potential].

All values are expressed as mean 7 s.e.m. Statistical significance was assessed

by two-tailed Student’s t-test.


ACKNOWLEDGMENTSWe thank S. Mok and C. Gebhardt for helpful discussions. This work wassupported by a Wellcome Trust Programme Grant and a Royal Society-WolfsonResearch Award to S.G.C.-C., and a US National Science Foundation grant toS.J.L. S.J.L. received a Wellcome Trust Travelling Fellowship.


Received 18 January; accepted 26 April 2005


1. Bliss, T.V. & Collingridge, G.L. A synaptic model of memory: long-term potentiation in thehippocampus. Nature 361, 31–39 (1993).

2. Bredt, D.S. & Nicoll, R.A. AMPA receptor trafficking at excitatory synapses. Neuron 40,361–379 (2003).

3. Sheng, M. & Kim, M.J. Postsynaptic signaling and plasticity mechanisms. Science 298,776–780 (2002).

4. Song, I. & Huganir, R.L. Regulation of AMPA receptors during synaptic plasticity. TrendsNeurosci. 25, 578–588 (2002).

5. Lei, S. & McBain, C.J. Two loci of expression for long-term depression at hippocampalmossy fiber–interneuron synapses. J. Neurosci. 24, 2112–2121 (2004).

6. Liu, S.Q. & Cull-Candy, S.G. Synaptic activity at calcium-permeable AMPA receptorsinduces a switch in receptor subtype. Nature 405, 454–458 (2000).

7. Geiger, J.R. et al. Relative abundance of subunit mRNAs determines gating and Ca2+

permeability of AMPA receptors in principal neurons and interneurons in rat CNS.Neuron 15, 193–204 (1995).

8. Jonas, P., Racca, C., Sakmann, B., Seeburg, P.H. & Monyer, H. Differences in Ca2+

permeability of AMPA-type glutamate receptor channels in neocortical neurons causedby differential GluR-B subunit expression. Neuron 12, 1281–1289 (1994).

9. Jonas, P. & Burnashev, N. Molecular mechanisms controlling calcium entry throughAMPA-type glutamate receptor channels. Neuron 15, 987–990 (1995).

10. Washburn, M.S., Numberger, M., Zhang, S. & Dingledine, R. Differential dependence onGluR2 expression of three characteristic features of AMPA receptors. J. Neurosci. 17,9393–9406 (1997).

11. Swanson, G.T., Kamboj, S.K. & Cull-Candy, S.G. Single-channel properties of recombi-nant AMPA receptors depend on RNA editing, splice variation, and subunit composition.J. Neurosci. 17, 58–69 (1997).

12. Kamboj, S.K., Swanson, G.T. & Cull-Candy, S.G. Intracellular spermine confers rectifi-cation on rat calcium-permeable AMPA and kainate receptors. J. Physiol. (Lond.) 486,297–303 (1995).

13. Bowie, D. & Mayer, M.L. Inward rectification of both AMPA and kainate subtypeglutamate receptors generated by polyamine-mediated ion channel block. Neuron 15,453–462 (1995).

14. Koh, D.S., Burnashev, N. & Jonas, P. Block of native Ca2+-permeable AMPA receptors inrat brain by intracellular polyamines generates double rectification. J. Physiol. (Lond.)486, 305–312 (1995).

15. Rozov, A. & Burnashev, N. Polyamine-dependent facilitation of postsynaptic AMPAreceptors counteracts paired-pulse depression. Nature 401, 594–598 (1999).

16. Greger, I.H., Khatri, L. & Ziff, E.B. RNA editing at Arg607 controls AMPA receptor exitfrom the endoplasmic reticulum. Neuron 34, 759–772 (2002).

17. Jia, Z. et al. Enhanced LTP in mice deficient in the AMPA receptor GluR2. Neuron 17,945–956 (1996).

18. Feldmeyer, D. et al. Neurological dysfunctions in mice expressing different levels of theQ/R site-unedited AMPAR subunit GluR-B. Nat. Neurosci. 2, 57–64 (1999).

19. Kawahara, Y. et al. Glutamate receptors: RNA editing and death of motor neurons.Nature 427, 801 (2004).

20. Colbourne, F., Grooms, S.Y., Suzanne Zukin, R.S.Z., Alastair, M., Buchan, A.M. &Bennett, M.V.L. Hypothermia rescues hippocampal CA1 neurons and attenuates down-regulation of the AMPA receptor GluR2 subunit after forebrain ischemia. Proc. Natl.Acad. Sci. USA 100, 2906–2910 (2003).

21. Liu, S. et al. Expression of Ca2+-permeable AMPA receptor channels primes cell death intransient forebrain ischemia. Neuron 43, 43–55 (2004).

22. Fuchs, E.C. et al. Genetically altered AMPA-type glutamate receptor kinetics in inter-neurons disrupt long-range synchrony of gamma oscillation. Proc. Natl. Acad. Sci. USA98, 3571–3576 (2001).

23. Toth, K. & McBain, C.J. Afferent-specific innervation of two distinct AMPA receptorsubtypes on single hippocampal interneurons. Nat. Neurosci. 1, 572–578 (1998).

24. Rubio, M.E. & Wenthold, R.J. Glutamate receptors are selectively targeted to post-synaptic sites in neurons. Neuron 18, 939–950 (1997).

25. Gardner, S.M., Trussell, L.O. & Oertel, D. Time course and permeation of synaptic AMPAreceptors in cochlear nuclear neurons correlate with input. J. Neurosci. 19, 8721–8729(1999).

26. Ogoshi, F. & Weiss, J.H. Heterogeneity of Ca2+-permeable AMPA/kainate channelexpression in hippocampal pyramidal neurons: fluorescence imaging and immunocyto-chemical assessment. J. Neurosci. 23, 10521–10530 (2003).

27. Keinanen, K. et al. A family of AMPA-selective glutamate receptors. Science 249, 556–560 (1990).

28. Petralia, R.S., Wang, Y.X., Mayat, E. & Wenthold, R.J. Glutamate receptor subunit2-selective antibody shows a differential distribution of calcium-impermeable AMPAreceptors among populations of neurons. J. Comp. Neurol. 385, 456–476 (1997).

29. Liu, S.J. & Cull-Candy, S.G. Activity-dependent change in AMPA receptor properties incerebellar stellate cells. J. Neurosci. 22, 3881–3889 (2002).

30. Srivastava, S. et al. Novel anchorage of GluR2/3 to the postsynaptic density by the AMPAreceptor-binding protein ABP. Neuron 21, 581–591 (1998).

31. Xia, J., Chung, H.J., Wihler, C., Huganir, R.L. & Linden, D.J. Cerebellar long-termdepression requires PKC-regulated interactions between GluR2/3 and PDZ domain-containing proteins. Neuron 28, 499–510 (2000).

32. Matsuda, S., Mikawa, S. & Hirai, H. Phosphorylation of serine-880 in GluR2 by proteinkinase C prevents its C terminus from binding with glutamate receptor-interactingprotein. J. Neurochem. 73, 1765–1768 (1999).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

33. Chung, H.J., Xia, J., Scannevin, R.H., Zhang, X. & Huganir, R.L. Phosphorylation of theAMPA receptor subunit GluR2 differentially regulates its interaction with PDZ domain-containing proteins. J. Neurosci. 20, 7258–7267 (2000).

34. Perez, J.L. et al. PICK1 targets activated protein kinase Calpha to AMPA receptorclusters in spines of hippocampal neurons and reduces surface levels of the AMPA-typeglutamate receptor subunit 2. J. Neurosci. 21, 5417–5428 (2001).

35. Hanley, J.G., Khatri, L., Hanson, P.I. & Ziff, E.B. NSF ATPase anda-/b-SNAPs disassemble the AMPA receptor-PICK1 complex. Neuron 34, 53–67(2002).

36. Lee, S.H., Liu, L., Wang, Y.T. & Sheng, M. Clathrin adaptor AP-2 and NSF interact withoverlapping sites of GluR2 and play distinct roles in AMPA receptor trafficking andhippocampal LTD. Neuron 36, 661–674 (2002).

37. Luthi, A. et al. Hippocampal LTD expression involves a pool of AMPARs regulated by theNSF-GluR2 interaction. Neuron 24, 389–399 (1999).

38. Nishimune, A. et al. NSF binding to GluR2 regulates synaptic transmission. Neuron 21,87–97 (1998).

39. Song, I. et al. Interaction of the N-ethylmaleimide-sensitive factor with AMPA receptors.Neuron 21, 393–400 (1998).

40. Noel, J. et al. Surface expression of AMPA receptors in hippocampal neurons is regulatedby an NSF-dependent mechanism. Neuron 23, 365–376 (1999).

41. Luscher, C. et al. Role of AMPA receptor cycling in synaptic transmission and plasticity.Neuron 24, 649–658 (1999).

42. Dong, H., O’Brien,, R.J., Fung, E.T., Lanahan, A.A., Worley, P.F. & Huganir, R.L. GRIP: asynaptic PDZ domain-containing protein that interacts with AMPA receptors. Nature386, 279–284 (1997).

43. Osten, P. et al. Mutagenesis reveals a role for ABP/GRIP binding to GluR2 in synapticsurface accumulation of the AMPA receptor. Neuron 27, 313–325 (2000).

44. Osten, P. et al. The AMPA receptor GluR2 C terminus can mediate a reversible, ATP-dependent interaction with NSF and a- and b-SNAPs. Neuron 21, 99–110 (1998).

45. Duprat, F., Daw, M., Lim, W., Collingridge, G. & Isaac, J. GluR2 protein-proteininteractions and the regulation of AMPA receptors during synaptic plasticity. Phil.Trans. R. Soc. Lond. B 358, 715–720 (2003).

46. Daw, M.I. et al. PDZ proteins interacting with C-terminal GluR2/3 are involved in a PKC-dependent regulation of AMPA receptors at hippocampal synapses. Neuron 28, 873–886 (2000).

47. Tardin, C., Cognet, L., Bats, C., Lounis, B. & Choquet, D. Direct imaging of lateralmovements of AMPA receptors inside synapses. EMBO J. 22, 4656–4665 (2003).

48. Linden, D.J. The expression of cerebellar LTD in culture is not associated with changes inAMPA-receptor kinetics, agonist affinity, or unitary conductance. Proc. Natl. Acad. Sci.USA 98, 14066–14071 (2001).

49. Terashima, A. et al. Regulation of synaptic strength and AMPA receptor subunitcomposition by PICK1. J. Neurosci. 24, 5381–5390 (2004).

50. Gardner, S.M. et al. Calcium-permeable AMPA receptor plasticity is mediatedby subunit-specific interactions with PICK1 and NSF. Neuron 45, 903–915 (2005).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Endocannabinoid signaling depends on the spatialpattern of synapse activation

Paıkan Marcaggi & David Attwell

The brain’s endocannabinoid retrograde messenger system decreases presynaptic transmitter release, but its physiological

function is uncertain. We show that endocannabinoid signaling is absent when spatially dispersed synapses are activated on

rodent cerebellar Purkinje cells but that it reduces presynaptic glutamate release when nearby synapses are active. This switching

of signaling according to the spatial pattern of activity is controlled by postsynaptic type I metabotropic glutamate receptors,

which are activated disproportionately when glutamate spillover between synapses produces synaptic crosstalk. When spatially

distributed synapses are activated, endocannabinoid inhibition of transmitter release can be rescued by inhibiting glutamate

uptake to increase glutamate spillover. Endocannabinoid signaling initiated by type I metabotropic glutamate receptors is

a homeostatic mechanism that detects synaptic crosstalk and downregulates glutamate release in order to promote

synaptic independence.

Plasticity of glutamatergic granule cell–Purkinje cell synapses is impor-tant for cerebellar learning1,2. Independent operation of these synapseswould allow information to be stored in individual synapse strengths,thereby increasing the number of motor programs that the cerebellumcan store1–3. However, the high synapse density in the cerebellum cancause crosstalk through glutamate spillover between synapses4. Whenglutamate diffuses out of a synapse, its effect on surrounding receptorsis expected to be larger when glutamate is also released at nearbysynapses because of non-linear summation of glutamate’s effects, eitherthrough the receptors’ dose-response curve or through local saturationof glutamate uptake. As each Purkinje cell dendritic tree receivesB100,000 granule cell inputs, of which only 50 need to be active toelicit an action potential5, physiological activity should involvesynapses that are relatively scattered and hence isolated from eachother. Despite that prediction, most studies of granule cell–Purkinje cellsynaptic transmission have been performed, for convenience, bystimulating the parallel fibers (the granule cell axons which run throughthe molecular layer in a parallel geometry) so that activated synapsesare adjacent.

We investigated the influence of the spatial pattern of synapticactivation on endocannabinoid-mediated plasticity6,7 of these synapsesby whole-cell clamping Purkinje cells in rodent cerebellar slices, asdescribed previously4, and comparing the effects of stimulation atdifferent locations (Fig. 1a). Stimulating either the granular layer orthe parallel fibers in the molecular layer activates the parallel fibersynapses onto Purkinje cells8, but because the stimulated ascendinggranule cell axons rise to different levels in the molecular layer (Fig. 1a),granular layer stimulation should activate synapses that are morespatially dispersed than occurs with molecular layer stimulation.Indeed, although the synapses recruited by granular layer stimulation

generate a fast AMPA receptor–mediated excitatory postsynapticcurrent (EPSCfast) similar to that evoked by molecular layer stimula-tion, they interact less because they are more isolated from eachother4. We show here that endocannabinoid signaling is absent whenspatially dispersed synapses are activated but that it reduces presynapticglutamate release when glutamate spillover occurs between nearbyactive synapses.

RESULTS

Endocannabinoid signaling depends on stimulation site

Tetanic stimulation of the parallel fibers activates type I meta-botropic glutamate receptors (mGluR1; refs. 9,10) and depressessynapses onto Purkinje cells by releasing endocannabinoids11–14. How-ever, whereas when parallel fibers were stimulated, tetani reliablydepressed the fast AMPA receptor–mediated excitatory postsynapticpotential (EPSPfast, initial amplitude 3.4 7 0.5 mV; n ¼ 8) for about20 s, and this was blocked by the CB1 receptor blocker AM251, incontrast, for granular layer stimulation evoking a similar EPSPfast

(3.9 7 0.8 mV), tetani reliably potentiated the EPSPfast (n ¼ 11),and this was unaffected by cannabinoid receptor block (Fig. 1b–e).Even when we paired granular layer stimulation with climbingfiber stimulation, which potentiates endocannabinoid-mediateddepression produced by parallel fiber stimulation15, we did not seedepression of the EPSC (Fig. 1e). Thus, when spatially isolatedparallel fiber synapses are activated, no significant endocannabinoidsignaling occurs.

Larger mGluR1 activation for parallel fiber stimulation

What causes the different effects of stimulating spatially adjacent versusspatially dispersed synapses? This was not the result of a different

Published online 1 May 2005; corrected 15 May 2005 (details online); doi:10.1038/nn1458

Department of Physiology, University College London, Gower Street, London, WC1E 6BT, UK. Correspondence should be addressed to P.M. ([email protected]).

776 VOLUME 8 [ NUMBER 6 [ JUNE 2005 NATURE NEUROSCIENCE

ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

density or effectiveness of cannabinoid or metabotropic receptors atthe synapses activated from the different stimulation sites, as thecannabinoid agonist WIN 55,212-2 and the mGluR1 agonist DHPGreduced EPSCs evoked by molecular layer and granular layer stimula-tion by the same amount (Fig. 1f–i). Rather, it resulted from a differentdegree of activation of mGluR1. For an EPSCfast of B600 pA at �70mV (B100 synapses active; see Methods), the slow mGluR1-mediatedEPSC (EPSCslow)10 evoked by granular layer stimulation was muchsmaller than that evoked in the same cell by molecular layer stimulationof a similar number4 of synapses (Fig. 2a,b), both at 27 1C in mouse(granular layer–evoked response was 15.4 7 4.6% of the size ofthe molecular layer–evoked response in the same cell, n ¼ 7, P ¼1.7 � 10�6) and at 35 1C in rat (granular layer–evoked response was5.5 7 1.7% of the molecular layer–evoked response, n ¼ 7,P ¼ 2.3 � 10�9).

Synaptic crosstalk activates mGluR1

To investigate why mGluR1 activation is larger for activation of spatiallyadjacent synapses, we analyzed how the mGluR1-mediated EPSCslow

varied as a function of the number of synapses activated by molecularlayer stimulation. The size of the fast AMPA receptor–mediated EPSC(EPSCfast) in response to a single stimulus (Fig. 2c) was used to assessthe number of synapses activated by a low or a high stimulationintensity4. The ratio of the variance to the mean of the EPSCfast

amplitude was independent of stimulus intensity (Fig. 2c, inset),consistent with the EPSCfast amplitude being proportional to thenumber of activated synapses4. NBQX (25 mM) was then applied toblock AMPA receptors, and trains of ten stimuli at 200 Hz (mimickinggranule cell activity in vivo16) at the same two intensities were used toevoke an EPSCslow (Fig. 2d). If all the synapses behaved independently,

the EPSCslow amplitude would be proportional to the number ofactivated synapses and hence to the EPSCfast. In fact, when the numberof activated synapses (and the size of EPSCfast) was increased, theEPSCslow amplitude increased disproportionately (Fig. 2c–e), and thedependence of EPSCslow amplitude on EPSCfast amplitude had anexponent of n ¼ 1.4 (Fig. 2f). Thus, synapses do not behave indepen-dently, and part of the EPSCslow is due to crosstalk between synapses.

We assessed quantitatively the importance of synaptic crosstalk foractivating mGluR1. We estimate the mean EPSCfast amplitude pro-duced by a single activated synapse to be B6 pA (see Methods). Wededuce from Figure 2f, therefore, that a single synapse activated bythe ten-stimulus train used here would produce a mean EPSCslow ofB34 fA. From this, we plotted a linear extrapolation (Fig. 2f,dotted line) predicting the EPSCslow amplitude that would occur ifsynapses did not interact. The low- and high-stimulus intensitiesused evoked an EPSCslow that was 4.7- and 8.3-fold higher thanthis prediction, showing that for molecular layer stimulation themajority of the EPSCslow is due to crosstalk. This provides an explana-tion for the much smaller EPSCslow seen when spatially dispersedsynapses are activated by granular layer stimulation (when crosstalkis minimal) than when parallel fibers are stimulated and crosstalkoccurs (Fig. 2a,b).

mGluR1 detects glutamate spillover

How does mGluR1 detect synaptic crosstalk? Crosstalk betweensynapses could occur intracellularly or extracellularly. The EPSCslow

may reflect direct activation of TRPC1 channels by mGluR1 (ref. 17),but molecular layer stimulation also induces an mGluR1-mediatedpostsynaptic calcium increase, which may increase the EPSCslow

10,18,19.We examined whether summation of internal calcium concentration

–2 0 –2 0

b

10 mV 10 mV

+2 s

–2 s

–2 s

+2 s

–10 10 20 300.0

0.5

1.0

1.5

c

Ctr

Time (s)–10 0 10 20 30

d e

Time (s)

ML stim GL stim

ML stim GL stima

WIN55,212-2

Ctr

WIN55,212-2

Ctr

f gML stim GL stim

200 pA

10 ms

200 pA

10 ms2 mV

40 ms2 mV

40 ms

Ctr200 pA

10 ms

DHPG

Ctr200 pA

10 ms

DHPGh iML stim GL stim

Ctr + CF stim

CtrAM251 2 µM

ML stim

GL stim

2 2

0Time (s)

Nor

mal

ized

EP

SP

0.0

0.5

1.0

1.5

Nor

mal

ized

EP

SPAM251 2 µM

Time (s)

Figure 1 Endocannabinoid-mediated plasticity seen for activation of nearby synapses is absent when stimulating spatially dispersed synapses.

(a) Stimulation sites in molecular layer (ML) and granular layer (GL). (b,c) Effect of a 10-pulse train at 200 Hz on EPSPfast evoked at �70 mV in the

same rat Purkinje cell by one stimulus, using ML (b) and GL (c) stimulation. Main traces: response to the train between EPSPfast responses to single

stimuli. Insets: EPSPs 2 s before and after train. (d) Suppression of EPSPfast after train (control, Ctr) for ML stimulation in eight cells, and its block by

the CB1 antagonist AM251 (five cells). (e) For GL stimulation, the train evokes a small potentiation (11 cells, filled circles) which is unaffected by AM251

(five cells). Pairing with mock-physiological climbing fiber (CF) stimulation15 (three pulses 100, 200 and 300 ms after the start of GL stimulation)

does not affect this (four cells). (f,g) Effect of the cannabinoid agonist WIN 55,212-2 (5 mM) on the EPSCfast at �70 mV evoked by ML (f) or GL

stimulation (g). The EPSCfast was reduced to 15.3 7 1.4% (s.e.m., four cells) and 19.0 7 2.3% (four cells) of control, respectively (not significantly

different, P ¼ 0.22). (h,i) Effect of the mGluR1 agonist DHPG (50 mM) on the EPSCfast at �70 mV evoked by ML (h) or GL stimulation (i). The EPSCfast

was reduced to 52.6 7 5.8% and 59.8 7 4.2% (ten cells) of control, respectively (P ¼ 0.29). DHPG suppression measured after 1 min in DHPG.

Temperature: 33 1C.

NATURE NEUROSCIENCE VOLUME 8 [ NUMBER 6 [ JUNE 2005 777

ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

([Ca2+]i) changes produced by adjacent synapses might mediate theEPSCslow potentiation produced by synaptic crosstalk. Increasing theintracellular EGTA concentration 20-fold (while maintaining free[Ca2+]i constant) should reduce mGluR1-evoked [Ca2+]i changesand thus reduce [Ca2+]i-mediated EPSCslow potentiation. However,after 30 min in whole-cell configuration, the EPSCslow when recordingwith 10 mM EGTA (148 7 24 pA, n ¼ 13) was similar (P ¼ 0.27) tothat seen using 0.5 mM intracellular EGTA (111 7 21 pA, n ¼ 12).Furthermore, the EPSCslow amplitude still increased non-linearly withEPSCfast amplitude (Fig. 2g). As previously reported10, thefaster Ca2+ buffer BAPTA reduced the EPSCslow amplitude by 64%(to 40 7 25 pA, P¼ 0.05, n¼ 8; see also ref. 20) after 30 min in whole-cell configuration. However, the EPSCslow amplitude still increasednon-linearly compared with EPSCfast amplitude (Fig. 2h). We concludethat summation of [Ca2+]i changes does not mediate the crosstalkdetected by mGluR1.

Next, we considered extracellular mediation of synaptic crosstalk.Two lines of evidence suggest that the detection of synaptic crosstalk bymGluR1 is due to glutamate spillover between synapses. First, thecharge transfer of the AMPA-mediated EPSCfast evoked by the stimulustrains used to evoke the EPSCslow also showed a supralinear depen-dence on the number of activated synapses (Fig. 3a,b), consistent withincreased spillover at higher stimulus intensities potentiating theglutamate receptor activation produced by the trains. Second, gluta-mate spillover is enhanced4 by suppression of the glial glutamatetransporters GLAST and GLT-1, and we found that the EPSCslow wasstrongly potentiated by this maneuver. Knockout of GLAST

does not affect the magnitude of the EPSCfast4, but we found that it

potentiated the EPSCslow 2.2-fold (from 75 7 27 pA in nine wild-typecells to 165 7 26 pA in six knockout cells, P ¼ 0.03; Fig. 3c) and thatfewer stimuli were needed to elicit a detectable EPSCslow (Fig. 3d).These differences were not due to a higher expression of mGluR1 or itsintracellular signaling, as t-ACPD, an mGluR1 agonist, evoked acurrent that was not significantly different (P ¼ 0.28) in Purkinjecells from knockout and wild-type mice (Fig. 3c, inset). Similarly,inhibiting GLT-1 with dihydrokainate in mice lacking GLAST potenti-ates the EPSCfast by only 10% (ref. 4), but here we found that itproduced a further 67 7 9% potentiation of the EPSCslow (n ¼ 3,P ¼ 0.017; Fig. 3c). Thus, the EPSCslow is strongly increased (3.7-foldfor size and 4.7-fold for charge transfer) when glutamate spillover isenhanced by suppression of glial glutamate transporters.

Promoting crosstalk rescues cannabinoid signaling

If mGluR1 activation and endocannabinoid signaling depend onglutamate spillover producing crosstalk between synapses, then pro-moting crosstalk would be expected to rescue the EPSCslow andendocannabinoid signaling that are normally greatly reduced whenspatially dispersed synapses are activated with granular layer stimula-tion. Consistent with this, although the EPSCslow was nearly undetect-able when spatially isolated synapses were activated by tetanic granularlayer stimulation (EPSCslow amplitude B10 pA in rat at 35 1C, Figs. 2band 4a), it was restored (Fig. 4b) when glutamate spillover wasenhanced by partially inhibiting glutamate transporters4. In the pre-sence of TBOA (200 mM), granular layer stimulation evoked a large

EPSCslowa EPSCfast b Mouse, 27 °C Rat, 35 °C

0

50

100

150

200

ML GL

EP

SC

slow

am

pl (

pA)

0

10

20

30

40

ML GL

P = 0.01 P = 0.02

20 pA

200 ms

GL

ML

GL

ML

10 ms

200 pA

EPSCfast EPSCslow

400 pA

500 ms

10 pA

Low stim

High stim

10 low stim at 200 Hz

10 high stim at 200 Hz

c d

5 ms

gf

e

Fra

ctio

nal i

ncre

ase

0

2

4

6

8

10

0

2

4

6

8

10

Fra

ctio

nal i

ncre

ase

Fra

ctio

nal i

ncre

ase

0

2

4

6

8

10

P = 0.03(n = 8)

P = 0.003(n = 11)

P = 0.002(n = 11)

EGTA 10 mM, Ca2+ 2 mM BAPTA 13 mM, Ca2+ 1 mMh

EPSC slow

EPSC fast

EPSC slow

EPSC fast

EPSC slow

EPSC fast

Var/meanof EPSCfast

(pA)0

10 P = 0.85(n = 18)

EPSCfast amplitude (pA)0 300 600 900 1,200

EP

SC

slow

am

plitu

de (

pA)

0

10

20

30

40

50

60

Low Highstim stim

EPSCslow = 0.0028 × EPSCfast1.4

Figure 2 Absence of endocannabinoid signalling

when activating spatially dispersed synapses

results from mGluR1 activation depending on

synaptic crosstalk. (a) Train-evoked EPSCslow (in

NBQX, 27 1C, �70 mV) for the same EPSCfast

amplitude evoked by interleaved ML and GL

stimulation in mouse. (b) Mean EPSCslow

amplitude for seven cells as in a (EPSCfast

amplitude 596 7 36 pA for ML and 572 7 44

pA for GL stimulation; not significantly different,

P ¼ 0.6) and for seven rat cells (35 1C, EPSCfast

amplitude 700 7 114 pA for ML and 667 7 67pA for GL stimulation, P ¼ 0.73). (c) AMPA

receptor–mediated EPSCfast at �70 mV in mouse

Purkinje cell evoked by two intensities of ML

stimulus (27 1C; EPSC amplitudes are in f). Inset:

variance/mean of the EPSCfast amplitude is

independent of stimulus size. (d) In same cell,

with same stimuli as in c, the EPSCslow evoked (in

25 mM NBQX, �70 mV) by ten stimuli at 200 Hz

increases more with stimulus strength than does

the EPSCfast. (e) Fractional change of EPSCslow

and EPSCfast. (f) Dependence of EPSCslow on

EPSCfast amplitude (proportional to number

of fibers stimulated). Dotted line predicts

relationship if synapses did not interact.

(g,h) Fractional increase of EPSCfast and EPSCslow

with Ca2+ buffering increased with EGTA

(g, 27 1C) or BAPTA (h, 351C), with stimulus

adjusted to increase EPSCfast from 310 7 20 to1,311 7 50 pA (g) or from 323 7 56 to 1,164

7 112 pA (h was in 0.2 mM NBQX to improve

voltage uniformity). All data except b from mouse.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

EPSCslow (1,6467 342 pA, n¼ 3) comparable in size to the one evokedby molecular layer stimulation (Fig. 4b), and the post-tetanic potentia-tion seen normally when stimulating the granular layer (Fig. 1e) wasconverted into a cannabinoid-mediated post-tetanic depression in allfive cells studied (Fig. 4c).

Suppressing crosstalk prevents mGluR1 activation

Conversely, if mGluR1-mediated cannabinoid signaling is largelyactivated by synaptic crosstalk between adjacent activated synapses,

and this signaling suppresses glutamate release sufficiently to preventsynaptic crosstalk, then subsequent activation of adjacent synapsesshould not activate mGluR1 significantly. To test this prediction,we activated about 300 adjacent synapses with molecular layer stimula-tion (four stimuli at 200 Hz) and recorded the resulting AMPAreceptor–mediated EPSCfast and mGluR1-mediated EPSCslow

(Fig. 4d, �10 s), before switching to current clamp mode andstimulating the synapses with a train of ten pulses at 200 Hz to activateendocannabinoid signaling. Ten seconds later (Fig. 4d, +10 s), the

Figure 4 Endocannabinoid-mediated plasticity

is rescued for stimulation of spatially dispersed

synapses by enhancing glutamate spillover,

and inhibits glutamate release sufficiently

to abolish crosstalk activation of mGluR1.

(a) Control response to ML and GL stimulus

trains in NBQX (�70 mV, EPSCfast amplitude

497 7 126 pA for ML and 560 7 133 pA for

GL, P ¼ 0.48; TBOA does not affect EPSCfast

amplitude4). (b) Response to same stimuli

with glutamate spillover enhanced by

blocking glutamate transporters with TBOA,

and with CPCCOEt present to block mGluR1.

(c) EPSPfast amplitude after a train for GL

stimulation in the absence (Ctr) and presence

of 200 mM TBOA: enhancing spillover rescues

the endocannabinoid-evoked depression

(five cells). AM251 (2 mM) blocked thedepression rescued in TBOA (eight cells).

(d) EPSCfast and EPSCslow (in 0.2 mM NBQX,

to improve voltage uniformity, which reduced

EPSCfast amplitude by 65 7 4%; five cells)

evoked by four stimuli (200 Hz) to the parallel

fibers, 10 s before (�10 s) and after (+10 s) a

ten-stimulus train (200 Hz) in current-clamp

mode. (e) After the train in d, suppression of

glutamate release by cannabinoid signaling reduced the first EPSCfast of a four-pulse train by 44 7 12% and reduced the peak EPSCfast produced by the

train by 36 7 6%, but it reduced the EPSCslow by 93 7 4% (four cells). (f,g) In the presence of AM251, there is little change in the EPSCfast or EPSCslow

(ten cells). All in rat, 33 1C.

a

Minimal number of pulses (at 200 Hz)to elicit a detectable EPSCslow

1 2 3 4 5 6 7

Per

cent

age

of c

ells

0

100d

c EPSCslow

GLAST–/–

GLAST –/–in DHK 200 µM

WT

GLAST –/–

WT

EPSCfast

1 nA

50 ms

Low stim

High stim

Amplitude of 1st EPSCfast of the train (pA)

0 300 600 900 1,200 EP

SC

fast

cha

rge

tran

sfer

follo

win

g tr

ain

(pC

)

0

40

80

120

160y = 0.0067*x1.42b

10 ms

200 pA

0.5 s

100 pA

WT

0.5 s

100 pA

2, 3

510 35

10

2

GLAST –/–

I AC

PD

(pA

)

I AM

PA

(nA

)

0

100

200

300

WT –/

–W

T–/–

0

1

2

3

50 ms 1

Figure 3 Glutamate spillover mediates synaptic crosstalk detection by

mGluR1. (a) AMPA receptor–mediated currents in response to parallel fiber

stimulus trains at two intensities (mouse Purkinje cell at 27 1C). Inset:

traces normalized to the amplitude of the first EPSCfast of the train,

demonstrating that post-train charge entry (shaded: gray, low stimulation;

black, high stimulation) increases more than the number of fibers

stimulated. (b) Dependence of charge transfer after the train on the

amplitude of the first EPSC (proportional to number of fibers stimulated).

Dotted line extrapolates relation calculated as if synapses did not interact

(as in Fig. 2f). (c) Effect of deleting GLAST (�/�) and of blocking GLT-1 with dihydrokainate (DHK) on the EPSCslow evoked in the presence of NBQX by trains

of parallel fiber stimuli that produce a similar amplitude EPSCfast (633 7 48 pA in nine wild-type cells; 617 7 36 pA in six knockout cells; P ¼ 0.8). WT,

wild-type control. Insets: the response of Purkinje cells at –70 mV to 0.5 mM AMPA (13 cells each) and 25 mM ACPD (15 cells each) is similar in wild-type and

knockout cells. (ACPD also activates group II mGluRs, but activating group II mGluRs generates no current in Purkinje cells25.) (d) Left: EPSCslow response (in

NBQX) of wild-type and GLAST knockout Purkinje cells to trains of two, three, five or ten stimuli applied to the parallel fibers at 200 Hz. Right: percentage ofnine wild-type and ten knockout cells that showed a detectable EPSCslow after different numbers of stimuli. Black bars, wild-type; white bars, knockout.

Time (s)0 10 20 30

Nor

mal

ized

EP

SP

0.0

0.5

1.0

1.5

cba

d

f

GL stim

TBOA

100 pA

0.5 s

GL stim

ML stim

500 pA

0.5 sGL stimTBOA 200 µM

ML stimTBOA 200 µM

ML stim TBOA + CPCCOEt 200 µMGL stim TBOA + CPCCOEt 200 µM

CtrTBOA + AM251

t = –10 s t = +10 s

t = –10 s t = +10 s

0.5 s

1 nA

0.5 s

1 nA

AM251 2 µM

0.00.20.40.60.81.0

Fra

ctio

nal c

hang

eF

ract

iona

l cha

nge

0.0

0.5

1.0

1.5

e

g AM251 2 µM

First Peak

EPSCfast

EPSCslow

EPSCfastEPSCfast

EPSCslow

First PeakEPSCfastEPSCfast

EPSCslow

–10


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

EPSCfast recorded in voltage-clamp mode was reduced, but theEPSCslow was almost abolished (Fig. 4e), and this abolition was blockedby AM251 (Fig. 4f,g). Thus, the mGluR1-cannabinoid signalingactivated by synaptic crosstalk does indeed reduce glutamate releasesufficiently to prevent crosstalk.

DISCUSSION

We have shown that endocannabinoid signaling occurs only whennearby synapses are activated. This dependence on the spatial pattern ofsynapse activity is a result of mGluR1 activation being stronglydependent on crosstalk between synapses: when active synapses arespatially isolated, little activation of mGluR1 is produced (Fig. 2),and hence little retrograde endocannabinoid signaling occurs(Fig. 1). However, mGluR1 is strongly activated in conditions enhan-cing crosstalk between synapses, such as excessive glutamate releasefrom adjacent synapses (occurring physiologically, or during experi-mental stimulation of the molecular layer) or when glutamatespillover is artificially increased by blocking glutamate uptake4,21.Synaptic crosstalk detection by mGluR1 results in a release ofendocannabinoids, which reduce presynaptic glutamate release for20 s, thus reducing synaptic crosstalk. This adjustment of glutamaterelease may not alter the overall level of excitation of the Purkinje cell,as retrograde cannabinoid signaling also suppresses GABA release22,23.We conclude that mGluR1-mediated crosstalk detection is a homeo-static mechanism promoting synaptic independence. It will be ofinterest to determine whether conditions can be found that generatesignificant mGluR1 activation and endocannabinoid signaling withoutadjacent synapses needing to release sufficient glutamate to generatesynaptic crosstalk.

It has recently been reported that whereas activating the climbingfiber input to Purkinje cells while stimulating the parallel fibersleads to long term depression (LTD) of the parallel fiber EPSC througha process dependent on mGluR1 activation2, no LTD is seen whenthe climbing fiber is stimulated while activating the granule cellaxons by stimulating in the granular layer24. The lack of LTD wasattributed to the preferential activation of synapses on the ascendinggranule cell axon, but the lack of LTD and cannabinoid signalingseen for granular layer stimulation is unlikely to reflect a lack ofmGluR1 at ascending synapses, as we have found that the mGluR1agonist DHPG depresses the synapses activated from the granularlayer by the same fraction as it depresses the synapses activated bystimulating the parallel fibers (Fig. 1h,i). Additionally, promotingglutamate spillover by blocking glutamate uptake rescues themGluR1-mediated EPSCslow and cannabinoid signaling when thegranular layer is stimulated (Fig. 4a–c). Our data suggest that thelack of LTD for granular layer stimulation may reflect the very limitedactivation of mGluR1 that is produced by this stimulation (Fig. 2a,b)owing to the lack of synaptic crosstalk when spatially dispersedsynapses are activated.

METHODSPreparation. Purkinje cells were whole-cell clamped in parasagittal cerebellar

slices from postnatal day 18 (P18) rats or P14–21 mice4 at 27 1C or 33–35 1C

(see text). Animal use was in accord with the UK Animals (Scientific

Procedures) Act (1986).

Solutions. External solution contained (in mM) NaCl 124, KCl 2.5, NaH2PO4

1, NaHCO3 26, CaCl2 3, MgCl2 1, glucose 10, GABAzine 0.01 (to block GABAA

receptors), bubbled with 95% O2/5% CO2. Pipette solution contained (in mM)

cesium gluconate 140, NaCl 4, HEPES 10, MgATP 4, Na3GTP 0.5, Cs2EGTA

0.5, CaCl2 0.1, pH adjusted to 7.3 with CsOH. When altering the Ca2+ buffer to

10 mM EGTA or 13mM BAPTA, the CaCl2 added was adjusted to maintain the

free calcium concentration ([Ca2+]i) at 30 nM (calculated with the program

MaxChelate), and the concentration of cesium gluconate was decreased to

maintain osmolarity. For current-clamp experiments (Figs. 1b–e, 4c–g), K+

replaced Cs+ in the pipette.

Electrophysiology. For EPSP and EPSCslow measurements, the series resistance

was not compensated, but it was o5 MO when we studied the effect of

different Ca2+ buffering power to ensure good cell dialysis. For EPSCfast

measurements, series resistance was compensated to o1 MO. Stimuli were

applied from a glass pipette placed in the slice below the Purkinje cell, either in

the molecular layer to activate parallel fibers directly or in the granular layer

(which will also activate a beam of parallel fiber activity8); stimulus trains were

applied every minute. The mean EPSCfast amplitude produced by a single

activated synapse (p� i1) at �70 mV was estimated as B6 pA from the ratio of

variance to mean measured for the EPSCfast amplitude with the internal

solution used here (Fig. 1a, inset) and the probability of release (p ¼ 0.48)

determined previously4, using variance/mean ¼ i1(1 � p), where i1 is the

current produced when a vesicle is released. For granular layer stimulation, the

rapid time to peak of the EPSCfast suggests that granule cells or ascending axons

were stimulated rather than mossy fibers, which would have led to a disynaptic,

delayed EPSCfast (as was seen when stimulating the white matter; data

not shown).

Statistics. Data are presented as mean 7 s.e.m. Statistical comparisons were by

Student’s t-test.

ACKNOWLEDGMENTSWe thank K. Tanaka for providing the knockout mice, and B. Barbour, A. Gibb,D. Rossi, A. Silver and M. Hamann for comments on the manuscript.Supported by the European Union, the Wellcome Trust and a Wolfson-RoyalSociety award.




1. Marr, D. A theory of cerebellar cortex. J. Physiol. (Lond.) 202, 437–470 (1969).2. Ito, M. Cerebellar long-term depression: characterization, signal transduction, and

functional roles. Physiol. Rev. 81, 1143–1195 (2001).3. Barbour, B. & Hausser, M. Intersynaptic diffusion of transmitter. Trends Neurosci. 20,

377–384 (1997).4. Marcaggi, P., Billups, D. & Attwell, D. The role of glial glutamate transporters in

maintaining the independent operation of juvenile mouse cerebellar parallel fibersynapses. J. Physiol. (Lond.) 552, 89–107 (2003).

5. Barbour, B. Synaptic currents evoked in Purkinje cells by stimulating individual granulecells. Neuron 11, 759–769 (1993).

6. Wilson, R.I. & Nicoll, R.A. Endocannabinoid signaling in the brain. Science 296,678–682 (2002).

7. Kreitzer, A.C. & Regehr, W.G. Retrograde signaling by endocannabinoids. Curr. Opin.Neurobiol. 12, 324–330 (2002).

8. Coutinho, V., Mutoh, H. & Knopfel, T. Functional topology of the mossy fiber-granule cell-Purkinje cell system revealed by imaging of intrinsic fluorescence in mouse cerebellum.Eur. J. Neurosci. 20, 740–748 (2004).

9. Batchelor, A.M. & Garthwaite, J. Novel synaptic potentials in cerebellar Purkinje cells:probable mediation by metabotropic glutamate receptors. Neuropharmacology 32,11–20 (1993).

10. Tempia, F., Miniaci, M.C., Anchisi, D. & Strata, P. Postsynaptic current mediated bymetabotropic glutamate receptors in cerebellar Purkinje cells. J. Neurophysiol. 80,520–528 (1998).

11. Brown, S.P., Brenowitz, S.D. & Regehr, W.G. Brief presynaptic bursts evoke synapse-specific retrograde inhibition mediated by endogenous cannabinoids. Nat. Neurosci. 6,1048–1057 (2003).

12. Levenes, C., Daniel, H. & Crepel, F. Retrograde modulation of transmitter release bypostsynaptic subtype 1 metabotropic glutamate receptors in the rat cerebellum.J. Physiol. (Lond.) 537, 125–140 (2001).

13. Kreitzer, A.C. & Regehr, W. Retrograde inhibition of presynaptic calcium influxby endogenous cannabinoids at excitatory synapses onto Purkinje cells. Neuron 29,717–727 (2001).

14. Maejima, T., Hashimoto, K., Yoshida, T., Alba, A. & Kano, M. Presynaptic inhibitioncaused by retrograde signal from metabotropic glutamate to cannabinoid receptors.Neuron 31, 463–475 (2001).

15. Brenowitz, S.D. & Regehr, W.G. Associative short-term synaptic plasticity mediated byendocannabinoids. Neuron 45, 419–431 (2005).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

16. Eccles, J.C., Llinas, R. & Sasaki, K. The mossy fiber-granule cell relay in thecerebellum and its inhibition by Golgi cells. Exp. Brain Res. 1, 82–101(1966).

17. Kim, S.J., Kim, Y.S., Yuan, J.P., Petralia, R.S., Worley, P.F. & Linden, D.J. Activation ofthe TRPC1 cation channel by metabotropic glutamate receptor mGluR1. Nature 426,285–291 (2003).

18. Takechi, H., Eilers, J. & Konnerth, A. A new class of synaptic response involving calciumrelease in dendritic spines. Nature 396, 757–760 (1998).

19. Batchelor, A.M. & Garthwaite, J. Frequency detection and temporally dispersed synapticsignal association through a metabotropic glutamate receptor pathway.Nature385, 74–77 (1997).

20. Kushmerick, C. et al. Retroinhibition of presynaptic Ca2+ currents by endocannabinoidsreleased via postsynaptic mGluR activation at a calyx synapse. J. Neurosci. 24,5955–5965 (2004).

21. Brasnjo, G. & Otis, T.S. Neuronal glutamate transporters control activation of post-synaptic metabotropic glutamate receptors and influence cerebellar long-term depres-sion. Neuron 31, 607–616 (2001).

22. Diana, M.A., Levenes, C., Mackie, K. & Marty, A. Short-term retrograde inhibition ofGABAergic synaptic currents in rat Purkinje cells is mediated by endogenous cannabi-noids. J. Neurosci. 22, 200–208 (2002).

23. Brenowitz, S.D. & Regehr, W.G. Calcium dependence of retrograde inhibition byendocannabinoids at synapses onto Purkinje cells. J. Neurosci. 23, 6373–6384 (2003).

24. Sims, R.E. & Hartell, N.A. Differences in transmission properties and suscepti-bility to long-term depression reveal functional specialization of ascending axon andparallel fiber synapses to Purkinje cells. J. Neurosci. 25, 3246–3257 (2005).

25. Neale, S.A., Garthwaite, J. & Batchelor, A.M. Metabotropic glutamate receptor subtypesmodulating neurotransmission at parallel fiber-Purkinje cell synapses in rat cerebellum.Neuropharmacology 41, 42–49 (2001).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Geometric and functional organization ofcortical circuits

Gordon M G Shepherd1,2,4, Armen Stepanyants2–4, Ingrid Bureau1,2, Dmitri Chklovskii2 & Karel Svoboda1,2

Can neuronal morphology predict functional synaptic circuits? In the rat barrel cortex, ‘barrels’ and ‘septa’ delineate an orderly

matrix of cortical columns. Using quantitative laser scanning photostimulation we measured the strength of excitatory projections

from layer 4 (L4) and L5A to L2/3 pyramidal cells in barrel- and septum-related columns. From morphological reconstructions

of excitatory neurons we computed the geometric circuit predicted by axodendritic overlap. Within most individual projections,

functional inputs were predicted by geometry and a single scale factor, the synaptic strength per potential synapse. This factor,

however, varied between projections and, in one case, even within a projection, up to 20-fold. Relationships between geometric

overlap and synaptic strength thus depend on the laminar and columnar locations of both the pre- and postsynaptic neurons, even

for neurons of the same type. A large plasticity potential appears to be incorporated into these circuits, allowing for functional

‘tuning’ with fixed axonal and dendritic arbor geometry.

The flow of excitation in cortical neural networks is largely determinedby stereotyped synaptic connections between populations of neurons1.Because functional projections require overlap of axons and dendrites,the shapes and locations of axonal and dendritic arbors provide animportant source of specificity1. The prevailing view holds thatfor excitatory neurons, this overlap directly defines synaptic circuits:where axons and dendrites are sufficiently close, synaptic connectionsoccur2–5. This anatomical approach has long been used to constructcortical wiring diagrams based on single cell reconstructions6–8, some-times in quantitatively explicit formulations4,9–11. These wiring dia-grams made on the basis of anatomy are assumed to representfunctional circuits4,6–8,11. However, it has proven difficult to testdirectly the relationships between geometric circuits (that is, structuralcircuits based on neuronal geometry) and functional circuits (that is,circuits assayed physiologically), due to the challenges of measuringboth in the same preparation.

Here, we investigated the relationship between neuronal geometryand functional synaptic connectivity in brain slices, exploiting theprecise laminar and columnar organization of the rat’s barrel cortex inconjunction with single-cell anatomical and functional analysis tools. L4barrels, which receive thalamocortical inputs carrying excitation fromindividual whiskers, can be visualized in both living and fixed slices12.Between the L4 barrels are the septa, associated with distinct thalamo-cortical13–15 and intracortical circuits16–18. Barrels and septa demarcatebarrel- and septum-related cortical columns spanning the vertical extentof cortex (Fig. 1a). The barrel grid thus allows one to align, average andcompare measurements from different brain slices and animals.

We used laser scanning photostimulation (LSPS) to measure thestrength of functional projections to individual neurons in L2/3. For

the same projections, we also reconstructed the axonal and dendriticarbor morphology of the excitatory neurons involved and usedcomputational geometry to measure the strength of the geometricprojections. This parallel approach revealed the structure-functionrelationships across multiple cortical projections, allowing us to testwhether neuronal morphology and the overlap of dendrites and axonspredicts functional circuits.

RESULTS

Functional projections to L2/3

To measure the strength of functional projections converging on L2/3pyramidal neurons, we needed to quantify the strengths of inputs frompopulations of excitatory neurons defined by laminar and columnarlocation in the barrel cortex. For this aim, LSPS by glutamate uncagingis an effective tool17,19–22. Presynaptic neurons at each stimulation sitein the slice are selectively excited close to their cell bodies (whileavoiding axons of passage), providing sub-laminar and sub-columnarresolution. Maps of synaptic input are rapidly generated by scanningthe beam to sample hundreds of sites while recording responses from asingle postsynaptic neuron. Other electrophysiological techniques arepoorly suited for our aims here. For example, pair recording (indivi-dually testing pairs of neurons for connections) is slow and inefficient,and it reveals the strength not of a projection, but of selectedpairwise connections within a projection; extracellular electrical stim-ulation is limited by low effective resolution, because axons of passageare stimulated.

We prepared slices of rat barrel cortex (4–5 weeks old) andrecorded functional input maps (spatial maps of excitatory synapticinput) for individual L2/3 pyramidal neurons using LSPS22 (Fig. 1).

Published online 8 May 2005; corrected 15 May 2005 (details online); doi:10.1038/nn1447

1Howard Hughes Medical Institute and 2Cold Spring Harbor Laboratory, Cold Spring Harbor, New York 11724, USA. 3Department of Physics and Center for InterdisciplinaryResearch on Complex Systems, Northeastern University, Boston, Massachusetts 02115, USA. 4These authors contributed equally to this work. Correspondence should beaddressed to K.S. ([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

We recorded from neurons in L2 and L3 above barrels and septa(Fig. 1a) while exciting clusters of neurons by photorelease ofglutamate in the focal spot of a UV laser beam on a 16 � 16pixel grid (Fig. 1b,d). The amplitudes of postsynaptic responsesindicate the strength of input to the recorded neuron from theregion of the brain slice excited by glutamate uncaging (see below).Control experiments established that LSPS input maps represent thesources of monosynaptic input with sub-laminar and sub-columnarresolution (B60 mm; see Methods; ref. 22).

At a particular spot on the grid (that is, a pixel at position x,y),photostimulation evokes a postsynaptic response (Qxy; Fig. 1c,d) that isproportional to the number of neurons stimulated, Ncell (equal to theproduct of the neuronal density, rcell, and the volume of excitedneurons, Vexc), the number of action potentials (APs) fired perstimulated neuron (SAP), and the average strength of the synapticconnection with the stimulated presynaptic neuron (qcon, defined hereas the postsynaptic charge per presynaptic neuron per action poten-tial21,22); that is,

Qxy ¼ ðrcellVexcSAPÞqcon ð1Þ

The term rcellVexcSAP gives the number of action potentials evokedin the excited presynaptic neurons and is therefore a measure ofpresynaptic excitation. Neuronal density, rcell, has previouslybeen determined23,24, and measurements of LSPS excitationprofiles (see Methods) provide Vexc and SAP. (Because Ncell ¼ rcellVexc,these measurements imply that photostimulation activated B54neurons in L4 and B34 in L5A.) Qxy is directly measured byLSPS. Therefore, LSPS, together with measurements of presy-naptic excitation, determines the strength of functional synapticprojections (qcon).

We grouped L2/3 pyramidal cells by laminar (L2 versus L3)and columnar (barrel- versus septum-related) location and averagedthe input maps (Qxy) within each group (Fig. 1e–h; see Methods).Maps were morphed to a standard barrel cortex template basedon the average dimensions of cytoarchitectonic landmarks (Fig. 1a,Supplementary Fig. 1). We restrict our attention to projectionsfrom L4 and L5A, the two main sources of translaminar input22.Neurons showed distinct spatial patterns of inputs depending ontheir particular location with respect to the columnar boundariesdefined by barrels and septa in L4. In barrel-related columns,cells in both L2 (Fig. 1e) and L3 (Fig. 1f) showed strong barrelinputs. In septum-related columns, the L5A inputs were muchstronger for L2 cells (Fig. 1g) than for L3 cells (Fig. 1h), despiteL2 cells being more distant targets; L4 inputs were weak (Fig. 1g,h). Inthe average map (Fig. 1g), but not always in the individual maps, thestrong focus of L5A input showed a small offset, extending fromdirectly below the septum anteromediad (rightward) under the neigh-boring barrel.

Responses from L5A clearly originate from L5A neurons, not fromactivation of L4 cells’ dendrites22. The density of L4 cells’ dendrites inL5A is extremely low, owing to the polarization of these dendritestowards barrel centers12, and excitation profiles (see Methods) showedthat photoexcitation of cells occurred only close to the soma and thuswithin the home layer. In addition, excitation profiles of L4 neuronslocated o50 mm from the L4/L5A border showed that the stimulationsites in L5A that were used for synaptic input mapping would not havecaused any spiking in L4 neurons22 (see Methods). Moreover, sites ofL5A input frequently occurred in isolation (that is, they did notconsistently abut sites with L4 inputs), and L5A inputs even exceededL4 inputs in the L5A-L2septum map (Fig. 1g), effectively excluding thepossibility that L4 dendrites generated the L5A signal.

L2/3

L4

L5A

e f g h

05

1015

05

1015

05

1015

05

1015

–30

0

Septum-relatedcolumn

Septum

Barrel

Pia

Barrel-relatedcolumn

x = 1.5 mm

y

c da b

–500Distance (µm)

0 500

Dis

tanc

e (µ

m)

–800

–600

–400

–200

0

200

400

600 L1

L2/3

L4

L5AL5B

L6

UV

z

2AP

0

L2/3

L4

L5A

Qxy

(pA

)

Qxy(pA)

L2barrel

cellsL2septum

cells L3septum

cellsL3barrel

cells

150

pA

Figure 1 LSPS maps of functional excitatory synaptic input (Qxy) to L2/3 neurons in different columnar positions. (a) Barrels and septa in L4 demarcate

columns. A barrel-related L2/3 pyramidal cell was reconstructed. (b) Blue dots (100 mm spacing) mark the LSPS mapping pattern. Average excitation profiles

of presynaptic cells in L4 and L5A are shown at right, indicating the resolution of photostimulation. (c) Input map for an L2/3 pyramidal neuron. Bar, 0.5 mm.

(d) During mapping, each UV flash stimulates action potentials in a small cluster of neurons (purple), some of which synaptically project to the postsynaptic

neuron (red). (e–h) Average input maps for L2/3 pyramidal cells grouped by columnar and laminar location. Plots below maps show horizontal profiles (100-mm

bins, mean 7 s.e.m.) of input from L4 (green line; data from region indicated by vertical green bar to left of map) and L5A (blue line). Barrels and laminar

boundaries drawn as dashed lines. Shown are maps for L2barrel cells (n ¼ 8, e), L3barrel cells (n ¼ 9, f); L2septum cells (n ¼ 8, g) and L3septum cells (n ¼ 7, h).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Thus, in barrel-related columns, excitatorysynaptic input arrives primarily through theL4-L2/3 projection, with some input fromL5A. There is little differentiation at thecircuit level between L2 and L3 pyramids. Inseptum-related columns, the functional orga-nization is different: the L5A-L2 projectionprovides the dominant source of input, andL3septum neurons receive only weak inputfrom L4 and L5A.

Geometric projections to L2/3

Functional projections require overlap ofdendritic and axonal arbors. Does axodendri-tic overlap predict the strength of functionalprojections? To quantify the geometric pro-jections underlying the functional projectionsconverging on pyramidal neurons in L2/3, weanalyzed reconstructions of axons and den-drites of the neurons involved in the functional projections (Fig. 2) soas to compute the geometric circuits. We labeled brain slice neurons ingranular and infragranular layers and reconstructed their axonalarbors. We also reconstructed dendritic arbors of supragranular pyr-amidal neurons.

First, to characterize the basic features of the morphological orga-nization of L4 and L5A neurons’ axonal arbors, we quantified the rawreconstructions in terms of length density2,18,21,25, averaged acrossmany axonal arbors (maps in Fig. 2e–h). Inspection of reconstructionsof excitatory neurons in L2–L6 (Fig. 2a–d) together with quantitativeanalysis of the spatial distribution of axonal length density (Fig. 2e–h)led to the following observations.

Layer 4 neurons (n ¼ 26 cells; Fig. 2c) sent copious axonalprojections into L2/3, overlapping with the territory of L2/3 pyramidalneurons’ dendrites (Fig. 2b) and thus providing a structural basis forthe L4-L2/3 projection21,25–28. These L4 axons were largely containedwithin the home column, and their density decreased monotonicallywith distance from L4. Consistent with previous observations inyounger animals21, L4 axons originating in septa (n ¼ 11 cells;Fig. 2f) were fairly dense, B50% compared with axons from barrels

(n ¼ 15; Fig. 2e) and were slightly broader horizontally in L1–3.The differences between these axons could reflect cell type differ-ences (pyramid-predominant in septum versus stellate-predominantin barrel)28.

Layer 5A neurons (n ¼ 23 cells; Fig. 2d) also sent axonal branchesthat ascended to L2. Although predicted by the LSPS maps, thesesupragranular branches have not previously been considered majorcomponents of these neurons’ axonal arbors29,30. Unlike ascending L4axonal branches, ascending L5A branches targeted L2: the density ofbranches that projected towards the pia first decreased and thenincreased, reaching peak values in L2 (Fig. 2d, arrow; Fig. 2g,h).These L5A axons also fanned out horizontally across multiple columns(Fig. 2d,g,h). In the home column in L1–L2, the density of L5A axonswas one-half that of L4 axons (49%). (Relative densities of L4 versusL5A axons were calculated as the ratio of the mean values for the twoaxon types in the upper region of either the home or side columns.) Inthe immediately adjacent side columns in L1–L2, L5A axons exceededL4 axons (159%), a difference that grew more pronounced at evengreater horizontal distances from the home column (for example, 500–1,000 mm) where L5A axons continued to extend branches but L4 axons

fe

g

dcb

L2/3 cells L4cells

L5Acells 4

3

2

1

3

2

1

5A

4

3

2

1

5A

a L4 stellateL2/3 pyramid L5B pyramid L6A cellL5A pyramid

1

2

3

4

5A

5B

6AxonsDendrites

h

dc

L4cells

L5Acells

3

2

1

4

3

2

1

5A

L4 stellateL2/3 pyramid L5B pyramid L6A cellL5A pyramid

1

2

3

4

5A

5B

6AxonsDendrites

0 8000

400

L4 axons,septum

L4 axons,barrel

0

800

0 800

L1

L2/3

L4L5A

L5B/6

0 8000

200

L5A axons,barrel-related

L1

L2/3

L4L5A

L5B/6

0 8000

200

L5A axons,septum-related

0 500

Density (µm–2)

Density (µm–2)Density(µm–2)

Figure 2 Axonal and dendritic structure.

(a) Examples of reconstructed neurons from

different layers projecting axons to L2/3.

(b) Dendrites of reconstructed L2/3 pyramidal

neurons (n ¼ 13), aligned by soma position.

Bar, 0.5 mm. (c) Axons of L4 neurons (n ¼ 26).

(d) Axons of L5A neurons (n ¼ 23). Ascending

branches reached peak densities in L2 (arrow).(e) Axons of L4barrel neurons (top, n ¼ 15). Axon

reconstructions were analyzed by length density

analysis (middle; average length density of axons).

Arbors were aligned by soma position (white

circles). Plot below length density map shows

horizontal profiles (100-mm bins, mean 7 s.e.m.)

of axonal density in the lower (vertical green bar)

and upper (vertical blue bar) regions of the

supragranular layers (L1–L3). Plot to right of map

shows vertical profiles of axonal density within

home column (horizontal black bar) and the

average of the two side columns (horizontal red

bar). White scale bar, 0.6 mm. (f) Axons of

L4septum neurons (n ¼ 11). (g) Axons of L5Abarrel

neurons (n ¼ 14). (h) Axons of L5Aseptum

neurons (n ¼ 9).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

did not (Fig. 2). Comparison of septum-related (n ¼ 9; Fig. 2h) andbarrel-related (n ¼ 14; Fig. 2g) L5A axonal arbors revealed similarprojection patterns. This was unexpected, because functional inputfrom L5A was strongly weighted towards L5Aseptum neurons, implyingthat L5Aseptum neurons selectively connect with L2 neurons in the samecolumn. A survey of cells in L5B and L6 revealed that they alsooccasionally sent axons to supragranular layers (Fig. 2a). However,these projections are infrequent29–32 (G.S. and K.S., unpublished data),consistent with weak input seen in the input maps.

Next, we used the three-dimensional geometry of the reconstruc-tions of presynaptic axons and postsynaptic dendrites to compute‘geometric input maps’ for L2/3 neurons (Methods; Figs. 3 and 4). Ouralgorithm for deriving these permitted direct and quantitative compar-ison to the functional inputs. We based our approach on the concept ofpotential synapses2,10,33. A potential synapse is defined as a location inthe neuropil where an axonal and dendritic branch are sufficiently closethat a synaptic connection could be made (Fig. 3a; see inset). Apotential synapse is a requirement for an actual synapse, and theratio of actual to potential synapses is in the range of 0.12–0.34 indifferent cortical tissues33. The number of potential synapses, Np,formed between the axon of a presynaptic neuron and the dendriteof a postsynaptic neuron is related to the functional connectionstrength (equation (1)):

qcon ¼ Npfqsyn ð2Þ

where f is the ratio of actual to potential synapses (filling fraction) andNp f is therefore the number of actual synapses. qsyn is the average

strength (charge per action potential) per actual synapse. Np representsthe ‘macroscopic’ connectivity at the level of axonal and dendritic arborgeometry, f represents the ‘microscopic’ connectivity at the level ofaxonal boutons and dendritic spines, and qsyn is the physiologicalsynaptic weight. Combining equation (1) and (2) gives

Qxy ¼ ðrcellVexcSAPÞNpfqsyn ð3Þ

Np depends on the densities and relative positions of axonal anddendritic arbors. In the example shown (Fig. 3a), there are fivepotential synapses for the particular offset of the two arbors. For anindividual pair of neurons it is straightforward to count the number ofpotential synapses directly, but this becomes computationally ineffi-cient when large numbers of neurons with complex, dense arbors areinvolved. However, the number of potential synapses between neuronscan also be computed from the product of dendritic and axonaldensities (Fig. 3b; see Methods). For reasons of computational effi-ciency, we thus used this alternative approach (Supplementary Meth-ods). The first-principle approach (identifying potential synapsesdirectly) and the theoretical method (computing from overlap densi-ties) gave statistically indistinguishable estimates of Np (Fig. 3c).Np provides the potential connectivity between pairs of positionally

defined neurons. To generate maps of geometric input (Gxy) from Np,we ‘activate’ clusters of presynaptic neurons at different locations in L4or L5A, and ‘record’ the resulting geometric input to L2/3 neurons atdifferent laminar and columnar locations (corresponding to the loca-tions of the LSPS-mapped cells). We use the same activation parameters(that is, rcellVexcSAP) that pertain to the LSPS mapping. This allows usto compute maps of geometric input (Gxy) that are expressed in termsof the number of potential synapses times the number of presynapticaction potentials per flash: Gxy ¼ NprcellVexcSAP (Supplementary

a

L1Potentialsynapses

Dendrite ofpostsynapticL2/3 neuron

Axon ofpresynapticneuron

L2/3

L4

L5A

–200 0 200Cell displacement, x (µm)

≤ 2 µm

0.0 0.1 0.2 0.3Density (µm–2)

b

ˆ,(r n )ρ0→ ˆ,(r n )ρ →

il

ir→

in

c

Pot

entia

l syn

apse

s (N

p)

First-principle calculationDensity calculation

0

2

4

6

8

400–400Cell displacement, x (µm)

–800 8000

Figure 3 Potential connectivity. (a) Potential synapses (open circles) are

points in the neuropil where axons (red) and dendrites (blue) are close

enough to make synaptic contact (inset). For these two neurons in these

locations, there were five potential synapses. (b) Arbor density profiles.

Skeleton density of a layer 2/3 pyramidal neuron dendritic arbor is denoted

by r0. Blow-up of the dendritic segment (left) explains the notation

(Supplementary Methods). The dendritic density profile, r, is obtained by

smearing the skeleton density. The sum projection of this profile (in mm�2)is represented by the map on the right. (c) Two methods for estimating the

number of potential synapses, Np, between two arbors (axons and dendrites

from a). Np depends on arbor morphologies and their relative displacement

along the cortical surface, x. First-principle calculation (gray): for each x,

the positions of both cells are randomly moved around their origins (with

Gaussian probability, s.d. ¼ 25 mm). Np is simply the average of this

distribution; error bars represent s.d. Density calculation (black): Np is

calculated from the estimated axon and dendrite densities.

L2/3

L4

L5A

a L2barrel

cells

0

100

200

300

b

0

100

200

300

c

2500

L2septum

cells

0

100

200

300

dL3septum

cells

0

100

200

300

Np AP

Gxy

(Np

AP

)

L3barrel

cells

Figure 4 Average maps of geometric input (Gxy) for L2/3 pyramidal cells grouped by columnar and laminar location. (a) Geometric input maps for L2barrel

pyramidal neurons. Small white circles represent soma positions of L4 and L5A neurons. Black regions: insufficient data. Plot shows average horizontal profiles

of Gxy for L4 (green line) and L5A (blue line). Dashed lines represent barrels and septa. (b) L3barrel cells. (c) L2septum cells. (d) L3septum cells.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Methods; Fig. 4). These geometric input maps predict the form of thefunctional input maps if the synaptic strength per potential synapse isdistributed in a spatially homogeneous manner (that is, if fqsyn

is constant).Geometric input maps were derived for the same four positionally

defined groups of L2/3 neurons that were functionally characterized(L2barrel, L3barrel, L2septum, L3septum; Fig. 4). For each group of post-synaptic L2/3 neurons, these maps show the spatial distribution ofgeometric input from L4 and L5A (‘stimulated’ using the sameexcitation parameters of resolution and intensity as in LSPS). Inspec-tion of these geometric input maps and the corresponding plots of thehorizontal profiles led to the following conclusions. In general, L2/3pyramidal neurons form vastly more potential synapses with L4 thanwith L5A axons. This applies for septum- as well as barrel-relatedcolumns and for adjacent as well as home columns. The horizontaldistributions of most projections were approximately bell-shaped andlimited to one column. The exceptions were the L5A-L2/3septum

projections (Fig. 4c,d): these had broader plateaus extending wellinto the adjacent barrel-related columns, reflecting a tendency forboth septum- and barrel-related L5A neurons axons to convergetowards L2septum pyramidal neurons.

Comparison of functional and geometric projections

To compare functional input maps and geometric input maps directly,we plotted the horizontal profiles against each other (that is, the datafor the green and blue traces below maps in Fig. 1 versus theircounterparts in Fig. 4) for each group of L2/3 neurons (Fig. 5;

Supplementary Fig. 2). Specifically, Qxy is plotted against Gxy, whereQxy/Gxy ¼ fqsyn; thus, these plots reveal the synaptic strength perpotential synapse across these projections. Steep slopes indicate strongfunctional connectivity per potential synapse.

Two input laminae (L4 and L5A) were evaluated for each of the fourgroups, so altogether eight projections were studied. Most of theindividual projections were well fit by a straight line (range of R2

values, 0.24–0.93; mean ¼ 0.72). In particular, all four of the projec-tions arising from L4 cells had R2 values of Z0.75 (Fig. 5a–d). TheL5A-L2septum projection (Fig. 5g; see also Fig. 5j) was poorly fit by astraight line (R2 ¼ 0.24). We conclude that within most projections thesynaptic strength per potential synapse was constant.

Comparing the slopes of the plots of functional versus geometricinput maps allowed us to test the hypothesis that the synaptic strengthper potential synapse is uniform across different intracortical circuitsbetween excitatory neurons; that is, whether the ratio of functional togeometric input is constant across different projections. If this were so,all the points for all the projections in these plots (Fig. 5a–h) should fallalong straight lines with identical slopes. The slopes of the regressionsshowed both similarities and differences across the eight projections.None of the four L5A-originating projections had significantly differentslopes (P4 0.05; Fig. 5j). Within the set of L4-originating projections,slopes differed up to fourfold (Fig. 5i); the L4-L2barrel projection hada significantly higher slope (Fig. 5a), and the L4-L3septum projectionhad a significantly lower slope (Fig. 5d) than the other projections(P o 0.05). Slopes were significantly higher (P o 0.05) for allprojections from L5A than from L4 (Fig. 5k) by a factor of 2.8 on

average. This observation implies highersynaptic strength per potential synapse forprojections from L5A than from L4.

To examine the spatial aspect of thesestructure-function comparisons in moredetail, we computed the average ratio offunctional to geometric input within thehome columns (Fig. 6a, gray bars; Fig. 6b,light brown bars; Supplementary Fig. 2) andcompared this to the average functional/geo-metric input ratio for the side columns (blackbars). This analysis was particularly revealingin septum-related projections (Fig. 6a), wherehome-column ratios varied 16-fold betweendifferent projections (compare L4-L3septum

and L5A-L2septum). Differences betweenhome- and side-column ratios within thesame projection were as high as fourfold(within L5A-L2septum). The maximum dif-ference observed across the entire data set wasa 28-fold disparity between home-columnL5A-L2septum and the right (anteromedial)side column of L4-L3septum. Unexpectedly,three of the eight projections (L4-L2septum,L4-L3septum, L5A-L2barrel) showed a clearleftward (posterolateral) ‘skew’, and two moreprojections showed a trend in this direction(L4-L3barrel, L5A-L3septum) (Fig. 6a,b, andSupplementary Fig. 2). Functional/geometricratios in barrel-related projections varied overa narrower range (Fig. 6b). Despite the widerange of ratios across the entire data set, theaverage functional/geometric ratios of sep-tum-related projections (Fig. 6a, dashed gray

0

5

10

0

5

10

a

R 2 = 0.93

L4→L2barrel L4→L3barrel

L5A→L2barrel L5A→L3barrel

b

0.91

d

0.75

c

0.81

L4→L2septum L4→L3septum

L5A→L2septum L5A→L3septume

0

5

10

0

5

10

0

5

10

0

5

10

0

5

10

0

5

10

0.82

f

0.86

h

0.43

g

0.24

0 20 40

0 100 200 100 2000 0100 200 0 100 200

i

0.71

All L4k

All L4

All L5A

0.53

jAll L5A

0 100 200 0 100 200

→ L2barrel

→ L3barrel

→ L3septum

→ L2septum

Avg. of all0

5

10

0

5

10

0

5

10

Geometric input, Gxy (Np AP)

Fun

ctio

nal i

nput

, Qxy

(pA

)

60 0 20 40 60

0 20 40 60

0 20 40 60 0 20 40 60

Figure 5 Comparison of geometric and functional projections. Horizontal profiles of the L4 and L5A

functional inputs (Qxy; data from plots in Fig. 1e–h) are plotted against the corresponding geometricinputs (Gxy; plots in Fig. 4a–d). In general, points further from the origin are from the home column, and

points closer to the origin are from adjacent columns. R2 values are shown. (a) L4-L2barrel projection.

(b) L4-L3barrel projection. (c) L4-L2septum projection. (d) L4-L3septum projection. (e) L5A-L2barrel

projection. Note different x-axis scale for L5A-originating projections. (f) L5A-L3barrel projection.

(g) L5A-L2septum projection. Arrow: region of peak functional input. (h) L5A-L3septum projection.

(i) All L4 data plotted together. (j) All L5A data plotted together. (k) L4 data (green) and L5A

data (blue) plotted together.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

line) did not differ from that of barrel-related projections (Fig. 6b,dashed gray line; P 4 0.05). We conclude that the average synapticstrength per potential synapse can vary widely between differentprojections, and in some cases there is strong spatial dependenceeven within a projection.

DISCUSSION

To investigate the relationships between neuronal geometry and func-tional synaptic connectivity in cortical circuits, we have used LSPS tomeasure the strength of functional excitatory synaptic projectionsonto individual L2/3 pyramidal neurons and morphological recon-structions of labeled neurons to quantify the axonal and dendriticgeometry underlying these functional circuits. By examining functionaland geometric projections originating from two laminae (L4 and L5A)and terminating onto pyramidal neurons located in two laminae(L2 and L3) and two columns (barrel- and septum-related), wehave characterized eight positionally defined intracortical projections.This dual approach allowed us to quantitatively compare thegeometric and functional organization of multiple cortical micro-circuits (Fig. 6c).

Neuronal geometry and functional connectivity

Does neuronal geometry predict functional connectivity? By recon-structing a statistically representative sample of axons from L4 and L5Aneurons and dendrites from L2 and L3 neurons in both the barrel andseptal columnar systems, we were able to quantify the neuronalgeometry underlying eight functional projections in a cortical circuit(Fig. 6c). We used these reconstructions (Fig. 2) to compute maps ofpotential connectivity (Fig. 3) and, subsequently, geometric inputmaps (Fig. 4). Direct comparison of geometric and functional maps(Figs. 5 and 6, Supplementary Fig. 2) demonstrated complex rela-tionships both within and between the microcircuits examinedhere (Fig. 6c).

Before discussing these comparisons in detail, it is useful to considerhow geometric and functional input maps are related. Geometric inputmaps are spatial maps of Gxy, the number of potential synapses timesthe number of presynaptic action potentials per flash. They can beconverted to functional input maps by multiplying by the ‘fillingfraction,’ f, which is the ratio of actual structural synapses per potentialstructural synapse33, and by the functional ‘synaptic weight,’ qsyn, whichis the functional synaptic efficacy (that is, synaptic charge per actionpotential) per actual synapse (equation (1)). Exact values for f andqsyn are not available for the projections studied here, but if theyare approximately constant within and across projections, the shapesof geometric and functional input maps should be identical. Thus,we can directly compare geometric and functional input maps toevaluate the hypothesis that functional circuits mirror the form of theunderlying geometric circuits; that is, that ‘function follows form’ inthese projections.

Our data are consistent with this ‘function follows form’ hypothesisin the following sense. Geometric inputs typically predict functionalinputs within individual projections (Fig. 5; the clear exceptionbeing the L5A-L2septum projection; see below). Moreover, whenall the projections arising from a particular layer (L4 or L5A) arepooled, spatial correlations between functional and geometric projec-tions remain high (Fig. 5i,j). Thus, for connections between particularlayers, functional projections were largely predicted by neuronalgeometry alone.

Our data are inconsistent with the ‘function follows form’ hypothesisin three specific ways. First, synaptic strength per potential synapse islower for all projections originating from L4 axons than for thoseoriginating from L5A (Fig. 5k). Thus, this relationship depends on thelaminar source of presynaptic axons. Second, this ratio of functional togeometric input varies between the four different L4 projections(Fig. 5i): it depends on the laminar and columnar position of thepostsynaptic neuron. Third, this ratio also varies within one projection,the L5A-L2septum projection, depending on the columnar source ofthe presynaptic axons. Thus, in the absence of functional data, neuronalgeometry alone cannot be used to predict the relative strengths ofprojections, and it thus provides only a crude estimate of functionalcortical circuits. Our results are consistent with the view that functionalprojections are substructures within the scaffolds of geometric projec-tions. In other words, circuits can be refined, perhaps through activity-dependent mechanisms, yielding within them inhomogeneous rela-tionships between synaptic strength and potential synapses.

Potential connectivity and Peters’ rule

Our potential connectivity analysis pertains directly to a structuralmodel for cortical circuit organization known as Peters’ rule3–5,11,34.This concept arose from the ultrastructural observation that synapticboutons of thalamocortical axons seem to form synapses with L4dendrites in proportion to their availability (that is, density) in the

L3septum

cells

a

c

L4L5A

Home

L sid

e

R si

de

L2L3

Septum-centered

0

0.1

0.2

0.3

Septum-related column Barrel-related column

BarrelSeptum

Gxy

Qxy

PROJECTION

L4

L3

L4

L2

L5A

L3

L5A

L2

b

L4L5A

Home

L sid

e

R si

de

L2L3

Barrel-centered

0

0.1

0.2

0.3

PROJECTION

L4

L3

L4

L2

L5A

L3

L5A

L2

L4 cellsL4 cells

L5A cells L5A cells

L2septum

cellsL2barrel

cells

L3barrel

cells

1.3/32

3.5/227

3.4/31

5.6/126

8.4/33

2.8/37

15.9/290

8.3/67

1.2/18

13.3/202

6.5/71

1.6/15

Figure 6 Ratios of functional and geometric projections in barrel- and

septum-related columns. (a) Average ratio of functional/geometric

connectivity (/Qxy/GxyS) for projections to L2septum and L3septum cells.

Horizontal bars in inset show subregions averaged in horizontal profiles; for

example, home-column data (gray) were averaged over region indicated by

gray horizontal bar, and side-column data (black) were averaged over region

indicated by black bars. Dashed gray line indicates the average value of these

septum-related ratios. (b) Qxy /Gxy ratios for projections to L2barrel and L3barrel

cells. Home-column data were averaged over region indicated by light brown

horizontal bar. Gray bar indicates average value of these barrel-related ratios.(c) Average values of functional and geometric connectivity (shown inside

the boxes as /QxyS//GxyS) for L2 and L3 pyramidal neurons in septum-

related columns (left) and barrel-related columns (right). All of the home-

column and a subset of the side-column Qxy /Gxy values are shown; ratios at

the top of each box are for side-column input from L5A, ratios in the middle

are for home-column L5A input, and lower ratios are for home-column L4

input. Representative reconstructions of dendritic arbors of major excitatory

cell types are shown in gray.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

neuropil5,34. According to Peters’ rule, structural synapses occur whereaxons (with synaptic boutons) and dendrites overlap, with the‘strength’ of a connection following, at least to a first approximation,from the density of the overlap. In more general and quantitativeformulations of Peters’ rule, the anatomical strength of a connectionbetween two classes of neurons can be computed as the product of thedensities of axonal boutons and dendrites4,11. Peters’ rule can thus beused to calculate connection strengths between different classes ofneurons and to generate cortical circuit diagrams based on anatomyalone4,11. Although Peters’ rule makes no explicit inference about thefunctional strengths of connections, it provides a blueprint of theimplied functional circuit if the synaptic strength per unit of axon-dendrite overlap (that is, per potential synapse) is assumed to beconstant on average; that is, if function directly follows form.

Our approach here for calculating geometric circuits, based oncomputing potential connectivity from neuronal geometry, is a versionof Peters’ rule: we estimate synaptic connectivity from the spatialdistributions (densities) of axons and dendrites. The main differenceis that we used densities of axons, not of axonal varicosities. With thesimplifying assumptions that axonal varicosities and synaptic strengthper potential synapse are homogeneously distributed and roughlysimilar for different types of axons, our ‘function follows form’hypothesis is effectively a functional version of Peters’ rule.

Specificity in cortical circuits

Cortical circuits show specificity on multiple levels. Axons and dendriteneed to overlap to make connections, and geometry by itself is thereforea source of specificity (Peters’ rule, as discussed previously)3–5,11,34.Physiological experiments have further revealed that within individualprojections, cortical neurons form non-random functional networks35–37.In addition, cell type–specific biases clearly have a role in establishingcortical connections3,20,38,39. Even in the case of projections betweenexcitatory neurons, there are suggestions that the shapes of axons anddendrites do not predict function40–43. Our data show quantitativelythat the relationship between structure and functional connectionstrength depends on the laminar and columnar positions of presynapticneurons and recipient neurons, even within the same cell classes. Thisdemonstrates that circuit organization can be strongly location-specificwithin a cell type.

Functional form of septum-related projections

The strong home-column and weak surrounding-column organizationof the L5A-L2 projection raises numerous questions. Are home-column inputs strengthened, or are side-column inputs weakened?Perhaps the former, as the functional input per potential synapsefor side-column inputs originating from L5A resembles that ofL4-originating home- and side-column inputs. How do home-columninputs become relatively strengthened? What purposes do the ‘weaksurround’ projection serve? Perhaps L5A acts as a plasticity-relatedlayer; functionally weaker ‘surround’ regions in L2 beyond the homecolumn could be a substrate for experience-dependent (long-term)circuit plasticity, analogous to the barn owl’s optic tectum44. Consistentwith this, POm (the thalamic nucleus innervating L5A) may beimportant for experience-dependent plasticity in the barrel cortex45.At the level of L2, a honeycomb mosaic has been described at theL1/L2 border, associated with plasticity markers such as zinc andNMDAR1 (ref. 46); perhaps L2barrel and L2septum neurons participatedifferently in this micromodularity. Alternatively, the weak surroundcould have different synaptic properties from the center or couldreflect areas where presynaptic L5A axons preferentially connectto either inhibitory interneurons or to the apical dendritic tufts of

deeper excitatory neurons (instead of to L2/3 neurons). In anycase, our findings indicate that strong intracolumnar functional pro-jections are a fundamental organizational feature of cortical circuits47,even when abundant anatomical substrates for non-columnar projec-tions exist.

In contrast to L5A-L2, the L4-L3septum projection is unexpectedlyweak compared with the high density of geometrically available L4axons. Could the low functional/geometric ratios in this projection(lower by a factor of 3.2 compared with the average of the three otherL4-based projections; Fig. 5a–d) reflect a role in experience-dependentplasticity? Previous data support this idea: in juvenile rats, L2/3septum

cells show an approximately twofold experience-dependent gain infunctional input from septal L4 (ref. 17) not attributable to axonalarbor reorganization by branch addition25. These observations supporta model in which some combination of functional synaptic plasticity(Dqsyn) and structural re-wiring at the level of individual spines andboutons (Df)33,48 produces functional re-wiring of synaptic projections(Dqcon) in which neuronal arbor geometry is fixed (that is, Np isconstant). Thus, the extremes among the projections studied here(L5A-L2septum and L4-L3septum) demonstrate that although therelationship between functional and geometric input tends to beconserved among different projections from particular layers, excep-tions can occur, and the dynamic range (that is, the plasticity potential)in this relationship is large.

METHODSElectrophysiology and photostimulation. Detailed methods have been pub-

lished previously17,21,22. Briefly, young adult rats (Sprague-Dawley, 26–36 d

postnatal) were used according to Cold Spring Harbor Laboratory’s Animal

Care and Use guidelines. Acute brain slices, 300 mm in thickness, were cut

perpendicular to barrel rows in chilled cutting solution (consisting of, in mM,

110 choline chloride, 25 NaHCO3, 25 D-glucose, 11.6 sodium ascorbate,

7 MgSO4, 3.1 sodium pyruvate, 2.5 KCl, 1.25 NaH2PO4, and 0.5 CaCl2),

and transferred to artificial cerebrospinal fluid (ACSF, consisting of, in mM,

127 NaCl, 25 NaHCO3, 25 D-glucose, 2.5 KCl, 4 MgCl2, 4 CaCl2, and

1.25 NaH2PO4, aerated with 95% O2/5% CO2) for incubation at 35 1C

for 30 min and thereafter at room temperature (22–23 1C). The slicing

orientation yielded slices in which up to five barrels were clearly discerned in

the living slice using bright-field optics. The only slices used were those in

which the vertical axes of neurons (for example, apical dendrites of pyramidal

neurons and primary descending axons) ran parallel to the slice plane, as seen

with interference contrast optics.

LSPS by glutamate uncaging was performed as described17,21,22. Briefly, with

caged glutamate (NI-glutamate, Sigma; 0.37 mM; ref. 49) in the ACSF, cells

were patched at depths of 50–100 mm (mean, 78 mm; no significant differences

between columnar or laminar subgroups) and recorded in voltage clamp at

room temperature. Intracellular solution consisted of (in mM) 120 KMeSO3,

20 CsCl, 4 NaCl, 10 HEPES, 1 EGTA, 4 Mg2ATP, and 0.3 Na2GTP, 14 sodium

phosphocreatine, 3 ascorbate, and 0.1 Alexa-594 (Molecular Probes). The beam

of an ultraviolet laser (DPSS Lasers) was flashed at sites within the mapping

pattern, a 16 � 16 array with 100-mm spacing between sites. The map covered a

2.3-mm2 square patch of cortex centered vertically on the L4/L5A boundary,

and horizontally on the midpoint of a barrel or septum. Input maps were

repeated 2–5 times for each cell. For analysis, the mean current during a

synaptic response window (8–100 ms after stimulus) was calculated. Thus, pixel

values represent synaptic charge (coulombs); however, for consistency with

prior studies and because synaptic current is more familiar, data are expressed

as pA. Each cell’s individual maps were averaged. Sites giving direct responses,

defined as events arising within 8 ms after stimulus43, were blanked (for

example, see black pixels in Fig. 1c). Because the boundary between L2 and L3

is not cytoarchitectonically apparent, we divided L2/3 in half vertically based on

the mean vertical position of the L2/3 neurons, 546 mm above the L4/L5A

border, and we use the terms ‘L2’ and ‘L3’ to refer to the upper and lower

regions of L2/3, respectively.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Calibration of photostimulation. Photostimulation was calibrated by measur-

ing with loose-seal voltage recording the excitation profiles of cells from

the major classes of excitatory neurons within the synaptic mapping area17,21,22.

Control experiments established that for the specific experimental conditions

used (such as beam profile and intensity, cage and divalent concentrations, and

animal age), there was reliable perisomatic photostimulation of presynaptic

cells while remaining well below the thresholds for photostimulation of distal

dendrites or photostimulation through synaptic (network, disynaptic) driving.

Detailed analysis of excitability parameters is presented elsewhere22. In parti-

cular, the average excitability of neurons, in terms of SAP, the average number of

action potentials per cell per flash (see equation (1)), was 1.84 7 0.34 for L4

cells (n¼ 27) and 2.59 7 0.30 for L5A cells (n¼ 27). Resolution, defined as the

mean distance from the soma of spike-eliciting sites, weighted by the number of

spikes per site, was 58 7 3 mm for L4 neurons and 64 7 2 for L5A neurons.

To gauge whether stimulation in L5A could generate spikes in L4 cells, we

also recorded excitation profiles from six cells in L4 located very close to L5A

(mean soma distance from L5A, 38 mm; range 26–52; ref. 22). Of all the spikes

generated, 98% came from stimulation at sites within L4, 2% from stimulation

at sites immediately adjacent (B13 mm) to the L4/5A border, and 0% from sites

in L5A located Z25 mm from the border.

Morphological reconstructions and analysis. Cells were loaded with biocytin

(3 mg/ml in the intracellular solution) during patch recording. Optimal

staining was achieved by patching for 2–10 min, slow retraction of the pipet

to form an outside-out patch, and incubation of the slice for 1–3 h in

oxygenated ACSF. Slices were fixed in 4% paraformaldehyde and rinsed in

phosphate buffer. For biocytin staining, after a quenching step (30 min in 1%

H2O2), samples were incubated overnight in ABC solution (ABC kit,

Vector Labs), and reacted in the presence of nickel. Samples were stained

for cytochrome oxidase for 1 h to visualize barrels and laminae. Samples

were mounted in DMSO50, and neurons were reconstructed using a 40�water-immersion objective lens and camera lucida system (Neurolucida,

MicroBrightField). Tracings were quantitatively analyzed using custom Matlab

(Mathworks) routines.

If L5A cells that were deeper in the slice had better-preserved (and more

laterally extending) axonal arbors, this would need to be factored into some of

the analyses. However, comparison of the axonal density extending horizontally

across L1–L3 of the shallower (37–73 mm, n ¼ 11) versus deeper (75–110 mm,

n ¼ 12) samples of L5A neurons showed nearly identical distributions, and

exponential fits to these data were not significantly different (P 4 0.05).

Computation of geometric projections. See Supplementary Methods for

details. Briefly, neuron tracings (axonal arbors of L4 and L5A neurons, and

dendritic arbors of L2/3 neurons) were shrinkage-corrected and morphed to a

standard barrel cortex template (based on average measured dimensions of

laminar and columnar cytoarchitectonic landmarks: that is, barrels, septa, and

laminae). The LSPS maps were similarly morphed for subsequent comparisons.

The number of potential synapses, Np, from L4 and L5A axons to individual

L2/3 neurons was computed based on the product of axonal and dendritic

arbor length density profiles (Supplementary Methods):

Npð~Ra; ~RdÞ ¼2s

Zrað~Ra; ~ra; naÞrdð~Rd; ~rd; ndÞ sinð dna; ndÞ��

dð~ra �~rdÞd3rad3rddOadOd

ð4Þ

where s is the average length of dendritic spines (a value of 2 mm is used),

ra;dð~Ra;d; ~ra;d; na;dÞ are the expected axonal and dendritic density profiles, and~Ra and ~Rd are the positions of the pre- and postsynaptic somata (see also Fig. 3).

Next, we convolve the numbers of potential synapses Np with the LSPS

excitability profiles of L4 and L5A neurons, SAP, to obtain the geometric input

maps of L2/3 cells, using Gxy ¼ rcell (VexcNp) � SAP (Supplementary Methods).


ACKNOWLEDGMENTSWe thank K. Zito and V. Scheuss for a critical reading of the manuscript,members of the Svoboda laboratory for useful discussions, J. Huang andC. Wu for access to their Neurolucida system and B.J. Burbach and C. Zhangfor technical assistance. Funded by the Howard Hughes Medical Institute

(G.S. and K.S.), US National Institutes of Health (D.C., A.S. and K.S.),Klingenstein Foundation (D.C.) and Human Frontier Science Program (I.B.).


Received 1 February; accepted 31 March 2005


1. Douglas, R.J. & Martin, K.A.C. Neuronal circuits of the neocortex. Annu. Rev. Neurosci.27, 419–451 (2004).

2. Uttley, A.M. The probability of neural connexions. Proc. R. Soc. Lond. B 144, 229–240(1955).

3. White, E.L. Cortical Circuits (Birkhauser, Boston, 1989).4. Braitenberg, V.B. & Schuz, A. Cortex: Statistics and Geometry of Neuronal Connectivity

(Springer, Berlin, 1998).5. Peters, A. Thalamic input to the cerebral cortex. Trends Neurosci. 2, 1183–1185

(1979).6. Ramon y Cajal, S. Sur la structure de l’ecorce cerebrale de quelques mamiferes. Cellule

7, 125–176 (1891).7. Gilbert, C.D. Microcircuitry of the visual cortex. Annu. Rev. Neurosci. 6, 217–247

(1983).8. Douglas, R.J. & Martin, K.A.C. A functional microcircuit for cat visual cortex. J. Physiol.

(Lond.) 440, 735–769 (1991).9. Kalisman, N., Silberberg, G. & Markram, H. Deriving physical connectivity from neuronal

morphology. Biol. Cybern. 88, 210–218 (2003).10. Hellwig, B. A quantitative analysis of the local connectivity between pyramidal neurons

in layers 2/3 of the rat visual cortex. Biol. Cybern. 82, 111–121 (2000).11. Binzegger, T., Douglas, R.J. & Martin, K.A.C. A quantitative map of the circuit of cat

primary visual cortex. J. Neurosci. 24, 8441–8453 (2004).12. Petersen, C.C.H. & Sakmann, B. The excitatory neuronal network of rat layer 4 barrel

cortex. J. Neurosci. 20, 7579–7586 (2000).13. Koralek, K.A., Jensen, K.F. & Killackey, H.P. Evidence for two complementary patterns

of thalamic input to the rat somatosensory cortex. Brain Res. 463, 346–351(1988).

14. Chmielowska, J., Carvell, G.E. & Simons, D.J. Spatial organization of thalamocorticaland corticothalamic projection systems in the rat SmI barrel cortex. J. Comp. Neurol.285, 325–338 (1989).

15. Lu, S.M. & Lin, R.C.S. Thalamic afferents of the rat barrel cortex: a light- and electron-microscopic study using Phaseolus vulgaris leucoagglutinin as an anterograde tracer.Somatosens. Mot. Res. 10, 1–16 (1992).

16. Kim, U. & Ebner, F.F. Barrels and septa: separate circuits in rat barrels field cortex.J. Comp. Neurol. 408, 489–505 (1999).

17. Shepherd, G.M.G., Pologruto, T.A. & Svoboda, K. Circuit analysis of experience-dependent plasticity in the developing rat barrel cortex. Neuron 38, 277–289(2003).

18. Brecht, M., Roth, A. & Sakmann, B. Dynamic receptive fields of reconstructed pyramidalcells in layers 3 and 2 of rat somatosensory barrel cortex. J. Physiol. 553, 243–265(2003).

19. Callaway, E.M. & Katz, L.C. Photostimulation using caged glutamate revealsfunctional circuitry in living brain slices. Proc. Natl. Acad. Sci. USA 90, 7661–7665(1993).

20. Dantzker, J.L. & Callaway, E.M. Laminar sources of synaptic input to cortical inhibitoryinterneurons and pyramidal neurons. Nat. Neurosci. 3, 701–707 (2000).

21. Bureau, I., Shepherd, G.M.G. & Svoboda, K. Precise development of functional andanatomical columns in the neocortex. Neuron 42, 789–801 (2004).

22. Shepherd, G.M.G. & Svoboda, K. Laminar and columnar organization of ascendingexcitatory projections to layer 2/3 pyramidal neurons in rat barrel cortex. J. Neurosci.(in the press).

23. Ren, J.Q., Aika, Y., Heizmann, C.W. & Kosaka, T. Quantitative analysis of neurons andglial cells in the rat somatosensory cortex, with special reference to GABAergic neuronsand parvalbumin-containing neurons. Exp. Brain Res. 92, 1–14 (1992).

24. Keller, A. & Carlson, G.C. Neonatal whisker clipping alters intracortical, butnot thalamocortical projections, in rat barrel cortex. J. Comp. Neurol. 412, 83–94(1999).

25. Bender, K.J., Rangel, J. & Feldman, D.E. Development of columnar topographyin the excitatory layer 4 to layer 2/3 projection in rat barrel cortex. J. Neurosci. 23,8759–8770 (2003).

26. Lubke, J., Egger, V., Sakmann, B. & Feldmeyer, D. Columnar organization of dendritesand axons of single and synaptically coupled excitatory spiny neurons in layer 4 of the ratbarrel cortex. J. Neurosci. 20, 5300–5311 (2000).

27. Petersen, C.C.H. & Sakmann, B. Functionally independent columns of rat somatosen-sory barrel cortex revealed with voltage-sensitive dye imaging. J. Neurosci. 21, 8435–8446 (2001).

28. Staiger, J.F. et al. Functional diversity of layer IV spiny neurons in rat somatosensorycortex: quantitative morphology of electrophysiologically characterized and biocytinlabeled cells. Cereb. Cortex (2004).

29. Gottlieb, J.P. & Keller, A. Intrinsic circuitry and physiological properties of pyramidalneurons in rat barrel cortex. Exp. Brain Res. 115, 47–60 (1997).

30. Manns, I.D., Sakmann, B. & Brecht, M. Sub- and suprathreshold receptive fieldproperties of pyramidal neurones in layers 5A and 5B of rat somatosensory barrel cortex.J. Physiol. (Lond.) 556, 601–622 (2004).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

31. Zhang, Z.W. & Deschenes, M. Intracortical axonal projections of lamina VI cells of theprimary somatosensory cortex in the rat: a single-cell labeling study. J. Neurosci. 17,6365–6379 (1997).

32. Burkhalter, A. Intrinsic connections of rat primary visual cortex: laminar organization ofaxonal projections. J. Comp. Neurol. 279, 171–186 (1989).

33. Stepanyants, A., Hof, P.R. & Chklovskii, D.B. Geometry and structural plasticity ofsynaptic connectivity. Neuron 34, 275–288 (2002).

34. Peters, A. & Feldman, M.L. The projection of the lateral geniculate nucleus to area 17 ofthe rat cerebral cortex. I. General description. J. Neurocytol. 5, 63–84 (1976).

35. Song, S., Sjostrom, P.J., Reigl, M., Nelson, S. & Chklovskii, D.B. Highly nonrandomfeatures of synaptic connectivity in local cortical circuits. PLoS Biol. 3, 1–13 (2005).

36. Yoshimura, Y., Dantzker, J.L. & Callaway, E.M. Excitatory cortical neurons form fine-scale functional networks. Nature 433, 868–873 (2005).

37. Kalisman, N., Silberberg, G. & Markram, H. The neocortical microcircuit as a tabularasa. Proc. Natl. Acad. Sci. USA 102, 880–885 (2005).

38. Gupta, A., Wang, Y. & Markram, H. Organizing principles for a diversity of GABAergicinterneurons and synapses in the neocortex. Science 287, 273–278 (2000).

39. Stepanyants, A., Tamas, G. & Chklovskii, D.B. Class-specific features of neuronal wiring.Neuron 43, 251–259 (2004).

40. Yabuta, N.H., Sawatari, A. & Callaway, E.M. Two functional channels from primary visualcortex to dorsal visual cortical areas. Science 292, 297–300 (2001).

41. Schubert, D., Kotter, R., Zilles, K., Luhmann, H.J. & Staiger, J.F. Cell type-specificcircuits of cortical layer IV spiny neurons. J. Neurosci. 23, 2961–2970 (2003).

42. Sawatari, A. & Callaway, E.M. Diversity and cell type specificity of local excitatoryconnections to neurons in layer 3B of monkey primary visual cortex. Neuron 25, 459–471 (2000).

43. Schubert, D. et al. Layer-specific intracolumnar and transcolumnar functionalconnectivity of layer V pyramidal cells in rat barrel cortex. J. Neurosci. 21, 3580–3592 (2001).

44. Linkenhoker, B.A. & Knudsen, E.I. Incremental training increases the plasticity of theauditory space map in adult barn owls. Nature 419, 293–296 (2002).

45. Kaas, J. & Ebner, F. Intrathalamic connections: a new way to modulate corticalplasticity? Nat. Neurosci. 1, 341–342 (1998).

46. Ichinohe, N., Fujiyama, F., Kaneko, T. & Rockland, K.S. Honeycomb-like mosaic atthe border of layers 1 and 2 in the cerebral cortex. J. Neurosci. 23, 1372–1382(2003).

47. Mountcastle, V.B. Modality and topographic properties of single neurons of cat’s somaticsensory cortex. J. Neurophysiol. 20, 408–434 (1957).

48. Trachtenberg, J.T. et al. Long-term in vivo imaging of experience-dependent synapticplasticity in adult cortex. Nature 420, 788–794 (2002).

49. Canepari, M., Nelson, L., Papageorgiou, G., Corrie, J.E. & Ogden, D. Photochemical andpharmacological evaluation of 7-nitroindolinyl-and 4-methoxy-7-nitroindolinyl-aminoacids as novel, fast caged neurotransmitters. J. Neurosci. Methods 112, 29–42 (2001).

50. Grace, A.A. & Llinas, R. Morphological artifacts induced in intracellularly stainedneurons by dehydration: circumvention using rapid dimethyl sulfoxide clearing.Neuroscience 16, 461–475 (1985).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Reversible blockade of experience-dependent plasticityby calcineurin in mouse visual cortex

Yupeng Yang1, Quentin S Fischer1, Ying Zhang2, Karsten Baumgartel3, Isabelle M Mansuy3 & Nigel W Daw1

Numerous protein kinases have been implicated in visual cortex plasticity, but the role of serine/threonine protein phosphatases

has not yet been established. Calcineurin, the only known Ca21/calmodulin-activated protein phosphatase in the brain, has

been identified as a molecular constraint on synaptic plasticity in the hippocampus and on memory. Using transgenic mice

overexpressing calcineurin inducibly in forebrain neurons, we now provide evidence that calcineurin is also involved in ocular

dominance plasticity. A transient increase in calcineurin activity is found to prevent the shift of responsiveness in the visual cortex

following monocular deprivation, and this effect is reversible. These results imply that the balance between protein kinases and

phosphatases is critical for visual cortex plasticity.

Brief monocular deprivation during the critical period of developmentdramatically alters neuronal responsiveness to subsequent stimu-lation of deprived and non-deprived eyes in the visual cortex. Thisexperience-dependent phenomenon is called ocular dominance plasti-city (ODP). Calcium influx through NMDA (N-methyl-D-aspartate)receptors is believed to be one of the initial steps of the mechanism ofODP1,2, followed by the activation of protein kinases and phosphatases.The function of protein kinases in ODP has been studied extensively3–5.However the contribution of protein phosphatases, known to beinvolved in hippocampal synaptic plasticity and memory6,7, has notbeen investigated.

Calcineurin is a serine/threonine protein phosphatase, highly sensi-tive to Ca2+ (Kd ¼ 0.1–1 nM), and is the only phosphatase activated byCa2+/calmodulin8. It is selectively enriched in pyramidal cells of theCNS9,10. In primary visual cortex, its expression and laminar distribu-tion are developmentally regulated and follow the inside-out pattern ofcortical maturation10. Calcineurin can regulate a wide array of sub-strates involved in brain plasticity by direct dephosphorylation orthrough activation of the downstream protein phosphatase 1 (PP1).Calcineurin and PP1 dephosphorylate specific sites on NMDA andAMPA (a-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid)receptors, thereby contributing to the mechanisms of long-termpotentiation (LTP), long-term depression (LTD) and depotentia-tion7,11–16. The pharmacological blockade of calcineurin impairsLTD17 and enhances the induction of LTP18 in visual cortex. Mechan-istically, it may serve in part to antagonize the cAMP-dependentprotein kinase A (PKA) by downregulating molecular substratesactivated by PKA6,19. It may also control PKA activity itself byinhibiting specific isoforms of the cAMP-producing adenylyl cyclase(AC9, ref. 20). In neurons, calcineurin and PKA are active simulta-neously, and their concerted action is facilitated by A kinase–anchoring

proteins (AKAP)21,22 through concomitant binding. Notably, previousresults have demonstrated that type II PKA23,24 and AKAP150 (Fischeret al., Soc. Neurosci. Abs. 37.9, 2003) are necessary for ODP. In additionto PKA, calcineurin and PP1 controls other ODP-related substrates thatinclude autophosphorylated Ca2+/calmodulin-dependent proteinkinase II (CaMKII; ref. 4), a kinase with a similar pattern of mRNAand protein expression as calcineurin in cortical and hippocampalstructures9,25, and the cAMP–response element binding protein(CREB) transcription factor26–28. In the visual cortex, CRE-mediatedtranscription is increased after monocular deprivation27, and blockingCREB activation prevents the loss of responses to the deprived eye28.Overall, these findings suggest the involvement of calcineurin in ODP.

To test this hypothesis, we took advantage of a line of transgenic miceexpressing an active form of calcineurin inducibly and reversibly in thebrain with the tetracycline-controlled transactivator (tTA) system19 andexamined ODP in these mutant mice. We observed that an excess ofcalcineurin activity in the visual cortex during the critical periodimpairs ODP in a reversible fashion. These findings indicate thatcalcineurin is critical for ODP and support the model that calcineurinnegatively regulates various forms of brain plasticity.

RESULTS

Calcineurin activity in the visual cortex

To confirm that calcineurin activity is increased in primary visualcortex (V1) of the calcineurin-overexpressing (CNO) mice, we per-formed phosphatase assays on extracts from binocular V1 at the peak ofthe critical period (postnatal day (P) 28–29). Assays showed a 48 7 7%increase in calcineurin activity in CNO mice compared with wild-typelittermates (Fig. 1a). This increase in calcineurin activity was lowerthan that observed in the adult hippocampus19 (112 7 9%), mostlikely due to a different efficiency and time course of transgene


1Department of Ophthalmology and Visual Science, Yale University School of Medicine, New Haven, Connecticut 06520 USA. 2Department of Chemical Engineering,University of Connecticut, Storrs, Connecticut 06269 USA. 3Brain Research Institute, Medical Faculty of the University of Zurich, Department of Biology of the Swiss FederalInstitute of Technology, 8057 Zurich, Switzerland. Correspondence should be addressed to Y.Y. ([email protected]).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

expression in these structures. As in hippocampus, calcineurinoverexpression in the visual cortex could be suppressed by doxy-cycline treatment, reflecting the full reversibility of the geneticmanipulation (Fig. 1a). Analysis of the subcellular localization ofcalcineurin overexpression revealed a significant increase in calci-neurin activity in cytoplasmic (46%) but not in synaptosomal ornuclear fractions in the mutant mice (Fig. 1b,c). This indicatesthat the calcineurin transgene concentrates in the cytoplasm inresting conditions.

Impaired ODP in CNO mutants

Four days of monocular deprivation during the critical period issufficient to induce a robust and saturating ocular dominance shiftto the open eye29. In wild-type littermates, neurons in the binocularzone of V1 were predominantly driven by the contralateral eye(Fig. 2a). After deprivation of the contralateral eye, the balance ofinputs shifted to the open ipsilateral eye (Fig. 2b). Longer periods ofdeprivation did not induce any further ocular dominance shift (t-test,4–5 d versus 6–7 d, P ¼ 0.58). In non-deprived CNO mice, corticalneurons showed similar ocular dominance distribution as wild-typelittermates (Fig. 2c). However, after monocular deprivation, nosignificant shift was observed in CNO mice (Fig. 2d), indicatingimpaired ODP. Ocular dominance distribution was still biased to thedeprived contralateral eye even after 7–8 d of deprivation (two of sevenmice), suggesting that the impairment in ODP was not because of areduced sensitivity of the mutant mice to deprivation. Furthermore, theimpairment was not because of a delay in the onset of the criticalperiod, as 4- to 5-d monocular deprivation starting at P33–34 (after thetypical critical period29) did not induce any apparent ocular dominanceshift in CNO or wild-type mice (Fig. 3).

The impairment in plasticity can be rescued

To further determine if the impairment in plasticity was a directconsequence of the increased calcineurin activity, we suppressedcalcineurin overexpression with doxycycline starting 7 d before mono-cular deprivation and examined whether this restored ODP in themutant mice (Fig. 4). In wild-type controls, ocular dominancedistribution was similar with or without doxycycline treatment inboth non-deprived and deprived groups (Fig. 4b,c; compare toFig. 2a,b). However, in mutants, doxycycline treatment before andduring monocular deprivation (leading to transgene suppression)restored ODP and induced a robust ocular dominance shift (Fig. 4e),similar to that seen in wild-type littermates. The rescue was not due to anon-specific effect of doxycycline, as doxycycline itself had no effect onODP in wild-type or CNO mice (Fig. 4b–d).

Weighted ocular dominance (WOD) scores for individual animalsare shown in Figure 4f. Without monocular deprivation (opensymbols), all mice had similar low WOD scores whether treated ornot treated with doxycycline (wild-type, 0.29 7 0.02; wild-type withdoxycycline, 0.27 7 0.01; CNO, 0.27 7 0.02; CNO with doxycycline,0.29 7 0.01; t-test; P ¼ 0.38, 0.88, 0.85, respectively), as in previousstudies23,24,29, indicating that calcineurin overexpression or doxycyclinetreatment had no effect on ocular dominance distribution in non-deprived animals. Following monocular deprivation (solid symbols),both wild-type groups showed a significant increase in WOD scores(with doxycycline ¼ 0.55 7 0.02, without doxycycline ¼ 0.55 7 0.02).CNO mice without doxycycline had only a small increase in WODscores (0.33 7 0.01; P ¼ 0.05 relative to non-deprived wild-type),

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

Nucleu

s

Synap

toso

mes

Cytopla

sm

Rat

io c

alci

neur

inA

/β–a

ctin WT

CNO*P = 0.042

*P < 0.01

NucleusSynaptosomesCytoplasm

β-actin

Calcineurin A

M C M C M C

CNO on doxCNOWT0

10

20

30

40

50

60

70

Cal

cine

urin

act

ivity

(nm

ol P

i/min

/mg)

a

b

c

7654321

76543217654321

7654321

Ocular dominance category Ocular dominance category

60

40

20

0

60

40

20

0

60

40

20

0

60

40

20

0

Per

cent

age

of c

ells

Per

cent

age

of c

ells

n = 7 mice

n = 6 mice

182 cells

176 cells

CNO: MD

n = 5 mice141 cellsCNO: ND

WT: MD

n = 5 mice147 cellsWT: NDa b

c d

Figure 2 ODP is impaired by calcineurin overexpression. (a) Ocular

dominance distribution in non-deprived (ND) wild-type (WT) mice. (b) After

monocular deprivation (MD) initiated on P24, ocular dominance distribution

shifts to the right in WT mice. (c) Without MD, CNO mice show an ocular

dominance distribution similar to WT mice. (d) After MD, CNO mice show

no ocular dominance shift.

Figure 1 Calcineurin activity is increased in the visual cortex of CNO mice.

(a) Enzyme activity in extracts from visual cortex of P28–29 CNO and wild-

type (WT) mice. WT: 36.8 7 1.8 nmol Pi min�1 mg�1, n ¼ 4; CNO mutants:

54.5 7 4.0 nmol Pi min�1 mg�1, n ¼ 3, P o 0.01; CNO on doxycycline:

36.2 7 5.4 nmol Pi min�1 mg�1, n ¼ 4, P 4 0.05. Doxycycline was

administered at least 7 d before tissue collection. (b) Representative

examples of bands observed on immunoblots in three homogenate

fractions from mouse cortex using an antibody against calcineurin A(M: CNO mutants, n ¼ 5; C: WT, n ¼ 3). (c) Relative calcineurin content

normalized to b-actin (43 kDa). For all measures, data is expressed as

mean 7 s.e.m., if not stated otherwise.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

which was significantly lower than in deprived wild-type littermates(t-test, P o 0.0001). However, with both monocular deprivation anddoxycycline treatment, WOD scores in CNO mice were similar to thosein wild-type littermates (0.56 7 0.01, t-test, P ¼ 0.62) but significantlydifferent from those in CNO mice without doxycycline (t-test,P o 0.0001), indicating a specific effect of calcineurin overexpressionthat was abolished by transgene suppression.

The impairment is not due to a developmental deficit

Calcineurin protein is not detectable until P4, although its mRNA isfound as early as P1 in rat brain30. To exclude the possibility that anincrease in calcineurin activity owing to transgene expression duringearly brain development (before the critical period) alters monoculardeprivation–induced plasticity, we suppressed calcineurin overexpres-sion at birth. Suppression was maintained either through development(‘dox on/on’ group), or calcineurin activity was elevated selectivelyduring the critical period (‘dox on/off ’ group)31 (Fig. 5). In CNO micewith endogenous levels of calcineurin (because of permanent transgenesuppression), ocular dominance distribution after deprivation shiftedto the open eye, similarly to that in wild-type littermates (Fig. 5a).However, after restoration of transgene expression, we did not observea shift in CNO mice after monocular deprivation (Fig. 5b, left). Bothwild-type groups showed identical WOD scores (Fig. 5c, dox on/on,0.56 7 0.02; dox on/off, 0.52 7 0.02). WOD values in the CNO groupswere significantly different (Fig. 5c, dox on/on, 0.567 0.03; dox on/off,0.35 7 0.01; t-test, P o 0.001), indicating that an overexpression ofcalcineurin during the critical period is sufficient to impair ODP (see

also Supplementary Fig. 1 online confirming that ODP is impaired bycalcineurin even with an intact onset of the critical period).

Further, WOD scores in doxycycline on/off conditions were similarto those in non-treated conditions for both CNO and wild-type groups(compare Fig. 5b with Fig. 2b,d; P ¼ 0.37 for wild-type, P ¼ 0.19 forCNO), suggesting that the impairment in ocular dominance shift inCNO mice was not due to any developmental anomaly induced beforethe critical period. Consistently, the retinotopic map of V1 (Fig. 6a),the size of receptive fields (Fig. 6b), response strength and signal-to-noise ratio (Fig. 6c) were normal in CNO animals.

Plasticity can be rescued in later age

To examine whether calcineurin overexpression may alter the closing ofthe critical period and, if so, whether plasticity may be rescued later inlife, we started doxycycline treatment at P38 and performed monoculardeprivation 1 week later (P45). Four to five days of deprivation did notinduce any shift in wild-type mice (Fig. 7a,c, WOD ¼ 0.28 7 0.03compared to WOD ¼ 0.26 (in ref. 29). However, in CNO mice, a lowbut significant shift was observed (Fig. 7b,c; WOD ¼ 0.43 7 0.06,P ¼ 0.003, t-test), suggesting that calcineurin overexpression inter-fered with the closing of the critical period and thereby extended thecritical period.

DISCUSSION

ODP is a form of sensory experience-dependent plasticity that has beenextensively studied but whose molecular mechanisms remain poorly

11882

WTCNOWTCNOP24 P33–34

**176

182

0.7

0.5

0.3

0.1

WO

D s

core

s

Figure 3 The impairment is not due to a delayed plasticity in CNO mutant

mice. Columns show the effect of MD initiated during (P24, left) or after

(P33–34, right) the critical period on WOD scores in CNO (n ¼ 7 and

3 mice, respectively) and WT mice (n ¼ 6 and 4 mice, respectively).

The number of cells is labeled above the column. **, P o 0.001, t-test.

Conventions as in Figure 2.

0.290.27

0.33

0.270.29

0.55 0.560.55

CNOWTCNOWTCNOWTCNOWTMD+doxND+doxND MD

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Wei

ghte

d oc

ular

dom

inan

ce s

core

s

Ocular dominance categoryOcular dominance category

7654321

76543217654321

7654321

60

40

20

0

60

40

20

0

Per

cent

age

of c

ells

Per

cent

age

of c

ells

CNO: MD + dox

WT: MD + dox

CNO: ND + dox164 cells

111 cells

137 cellsn = 6 mice

n = 4 mice

WT: ND + dox89 cells

n = 3 mice

n = 5 mice

60

40

20

0

60

40

20

0

P28–32P24P17P0

DoxycyclineMD or ND

b

a

c

d e

f

Figure 4 Plasticity is rescued by transgene suppression in CNO mice.

(a) Suppression of calcineurin overexpression by doxycycline treatment

starting on P17. (b) Doxycycline-treated WT littermates without deprivation

(ND). (c) Doxycycline-treated WT littermates subjected to MD.

(d) Doxycycline-treated CNO mice without deprivation. (e) After MD,

doxycycline-treated CNO mice show a dramatic shift to the open eye. (f) WOD

scores for individual animals. Without doxycycline treatment, non-deprived

animals (open symbols) show low WOD scores (WT, open circles; n ¼ 5 mice;

CNO, open triangles; n ¼ 5 mice). After MD (filled symbols), WT mice show

increased WOD scores (filled circles, n ¼ 6 mice), wheareas CNO mice show

impaired plasticity (filled triangles, n ¼ 7 mice). After doxycycline treatment,

WOD scores of non-deprived WT and CNO animals have no significant change

(WT, open squares; n ¼ 3 mice; CNO, open inverted triangles; n ¼ 5 mice).After MD, doxycycline-treated CNO animals show high WOD scores (filled

inverted triangles; n ¼ 6 mice) similar to those in WT littermates (filled

squares, n ¼ 4 mice; P o 0.001). Each symbol represents the WOD score

for a single mouse. Mean values are shown by horizontal bars with numbers.

Conventions as in Figure 2.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

understood. This study provides evidence that the protein phosphatasecalcineurin is involved in ODP by showing that an excess of calcineurinblocks the shift in ocular dominance normally induced by monoculardeprivation. The impairment was induced by a transient increase incalcineurin activity during the critical period and could be rescued byrestoring normal calcineurin activity, owing to the inducibility andreversibility of the genetic manipulation. These results are consistentwith previous studies showing that downregulation of protein kinaseactivity impairs ODP3–5 and emphasize the importance of the interplaybetween protein kinases and phosphatases in this form of plasticity.

As calcineurin is associated with PKA through anchoring proteins, itis likely to affect similar substrates during ODP. One of these substratesmay be inhibitor-1 (I-1), a PP1 inhibitor activated by phosphorylationby PKA and blocked by dephosphorylation by calcineurin. Oncecalcineurin is stimulated by Ca2+ influx through NMDA receptors, itcan activate PP1 via relief of I-1–mediated inhibition. Calcineurin canalso stimulate the striatal-enriched protein tyrosine phosphatase(STEP), another protein phosphatase that is inactive in basal condi-

tions32. A potential molecular cascade involved in ODP may thusengage calcineurin and PKA in a gate that controls PP1 activity and, inturn, the autophosphorylation of CaMKII (ref. 4,33), a critical player inODP. Calcineurin and PKA may also gate the ERK pathway bymodulating ERK activity through direct phosphorylation or dephos-phorylation34, or by interfering with its translocation through STEP32.An ultimate component of the cascade may implicate CREB andCREB-dependent gene expression in the nucleus, as ODP requiresCREB28 activation and protein synthesis35. Control of CREB-mediatedtranscription depends on CREB phosphorylation or dephosphoryla-tion at Ser133 by protein kinases including PKA, CaMKIV, MAPK andpossibly the protein phosphatase PP1, possibly activated by calci-neurin26,36. Calcineurin also modulates intracellular Ca2+ by reducingNMDA receptor current decay time and regulating intracellular Ca2+

release37, thereby controlling other Ca2+-dependent enzymes. Thus,calcineurin may be involved in the control of signaling from thesynapse to the nucleus during ODP. This is also consistent with thefact that in our mutants calcineurin is essentially overexpressed incytoplasmic fractions and not at synapses and is therefore more likelyto perturb intracellular signaling pathways than membrane receptors.

Recent work suggests that ODP induction and expression machinerymay be different. Benzodiazepines restore the onset of the criticalperiod, which is altered in GAD65�/� mice, but appear not to intervenein the expression of ODP38. Further, GABAA a1 knock-in mice shownormal expression of ODP, although the transition to the precociouscritical period by benzodiazepines is impaired39 because benzodiaze-pines act through this subunit of GABA receptors. Thus, GABAergicinhibition seems to be uniquely responsible for the inductionmachinery. In our mutant mice, however, ODP is impaired by an

CNO WTCNO WT

MD + dox on/offMD + dox on/on

0.1

0.2

0.3

0.4

0.5

0.6

0.7

WO

D s

core

s

0.35

0.520.560.56

1 2 3 4 5 6 7

1 2 3 4 5 6 71 2 3 4 5 6 7

1 2 3 4 5 6 7

Ocular dominance categoryOcular dominance category

60

40

20

0

60

40

20

0

60

40

20

0

60

40

20

0

Per

cent

age

of c

ells

Per

cent

age

of c

ells

n = 4 mice116 cells

WT: dox on/off

n = 4 mice121 cells

WT: dox on/on

n = 4 mice104 cells

CNO: dox on/on

n = 3 mice75 cells

CNO: dox on/off

P0

P0

P17P24

P24

P28–32

P28–32

MD

MDDoxycycline

Doxycycline

a

b

c

CNOWTCNOWT CNOWT0

1

2

3

0

1

2

3

Sig

nal/n

oise

rat

io

Res

pons

e st

reng

th

0

5

10

15

RF

siz

e (d

eg)

0.60.40.20.0

L–M position (mm)

0

10

20

30

RF

azi

mut

h (d

eg)

WT CNO

0.6

0.8

1.0

Correlationcoefficients

a

cbFigure 6 Response properties are normal in CNO mutant mice.

(a) Retinotopic mapping in WT mice (open circles, n ¼ 15 units in three

mice) is similar to that in CNO animals (+; n ¼ 20 units in four mice). Each

symbol represents the average receptive field azimuth of a single penetration.

Inset: correlation coefficients for three WT and four CNO regressions

(P ¼ 0.5, t-test). (b) Size of receptive field (RF; WT: n ¼ 118 cells in fourmice; CNO: n ¼ 128 cells in five mice) is unchanged (P ¼ 0.2, t-test).

(c) Response strength (left) and ratio of signal and noise (right) are similar in

WT and CNO animals (WT, n ¼ 118 cells; CNO, n ¼ 128 cells, P 4 0.1 for

both, t-test). Conventions as in Figure 2.

Figure 5 Calcineurin overexpression only during the critical period impairs

ODP. (a) Doxycycline was administered to CNO and WT mice from P0 to the

date of recording (‘dox on/on’). MD induced a similar ocular dominance shift

in CNO mutants (left) and WT littermates (right). (b) Doxycycline was

removed 1 week before MD (P17, ‘dox on/off’). No shift was observed in CNO

mice (left), and plasticity was normal in WT mice (right). (c) WOD scores for

individual animals. In dox on/on group, MD increases WOD scores in both

CNO (filled circles) and WT (open triangles) mice. In contrast, CNO mice inthe MD dox on/off group (filled circles) have significantly lower WOD scores

than WT littermates (filled triangles, P o 0.001, t-test). Each symbol

represents the WOD score for a single mouse. Mean values are shown by

horizontal bars with numbers. Conventions as in Figure 2.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

increase in calcineurin activity with an intact onset of the critical period(Supplementary Fig. 1, Fig. 5). Therefore, calcineurin may regulateODP through mechanisms other than GABA inhibition during thecritical period. Our results also show that calcineurin overexpressionmay disturb the closure of the critical period. The mechanisms ofclosure are not well understood but may involve the maturation ofextracellular matrix40 and intracortical myelination. Recent experi-ments reveal that ODP persists into adulthood in mice deficient formyelin-associated protein Nogo-A or its receptor (A.W. McGee, Y.Y.,N.W.D. and S.S. Strittmatter, unpublished data). Thus, it could be ofinterest to examine whether the synthesis of these molecules is alteredin CNO mice, as well as in other kinase-deficient mutant mice.

Short deprivation episodes during the critical period can inducestructural changes in the brain. For instance, spine motility is signifi-cantly increased after monocular deprivation41. In the hippocampus,calcineurin is known to induce shrinkage of dendritic spines, possiblyby dephosphorylating cofilin and perturbing cofilin-actin polymeriza-tion42. Together with PKA, it is also believed to control neuriteextension and axonal regeneration43,44. Therefore, it is possible thatcalcineurin overexpression in our model impairs visual cortex plasticityby interfering with structural changes necessary for such plasticity.Further investigations would be necessary to examine this point.

ODP has been suggested to share mechanisms with LTD in the visualcortex45. However, previous work showed that calcineurin inhibitionby FK506 impairs LTD in visual cortex17, and calcineurin activityincreases during LTD in vivo. Calcineurin overexpression in our mutantmice did not change LTD in layer II/III in visual cortex (Supplemen-tary Fig. 2), consistent with its reported lack of effect on LTD in thehippocampus in the same mice7. This suggests that LTD and ODP maynot be directly related or have a complex relationship as suggested by

several other studies23,46,47. It may also be explained by the existence ofseveral forms of LTD in different cortical layers (see review, ref. 48).

Finally, we did not observe any ODP in older wild-type animals(Fig. 3), similar to results from other single-unit studies4,29. Thiscontrasts, however, with recent reports describing the existence of anadult form of visual plasticity measured by visually-evoked poten-tials49,50. This adult plasticity is strongly suppressed by nembutal50 andmay therefore not be observed in our experimental conditions.

METHODSGeneration and maintenance of Tet-CN279 transgenic mice. All procedures

used in this study were approved by the Animal Care and Use Committee at

Yale University and conform to the guidelines of the National Institutes of

Health and The Society for Neuroscience. Tet-CN279 transgenic mice (TetO

promoter-DCaM-AI transgene) were generated by microinjection of a linear

DNA construct into fertilized one-cell eggs as previously described19. The

founder mouse was backcrossed to C57BL6/J mice for 9–10 generations to

generate heterozygous offspring and then crossed to heterozygous CaMKIIapromoter-tTA mice to generate Tet-CN279 mice. Genotyping was performed by

PCR on tail DNA. Tet-CN279 double mutants (CNO) and wild-type littermates

were used. The animals were maintained in the facility according to standard

protocols. CNO and wild-type littermates were fed with normal food or with

food supplemented with 2 mg kg�1 of doxycycline (Research Diets).

In vivo electrophysiology. Electrophysiological recordings were performed

under nembutal/chlorprothixene (50 mg kg�1, i.p., Abott Laboratories;

10 mg kg�1, i.m., Sigma) anesthesia using standard procedures24. Atropine

(20 mg kg�1 s.c., Optopics) was injected to reduce secretions and parasympa-

thetic effects of anesthetic agents, and dexamethasone (4 mg kg�1 s.c.,

American Reagent Laboratories) was administered to reduce cerebral edema.

Mice were placed in a stereotaxic device, and a tracheal tube and intraperitoneal

cannulae were inserted. A craniotomy was made over the right visual cortex,

and agar was applied to enhance recording stability and prevent desiccation.

Eyelids were removed from both eyes, and corneas were protected thereafter by

frequent application of silicon oil. Body temperature was maintained at 37 1C

by a homeostatically controlled heating pad. Heart rate and respiration were

monitored continuously.

Four to six cells (490 mm apart) through the full thickness of the cortex

were evaluated in each of four to six penetrations spaced evenly (at least

200 mm apart) crossing the binocular region (azimuth o251) of area 17 to avoid

sampling bias. Cells were assigned to ocular dominance categories according to

the seven-category scheme of Hubel and Wiesel. Ocular dominance histograms

were constructed and WOD scores were calculated for each mouse with the

formula: WOD ¼ (1/6G2 + 2/6G3 + 3/6G4 + 4/6G5 + 5/6G6 + G7)/N, where

Gi is the number of cells in ocular dominance groups, and N is the total

number of cells. Normal mice have an average WOD of about 0.28; that is, they

are dominated by the contralateral eye. Response quality was assessed by rating

the level of visually driven and spontaneous activity, each on a three-point scale

(1 ¼ low, 3 ¼ high).

Monocular deprivation. Lid suture of the left eye was performed under 1–2%

halothane anesthesia on postnatal day 24 (P24) or P33 for all mice receiving

deprivation. Lid margins were trimmed and lids sutured together using 6-0 silk.

Experiments were performed blind to genotype and drug treatment.

Phosphatase assay. Animals were killed by decapitation after anesthesia with

halothane. Phosphatase assays were performed using an assay kit (Calbiochem).

Pooled binocular visual cortices (L2–4 mm, P0–2mm) were homogenized and

centrifuged. After desalting to remove free phosphates, supernatants were

diluted in 50mM Tris (pH 7.5), 1 mM DTT, 100 mM EDTA, 100 mM EGTA,

0.2% NP-40 and were incubated at 30 1C for 30 min in reaction buffer. Okadaic

acid (1 mM) was added to inhibit PP1 and PP2A activity. PKA regulatory

subunit type II was used as substrate (DLDVPIPRFDRRV-pSer-VAAE). Calci-

neurin activity was expressed in nmol Pi released min�1 mg�1 protein. The

protein concentration was measured using a Biorad protein assay. All measures

were performed in triplicate.

MD

CNO+doxWTWT

ND

0.1

0.2

0.3

0.4

0.5

0.6 P45

WO

D s

core

s

7654321

7654321

Ocular dominance category

Ocular dominance category

0

10

20

30

40

50

Per

cent

age

of c

ells

0

10

20

30

40

50

Per

cent

age

of c

ells

CNO: dox + MD at P45

WT: MD at P45

4 mice102 cells

WOD = 0.43

4 mice118 cells

WOD = 0.28a

b

c

Figure 7 Restoring normal calcineurin action induces ODP in older mice.

(a) Brief MD does not induce an ocular dominance shift in WT mice (P45).

(b) Doxycycline treatment at P38 (7 d before MD) results in an ocular

dominance shift in CNO mice (P o 0.0001, w2 test). (c) Average WOD score

for non-deprived and deprived WT mice and deprived doxycycline-treated

CNO mice at P45. Conventions as in Figure 2.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Immunoblotting. Homogenates from cortex were subjected to sucrose gradi-

ent centrifugation, and cytoplasmic, nuclear and crude synaptosomal fractions

were collected. Samples were separated by 10% SDS-PAGE and then transferred

to nitrocellulose membranes. After 1 h blocking (2% goat serum) at 22 1C,

membranes were incubated with 1:6,000 anti-calcineurin (Chemicon, ab1695)

and 1:3,000 anti–b actin (Sigma, A-5316) for 1 h, washed, then further

incubated in 1:6,000 anti-rabbit (Upstate Biotechnology 12-348) and 1:3,000

anti-mouse horseradish peroxidase–conjugated secondary antibody (Upstate

Biotechnology 12-349). Horseradish peroxidase was detected by adding 300 ml

chemiluminescence reagent (Perkin Elmer Western Lightning) and exposing

membranes to Kodak MR films. Quantification was conducted using the Image

software. Immunoblots were prepared in duplicate or triplicate, and results

were averaged.


ACKNOWLEDGMENTSThis work was supported by Public Health Service Grant R01 EY00053 andthe Connecticut Lions Eye Research Foundation. I.M.M. is supported by theUniversity of Zurich, the Swiss Federal Institute of Technology, the SwissNational Science Foundation, Swiss National Centre for Competence in Research‘Neural Plasticity and Repair’, European Molecular Biology Organization, HumanFrontier Science Program. N.W.D. is a Senior Science Investigator of Research toPrevent Blindness. We thank D. Winder for providing the mice, A. LaRue forhelp with PCR, R. Munton for help with synaptosomal preparations, and Y. Raofor help with the LTD measurements.




1. Roberts, E.B., Meredith, M.A. & Ramoa, A.S. Suppression of NMDA receptor functionusing antisense DNA block ocular dominance plasticity while preserving visualresponses. J. Neurophysiol. 80, 1021–1032 (1998).

2. Daw, N.W. et al. Injection of MK-801 affects ocular dominance shifts more than visualactivity. J. Neurophysiol. 81, 204–215 (1999).

3. Beaver, C.J., Ji, Q., Fischer, Q.S. & Daw, N.W. Cyclic AMP-dependent protein kinasemediates ocular dominance shifts in cat visual cortex. Nat. Neurosci. 4, 159–163(2001).

4. Taha, S., Hanover, J.L., Silva, A.J. & Stryker, M.P. Autophosphorylation of alphaCaMKIIis required for ocular dominance plasticity. Neuron 36, 483–491 (2002).

5. Di Cristo, G. et al. Requirement of ERK activation for visual cortical plasticity. Science292, 2337–2340 (2001).

6. Mansuy, I.M. Calcineurin in memory and bidirectional plasticity. Biochem. Biophys. Res.Commun. 311, 1195–1208 (2003).

7. Winder, D.G. & Sweatt, J.D. Roles of serine/threonine phosphatases in hippocampalsynaptic plasticity. Nat. Rev. Neurosci. 2, 461–474 (2001).

8. Klee, C.B., Crouch, T.H. & Krinks, M.H. Calcineurin: a calcium- and calmodulin-bindingprotein of the nervous system. Proc. Natl. Acad. Sci. USA 76, 6270–6273 (1979).

9. Goto, S. et al. Cellular colocalization of calcium/calmodulin-dependent protein kinase IIand calcineurin in the rat cerebral cortex and hippocampus. Neurosci. Lett. 149,189–192 (1993).

10. Goto, S., Singer, W. & Gu, Q. Immunocytochemical localization of calcineurin in theadult and developing primary visual cortex of cats. Exp. Brain Res. 96, 377–386(1993).

11. Lieberman, D.N. & Mody, I. Regulation of NMDA channel function by endogenousCa(2+)-dependent phosphatase. Nature 369, 235–239 (1994).

12. Tong, G., Shepherd, D. & Jahr, C.E. Synaptic desensitization of NMDA receptors bycalcineurin. Science 267, 1510–1512 (1995).

13. Ehlers, M.D. Reinsertion or degradation of AMPA receptors determined by activity-dependent endocytic sorting. Neuron 28, 511–525 (2000).

14. Lee, H.K., Barbarosie, M., Kameyama, K., Bear, M.F. & Huganir, R.L. Regulation ofdistinct AMPA receptor phosphorylation sites during bidirectional synaptic plasticity.Nature 405, 955–959 (2000).

15. Genoux, D. et al. Protein phosphatase 1 is a molecular constraint on learning andmemory. Nature 418, 970–975 (2002).

16. Morishita, W. et al. Regulation of synaptic strength by protein phosphatase 1. Neuron32, 1133–1148 (2001).

17. Torii, N., Kamishita, T., Otsu, Y. & Tsumoto, T. An inhibitor for calcineurin, FK506,blocks induction of long-term depression in rat visual cortex. Neurosci. Lett. 185, 1–4(1995).

18. Funauchi, M., Haruta, H. & Tsumoto, T. Effects of an inhibitor for calcium/calmodulin-dependent protein phosphatase, calcineurin, on induction of long-term potentiation inrat visual cortex. Neurosci. Res. 19, 269–278 (1994).

19. Mansuy, I.M., Mayford, M., Jacob, B., Kandel, E.R. & Bach, M.E. Restricted andregulated overexpression reveals calcineurin as a key component in the transition fromshort-term to long-term memory. Cell 92, 39–49 (1998).

20. Antoni, F.A. et al. Ca2+/calcineurin-inhibited adenylyl cyclase, highly abundant inforebrain regions, is important for learning and memory. J. Neurosci. 18, 9650–9661(1998).

21. Coghlan, V.M. et al. Association of protein kinase A and protein phosphatase 2B with acommon anchoring protein. Science 267, 108–111 (1995).

22. Oliveria, S.F., Gomez, L.L. & Dell’Acqua, M.L. Imaging kinase–AKAP79–phosphatasescaffold complexes at the plasma membrane in living cells using FRET microscopy.J. Cell Biol. 160, 101–112 (2003).

23. Rao, Y. et al. Reduced ocular dominance plasticity and long-term potentiation indeveloping visual cortex of protein kinase A RIIa mutant mice. Eur. J. Neurosci. 20,837–842 (2004).

24. Fischer, Q.S. et al. Requirement for the RIIbeta isoform of PKA, but not calcium-stimulated adenylyl cyclase, in visual cortical plasticity. J. Neurosci. 24, 9049–9058(2004).

25. Goto, S. et al. Cellular localization of type II Ca2+/calmodulin-dependent protein kinasein the rat basal ganglia and intrastriatal grafts derived from fetal striatal primordia, incomparison with that of Ca2+/calmodulin-regulated protein phosphatase, calcineurin.Neuroscience 62, 695–705 (1994).

26. Bito, H., Deisseroth, K. & Tsien, R.W. CREB phosphorylation and dephosphorylation:a Ca(2+)- and stimulus duration-dependent switch for hippocampal gene expression.Cell 87, 1203–1214 (1996).

27. Pham, T.A., Impey, S., Storm, D.R. & Stryker, M.P. CRE-mediated gene transcriptionin neocortical neuronal plasticity during the developmental critical period. Neuron 22,63–72 (1999).

28. Mower, A.F., Liao, D.S., Nestler, E.J., Neve, R.L. & Ramoa, A.S. cAMP/Ca2+ responseelement-binding protein function is essential for ocular dominance plasticity.J. Neurosci. 22, 2237–2245 (2002).

29. Gordon, J.A. & Stryker, M.P. Experience-dependent plasticity of binocular responses inthe primary visual cortex of the mouse. J. Neurosci. 16, 3274–3286 (1996).

30. Polli, J.W., Billingsley, M.L. & Kincaid, R.L. Expression of the calmodulin-dependent protein phosphatase, calcineurin, in rat brain: developmental patternsand the role of nigrostriatal innervation. Brain Res. Dev. Brain Res. 63, 105–119(1991).

31. Robertson, A., Perea, J., Tolmachova, T., Thomas, P.K. & Huxley, C. Effects of mousestrain, position of integration and tetracycline analogue on the tetracycline conditionalsystem in transgenic mice. Gene 282, 65–74 (2002).

32. Paul, S., Nairn, A.C., Wang, P. & Lombroso, P.J. NMDA-mediated activation of thetyrosine phosphatase STEP regulates the duration of ERK signaling. Nat. Neurosci. 6,34–42 (2003).

33. Giese, K.P., Fedorov, N.B., Filipkowski, R.K. & Silva, A.J. Autophosphorylation atThr286 of the alpha calcium-calmodulin kinase II in LTP and learning. Science 279,870–873 (1998).

34. Cancedda, L. et al. Patterned vision causes CRE-mediated gene expression in the visualcortex through PKA and ERK. J. Neurosci. 23, 7012–7020 (2003).

35. Taha, S. & Stryker, M.P. Rapid ocular dominance plasticity requires cortical but notgeniculate protein synthesis. Neuron 34, 425–436 (2002).

36. Groth, R.D., Dunbar, R.L. & Mermelstein, P.G. Calcineurin regulation of neuronalplasticity. Biochem. Biophys. Res. Commun. 311, 1159–1171 (2003).

37. Cameron, A.M. et al. Calcineurin associated with the inositol 1,4,5-trisphosphatereceptor-FKBP12 complex modulates Ca2+ flux. Cell 83, 463–472 (1995).

38. Iwai, Y., Fagiolini, M., Obata, K. & Hensch, T.K. Rapid critical period induction by tonicinhibition in visual cortex. J. Neurosci. 23, 6695–6702 (2003).

39. Fagiolini, M. et al. Specific GABAA circuits for visual cortical plasticity. Science 303,1681–1683 (2004).

40. Berardi, N., Pizzorusso, T. & Maffei, L. Extracellular matrix and visual cortical plasticity:freeing the synapse. Neuron 44, 905–908 (2004).

41. Oray, S., Majewska, A. & Sur, M. Dendritic spine dynamics are regulated bymonocular deprivation and extracellular matrix degradation. Neuron 44, 1021–1030(2004).

42. Zhou, Q., Homma, K.J. & Poo, M.M. Shrinkage of dendritic spines associated with long-term depression of hippocampal synapses. Neuron 44, 749–757 (2004).

43. Lautermilch, N.J. & Spitzer, N.C. Regulation of calcineurin by growth cone calciumwaves controls neurite extension. J. Neurosci. 20, 315–325 (2000).

44. Fansa, H. et al. Stimulation of Schwann cell proliferation and axonal regeneration byFK 506. Restor. Neurol. Neurosci. 16, 77–86 (2000).

45. Heynen, A.J. et al. Molecular mechanism for loss of visual cortical responsivenessfollowing brief monocular deprivation. Nat. Neurosci. 6, 854–862 (2003).

46. Hensch, T.K. et al. Comparison of plasticity in vivo and in vitro in the developing visualcortex of normal and protein kinase A RIbeta-deficient mice. J. Neurosci. 18, 2108–2117 (1998).

47. Renger, J.J. et al. Experience-dependent plasticity without long-term depression by type2 metabotropic glutamate receptors in developing visual cortex. Proc. Natl. Acad. Sci.USA 99, 1041–1046 (2002).

48. Daw, N., Rao, Y., Wang, X.F., Fischer, Q. & Yang, Y. LTP and LTD vary with layer in rodentvisual cortex. Vision Res. 44, 3377–3380 (2004).

49. Sawtell, N.B. et al. NMDA receptor-dependent ocular dominance plasticity in adultvisual cortex. Neuron 38, 977–985 (2003).

50. Pham, T.A. et al. A semi-persistent adult ocular dominance plasticity in visual cortex isstabilized by activated CREB. Learn. Mem. 11, 738–747 (2004).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Ovarian cycle–linked changes in GABAA receptorsmediating tonic inhibition alter seizure susceptibilityand anxiety

Jamie L Maguire, Brandon M Stell, Mahsan Rafizadeh & Istvan Mody

Disturbances of neuronal excitability changes during the ovarian cycle may elevate seizure frequency in women with catamenial

epilepsy and enhance anxiety in premenstrual dysphoric disorder (PMDD). The mechanisms underlying these changes are

unknown, but they could result from the effects of fluctuations in progesterone-derived neurosteroids on the brain. Neurosteroids

and some anxiolytics share an important site of action: tonic inhibition mediated by d subunit–containing GABAA receptors

(dGABAARs). Here we demonstrate periodic alterations in specific GABAAR subunits during the estrous cycle in mice, causing

cyclic changes of tonic inhibition in hippocampal neurons. In late diestrus (high-progesterone phase), enhanced expression of

dGABAARs increases tonic inhibition, and a reduced neuronal excitability is reflected by diminished seizure susceptibility and

anxiety. Eliminating cycling of dGABAARs by antisense RNA treatment or gene knockout prevents the lowering of excitability

during diestrus. Our findings are consistent with possible deficiencies in regulatory mechanisms controlling normal cycling of

dGABAARs in individuals with catamenial epilepsy or PMDD.

Fluctuations in neuronal excitability and anxiety levels during theovarian cycle1,2 have been attributed to cyclic changes in endogenousneurosteroid levels1–3. Albeit indirectly, previous reports have suggestedthat changes occur in GABAARs during the ovarian cycle4–6, but thesestudies relied on agonist binding and benzodiazepine sensitivity, whichdo not distinguish between specific subunit combinations of thesereceptors. The brain contains dozens of GABAAR subtypes assembledfrom different subunits that define their kinetics and pharmacology7,8.Of the many GABAAR assemblies, only a few are sensitive to physio-logical concentrations of neurosteroids, limiting the number of sub-types that are ideally suited to mediate changes in neuronal excitabilityduring the ovarian cycle. The dGABAARs, when expressed in expressionsystems, have a high sensitivity to neurosteroids9–11 and the tonicinhibition mediated by these receptors is a preferred target of neuro-steroid action12. The same receptors may also be involved in anxiety,because they are highly ethanol sensitive13,14 and anxiolytic concentra-tions of ethanol preferentially enhance tonic inhibition mediated bydGABAARs15.

The control of neuronal excitability by tonic inhibition involvesthe generation of a steady conductance that is pivotal in reducingthe gain of neuronal input-output functions16. Tonic inhibition is ‘on’most of the time, as it is activated by the ambient low levels of GABApresent in the extracellular space16. Enhancing tonic inhibition byphysiological concentrations of neurosteroids is an effective meansof reducing neuronal excitability12. In contrast, phasic (synaptic)inhibition is intermittently activated at GABAergic synapses whenever

presynaptic terminals release transmitter16. Because phasic inhibition ismediated by g subunit–containing GABAARs, it is not particularlysensitive to physiological levels of neurosteroids12 or to anxiolyticethanol concentrations15.

The high sensitivity of dGABAARs and the tonic inhibition mediatedby these receptors in response to neurosteroids12 and to anxiolyticconcentrations of ethanol15 suggests that this type of inhibition may becrucial in the alterations of neuronal excitability and anxiety associatedwith changes in neurosteroid levels over the ovarian cycle. Little isknown about possible alterations in specific neurosteroid-sensitiveGABAAR subunits during the ovarian cycle, but changes in seizurethreshold consistent with altered neuronal excitability have beenassociated with various stages of the cycle17,18. This may be relevantto the human condition of catamenial epilepsy, in which 75% ofepileptic women report seizure exacerbation related to specific stagesof their menstrual cycle2. The three different forms of catamenialepilepsy—seizures arising at the time of ovulation, seizures arisingimmediately before and during menstruation and seizures arising inpatients with inadequate luteal-phase cycles—all occur when proges-terone levels are low2. The anxiety and dysphoric disorder PMDDfollows a similar pattern of symptom manifestation, occurring beforeand/or during menstruation1.

Here we examined changes in tonic inhibition, seizure susceptibilityand anxiety along with the expression pattern of the highly neuroster-oid-sensitive dGABAARs over the estrous cycle in mice. We show thatspecific GABAAR subunits are periodically reorganized during the


Departments of Neurology and Physiology, The David Geffen School of Medicine, University of California Los Angeles, 710 Westwood Plaza, Los Angeles, California 90095,USA. Correspondence should be addressed to I.M. ([email protected]).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

cycle, resulting in changes in tonic inhibition and neuronal excitability.These changes may be responsible for the observed cyclic alterations inseizure susceptibility and anxiety during the ovarian cycle.

RESULTS

Determination of the estrous cycle in mice

Different stages of the estrous cycle were accurately determined byanalyzing cellular profiles of vaginal smears and by measuring vaginalimpedance (Fig. 1). The average cycle length in C57/Bl6 mice was 7.0 70.35 d (Fig. 1). Plasma progesterone concentrations determined byimmunoassay were correlated to the cycle stage. Circulating progester-one levels were significantly higher at late diestrus (11.06 7 0.76ng ml�1) than in estrus (2.38 7 0.29 ng ml�1; Fig. 1; n ¼ 10; P o0.05). All estrous cycle–related experiments described were carried out inmice with well-established ovarian cycles. Two stages of the estrous cyclewere chosen for comparison: the estrus phase, when estrogen levels arehigh and progesterone levels are low during the day, and the late diestrusphase, when progesterone levels peak and estrogen levels are low19.

Altered GABAAR subunit expression during the ovarian cycle

To determine the effects of the ovarian cycle on GABAAR subunitcomposition, we carried out western blot analysis on membranefractions isolated from the hippocampi of mice in estrus and latediestrus. We measured the abundance of three GABAAR subunits, a4,g2 and d, directly in samples containing the same amount of totalprotein to ensure consistent comparisons across the ovarian cycle.Hippocampal membrane fractions expressed significantly differentlevels of GABAAR subunits in the estrus as compared to the late diestrusphase. The optical density of dGABAAR blots decreased from 0.53 70.01 OD per 100 mg of total protein during late diestrus to 0.37 7 0.04OD per 100 mg at estrus (Fig. 2; n ¼ 6; P o 0.05). GABAAR g2 subunitexpression showed a pattern complementary to that for d subunits. Theoptical density of GABAAR g2 subunits was 0.34 7 0.01 OD per 100 mgof total protein during late diestrus and increased to 0.51 7 0.02 OD

per 100 mg of total protein during estrus (Fig. 2; n¼ 6; Po 0.05). Theseresults demonstrate dynamic alterations in GABAAR subunit composi-tion over the estrous cycle, whereby elevated progesterone levelscorrelated with increased dGABAAR expression and decreasedg2GABAAR expression. In accordance with suggested rules of GABAARpartnership20,21, a4 subunits coassemble with d subunits21,22 or poten-tially with g2 subunits21,23. Moreover, loss of one subunit may result inan increased assembly of the a4 subunits with another partner23.Therefore, it is not surprising that alterations in GABAAR d and g2subunit expression over the estrous cycle were not paralleled by changesin the expression of a4 subunits. Expression of GABAAR a4 subunitwas 0.54 7 0.01 OD per 100 mg for mice in estrus and 0.55 7 0.01 ODper 100 mg for those in late diestrus (Fig. 2; n ¼ 6; P 4 0.05).

Altered GABAAR-mediated tonic inhibition during the cycle

Functional changes in GABAergic inhibition over the cycle wereassessed by recording tonic and phasic inhibitory currents in dentategyrus granule cells (DGGCs) and CA1 pyramidal cells, two cell types inwhich tonic inhibition is mediated by distinct GABAARs12,24. Wemeasured a twofold higher tonic conductance in DGGCs from micein late diestrus as compared to estrus (Fig. 3). The average tonicconductance normalized to whole-cell capacitance was 29.8 7 5.5pS pF�1 in mice in estrus and 57.6 7 10.2 pS pF�1 during late diestrus(Fig. 3; n ¼ 18 cells; n ¼ 6 mice; P o 0.05). However, there was nosignificant difference in the tonic conductance recorded in CA1pyramidal cells (estrus: 38.1 7 8.5 pS pF�1; diestrus: 29.9 7 2.7pS pF�1; Fig. 3; n ¼ 10 cells; n ¼ 5 mice; P ¼ 0.38). The absence ofchanges in the CA1 region, where dGABAAR expression is low22,25 andthese subunits do not contribute to the tonic current in this region12,24,is consistent with a selective regulation of d subunits during the ovariancycle. The idea that other GABAAR subunits are not substantiallyaltered during the cycle is also supported by the lack of significant

Estrus Diestrus

5 µm 5 µm

20

15

10

5

02 4 6

Day of monitoring

Vag

inal

impe

denc

e (k

Ohm

)

8 10 12 14

2.38ng/ml

11.06ng/ml

Pro

estr

us

Est

rus

Met

estr

us

Die

stru

s

Pro

estr

us

Est

rus

Met

estr

us

Die

stru

s

Pro

estr

us

a

b

Estrushippo

Diestrushippo

~73 kDaδ subunit

γ2 subunit

α4 subunit

~47 kDa

~67 kDa

0.6*

0.5

0.4

OD

/100

µg

prot

ein

0.3

0.2

0.1

0δ γ2 α4

Estrus Diestrus

*

a b

Figure 2 Alterations in abundance of GABAAR subunits over the estrous

cycle. (a) Representative immunoblots of total membrane protein isolated

from the hippocampi of mice at estrus or late diestrus. (b) dGABAAR levels

in membrane fractions are expressed as OD per 100 mg of total protein.

Membrane dGABAARs were more abundant, but g2 subunit expression was

lower, during late diestrus than during estrus. There were no differences in

a4 subunit levels between mice in estrus and in late diestrus. *P o 0.05,

unpaired t-test, compared to estrus (n ¼ 5–7 for each experimental group).

Figure 1 Stages of the estrous cycle in C57/Bl6 mice. (a) In Giemsa stains,

the cellular profile of smears obtained during estrus is characterized by the

presence of irregularly shaped cornified cells, whereas the hallmark of

diestrus is the presence of leukocytes and a sparse number of parabasal

cells. (b) The graph shows the average (7 s.e.m.) values of the vaginal

mucosa electrical resistance in eight mice (gray lines) over two complete

cycles. The fit is a sinusoid (thick black line) with the period 7.0 7 0.35 d.

The phases of the cycle are indicated in relationship to the sinusoid, andarrows show the measured levels of progesterone at the two stages relevant to

our experiments.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

changes in the kinetics or frequency of spontaneous inhibitory post-synaptic currents (sIPSCs) observed in either DGGCs or CA1 pyrami-dal cells over the cycle (Fig. 3). In DGGCs sIPSCs, frequency was 2.487 0.69 Hz in estrus and 1.69 7 0.48 Hz in late diestrus (Fig. 3; n ¼ 18cells; n ¼ 6 mice; P ¼ 0.93). The peak amplitude of sIPSCs recordedin diestrus was 114.5 7 20.5 pA as compared to 90.6 7 13.4 pA inestrus. The weighted decay time constant (tw) in diestrus was 15.1 70.91 ms as compared to 16.0 7 0.82 ms in estrus (Fig. 3). Neitherof these differences was statistically significant (P = 0.33 and P = 0.44,respectively, two-sample paired t-test).

Altered seizure susceptibility during the ovarian cycle

To investigate the effects of GABAAR alterations and the resultingchanges in tonic GABAergic inhibition on neuronal excitability, wemeasured seizure susceptibility in response to an intraperitoneal (i.p.)injection of 15 mg kg�1 kainic acid in mice in estrus and diestrus.Consistent with the upregulation dGABAARs during late diestrus andthe consequent increase in tonic inhibition, we observed a decrease in

seizure susceptibility at late diestrus. The latency to seizure onset,measured as the elapsed time to the first seizure after the mice wereinjected with kainic acid, was nearly twofold longer in diestrus than inestrus (Fig. 4, Table 1; n ¼ 9; P o 0.05; see also next section). Mice inestrus spent significantly more cumulative time seizing during 120-minrecording sessions than did mice in diestrus. Mice in estrus hadcumulative seizure times of 79 7 1.5% of 120 min (Fig. 4, Table 1;see also next section), whereas those in late diestrus underwent seizuresduring only 45.2 7 6.8% of the total observation time (Fig. 4, Table 1;n ¼ 9; P o 0.05; see also next section). The average duration of theindividual electrographic events was longer in estrus (9.4 7 1.7 min)than in late diestrus (1.7 7 0.38 min; also see Fig. 6). These resultsare consistent with the idea that alterations in tonic GABAergicinhibition over the estrous cycle result in alternating periods ofincreased seizure susceptibility (during estrus) and periods of seizureresistance (during diestrus).

To confirm the greater dGABAAR expression during diestrus as com-pared to estrus by pharmacologic means, we examined the effects of anon-sedative dose26,27 of 4,5,6,7-tetrahydroisothiazolo-[5,4-c]pyridin-3-ol (THIP or gaboxadol) on seizure susceptibility at the two stages ofthe estrous cycle. In a separate set of experiments (SupplementaryFig. 1) using wild-type and dGABAAR-null mutant (Gabrd�/�)mice, 5 mM THIP produced a much larger tonic current in wild-typeDGGCs than did 5 mM GABA, consistent with its higher efficacy at

Estrus

EstrusEstrus

Diestrus

Diestrus

Diestrus

Dentate

Dentate

CA1

CA1

10 s

0.5 s

25 pA

100 pA

20 ms

40 pA

150

100

50

0

Con

duct

ance

(pS

/pF

)

a

b c*

Estrus Diestrus Estrus + THIP Diestrus + THIP Diestrus +antisense

Diestrus +Gabrd –/–

1 mV

50 s

1 mV

1 s

1 mV

50 s

1 mV

1 s

1 mV

50 s

1 mV

1 s

a b c

Figure 4 Electrographic seizures vary with stages of the estrous cycle and are affected by alterations in dGABAARs. Seizures in response to a single i.p.

injection of 15 mg kg–1 kainic acid were recorded in female mice by hippocampal EEG (shown here band-pass filtered at 0.1–250 Hz). (a) More severe seizure

episodes are seen in the representative traces from mice in estrus as compared to mice in diestrus. (b) The dGABAAR-specific agonist THIP, injected i.p. at a

dose of 10 mg kg–1 30 min before kainic acid injection, attenuates seizures more effectively during late diestrus than during estrus. (c) The involvement of

dGABAAR is shown by the exacerbation of the kainic acid–induced seizures during diestrus in dGABAAR antisense mRNA–treated mice (left) and in Gabrd�/�

mice (right), as evidenced by comparison of these EEG traces to that from a control mouse during diestrus (a).

Figure 3 Tonic conductance is elevated during late diestrus in dentate gyrus

granule cells. Whole-cell patch clamp recordings were performed on dentate

gyrus granule cells (Dentate) and CA1 pyramidal cells (CA1) with the

investigator blinded as to the origin of the slices. (a) Representative

recordings from mice in estrus and in late diestrus indicate that tonic

conductance in DGGC cells is higher in late diestrus, but show no estrous

cycle–dependent changes in tonic currents in CA1 pyramidal cells. Dashed

lines indicate the basal current in the presence of saturating concentrationsof the GABAAR antagonist SR-95531 perfused during the time indicated by

the horizontal black bars. (b) Bar graphs of the average tonic conductance

(normalized for cell capacitance) values in DGGCs show that this is

significant higher in late diestrus than in estrus, whereas no difference was

detected in CA1 pyramidal cells (*P o 0.05, t-test, compared to estrus;

n ¼ 18 cells, 6 mice). (c) Recordings of sIPSCs (phasic currents) show no

changes in frequency, peak amplitude or tw over the estrous cycle. Two

superimposed averaged sIPSCs, each recorded in a DGGC at a different

stage of the estrous cycle, show no changes in the kinetics of the

spontaneous events.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

dGABAARs10. In contrast, there were no differences between THIP andGABA in activating a residual tonic current in Gabrd�/� mice, and thislow concentration of THIP had no effect on the properties of sIPSCs(Supplementary Fig. 1). Our findings on seizure susceptibility demon-strate that, as compared to estrus, THIP was more effective inattenuating seizure activity during late diestrus (Table 1) when thereare more dGABAARs. THIP increased the latency to seizure onset byonly 1.5-fold (1.3 7 0.1 to 2.0 7 0.6 min) in estrus but by more than6-fold (3.2 7 0.7 to 21.6 7 6.3 min) in late diestrus (Fig. 4, Table 1;n ¼ 9; P o 0.05). THIP was also less effective during estrus indecreasing the fraction of time seizing, causing a decrease from 79 71.5% to 73.5 7 1.5% in estrus as compared to a reduction from 45.2 76.8% to 22.67 1.1% in late diestrus (Fig. 4, Table 1; n ¼ 10; P o 0.05).These results demonstrate that the dGABAAR agonist THIP10 is moreeffective during late diestrus, when dGABAAR expression is high.Apparently, doses of THIP that do not cause sedation are most effectiveas anticonvulsants when there is an increased dGABAAR expressionduring late diestrus, presumably because the drug augments the tonicconductance mediated by these receptors and thus further decreasesseizure susceptibility.

Reduction of dGABAARs prevents ovarian cycle–related changes

To ascertain whether the changes seen in tonic inhibition and excit-ability over the estrous cycle result from specific alterations indGABAARs, we investigated cycle-linked changes in seizure suscep-tibility and tonic inhibition in Gabrd�/� mice28 and in wild-type micetreated with dGABAAR antisense oligonucleotides. To demonstrate

knockdown of dGABAARs at diestrus after intracerebroventricularadministration of dGABAAR antisense mRNA, we compared dGA-BAAR expression in hippocampal membranes of mice treated withdGABAAR missense mRNA or with dGABAAR antisense mRNA. Micewere treated with either 5 nmol dGABAAR missense mRNA or 5 nmoldGABAAR antisense mRNA at estrus, and analysis was performedapproximately 2–3 d later, when the mice reached the late diestrusphase. Antisense treatment significantly decreased dGABAAR expres-sion at this phase, from 0.53 7 0.01 OD per 100 mg to 0.34 7 0.01 ODper 100 mg, a level comparable to the expression during estrus (Fig. 5;n ¼ 7; P o 0.05). However, for mice in late diestrus, there was nosignificant difference in dGABAAR expression between negative controlRNA– (0.53 7 0.01 OD per 100 mg) and dGABAAR missense mRNA–treated mice (0.54 7 0.01 OD per 100 mg) (Fig. 5; n ¼ 4). In dGABAARantisense mRNA–treated mice at late diestrus, the optical density of g2subunits was 0.37 7 0.02 OD per 100 mg, which is not significantlydifferent from the 0.39 7 0.01 OD per 100 mg in dGABAAR missensemRNA–treated mice (P ¼ 0.13) or the 0.34 7 0.01 OD per 100 mg incontrol mice (Fig. 5; n ¼ 7; P ¼ 0.23).

We carried out whole-cell patch-clamp recordings performed onDGGCs in hippocampal slices prepared from mice in late diestrus thathad been treated at estrus with 5 nmol negative control RNA or 5 nmoldGABAAR antisense mRNA, and thereby confirmed that tonicGABAergic inhibition was attenuated after dGABAAR antisensemRNA administration. Antisense treatment decreased the tonic con-ductance in DGGCs: in mice in late diestrus, the conductance was 89.97 18.4 pS pF�1 in those treated with negative control RNA and 22.0 76.4 pS pF�1 in those treated with antisense mRNA (Fig. 5; n ¼ 7; P o0.05). Thus, preventing upregulation of dGABAARs that normallyoccurs at late diestrus blocked the enhancement of tonic GABAergicinhibition. We did not observe significant changes in the kinetics orfrequency of spontaneous inhibitory postsynaptic currents (sIPSCs) ofDGGCs in mice treated with negative control or dGABAAR antisensemRNA. In DGGCs, the frequency of sIPSCs was 0.78 7 0.35 Hz fornegative control mRNA–treated mice and 0.63 7 0.11 Hz for antisensemRNA–treated mice (P ¼ 0.37). The peak amplitude of sIPSCsrecorded from mice treated with dGABAAR antisense mRNA was102.9 7 14.1 pA, as compared to 93.1 7 15.4 pA in mice givennegative control mRNA. The tw was 10.0 7 0.61 ms, as compared to11.1 7 0.67 ms in mice given negative control mRNA. Neither of thesedifferences was statistically significant (P ¼ 0.64 and P ¼ 0.25,respectively, two-sample paired t-test).

Diestrusmissense Diestrus missense

Diestrusantisense

Diestrusneg control

Diestrusnegative control

Diestrusantisense

Diestrusantisense

Diestrus antisense

δ subunit~73 kDa

γ2 subunit~47 kDa

0.6

120

100

80

60

40

20

0

0.4

0.2

0δ γ 2

OD

/100

µg

prot

ein

Con

duct

ance

(pS

/pF

)

*

*

20 pA10 ms

a b c d

Figure 5 Decreased dGABAAR expression and tonic GABAergic inhibition after dGABAAR antisense mRNA treatment. (a) Treatment with dGABAAR antisensemRNA results in lower membrane dGABAAR expression, measured by western blotting, as compared to that in mice treated with dGABAAR missense mRNA.

Representative immunoblots from dGABAAR missense– or antisense mRNA–treated diestrus-phase mice demonstrate decreased dGABAAR expression after

antisense treatment. (b) Densitometric quantification of changes in dGABAAR expression in mice in diestrus after dGABAAR antisense or missense treatment,

as compared to control mice in estrus and diestrus (*P o 0.05, ANOVA; n ¼ 4–6 for each experimental group). (c) Representative whole-cell recordings of

tonic currents from DGGCs of mice treated with negative control mRNA and dGABAAR antisense mRNA. Dashed lines indicate basal current (as in Fig. 3).

(d) Tonic GABAergic inhibition is significantly lower in DGGCs of mice in late diestrus 2 d after intraventricular administration of dGABAAR antisense mRNA

as compared to those treated with negative control RNA (*P o 0.05, t-test, n ¼ 17 cells from seven mice for each experimental group).

Table 1 Efficacy of THIP in attenuating seizure susceptibility is

altered over the estrous cycle

Seizure duration

(min)

% time

seizing

Latency to seizure

onset (min) n

Diestrus 1.7 7 0.4 45.2 7 6.8 3.2 7 0.7 4

Estrus 9.4 7 1.7* 81.7 7 2.7* 1.3 7 0.1* 5

Diestrus + THIP 3.1 7 0.7 22.6 7 1.1 27.4 7 4.6 5

Estrus + THIP 6.9 7 0.3* 73.5 7 1.5* 2.0 7 0.6* 5

Seizure susceptibility in response to 15 mg kg�1 i.p. kainic acid was analyzed in wild-type micein estrus and diestrus. The three seizure parameters were measured in mice in estrus anddiestrus with or without administration of 10 mg kg�1 THIP 30 min before kainic acid. * denotesstatistically significant difference (P o 0.05) compared to diestrus.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

To determine whether the decreased seizure susceptibility observedduring diestrus was a direct result of dGABAAR upregulation, weinvestigated seizure susceptibility during diestrus in Gabrd�/� miceand in mice treated with dGABAAR antisense mRNA. Knockdown ofdGABAAR expression in mice in diestrus after antisense treatmentresulted in an increase in seizure susceptibility to a level comparable tothat observed during estrus (Fig. 6). In response to i.p. administrationof 15 mg kg�1 kainic acid, mice in late diestrus pretreated withdGABAAR antisense mRNA spent a significantly larger fraction oftime seizing (81.4 7 3.5%) than those pretreated with dGABAARmissense mRNA (44.7 7 2.7%) (Fig. 6, n ¼ 4; P o 0.05). Antisensetreatment also decreased the latency to seizure onset in late diestrus,from 3.6 7 0.7 min in missense mRNA–treated mice to 1.2 7 0.1 minin dGABAAR antisense mRNA–treated mice (Fig. 6, n ¼ 4; P o 0.05).We measured seizure susceptibility in mice in diestrus only in the totalabsence of dGABAARs. Estrus-phase mice of this seizure-prone phe-notype were not analyzed owing to their already elevated seizuresusceptibility. In the diestrus phase, homozygous Gabrd�/� and hetero-zygous Gabrd+/� mice spent 89.2 7 2.2% and 88.6 7 2.1% of the timeundergoing seizures, respectively, as compared to 45.2 7 6.8% forwild-type mice (Fig. 6; n ¼ 3–5; P o 0.05). In diestrus, the latency toseizure onset was 3.2 7 0.7 min in wild-type mice but only 1.2 7 0.2min in Gabrd�/� mice and to 1.4 7 0.2 min in Gabrd+/� mice (Fig. 6;n ¼ 3–5; P o 0.05). Our findings seem to indicate that the lowerdGABAAR expression during estrus underlies the increased seizuresusceptibility in wild-type mice during estrus, as it does the suscept-ibility in Gabrd�/� mice and in dGABAAR antisense mRNA–treatedmice during diestrus. A similar antisense or missense mRNA treatmentdid not affect seizure susceptibility in mice at the estrus stage of thecycle (Fig. 6). For mice in estrus, the fraction of time spent seizing perrecording session did not differ between those treated with antisensemRNA (76.5 7 6.4%) and those treated with missense mRNA (73.8 71.52%). The latency to onset of electrographic seizures and the averageduration of individual seizure episodes also did not differ between the

two groups (latency 1.5 7 0.19 min, duration 10.7 7 3.1 min forantisense mRNA–treated mice; latency 1.1 7 0.01 min, duration 8.0 71.4 min for missense mRNA–treated mice).

Altered levels of anxiety during the ovarian cycle

In light of the specific effects of anxiolytic ethanol concentrations ondGABAARs13–15 and on the tonic inhibition mediated by these recep-tors15, together with the changes in mood and anxiety during PMDD1

and associated changes in the effects of ethanol29,30, we wanted toexamine the levels of anxiety in mice at different stages of the ovariancycle. We used an elevated plus-maze to measure anxiety levels31 infemale mice with well-established ovarian cycles. Mice in late diestrus, atime when the dGABAARs are upregulated, showed decreased anxiety.Individual mice tested throughout the course of their cycles spent moretime in the open arms of the maze during diestrus than during estrus(Table 2). In contrast, mice in estrus spent more time in the closedarms than mice in diestrus (Table 2). Mice in diestrus spent signifi-cantly more time in the center of the maze (diestrus: 9.26 7 1.83%;estrus: 4.91 7 0.57%; n ¼ 6; P o 0.05) and made significantly moretotal arm entries (diestrus: 37.47 7 3.64; estrus: 25.00 7 2.98; n ¼ 6;P o 0.05). Although the number of total entries was significantlyhigher during late diestrus, analysis of covariance in all experimentalgroups (two groups of females and two groups of males) did not showany significant relationship (P ¼ 0.37) between the number of totalentries and the fraction of entries into open or closed arms. Male micetested daily on the elevated plus-maze over a period of 15 d did notshow fluctuations in anxiety levels. The average fraction of time spentby male mice in the open arms during three successive test periods

Table 2 Effect of the estrous cycle on anxiety evaluated using the

elevated plus-maze

Open arms Closed arms

Entries (% total) Time (% total) Entries (% total) Time (% total)

Estrus 5.68 7 1.11* 1.36 7 1.05* 44.53 7 1.06* 93.71 7 0.69*

Diestrus 9.69 7 2.29 3.80 7 0.32 40.08 7 1.04 86.90 7 1.68

*denotes significant difference (P o 0.05) between mice in estrus and in diestrus (paired t-test;n ¼ 6).

Diestrus Estrus

Diestrus Estrus

Diestrus Estrus

100

75

50

25

0

5

4

3

2

1

0

15

10

5

0

WT WT–/–Gabrd

+/–Gabrd

Anti AntiMis Mis

WT WT–/–Gabrd

+/–Gabrd

Anti AntiMis Mis

WT WT–/–Gabrd

+/–Gabrd

Anti AntiMis Mis

Per

cent

tim

e se

izin

gpe

r 12

0 m

inLa

tenc

y to

onse

t (m

in)

Sei

zure

dura

tion

(min

)* * * * *

*

* *** * *

* **

**

*

a

b

c

Figure 6 Characteristics of kainic acid–induced seizures as a function of

dGABAAR expression during the estrous cycle and after experimental

manipulation of these subunits. (a) The cumulative percent seizure time per120 min after kainic acid administration was lower in diestrus than in estrus

for wild-type mice, but not Gabrd�/� or heterozygous Gabrd+/� mice or mice

treated with dGABAAR antisense mRNA. Mice treated with missense mRNA

showed characteristics similar to those of late diestrus controls. Estrus wild-

type mice were similar to estrus missense mRNA– or antisense mRNA–

treated mice (n ¼ 3–6 mice; *P o 0.05 from unlabeled groups; post-ANOVA

multiple comparisons). (b) The latency to seizure onset was longer in diestrus

than estrus. However, in late diestrus Gabrd�/� or Gabrd+/� mice, or mice

treated with dGABAAR antisense mRNA, had shorter latencies than controls.

During late diestrus, missense mRNA–treated mice were similar to late

diestrus wild-type controls. In the estrus phase, wild-type, missense mRNA–

treated and antisense mRNA–treated mice showed similar latencies to onset.

(n ¼ 3–6 mice; *P o 0.05 as in a). (c) The average duration of single

electrographic seizures was shorter in diestrus than in estrus. Gabrd�/�,

Gabrd+/� and dGABAAR antisense mRNA–treated mice had longer seizure

durations than diestrus controls. In diestrus, missense mRNA–treated mice

were similar wild-type controls. In estrus, wild-type control, missense mRNA–

and antisense mRNA–treated mice had similar seizure durations (n ¼ 3–6mice; *P o 0.05 as in a).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

separated by 7 d was 3.67 7 1.09% (n ¼ 9). This was not significantlydifferent from the values obtained for the same male mice tested duringthe middle of the two 7-d periods to resemble the tests done in females(4.41 7 1.25%; n ¼ 9; P ¼ 0.67, paired t-test). Our data are consistentwith fluctuating anxiety levels in female mice related to the stages of theestrous cycle. Mice in diestrus, when levels of dGABAARs are high, havelower anxiety levels than mice in estrus. The attenuated tonic inhibitionduring estrus resulting from decreased dGABAAR expression corre-sponds to a stage of enhanced anxiety during this phase of cycle.

DISCUSSION

We have identified specific GABAAR subunits that undergo fluctua-tions over the estrous cycle, resulting in changes in tonic inhibitionparalleled by commensurate alterations in seizure susceptibility andanxiety levels. During late diestrus, when circulating levels of proges-terone are high, enhanced expression of dGABAARs and a consequentincrease in tonic inhibition are accompanied by decreased seizuresusceptibility and lowered anxiety. Until now, estrous cycle–relatedalterations in excitability have been attributed to direct effects ofneurosteroids on GABAARs. Our study is the first to focus on cellularand molecular events underlying changes in seizure susceptibilityand anxiety over the course of the cycle, demonstrating cycle-dependent changes in the expression of GABAAR d and g2 subunits.In the future, the precise mechanisms responsible for cycle-related alterations of GABAAR subunits will need to be determined,as there may be links between the subunit changes and fluctuationsin hormone levels.

Estrous cycle related changes in GABAARs

Our findings are consistent with previous studies showing alterationsin muscimol binding and changes in neurosteroid and benzodiazepinesensitivities over the cycle4–6,32. Increased sensitivity to neurosteroids32

and decreased efficacy of benzodiazepines4–6 have been reported atdiestrus, consistent with our findings of increased neurosteroid-sensi-tive10,11 d and decreased benzodiazepine-sensitive, neurosteroid-insen-sitive10,11 g2 subunit expression at diestrus. Previous studies haveshown that withdrawal of female rats from 3-week-long exogenousprogesterone administration results in marked alterations in GABAARsand phasic inhibition in the CA1 region, including increased a4 and dsubunit expression and decreased benzodiazepine sensitivity33,34. Theincreased dGABAAR expression in our study occurs at a time whenprogesterone levels are relatively high during diestrus. Therefore, itseems unlikely that progesterone withdrawal might be responsible forthe changes we observed. We did not find any changes in a4 subunitexpression, but our data demonstrate tightly regulated expression ofGABAAR d and g2 subunits over the course of the estrous cycle,resulting in specific changes in tonic inhibition, seizure susceptibilityand anxiety levels. It seems unlikely that additional hippocampalGABAAR subunits undergo substantial alterations during the cycle,as there were no alterations of the phasic inhibition (sIPSCs) in DGGCsor CA1 pyramidal cells or in the tonic inhibition in the CA1. Thespecificity of the effects of dGABAAR antisense mRNA also diminishthe likelihood that substantial ovarian cycle–related changes occur inother GABAAR subunits.

The complementary regulation of g2 and d subunit expressionduring the estrous cycle is reminiscent of the upregulation of g2subunits in Gabrd�/� mice23. Because g2 subunits are found mostlyat synapses35, we also expected to see changes in the phasic (synaptic)component of inhibition over the estrous cycle. As phasic inhibitionremained constant, g2 subunits may have replaced d subunits at estrussolely at extrasynaptic sites where the continuous presence of GABA

prevented the g2GABAARs from generating tonic currents owing todesensitization10,11. Alternatively, an increase in g2 subunits at den-dritic synapses where spontaneous GABA release is extremely low36

may have gone undetected because we recorded sIPSCs of predomi-nantly somatic origin. The abundance of these two subunits maybe independently regulated by a common ovarian cycle–relatedmechanism: when the increase in d subunits during late diestrus wasstalled by antisense mRNA, g2 subunit levels still decreased.

Tonic inhibition during the estrous cycle

In many neurons in which a tonic current is activated by the GABAlevels present in the extracellular space8,16, the overall charge carried bythe activation of tonically active GABAARs can be more than threetimes larger than that produced by phasic inhibition8. Experimentaland theoretical studies indicate that a continuously active inhibitoryconductance significantly affects excitability and gain control bothin vitro and in vivo16.

Selectively increasing the tonic inhibition of DGGCs by abouttwofold with neurosteroids results in an overall reduction in theneuronal excitability of the dentate gyrus12. The similar twofold largertonic inhibition in diestrus as compared to estrus should substantiallyreduce the throughput of the dentate gyrus during late diestrus. Thisdecreased throughput could be further diminished at diestrus by thepotentiating actions of progesterone-derived neurosteroids on theelevated number of dGABAARs. It is unlikely that elevated neurosteroidlevels could have persisted in acute hippocampal slices to account forthe enhanced tonic currents found in late diestrus. A recent studyinvestigating the relationship between ethanol and neurosteroideffects37 has shown that inhibiting neurosteroid synthesis in acutebrain slices had no bearing on neurosteroid-dependent effects.

The magnitude of the tonic current at late diestrus is comparable tothat found in male mice12. This finding also matches well with thesimilar anxiety levels of male mice and females at late diestrus. As thedentate gyrus is not the only brain area involved in anxiety, futurestudies will need to focus on ovarian cycle–related dGABAAR expres-sion and tonic inhibition in other brain areas enriched in dGABAARs.

Seizure susceptibility during the estrous cycle

Exogenous progesterone administration has anticonvulsant effects inwomen with catamenial epilepsy as well as in experimental models ofepilepsy38–40. Likewise, progesterone withdrawal or inhibition of pro-gesterone metabolism exacerbates seizure susceptibility41,42. Yet fewstudies have focused on the effect of endogenous alterations in neuro-steroid levels on seizure susceptibility. Our study clearly shows changesin seizure susceptibility over the estrous cycle: seizure susceptibility isdecreased during late diestrus, when progesterone levels are elevated anddGABAARs are increased resulting in enhanced tonic inhibition.

In the mouse pilocarpine model of temporal-lobe epilepsy,dGABAAR expression in DGGCs becomes progressively reduced43.In addition, loss-of-function mutations in dGABAARs have beenidentified in patients with generalized epilepsy with febrile seizuresplus44. Thus, dGABAARs may be an important therapeutic target forthe treatment of epilepsy. Our studies show a clear attenuation inseizure activity by the GABAAR agonist THIP, which at low concentra-tions is a preferred agonist at dGABAARs10. This drug is presentlyundergoing clinical trials as a deep sleep–inducing hypnotic45, butlower doses of THIP may offer a safe and effective treatment forpatients with various forms of epilepsy. Other potential treatments mayinclude the GABA uptake inhibitor tiagabine (Gabitril), as inhibition ofGABA uptake specifically enhances tonic inhibition46. However, asshown by the lack of an effect for THIP on seizure susceptibility


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

during estrus, such drugs may be ineffective if an impaired upregula-tion of dGABAARs during the luteal phase is responsible for theincreased seizure frequency.

Anxiety during the estrous cycle

Neurosteroid modulation of GABAARs is thought to have a role indepression, anxiety, and mood disorders associated with PMDD1.Patients with PMDD have a much higher incidence of catamenialepilepsy than the general population47, and show increased luteal-phase excitability2,48, suggesting shared pathogenesis. Women withPMDD lack the enhancement of GABAergic inhibition normallypresent during the luteal phase32,48 when progesterone-derived neuro-steroid levels are high. Thus, a failure in neurosteroid-sensitive inhibi-tion may underlie both dysphoria and increased seizure susceptibility.According to our findings, catamenial epilepsy and PMDD may bothresult from a disturbance in the normal hormone-dependent potentia-tion of dGABAAR-mediated inhibition during the menstrual cycle.These patients may benefit from treatments targeting the enhancementof neurosteroid-sensitive tonic GABAergic inhibition. Consistent withthis hypothesis, women with PMDD have decreased sensitivity toethanol29, a drug that selectively enhances tonic GABAergic inhibitionmediated by dGABAARs15. Consequently, women with PMDD mayhave a deficit in dGABAAR-mediated tonic inhibition during the lutealphase, perhaps secondary to an impaired expression of the dGABAARssensitive to neurosteroids and ethanol.

Our findings provide new insight into ovarian cycle–related changesin neuronal excitability. These estrous cycle–related changes inGABAergic inhibition may underlie the increased seizure susceptibilityand anxiety in catamenial epilepsy and PMDD, respectively. Themolecular mechanisms underlying the cyclic regulation of GABAARsubunits require further elucidation and are expected to uncover newpotential therapeutic targets.

METHODSEstrous cycle detection. The estrous cycle of adult C57/Bl6 mice (10–14 weeks

of age) (Harlan) was determined by cellular profile analysis (Giemsa staining,

Fisher Diagnostics) in vaginal smears and by measuring the electrical resistance

of the vaginal mucosa (Estrus cycle monitor EC40, Fine Science Tools). Only

mice showing regular estrous cycles were used (Fig. 1).

Progesterone level determination. Trunk blood was collected from wild-type

mice in estrus and late diestrus. Duplicate 25-ml plasma samples were measured

using a spectrophotometer (at 450 nm) against a standard curve according to

the manufacturer’s specification (Progesterone-HRP enzyme immunoassay test

kit, BioCheck).

Whole-cell recordings. Whole-cell recordings were performed at 21–23 1C on

visually identified DGGCs and CA1 pyramidal cells (Vh ¼ �60 mV) in coronal

hippocampal slices 350 mm thick as previously described12,15 (Supplementary

Methods). Measurements and data analysis were performed as previously

described12,46 (Supplementary Methods). The experimenter was blinded to

the stage of the estrous cycle throughout the recordings and analyses.

Western blot analysis. Mice were anesthetized with halothane and killed by

cervical dislocation. Hippocampi were homogenized in a buffer containing

50 mM Tris-HCl, 5 mM EDTA, 10 mM EGTA and 0.5 mM dithiothreitol along

with protease inhibitors (Complete Mini, Roche). The homogenate was

centrifuged at 100,000g for 30 min at 41 C; the pellet was resuspended in

homogenization buffer containing 1% Triton X-100, incubated on ice for 1 h

and centrifuged at 100,000g for 30 min at 4 1C. The supernatant was then

collected as the membrane fraction.

Protein concentrations were determined using the DC Protein Assay (Bio-

Rad). Protein (100 mg) was loaded onto a 12% SDS-polyacrylamide gel,

subjected to electrophoresis and transferred to a pure nitrocellulose membrane

(Amersham). The membrane was blocked in 10% nonfat milk and probed with

polyclonal antibodies specific for a4 (1:5,000) or d (1:5,000), both gifts from

W. Sieghart, or g2 (1:10,000) from Novus. The blots were incubated with

peroxidase-labeled anti–rabbit IgG (1:2,000, Vector Laboratories) and immuno-

reactive proteins were visualized using enhanced chemiluminescence (Amer-

sham). Optical density was determined using the NIH Image J software. Pixel

intensities were converted to optical density (OD) using the calibration curve of

the software, and background-subtracted values were expressed as OD per

100 mg total protein.

Electroencephalogram (EEG) recordings. Age-matched adult C57/Bl6 mice

(Harlan) were anesthetized with 100 mg kg�1 ketamine, 5.2 mg kg�1 xylazine

and 1.0 mg kg�1 acepromazine according to a protocol approved by the UCLA

Chancellor’s Animal Research Committee. A hippocampal depth electrode

(Plastics One) was placed 2.2 mm posterior to bregma and 1.7 mm lateral to

the midline at a depth of 2.0 mm. The electrode was fixed to the skull using

dental cement and the mouse was allowed to recover for 48 h. EEG recordings

were started 10 min before an i.p. injection of 15 mg kg�1 kainic acid (Sigma)

and continued for more than 120 min. Mice treated with 10 mg kg�1 THIP

(Sigma) were injected i.p. 30 min before kainic acid treatment. Recordings were

band-pass filtered between 0.1 and 200 Hz (8-pole Bessel, Frequency Devices)

and were sampled at 1 kHz using an in-house Labview-based (National

Instruments) software. Electrographic seizure events were defined as changes

in the amplitude and frequency of the EEG activity, and their duration was

calculated by the software. Measures of seizure susceptibility were the seizure

latency, the cumulative time seizing expressed as a fraction (%) of the total

recording time, and the average duration of individual electrographic events.

Seizure latency was defined as the time elapsed from the injection of kainic acid

to the start of the first electrographic seizure. The fraction of total time spent in

seizures (% time seizing) was calculated as the cumulative time of all

seizure activity during a 120-min recording period divided by 120 min. The

durations of individual electrographic events were measured from the start

of the repetitive EEG pattern until return to baseline. The average time of

these events observed over 120 min was calculated to obtain the average seizure

duration. The number of electrographic events over the 120-min period can

be calculated by dividing the (% time seizing � 120 min) by the average

seizure duration.

Antisense generation and administration. Antisense oligodeoxynucleotides

(ODNs) were designed complementary to the mouse dGABAAR mRNA. The

20-mer ODN, CGT TTG TAC CTT ATG TGG TA, does not show any

significant homology with any other sequence identified in GenBank. A 20-

mer ODN missense mRNA was also constructed as AT GGT GTA TTC CAT

GTT TGC, and did not show any significant homology with any other sequence

identified in GenBank. A 20-mer negative control RNA with no significant

sequence similarity to mouse, rat or human gene sequences (Silencer Negative

Control siRNA, Ambion) was used as a negative control in the whole-cell

tonic recording experiments. ODNs (5 nmol) were injected in a 5 ml volume

of sterile saline into the lateral ventricle at coordinates: 0.6 mm posterior

to bregma and 1.1 mm lateral to the midline at a depth of 2.0 mm. The mice

were injected with ODNs at estrus and allowed to recover for 48 h until the

diestrus phase.

Behavioral anxiety test. Mice were tested for 10 min on an elevated (57 cm)

plus-maze every day for 14 d during the light period (13:00–6:00 h). At the

beginning of each test, each mouse was placed in the central platform facing an

open arm. The activity of the mice was evaluated based on the number of

entries into the closed arms and open arms and the percentage of time spent in

the open versus closed arms. After blind testing and analysis, vaginal smears

and impedance monitoring were performed to correlate the behavior to the

stages of the estrous cycle.

All data are presented as mean 7 s.e.m. Significance was considered to be

P o 0.05, as determined by paired and unpaired t-tests and by ANOVA,

as indicated. Further details of methods are available online in the

Supplementary Methods.

Accession code. GenBank: dGABAAR mRNA, LocusID 14403.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e


ACKNOWLEDGMENTSWe thank W. Sieghart (University of Vienna) for the gift of some ofthe antibodies used in this study. This work was supported by US NationalInstitutes of Health grants NS30549 and NS02808 and by the CoelhoEndowment to I.M. J.M was also supported by the Training Program inNeural Repair (T32 NS07449).


Received 20 December 2004; accepted 21 April 2005


1. Backstrom, T. et al. Pathogenesis in menstrual cycle–linked CNS disorders. Ann. NYAcad. Sci. 1007, 42–53 (2003).

2. Herzog, A.G., Klein, P. & Ransil, B.J. Three patterns of catamenial epilepsy. Epilepsia38, 1082–1088 (1997).

3. Rogawski, M.A. Progesterone, neurosteroids, and the hormonal basis of catamenialepilepsy. Ann. Neurol. 53, 288–291 (2003).

4. Martin, J.V. & Williams, D.B. Benzodiazepine binding varies with stage of estrous cycle inunwashed membranes from mouse brain. Life Sci. 57, 1903–1909 (1995).

5. Molina-Hernandez, M., Contreras, C.M. & Tellez-Alcantara, P. Diazepam increases thenumber of punished responses in a conflict-operant paradigm during late proestrus andestrus in the Wistar rat. Neuropsychobiology 43, 29–33 (2001).

6. Reddy, D.S. & Kulkarni, S.K. Sex and estrous cycle–dependent changes in neurosteroidand benzodiazepine effects on food consumption and plus-maze learning behaviors inrats. Pharmacol. Biochem. Behav. 62, 53–60 (1999).

7. Hevers, W. & Luddens, H. The diversity of GABAA receptors—pharmacological andelectrophysiological properties of GABAA channel subtypes. Mol. Neurobiol. 18, 35–86(1998).

8. Mody, I. & Pearce, R.A. Diversity of inhibitory neurotransmission through GABAA

receptors. Trends Neurosci. 27, 569–575 (2004).9. Belelli, D., Casula, A., Ling, A. & Lambert, J.J. The influence of subunit composition on

the interaction of neurosteroids with GABAA receptors. Neuropharmacology 43, 651–661 (2002).

10. Brown, N., Kerby, J., Bonnert, T.P., Whiting, P.J. & Wafford, K.A. Pharmacologicalcharacterization of a novel cell line expressing human a4b3d GABAA receptors. Br. J.Pharmacol. 136, 965–974 (2002).

11. Wohlfarth, K.M., Bianchi, M.T. & Macdonald, R.L. Enhanced neurosteroid potentiationof ternary GABAA receptors containing the d subunit. J. Neurosci. 22, 1541–1549(2002).

12. Stell, B.M., Brickley, S.G., Tang, C.Y., Farrant, M. & Mody, I. Neuroactive steroids reduceneuronal excitability by selectively enhancing tonic inhibition mediated by d subunit–containing GABAA receptors. Proc. Natl. Acad. Sci. USA 100, 14439–14444 (2003).

13. Sundstrom-Poromaa, I. et al. Hormonally regulated a4b2d GABAA receptors are a targetfor alcohol. Nat. Neurosci. 5, 721–722 (2002).

14. Wallner, M., Hanchar, H.J. & Olsen, R.W. Ethanol enhances a4b3d and a6b3dg-aminobutyric acid type A receptors at low concentrations known to affect humans.Proc. Natl. Acad. Sci. USA 100, 15218–15223 (2003).

15. Wei, W.Z., Faria, L.C. & Mody, I. Low ethanol concentrations selectively augment thetonic inhibition mediated by d subunit–containing GABAA receptors in hippocampalneurons. J. Neurosci. 24, 8379–8382 (2004).

16. Farrant, M. & Nusser, Z. Variations on an inhibitory theme: phasic and tonic activation ofGABAA receptors. Nat. Rev. Neurosci. 6, 215–229 (2005).

17. Edwards, H.E., Burnham, W.M., Mendonca, A., Bowlby, D.A. & MacLusky, N.J. Steroidhormones affect limbic after discharge thresholds and kindling rates in adult femalerats. Brain Res. 838, 136–150 (1999).

18. Smith, M.J., Adams, L.F., Schmidt, P.J., Rubinow, D.R. & Wassermann, E.M. Effects ofovarian hormones on human cortical excitability. Ann. Neurol. 51, 599–603 (2002).

19. Guerra-Araiza, C., Villamar-Cruz, O., Gonzalez-Arenas, A., Chavira, R. & Camacho-Arroyo, I. Changes in progesterone receptor isoforms content in the rat brain duringthe oestrous cycle and after oestradiol and progesterone treatments. J. Neuroendocrinol.15, 984–990 (2003).

20. Jones, A. et al. Ligand-gated ion channel subunit partnerships: GABAA receptor a6subunit gene inactivation inhibits d subunit expression. J. Neurosci. 17, 1350–1362(1997).

21. Sur, C. et al. Preferential coassembly of a4 and d subunits of the g-aminobutyric acidA

receptor in rat thalamus. Mol. Pharmacol. 56, 110–115 (1999).

22. Pirker, S., Schwarzer, C., Wieselthaler, A., Sieghart, W. & Sperk, G. GABAA receptors:immunocytochemical distribution of 13 subunits in the adult rat brain. Neuroscience101, 815–850 (2000).

23. Peng, Z. et al. GABAA receptor changes in d subunit–deficient mice: altered expressionof a4 and g2 subunits in the forebrain. J. Comp. Neurol. 446, 179–197 (2002).

24. Caraiscos, V.B. et al. Tonic inhibition in mouse hippocampal CA1 pyramidal neurons ismediated by a5 subunit–containing, g-aminobutyric acid type A receptors. Proc. Natl.Acad. Sci. USA 101, 3662–3667 (2004).

25. Wei, W.Z., Zhang, N.H., Peng, Z.C., Houser, C.R. & Mody, I. Perisynaptic localization of dsubunit–containing GABAA receptors and their activation by GABA spillover in themouse dentate gyrus. J. Neurosci. 23, 10650–10661 (2003).

26. Agmo, A. & Soria, P. GABAergic drugs and sexual motivation, receptivity andexploratory behaviors in the female rat. Psychopharmacology (Berl.) 129, 372–381(1997).

27. Broadbent, J. & Harless, W.E. Differential effects of GABAA and GABAB agonists onsensitization to the locomotor stimulant effects of ethanol in DBA/2 J mice. Psycho-pharmacology (Berl.) 141, 197–205 (1999).

28. Mihalek, R.M. et al. Attenuated sensitivity to neuroactive steroids in g-aminobutyratetype A receptor d subunit knockout mice. Proc. Natl. Acad. Sci. USA 96, 12905–12910(1999).

29. Nyberg, S., Wahlstrom, G., Backstrom, T. & Sundstrom, P.I. Altered sensitivity to alcoholin the late luteal phase among patients with premenstrual dysphoric disorder. Psycho-neuroendocrinology 29, 767–777 (2004).

30. Smith, S.S., Ruderman, Y., Hua, G.Q. & Gulinello, M. Effects of a low dose of ethanol inan animal model of premenstrual anxiety. Alcohol 33, 41–49 (2004).

31. Belzung, C. & Griebel, G. Measuring normal and pathological anxiety-like behaviour inmice: a review. Behav. Brain Res. 125, 141–149 (2001).

32. Sundstrom, I. et al. Patients with premenstrual syndrome have a different sensitivity to aneuroactive steroid during the menstrual cycle compared to control subjects. Neuroen-docrinology 67, 126–138 (1998).

33. Gulinello, M., Gong, Q.H. & Smith, S.S. Progesterone withdrawal increases the a4subunit of the GABAA receptor in male rats in association with anxiety and alteredpharmacology—a comparison with female rats. Neuropharmacology 43, 701–714(2002).

34. Smith, S.S. et al. GABAA receptor a4 subunit suppression prevents withdrawal proper-ties of an endogenous steroid. Nature 392, 926–930 (1998).

35. Somogyi, P., Fritschy, J.M., Benke, D., Roberts, J.D.B. & Sieghart, W. The g2 subunit ofthe GABAA receptor is concentrated in synaptic junctions containing the a1 and b2/3subunits in hippocampus, cerebellum and globus pallidus. Neuropharmacology 35,1425–1444 (1996).

36. Soltesz, I., Smetters, D.K. & Mody, I. Tonic inhibition originates from synapses close tothe soma. Neuron 14, 1273–1283 (1995).

37. Sanna, E. et al. Brain steroidogenesis mediates ethanol modulation of GABAA receptoractivity in rat hippocampus. J. Neurosci. 24, 6521–6530 (2004).

38. Herzog, A.G. Progesterone therapy in women with epilepsy: a 3-year follow-up. Neurol-ogy 52, 1917–1918 (1999).

39. Reddy, D.S. & Rogawski, M.A. Enhanced anticonvulsant activity of neuroactive steroidsin a rat model of catamenial epilepsy. Epilepsia 42, 337–344 (2001).

40. Reddy, D.S., Castaneda, D.C., O’Malley, B.W. & Rogawski, M.A. Anticonvulsant activityof progesterone and neurosteroids in progesterone receptor knockout mice. J. Pharma-col. Exp. Ther. 310, 230–239 (2004).

41. Reddy, D.S., Kim, H.Y. & Rogawski, M.A. Neurosteroid withdrawal model of perimen-strual catamenial epilepsy. Epilepsia 42, 328–336 (2001).

42. Smith, S.S. Withdrawal properties of a neuroactive steroid: implications for GABAA

receptor gene regulation in the brain and anxiety behavior. Steroids 67, 519–528(2002).

43. Peng, Z.C., Huang, C.S., Stell, B.M., Mody, I. & Houser, C.R. Altered expression of the dsubunit of the GABAA receptor in a mouse model of temporal lobe epilepsy. J. Neurosci.24, 8629–8639 (2004).

44. Dibbens, L.M. et al. GABRD encoding a protein for extra- or peri-synaptic GABAA

receptors is a susceptibility locus for generalized epilepsies. Hum. Mol. Genet. 13,1315–1319 (2004).

45. Lancel, M., Wetter, T.C., Steiger, A. & Mathias, S. Effect of the GABAA agonist gaboxadolon nocturnal sleep and hormone secretion in healthy elderly subjects. Am. J. Physiol.Endocrinol. Metab. 281, E130–E137 (2001).

46. Nusser, Z. & Mody, I. Selective modulation of tonic and phasic inhibitions in dentategyrus granule cells. J. Neurophysiol. 87, 2624–2628 (2002).

47. Shuman, S. et al. Premenstrual dysphoric disorder in women with partial epilepsy.Epilepsia 44 (Suppl. 9): 292 (2003).

48. Smith, M.J., Adams, L.F., Schmidt, P.J., Rubinow, D.R. & Wassermann, E.M. Abnormalluteal phase excitability of the motor cortex in women with premenstrual syndrome. Biol.Psychiatry 54, 757–762 (2003).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Dopaminergic modulation of limbic and cortical driveof nucleus accumbens in goal-directed behavior

Yukiori Goto & Anthony A Grace

Goal-directed behavior is believed to involve interactions of prefrontal cortical and limbic inputs in the nucleus accumbens

(NAcc), and their modulation by mesolimbic dopamine (DA) seems to be of primary importance in NAcc function. Using in vivo

electrophysiological recordings simultaneously with DA system manipulation in rats, we show that tonic and phasic DA release

selectively modulates hippocampal and prefrontal cortical inputs through D1 and D2 receptors, respectively. In addition, we also

found that D1 activation and D2 inactivation in the NAcc produced behaviorally selective effects (learning versus set shifting

of response strategy) that correspond to specific afferents. These results suggest that the dynamics of DA release regulate the

balance between limbic and cortical drive through activation and inactivation of DA receptor subtypes in the accumbens, and

this regulates goal-directed behavior.

The NAcc has been described as the limbic-motor interface1. It receivesconvergent synaptic inputs from the prefrontal cortex (PFC) andlimbic structures such as the hippocampus and amygdala2. This ledto the idea that the NAcc mediates goal-directed behavior by integrat-ing hippocampus-dependent contextual information and amygdala-dependent affective information with PFC cognitive functions to selectappropriate behavioral responses1,3.

The NAcc is also the target of a dense dopaminergic innervationarising from the ventral tegmental area (VTA)4, which is involved inincentive motivation. Therefore, DA is critical in the NAcc in modulat-ing goal-directed behavior5,6. Afferents to VTA DA neurons cause themto show two activity states, a slow tonic firing and phasic burst firing7,and these activity patterns differentially affect the dynamics of DArelease in the NAcc8. However, the functional impact of tonic andphasic DA release on afferent regulation of the NAcc is unknown.

Our previous study has shown that different basal ganglia nucleiselectively regulate the different activity states of VTA DA neurons andconsequently DA release in the NAcc8. Thus, ventral pallidum inactiva-tion was found to increase the number of tonically active DA neuronsin the VTA, causing an increase in tonic extracellular DA levels in theNAcc, whereas pedunculopontine tegmental (PPTg) activation causedincreased burst spike firing in those spontaneously active DA neurons,leading to phasic DA release.

This study used selective modulation of tonic and phasic DAtransmission to examine dopamine’s effect on hippocampus andPFC synaptic integration within the NAcc. We examined the functionalconsequence of this interaction using behavioral models that dependon intact hippocampus and PFC afferent inputs into the NAcc, and weexamined the impact of these afferents on goal-directed behavior.Specifically, we evaluated behavioral flexibility and spatial learning,

cognitive functions that are critical for goal-directed behavior, inanimals in which hippocampus-NAcc or PFC-NAcc informationprocessing was selectively disrupted using a disconnection method.This model used unilateral inactivation of the afferent structure withcontralateral disruption of the same input by DA system manipulationand examined the impact on the ability to learn and switch betweentwo spatial maze tasks requiring different response strategies. We foundthat tonic and phasic DA release selectively modulates PFC andhippocampus synaptic inputs in the NAcc through D2 and D1receptors, respectively, to affect goal-directed behavior.

RESULTS

Selective D1 and D2 modulation of hippocampus and PFC inputs

We placed stimulation electrodes in the PFC and hippocampus andrecorded extracellular local field potentials in the NAcc (Figs. 1 and2a–c). DA agonists and antagonists were perfused via microdialysisprobes placed next to the recording electrodes. Single-pulse stimulationin the hippocampus and PFC evoked characteristic field potentialresponses (Fig. 1b), and we measured response amplitude before andafter DA agonist or antagonist infusion. Administration of the D1agonist SKF38393 (SKF; 10 mM) selectively facilitated hippocampus-evoked, but not PFC-evoked, responses (hippocampus response, +39.17 16.2% (mean 7 s.d.) at stimulation intensity 1.0 mA, P o 0.01,paired t-test comparing measurements before and after drug admin-istration; PFC response, +1.01 7 5.3%, n ¼ 8; Fig. 3a, Table 1).Administration of the D1 antagonist SCH23390 (SCH; 10 mM) had noeffect on either hippocampus- or PFC-evoked responses (hippocampusresponse, +2.4 7 7.3%; PFC response, �1.6 7 9.0%, n ¼ 8; Fig. 3b,Table 1). In contrast, the D2 agonist quinpirole (10 mM) attenuated,and the D2 antagonist eticlopride (20 mM) facilitated, PFC-evoked


Departments of Neuroscience, Psychiatry and Psychology, University of Pittsburgh, Pittsburgh, Pennsylvania 15260, USA. Correspondence should be addressed to Y.G.([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

responses, respectively (with quinpirole, �27.6 7 14.7%, P o 0.01,n ¼ 7; with eticlopride, +43.3 7 20.7%, P o 0.01, n ¼ 8; Fig. 3c,d;Table 1). However, neither the D2 agonist nor the D2 antagonistaffected hippocampus-evoked responses (with quinpirole, �0.61 713.2%; with eticlopride, �2.4 7 5.0%; Fig. 3c,d; Table 1). These resultssuggest that D1 and D2 receptor activation or inactivation selectivelyshifts the balance between hippocampus and PFC synaptic inputs.

Tonic and phasic DA modulation of hippocampus and PFC inputs

D1 and D2 receptors show comparatively low and high affinities to DA,respectively9. It has been suggested that D1 receptor activation requiresphasic DA release but that D2 receptors are continuously stimulatedwith basal, tonic DA release7. Therefore, selective modulation ofhippocampus and PFC inputs into the NAcc via D1 and D2 receptorsmay be related to phasic and tonic DA release. To examine thispossibility, we implanted infusion cannulae into the ventral pallidumor PPTg, and we activated or inactivated these areas with GABA agonistor antagonist infusion (Fig. 2d,e). We have reported previously8 thatPPTg activation by the GABAA antagonist bicuculline increases DAneuron burst firing but does not affect the tonic population activity(that is, the number of neurons firing). This leads to a correspondingincrease in high-amplitude phasic DA release in the NAcc. In contrast,ventral pallidum inactivation by the GABAA/B agonists muscimol andbaclofen increases tonic DA neuron population activity and lower-amplitude tonic DA release in the NAcc. In the present study, bicucul-line administration into the PPTg selectively facilitated only thehippocampus-evoked responses (hippocampus response, +47.1 717.6%, P o 0.01; PFC response, �4.0 7 2.7%, n ¼ 8; Fig. 4a,Table 1). This facilitation was blocked by D1 antagonist pretreatmentadministered locally to the NAcc through the microdialysis probe(hippocampus response, examined at a stimulation intensity of

1.0 mA, at which the largest difference in amplitude was observed beforeand after drug administration: +0.67 7 7.7%, n ¼ 5, Fig. 5a), but notby infusion of the D2 antagonist (+20.8 7 6.2%, n ¼ 6; Fig. 5a).However, the facilitation observed during D2 antagonist administra-tion at the same intensity (1.0 mA) was significantly smaller than thatproduced without D2 antagonist pretreatment (P o 0.05, unpairedt-test). D2 antagonist pretreatment alone facilitated PFC inputs (+25.67 7.6%, n ¼ 6). Inactivation of the PPTg by a mixture of the GABAagonists muscimol and baclofen did not affect either hippocampus- orPFC-evoked responses (hippocampus response, +1.8 7 1.8%; PFCresponse, +1.0 7 2.8%, n ¼ 8; Fig. 4b, Table 1). In contrast, ventralpallidum inactivation by GABA agonists attenuated PFC-evokedresponses (PFC response, �24.9 7 12.9%, P o 0.01, n ¼ 8; Fig. 4c,Table 1), and ventral pallidum activation by a GABA antagonistfacilitated PFC-evoked responses (PFC response, +35.3 7 19.4%,P o 0.01, n ¼ 8; Fig. 4d, Table 1) without affecting hippocampus-evoked responses (hippocampus response, �0.29 7 3.9% and +0.187 2.7% for ventral pallidum inactivation and activation, respectively).This attenuation of PFC-evoked responses by ventral pallidum inacti-vation with GABAA/B agonists was prevented (in fact, PFC-evokedresponses were facilitated) by prior infusion of the D2 antagonist intothe NAcc (PFC response, +20.6 7 9.4%, n ¼ 5, Fig. 5b) but not byinfusion of the D1 antagonist (�34.5 7 13.2%, n ¼ 6, Fig. 5b). Takentogether, these results suggest that phasic DA release facilitates hippo-campus inputs through activation of D1 receptors, shifting the balancein favor of limbic inputs over cortical afferents. In contrast, a suppres-sion of tonic DA release shifts the balance towards PFC predominancein the NAcc through decreased stimulation of D2 receptors. Indeed,such a condition may occur when administering D2-selective anti-psychotic drugs, which would increase DA release and stimulation ofD1 receptors while attenuating tonic D2 stimulation.

DA

10 ms

100 µV

P2

N1

N2

P2

N1

P1

a b

HPC

PFCLV

VTA

VP

NAcc

HPC

GluGABA

PPTg

Glu+Ach

PFC

Figure 1 Local field potential responses in

the NAcc evoked by PFC and hippocampus

stimulation. (a) Placement of stimulation

electrode (double black lines) into the PFC and

hippocampus, microdialysis (thick black line)

and recording electrode (cone) into the NAcc,

and infusion cannulae (solid colored lines) into

the ventral pallidum and PPTg. VP, ventralpallidum; HPC, hippocampus. (b) Representative

traces of hippocampus- and PFC-evoked

responses. Traces of five individual responses,

left, and average of ten responses, right.

Hippocampus-evoked responses consist of a

series of positive and negative shifts (P1, N1,

P2, N2). Amplitude was defined as P1–N1.

PFC-evoked responses consist of a negative shift

followed by a prolonged positive shift. Amplitude

was defined as N1 minus baseline measured

1 ms before stimulation.

+0.2+1.6+1.3+1.0+3.2 +3.7+2.7 mma b c ed–5.2 –5.8–5.5 –0.3 –0.1 –8.2–8.0–7.8

Figure 2 Graphs illustrating the range of the intracranial placements around the targeted regions in electrophysiological and behavioral experiments. (a) PFC.

(b) Hippocampus. (c) NAcc (light gray, recording electrode; dark gray, dialysis probe placements). (d) Ventral pallidum. (e) PPTg.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Pre- versus postsynaptic DA modulation of inputs

We further investigated whether these selective responses to DAinvolved presynaptic or postsynaptic mechanisms. Modulation ofhippocampus-evoked responses with PPTg activation, hippocampus-evoked responses with D1 agonist infusion, PFC-evoked responses withventral pallidum activation and inactivation and PFC-evoked respon-ses with D2 agonist and antagonist infusion were examined with arank-order analysis in which all individual responses were alignedfrom the smallest to the largest and were compared before andafter manipulation. The slope of the linear regression line fitted to

hippocampus-evoked responses was signifi-cantly larger than 1.0 with PPTg activation(slope a ¼ 1.62 7 0.27, P o 0.05, pairedt-test, n ¼ 8; Fig. 6a,c) and D1 agonistadministration (a ¼ 1.57 7 0.39, P o 0.05,n ¼ 8; Fig. 6b,c), suggesting that D1-depen-dent phasic DA modulation of hippocampus-evoked responses was more effective in largerresponses than in smaller ones. In contrast, theslopes of the linear regression lines for all PFC-evoked responses were not different from 1.0(with ventral pallidum activation, a ¼ 1.02 70.19, n ¼ 8; with ventral pallidum inactiva-tion, a¼ 0.94 7 0.30, n¼ 8; with D2 agonist,a ¼ 0.91 7 0.23, n ¼ 8; with D2 antagonist,a¼ 1.07 7 0.21, n¼ 8; Fig. 6a–c), suggestingthat D2-dependent tonic DA modulation of

PFC-evoked responses is independent of the size of the responses.Thus, these results suggest that DA modulation of hippocampus- andPFC-evoked responses involves different mechanisms.

We examined variance analyses10,11 and calculated putative prob-ability of synaptic release for modulation of hippocampus-evokedresponses with PPTg activation and PFC-evoked responses with ventralpallidum activation and inactivation (Fig. 6d–g). Based on theseanalyses, it was estimated that the probability of synaptic release wasnot altered in hippocampus-evoked responses before and after PPTgactivation (Fig. 6f,g). However, probabilities of synaptic release in

a b

**

500

100

200

300

400

000.20.40.60.81.0

500

100

200

300

400

000.20.40.60.81.0

200

600

400

000.20.40.60.81.0

200

600

400

000.20.40.60.81.0

(i) (ii) (i)+(ii)(i) (ii) (i)+(ii)(i) (ii) (i)+(ii)(i) (ii) (i)+(ii)

c d

00.20.40.60.81.0

**

200

600

400

000.20.40.60.81.0

500

100

200

300

400

000.20.40.60.81.0

(i) (ii) (i)+(ii)(i) (ii) (i)+(ii)(i) (ii) (i)+(ii)(i) (ii) (i)+(ii)

100

200

300

400

000.20.40.60.81.0

*

100

*

0.20.40.60.81.0

Stimulation (mA)

*

Per

cent

age

ch

ange

100

0

0.20.40.60.81.0

100

0

0.20.40.60.81.0

100

0

0.20.40.60.81.0

100

0

100

0

0.20.40.60.81.0

100

0 * *

0.20.40.60.81.0

100

0

0.20.40.60.81.0

100

0

** * *

0.20.40.60.81.0

200

600

Am

plitu

de (

µV)

Am

plitu

de (

µV)

400

0

SKF SCH

QIN ETI

HPC HPC

HPC HPC

PFC PFC

PFC PFC

+

+

++

+

+

Stimulation (mA)

Per

cent

age

ch

ange

–100 _100_100_100

–100 –100–100–100

BasalDrug

0

100

200

300

400

0.20.40.60.81.0 0.20.40.60.81.00

100

200

300

400

0.20.40.60.81.0

500

100

200

300

400

0

500

100

200

300

400

00.20.40.60.81.0

0.20.40.60.81.0

500

100

200

300

400

00.20.40.60.81.0

500

100

200

300

400

00.20.40.60.81.0

0

100

200

300

400

0.20.40.60.81.0

500

100

200

300

400

0

Figure 3 Selective modulation of hippocampus-

and PFC-evoked field potential responses in the

NAcc by local infusion of D1 and D2 agonists and

antagonists. (a) Effects of D1 agonist SKF38393

on hippocampus (HPC)- and PFC-evoked

responses. Top: averages of ten responses before

(i) and after (ii) drug administration. Scale bar:

0.2 mV and 20 ms. Examples of individual(second row) and averaged (third row) amplitude

facilitation in hippocampus-evoked, but not in

PFC-evoked, responses are shown (stimulation

intensities tested were 0.2–1.0 mA). Bottom:

summary of results as percentage change of

amplitude before and after drug administration.

Error bars indicate s.d. (*, P o 0.01; +,

P o 0.05, paired t-test comparing measurements

before and after drug perfusion). (b) D1 antagonist

SCH23309 had no effect on both hippocampus-

and PFC-evoked responses. Scale: 0.3 mV and

10 ms. Panels as in a. (c) D2 agonist quinpirole

(QIN) attenuated PFC-evoked, but not

hippocampus-evoked, responses. Scale: 0.2 mV

and 10 ms. Panels as in a. (d) D2 antagonist

eticlopride (ETI) facilitated PFC-evoked, but not

hippocampus-evoked, responses. Scale bars:

0.2 mV and 10 ms. Panels as in a.

Table 1 The effects of DA system manipulations on hippocampus- and PFC-evoked responses recorded in the NAcc

Treatment D1+ (SKF) D1� (SCH) D2+ (QIN) D2� (ETI) Phasic DA+ (PPTg + bic) Phasic DA� (PPTg + mus) Tonic DA+ (VP + mus) Tonic DA� (VP + bic)

HPC inputs m N.C. N.C. N.C. m N.C. N.C. N.C.

PFC inputs N.C. N.C. k m N.C. N.C. k m

+, activation/increase; –, inactivation/decrease; m, facilitation; k, attenuation; N.C., no change. QIN, quinpirole; ETI, eticlopride; mus, muscimol; bic, bicuculline; SKF, D1 agonistSKF38393; SCH, D1 antagonist SCH23390. VP, ventral pallidum; HPC, hippocampus.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

PFC-evoked responses were significantlydecreased with ventral pallidum inactivationand increased with ventral pallidum activa-tion, especially at higher intensity of electricalstimulation (Fig. 6f,g). These results suggestthat D1-dependent, hippocampus-evokedresponses may involve a postsynapticmechanism, whereas DA modulation ofD2-dependent, PFC-evoked responses mayinvolve a presynaptic mechanism.

DA modulation of inputs in goal-directed

behavior

We investigated whether this interaction wasreflected by behaviorally specific alterationswithin this system using a plus maze with anasymmetrical infusion procedure (Fig. 7a). Inthe tasks, animals made turns to obtain rewards based on one of twocriteria: (i) a visual cue placed in the maze (visual cue task or VCT) or(ii) response direction (response direction task or RDT). After ratsreached response criterion in VCT, the task was switched to RDT untilcriterion performance level was again reached. One group of ratsreceived unilateral inactivation of the hippocampus with lidocaineinfusion combined with either contralateral injection of saline (‘hip-pocampus-saline’, n ¼ 6), the D1 antagonist (‘hippocampus-D1’,

n ¼ 6) or the D2 agonist (‘hippocampus-D2’, n ¼ 5) into the NAcc 5min before the session started. Another group of rats received unilaterallidocaine inactivation of the PFC with either saline (PFC-saline, n¼ 6),the D1 antagonist (PFC-D1, n¼ 5) or the D2 agonist (PFC-D2, n¼ 6)injection into the contralateral NAcc.

The control rats (combined data from PFC-saline and hippocam-pus-saline animals, which showed no difference in response) required50.3 7 5.3 trials in VCT and 70.8 7 10.7 trials in RDT to reach

*

00.20.40.60.81.0

200

400

600

000.20.40.60.81.0

200

400

600

0

00.20.40.60.81.0

200

400

600

0

**

00.20.40.60.81.0

200

400

600

0

00.20.40.60.81.0

100

200

300

400

0

00.20.40.60.81.0

500

100

200

300

400

0

* **

*

00.20.40.60.81.0

500

100

200

300

400

0

a b

d

(i) (ii) (i)+(ii) (i) (ii) (i)+(ii)

(i) (ii) (i)+(ii)(i) (ii) (i)+(ii) (i) (ii) (i)+(ii)

(i) (ii) (i)+(ii)(i) (ii) (i)+(ii)

c(i) (ii) (i)+(ii)

0.20.40.60.81.0

100

0

0.20.40.60.81.0

100

0

0.20.40.60.81.0

100

0

0.20.40.60.81.0

100

0

0.20.40.60.81.0

100

0

0.20.40.60.81.0

100

0

* * *

00.20.40.60.81.0

200

400

600

0

PPTg activation

HPC PFC HPC PFC

VP inactivationHPC PFC

VP activationHPC PFC

PPTg inactivation

++

+

+

+

Per

cent

age

Am

plitu

de (

µV)

chan

ge

100

0

* +*

Stimulation (mA)

0.20.40.60.81.0

0.20.40.60.81.0

* * * +

100

0

–100 –100 _100–100

–100 _100–100–100

Per

cent

age

c

hang

e

Stimulation (mA)

0.20.40.60.81.0

200

400

600

0

BasalDrug

500

100

200

300

400

00.20.40.60.81.0 0.20.40.60.81.0

500

100

200

300

400

00.20.40.60.81.0

100

200

300

0

0.20.40.60.81.0

500

100

200

300

400

00.20.40.60.81.0

500

100

200

300

400

00.20.40.60.81.0

500

100

200

300

400

00.20.40.60.81.0

500

100

200

300

400

0

Am

plitu

de (

µV)

Figure 4 Selective modulation of hippocampus-

and PFC-evoked responses in the NAcc by

activation and inactivation of the ventral pallidum

and PPTg. (a) Hippocampus (HPC)- and PFC-

evoked responses with PPTg activation (that is,

increased phasic DA). Top: averages of ten

responses before (i) and after (ii) drug

administration. Scale bars: 0.3 mV and 20 ms.Example of individual (second row) and averaged

(third row) amplitude facilitation in hippocampus-

evoked, but not in PFC-evoked, responses are

shown (stimulation intensities tested were

0.2–1.0 mA). Bottom: summary of results as

percentage change of amplitude before and after

drug administration. Error bars indicate s.d.

(*, P o 0.01: +, P o 0.05; paired t-test

comparing before and after drug perfusion).

(b) PPTg inactivation has no effect on both

hippocampus- and PFC-evoked responses. Scale:

0.4 mV and 10 ms. Panels as in a. (c) Ventral

pallidum (VP) inactivation (increased tonic DA)

attenuates PFC-evoked, but not hippocampus-

evoked, responses. Scale: 0.2 mV and 10 ms.

Panels as in a. (d) Ventral pallidum activation

(decreased tonic DA) facilitates PFC-evoked,

but not hippocampus-evoked, responses. Scale:

0.2 mV and 10 ms. Panels as in a.

a b

Basal +SCH +ETI

VP inactivationPFC (1.0 mA)

100

50

0

PPTg activationHPC (1.0 mA)

100

Basal +SCH +ETI

*50

0

Per

cent

age

chan

ge

–50

–100

–50

–100

Figure 5 The effects of D1 and D2 antagonists on modulation of

hippocampus (HPC)- and PFC-evoked responses in the NAcc by activation

and inactivation of the ventral pallidum (VP) and PPTg. (a) Blockade ofphasic DA effects with SCH23390 and eticlopride pretreatments for

hippocampus (HPC) response. ‘Basal’ is the percentage increase shown in

Figure 4a. Error bars indicate s.d. (*, P o 0.05; unpaired t-test comparing

measurements with and without D2 antagonist pretreatment at stimulation

intensity of 1.0 mA). (b) Blockade of tonic DA effects with SCH and

eticlopride pretreatments for PFC response. ‘Basal’ is the percentage increase

shown in Figure 4c.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

criterion (Fig. 7b). Perseverative errors made at task switching were6.5 7 2.2 trials (Fig. 7c). Hippocampus-D1 rats required a significantlylarger number of trials to reach criterion in both VCT and RDT thandid control animals (hippocampus-D1, 75.3 7 7.2 trials in VCT,88.5 7 5.5 trials in RDT; one-way ANOVA with Dunnett’s post hoccomparison, P o 0.05), but hippocampus-D2 rats did not (hippo-campus-D2, 50.8 7 3.0 trials in VCT, 67.8 7 5.3 trials in RDT;Fig. 7b). However, perseverative errors in both hippocampus-D1 andhippocampus-D2 rats were not different from control rats (hippocam-pus-D1, 5.5 7 1.0 trials; hippocampus-D2, 5.6 7 1.7 trials; Fig. 7c). Incontrast, PFC-D2 rats, but not PFC-D1 rats, showed a distinct increasein perseverative errors at task switching (PFC-D1, 5.47 1.7 trials; PFC-D2, 12.8 7 1.9 trials; Po 0.05; Fig. 7c). This increase of perseverativeerrors resulted in a small but significantly larger number of trialsrequired to reach criterion in RDT in PFC-D2 animals (PFC-D1,65.6 7 6.0 trials; PFC-D2, 80.7 7 5.1 trials; P o 0.05; Fig. 7b),although the number of trials that both PFC-D1 and PFC-D2 ratsrequired to reach criterion for the task before switching (for example,in VCT) did not differ from control rats (PFC-D1, 47.8 7 4.5 trials;PFC-D2, 47.5 7 2.6 trials; Fig. 7b). These results suggest that D1-

mediated hippocampus-NAcc information processing and D2-mediated PFC-NAcc information processing may mediate differentaspects of goal-directed behavior.

DISCUSSION

We have shown that tonic and phasic DA release selectively shifts thebalance between hippocampus and PFC synaptic inputs in the NAccand mediates cognitive functions required for goal-directed behavior.Our results suggest that phasic DA release selectively facilitates hippo-campus inputs via D1 receptor activation, which could affect spatiallearning through hippocampus-NAcc information processing. In con-trast, either an increase or a decrease in tonic DA release selectivelyimpacts PFC but not hippocampus inputs, producing an attenuationor facilitation of PFC inputs, respectively, through D2 receptors. SuchD2-mediated PFC input facilitation provides a potent modulatoryaction over PFC-NAcc information processing that mediates set shift-ing response strategies.

In this study, we measured local field potentials to examine thesynaptic response evoked by hippocampus and PFC stimulation. Theamplitude of the field potential, defined as P1-N1 and N1-baseline for

a

b

SKFQIN ETI

PPTg Acti

vatio

n

VP Inac

tivat

ion

VP Acti

vatio

n

a (S

lope

)

0

2.0

1.5

1.0

0.5

2.5

r = 0.99a = 1.03

VPActivation:

r = 0.98 a = 1.03

VPInactivation:

r = 0.99a = 1.41

PPTgActivation:

r = 0.99a = 0.97

QIN:

r = 0.98a = 1.02

ETI:

r = 0.98a = 1.48

SKF:

Basal +Drug

1.00.60.4 0.80.2Stimulation (mA)

Pr

1.0

0.8

0.6

0.4

0.2

01.00.60.4 0.80.2

1.0

0.8

0.6

0.4

0.2

01.00.60.4 0.80.2

1.0

0.8

0.6

0.4

0.2

0

r = 0.97

r = 0.96

500

400

300

200

100

04003002001000800

r = 0.98

r = 0.91

1/ (

CV

)2

600

450

300

150

06004002000

r = 0.95

r = 0.91

400

300

200

100

05004003002001000

d

c

e

f

g

1.00.60.4 0.80.2

100

0

* * **

1.00.60.4 0.80.2

100

0 * * *

1.00.60.4 0.80.2

Stimulation (mA)

100

0

Per

cent

age

ch

ange

* *

HPC (PPTg activation)

V –

Vm

(µV

)

PFC (VP inactivation) PFC (VP activation)

–100 –100–100

100

0

–1001.0 0.6 0.40.8 0.2

Stimulation (mA)1.0 0.6 0.40.8 0.2

100

0

–1001.0 0.6 0.40.8 0.2

100

0

–100

400 500 600 700300300

400

500

600

700

400 500 600 700300

Am

plitu

de-a

fter

(µV

)

Amplitude-before (µV)

400

500

600

700

300

Mean amplitude (µV)

Figure 6 Analyses of pre- versus postsynaptic DA modulation of hippocampus- and PFC-evoked responses. (a) Rank-order analysis of hippocampus- and PFC-

evoked responses. Examples of 100 hippocampus-evoked responses with D1 agonist (SKF; open black square) and PFC-evoked responses with D2 agonist

(QIN; open light gray triangle) and with antagonist (ETI; open dark gray circle) at a stimulation intensity of 1.0 mA. The linear regression lines were fitted toall cases. r denotes the correlation coefficient, and a denotes the slope of the regression line. (b) Ranked-order analysis of hippocampus- and PFC-evoked

responses modulated by ventral pallidum (VP) or PPTg activation or inactivation. Examples of 100 hippocampus-evoked responses with PPTg activation

(open black square) and PFC-evoked responses with ventral pallidum inactivation (open light gray triangle) and activation (open dark gray circle) at a

stimulation intensity of 1.0 mA. The linear regression lines were fitted to all cases. (c) Summary of results showing the slopes of the linear regression lines in

a and b. Error bars indicate s.d. (d–g) Variance analysis of DA modulation of hippocampus- and PFC-evoked responses. Variability of responses is illustrated by

subtracting mean amplitude at variable stimulation intensities (d), and e shows the relationship between mean amplitude at 0.2–1.0 mV stimulation intensity.

Variability of individual response amplitudes was quantified by calculating the coefficient of variation (CV). In all cases, second order polynomial curves were

fitted. Based on these curve fittings, the probability of synaptic release and its modulation by DA were estimated (f). Results are summarized in g.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

hippocampus- and PFC-evoked responses, respectively, was used as thedependent measure, since these field potential shifts are most likely toreflect excitatory monosynaptic responses evoked by electrical stimula-tion. Furthermore, these responses showed a similar latency andduration as those of the evoked synaptic responses observed duringintracellular recordings in NAcc neurons in our previous study12.

The NAcc is comprised of two areas, the core and shell. We (Y. Goto& A.A. Grace, Biol. Psychiatry 55 (suppl.), 743, 2004) and others13 havereported that the polarity of the field potential responses evoked in theshell by hippocampus and PFC are opposite in direction whencompared with those evoked in the core. We used such reversal inpolarity to ensure that the recordings performed in this study werelocated only within the core subregion. Moreover, although the coreand shell show functional differences14, we found that DA affectssynaptic plasticity induced at hippocampus and PFC inputs into theNAcc in a similar manner in both regions. We did not examine whetherother limbic inputs, such as those from the amygdala2, show a similarmodulation by DA; although our previous study suggests that theseinputs are also sensitive to D1 receptor stimulation15. The fieldpotential also showed a delay in recovery of the P1-N1 component inthe hippocampus-evoked responses with D1 stimulation in most cases,which may be due to a facilitation of NMDA (N-methyl-D-aspartate)receptor–mediated responses produced by D1 stimulation16. Moreover,the P2 component in the hippocampus-evoked responses was slightlyenhanced with eticlopride; such a condition is consistent with the P2component of the hippocampus-evoked response occurring seconda-rily to activation of the PFC by hippocampus stimulation.

Our results demonstrate that tonic and phasic DA release selectivelyaffect hippocampus and PFC inputs. Although overflow of phasicallyreleased DA could potentially contribute to tonic DA levels, this was notlikely the case, as (i) phasic DA release did not affect both hippocampusand PFC afferents, and (ii) PPTg-modulated DA neuron burst firingdoes not affect extracellular tonic DA levels unless DA uptake isblocked8. Previous studies have shown that overflow during phasicDA release is very brief and spatially limited, although the stimulationtechnique used caused activation of numerous DA fibers and thus was

likely to have affected tonic DA release as well17. Therefore, if phasic DArelease does contribute to extracellular DA levels, it may be that thesignal is too brief to sufficiently activate D2 receptors in the samemanner as does tonic DA release. Administration of the D2 antagonistduring phasic DA activation, although it did not block the facilitationof hippocampus inputs, caused a significant attenuation of this facil-itation. There are several potential explanations for this effect. Forexample, activation of PFC inputs attenuates hippocampus-mediatedplasticity in the NAcc (Y. Goto & A.A. Grace, Biol. Psychiatry 55(suppl.), 743, 2004). Therefore, D2 antagonist–mediated facilitationof PFC inputs may induce secondary attenuation of hippocampusafferent drive. We do not believe that the decrease in phasic DAeffects was due to non-specific effects of the antagonists, given thatthe D2 antagonist shows approximately 10,000-fold higher affinityfor D2 than for D1 receptors18 and that infusion of the same doseof this antagonist in a previous study19 did not interfere withD1-mediated responses.

We used two types of behavioral tests to examine the functionalimpact of DA modulation of hippocampus and PFC inputs into theNAcc: VCT, which required an allocentric response strategy, and RDT,which required an egocentric response strategy. Allocentric and ego-centric spatial learning are reported to be mediated separately by thehippocampus and striatum20. We have shown here that both of theselearning strategies are also dependent on the interaction between thehippocampus and NAcc and modulation of the interaction by phasicDA-mediated stimulation of D1 receptors. In contrast, studies haveshown that the PFC and its interaction with the striatum are involved inbehaviors related to set shifting and response inhibition21–23. Our resultsshow that this PFC-mediated behavior is also dependent on tonic DArelease and D2 receptor stimulation within the NAcc, with tonic DArelease suppressing PFC-dependent set shifting behaviors. Although wetested task switching only from the VCT to the RDT, other studies24

using similar tasks and task switching have shown that reversing theorder of the task does not affect the results of the experiments.

Also, the results of the behavioral studies were not influenced byinducing deficits in motor function or motivation, as similar asym-metric disconnection studies of limbic-NAcc and PFC-NAcc pathwayshave not reported a disruption in motor control or motivationto reward as significant factor in their results25–27. Indeed, drugsor lesions typically must be given bilaterally in these regions in orderto disrupt motor performance or motivated behavior. Furthermore,we did not observe motor slowing or differences in the time that

PFC PFC

HPC HPCD1 D1

D2 D2

PF

C: l

idoc

aine

HP

C: l

idoc

aine

NAcc: D2 agonist NAcc: D1 antagonist

HPC-NAcc

PFC PFC

HPC HPCD1 D1

D2 D2

PFC-NAccHPC-NAcc

PFC PFC

HPC HPCD1 D1

D2 D2

PFC PFC

HPC HPCD1 D1

D2 D2

PFC-NAcc

a

b c

NAcc

PFC-NAccHPC-NAcc

NAcc NAcc NAcc

NAcc NAcc NAcc NAcc

+

HPC

-sal

PFC

-sal

HPC

-sal

PFC

-sal

HPC

-D1

HPC

-D1

HPC

-D2

HPC

-D2

PFC

-D1

PFC

-D1

PFC

-D2

PFC

-D2

VCT RDT120

90

60

30

0

Num

ber

of tr

ials

*** 12

0

4

8

16

Per

seve

rativ

e er

rors

*

Figure 7 The effects of hippocampus-NAcc and PFC-NAcc functional

disconnection by unilateral inactivation of the hippocampus or PFC combined

with contralateral D1 antagonist or D2 agonist infusion into the NAcc on

performance in the plus maze tasks. (a) Experimental design.

Electrophysiological data suggest that lidocaine infusion into the PFC

combined with D2 agonist, but not D1 antagonist, infusion into the NAcc

should disrupt PFC-NAcc afferent inputs. On the other hand, lidocaine

infusion into the hippocampus (HPC) combined with D1 antagonist, but notD2 agonist, infusion into the NAcc should disrupt hippocampus-NAcc

afferent inputs. (b) Number of trials that animals required to reach criterion

in visual cue task (VCT) and response direction task (RDT). The number of

trials in hippocampus-D1 animals was significantly increased both in VCT

and RDT (*, **, P o 0.05 compared with VCT and RDT, respectively, in

control group, one-way ANOVA with Dunett’s post hoc comparison). The

number of trials in RDT in PFC-D2 animals was also significantly increased.

Error bars indicate s.d. (+, P o 0.05 compared to RDT in control group).

(c) Perseverative errors are significantly increased only in PFC-D2 animals

(*, P o 0.05 compared with control), indicating that increased number

of trials in DRT in PFC-D2 animals is due to perseverative errors at

task switching.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

animals took to reach the reward when comparing the control andexperimental groups.

Our results suggest that a tonic, basal level of DA release maintainsthe balance between limbic and cortical inputs in the NAcc, andtherefore both increases and decreases in DA release exert a potenteffect on information processing in this region (Supplementary Note,Supplementary Fig. 1). Disruption of this balance may be one ofthe central components in the pathophysiology of schizophrenia28,29,and antipsychotic drugs may achieve their therapeutic effectsthrough restoration of the balance between limbic and cortical inputsin the NAcc30–33.

METHODSRecording. All experiments were conducted in accordance with the US

National Institutes of Health Guide for the Care and Use of Laboratory Animals

and were approved by the University of Pittsburgh Institutional Animal Care

and Use Committee.

In vivo field potential recordings were conducted in adult male Sprague-

Dawley rats (250–350 g). Rats were anesthetized with chloral hydrate (400 mg/

kg) and placed in a stereotaxic apparatus. Extracellular field potential electrodes

pulled from glass micropipettes and filled with 2 M NaCl were lowered into the

NAcc core at a 101 angle (Figs. 1a and 2c). Field potential signals were

amplified 1,000 times with an AC amplifier and were band-pass filtered at 0.1–

100 Hz. Recordings were digitized with an interface board at 10 kHz and were

analyzed off-line using custom-made software (Neuroscope).

Concentric bipolar stimulation electrodes were placed in the hippocampus

(ventral CA1/subiculum) and PFC (prelimbic cortex; Figs. 1a and 2a,b).

Single-current pulses (0.2 ms; 0.2–1.0 mA) were delivered every 30 s alternately

to the hippocampus and PFC.

Drug administration. DA agonists and antagonists were administered via

reverse microdialysis19. The probes were located within 500 mm of the tip of the

recording electrodes (Fig. 1a, Fig. 2c). Dialysis probes (2 mm exposed

membrane) were advanced into the NAcc at the rate of 3–5 mm/s to minimize

damage to brain tissues. Drugs were dissolved in artificial cerebrospinal fluid

(aCSF; SKF38393, 10 mM; SCH23390, 10 mM; quinpirole, 10 mM; eticlopride,

20 mM). We continuously perfused aCSF throughout the experiments, and we

switched to drug-containing solutions during recordings.

Ventral pallidum and PPTg activation and inactivation were conducted by

local injection of GABA antagonist and agonists through infusion cannulae

(Figs. 1a and 2d,e)8. GABA agonists (a mixture of 0.2 mg of the GABAA agonist

muscimol and 0.2 mg GABAB agonist baclofen) and a GABAA antagonist

(0.1 mg bicuculline methiodide) were dissolved in 0.5 ml aCSF. Guide cannulae

(28-gauge) were lowered into the ventral pallidum or PPTg and were attached

on the skull by dental cement. Infusion cannulae (30-gauge) were inserted into

the guide cannulae for drug injection. After 2 min of injection, cannulae

remained in place for an additional 3 min before removal.

The drug doses employed were identical to those that we had previously

found to modulate tonic and burst spike firing in VTA DA neurons (and conse-

quently DA release in the NAcc)8. The doses of DA agonists/antagonists admini-

stered via reverse microdialysis have also been shown to selectively produce

receptor subtype-specific alterations of neuronal activity in striatal neurons19.

Behavioral test. Strategy learning and switching in goal-directed behavior were

tested with a visual cue task (VCT) and a response direction task (RDT) using a

plus maze, as described in other studies24.

After 3 d of intensive handling (10 min each day) and another 3 d of

habituation to the maze, intracranial cannulae implantation was done. Rats

were anesthetized with sodium pentobarbitol (50 mg/kg), and 28-gauge guide

cannulae were placed in the left hemisphere of either the PFC or hippocampus,

and the right hemisphere of the NAcc at similar regions where electrophysio-

logical recordings were conducted (Fig. 2a–c). After 1 week of recovery, these

animals were subjected to behavioral tests. During handling and maze

habituation, animals were food-restrained to maintain about 85% of normal

body weight.

VCT was assigned first, during which animals had to make a right or left

turn toward the arm of the maze where the visual cue was placed to obtain

rewards. After performance reached criterion (ten consecutive correct trials in

one session; one session consisted of 12 trials; two sessions per day), the task

was switched to RDT, in which animals had to make a turn based on direction

(animals had to make a left or right turn regardless of the visual cue), and

sessions were continued until performance criterion was reached. Perseverative

errors were defined as the number of error trials until animals made the first

correct turn in the trial in which the visual cue was placed in the arm opposite

to the direction of the turn.

All drugs were administered via 30-gauge infusion cannulae at 0.2 ml/min,

beginning 5 min before the first sessions of the day. Lidocaine (20 mg/0.5 ml

aCSF) was infused into the PFC or hippocampus, and either saline, the D1

antagonist (SCH23390, 1.0 mg/ 0.5 ml) or the D2 agonist (quinpirole, 10 mg/ 0.5

ml) was infused into the NAcc. The doses of drugs were based on those used

previously to selectively alter behaviors26,34. Animals usually completed two

sessions (24 trials per day, with 10-s intertrial intervals) within 30 min and took

3–4 d to reach criterion. Thus, the effects of the lidocaine administration

should have persisted throughout the time period of the sessions for each day.


ACKNOWLEDGMENTSWe thank N. Macmurdo and C. Smolak for technical assistance, B. Lowry fordata acquisition software and A. West and M. Takita for suggestions on reversemicrodialysis technique. This work was supported by US National Instituteof Mental Health MH57440 (A.A.G.) and a National Alliance for Research onSchizophrenia and Depression (NARSAD) Young Investigator Award (Y.G.).Y.G. is a NARSAD Essel Investigator.




1. Mogenson, G.J., Jones, D.L. & Yim, C.Y. From motivation to action: functional interfacebetween the limbic system and the motor system. Prog. Neurobiol. 14, 69–97 (1980).

2. Groenewegen, H.J., Wright, C.I., Beijer, A.V. & Voorn, P. Convergence and segregation ofventral striatal inputs and outputs. Ann. NY Acad. Sci. 877, 49–63 (1999).

3. Grace, A.A. Gating of information flow within the limbic system and the pathophysiologyof schizophrenia. Brain Res. Brain Res. Rev. 31, 330–341 (2000).

4. Voorn, P., Jorritsma-Byham, B., Van Dijk, C. & Buijs, R.M. The dopaminergic innervationof the ventral striatum in the rat: a light- and electron-microscopical study withantibodies against dopamine. J. Comp. Neurol. 251, 84–99 (1986).

5. Berridge, K.C. & Robinson, T.E. What is the role of dopamine in reward: hedonicimpact, reward learning, or incentive salience? Brain Res. Brain Res. Rev. 28, 309–369 (1998).

6. Kelley, A.E. & Berridge, K.C. The neuroscience of natural rewards: relevance to addictivedrugs. J. Neurosci. 22, 3306–3311 (2002).

7. Grace, A.A. Phasic versus tonic dopamine release and the modulation of dopaminesystem responsivity: a hypothesis for the etiology of schizophrenia. Neuroscience 41,1–24 (1991).

8. Floresco, S.B., West, A.R., Ash, B., Moore, H. & Grace, A.A. Afferent modulation ofdopamine neuron firing differentially regulates tonic and phasic dopamine transmission.Nat. Neurosci. 6, 968–973 (2003).

9. Creese, I., Sibley, D.R., Hamblin, M.W. & Leff, S.E. The classification of dopaminereceptors: relationship to radioligand binding. Annu. Rev. Neurosci. 6, 43–71 (1983).

10. Humeau, Y., Popoff, M.R., Kojima, H., Doussau, F. & Poulain, B. Rac GTPase plays anessential role in exocytosis by controlling the fusion competence of release sites.J. Neurosci. 22, 7968–7981 (2002).

11. Foster, K.A. & Regehr, W.G. Variance-mean analysis in the presence of a rapid antagonistindicates vesicle depletion underlies depression at the climbing fiber synapse. Neuron43, 119–131 (2004).

12. Goto, Y. & O’Donnell, P. Timing-dependent limbic-motor synaptic integration in thenucleus accumbens. Proc. Natl Acad. Sci. USA 99, 13189–13193 (2002).

13. Albertin, S.V., Mulder, A.B., Tabuchi, E., Zugaro, M.B. & Wiener, S.I. Lesions of themedial shell of the nucleus accumbens impair rats in finding larger rewards, but sparereward-seeking behavior. Behav. Brain Res. 117, 173–183 (2000).

14. Di Chiara, G. Nucleus accumbens shell and core dopamine: differential role in behaviorand addiction. Behav. Brain Res. 137, 75–114 (2002).

15. Charara, A. & Grace, A.A. Dopamine receptor subtypes selectively modulate excitatoryafferents from the hippocampus and amygdala to rat nucleus accumbens neurons.Neuropsychopharmacology 28, 1412–1421 (2003).

16. Greengard, P., Allen, P.B. & Nairn, A.C. Beyond the dopamine receptor: the dopami-neRPP-32/protein phosphatase-1 cascade. Neuron 23, 435–447 (1999).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

17. Phillips, P.E. & Wightman, R.M. Extrasynaptic dopamine and phasic neuronal activity.Nat. Neurosci. 7, 199 (2004).

18. Hall, H., Kohler, C. & Gawell, L. Some in vitro receptor binding properties of [3H]eti-clopride, a novel substituted benzamide, selective for dopamine-D2 receptors in the ratbrain. Eur. J. Pharmacol. 111, 191–199 (1985).

19. West, A.R. & Grace, A.A. Opposite influences of endogenous dopamine D1 and D2receptor activation on activity states and electrophysiological properties of striatalneurons: studies combining in vivo intracellular recordings and reverse microdialysis.J. Neurosci. 22, 294–304 (2002).

20. Kesner, R.P. & Rogers, J. An analysis of independence and interactions of brainsubstrates that subserve multiple attributes, memory systems, and underlying pro-cesses. Neurobiol. Learn. Mem. 82, 199–215 (2004).

21. Meck, W.H. & Benson, A.M. Dissecting the brain’s internal clock: how frontal-striatalcircuitry keeps time and shifts attention. Brain Cogn. 48, 195–211 (2002).

22. Roberts, A.C. et al. 6-Hydroxydopamine lesions of the prefrontal cortex in monkeysenhance performance on an analog of the Wisconsin Card Sort Test: possible interactionswith subcortical dopamine. J. Neurosci. 14, 2531–2544 (1994).

23. Monchi, O., Petrides, M., Petre, V., Worsley, K. & Dagher, A. Wisconsin card sortingrevisited: distinct neural circuits participating in different stages of the task identified byevent- related functional magnetic resonance imaging. J. Neurosci. 21, 7733–7741(2001).

24. Ragozzino, M.E., Ragozzino, K.E., Mizumori, S.J. & Kesner, R.P. Role of the dorsomedialstriatum in behavioral flexibility for response and visual cue discrimination learning.Behav. Neurosci. 116, 105–115 (2002).

25. Floresco, S.B., Seamans, J.K. & Phillips, A.G. Selective roles for hippocampal,prefrontal cortical, and ventral striatal circuits in radial-arm maze tasks with or withouta delay. J. Neurosci. 17, 1880–1890 (1997).

26. Floresco, S.B. & Phillips, A.G. Dopamine and hippocampal input to the nucleusaccumbens play an essential role in the search for food in an upredictable environment.Psychobiol 27, 277–286 (1999).

27. Christakou, A., Robbins, T.W. & Everitt, B.J. Prefrontal cortical-ventral striatal interac-tions involved in affective modulation of attentional performance: implications forcorticostriatal circuit function. J. Neurosci. 24, 773–780 (2004).

28. Meyer-Lindenberg, A. et al. Reduced prefrontal activity predicts exaggerated striataldopaminergic function in schizophrenia. Nat. Neurosci. 5, 267–271 (2002).

29. Csernansky, J.G. & Bardgett, M.E. Limbic-cortical neuronal damage and the pathophy-siology of schizophrenia. Schizophr. Bull. 24, 231–248 (1998).

30. Grunder, G., Carlsson, A. & Wong, D.F. Mechanism of new antipsychotic medications:occupancy is not just antagonism. Arch. Gen. Psychiatry 60, 974–977 (2003).

31. Creese, I., Burt, D.R. & Snyder, S.H. Dopamine receptor binding predicts clinical andpharmacological potencies of antischizophrenic drugs. Science 192, 481–483 (1976).

32. Seeman, P. Dopamine receptors and the dopamine hypothesis of schizophrenia.Synapse 1, 133–152 (1987).

33. Wong, D.F. et al. Positron emission tomography reveals elevated D2 dopamine receptorsin drug-naive schizophrenics. Science 234, 1558–1563 (1986).

34. Swanson, C.J., Heath, S., Stratford, T.R. & Kelley, A.E. Differential behavioral responsesto dopaminergic stimulation of nucleus accumbens subregions in the rat. Pharmacol.Biochem. Behav. 58, 933–945 (1997).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Instructive signals for motor learning from visualcortical area MT

Megan R Carey1,2, Javier F Medina1 & Stephen G Lisberger1

Sensory error signals have long been proposed to act as instructive signals to guide motor learning. Here we have exploited the

temporal specificity of learning in smooth pursuit eye movements and the well-defined anatomical structure of the neural circuit

for pursuit to identify a part of sensory cortex that provides instructive signals for motor learning in monkeys. We show that

electrical microstimulation in the motion-sensitive middle temporal area (MT) of extrastriate visual cortex instructs learning in

smooth eye movements in a way that closely mimics the learning instructed by real visual motion. We conclude that MT provides

instructive signals for motor learning in smooth pursuit eye movements under natural conditions, suggesting a similar role for

sensory cortices in many kinds of learned behaviors.

We learn from our mistakes. Incorrect movements generally arefollowed by sensory feedback indicating that an error has been made.Many learning theories1–3 propose that neural representations ofsensory error signals are the teachers for motor learning. Yet we under-stand only poorly how and where sensory error signals are convertedinto neural instructive signals that induce behavioral learning.

We have used smooth pursuit eye movements of rhesus monkeys tostudy the neural mechanisms of motor learning. Pursuit allowsprimates to track smoothly moving targets by matching eye velocityto target velocity4,5, thereby minimizing image motion relative to theeye. If a target changes direction or speed unexpectedly, then the delayin visual-motor processing for pursuit will result in image motionbecause there is a delay of 100 ms or more before eye velocity reacts tothe new target velocity. If the same change in the direction or speed oftarget motion occurs repeatedly at a consistent time, then the visual tomotor transformation for pursuit undergoes experience-dependentlearning. Over time, the pursuit response to the initial target motionis altered so that it anticipates the change in target motion6–8. Thus,pursuit shows associative learning that can be instructed by a consistentchange in target trajectory. In real life, continual pursuit learningensures excellent tracking of naturally-occurring object motion.

In the present paper, we have asked whether representations of imagemotion in area MT provide instructive signals for pursuit learning. MTneurons are selective for direction and speed of retinal image motion aswell as a variety of other features of the visual stimulus9,10. Prior studieshave shown that visual motion signals for pursuit are represented inMT11–13 and that microstimulation of clusters of MT neurons withshared direction preferences has a directional effect on pursuit eyemovements14,15 and perceptual judgments16,17. Here, we move beyondthese previous observations to show that microstimulation createsinstructive signals capable of driving learning in pursuit. Further, we

demonstrate that the learning evoked by electrical stimulation in MTclosely resembles that evoked by real visual motion, suggesting thatactivation of MT provides a powerful instructive signal for pursuitlearning under natural conditions.

RESULTS

Learning induced by microstimulation in MT

We asked whether microstimulation in area MT, when consistentlyassociated with motion of a pursuit target, would be sufficient toinstruct directional learning of pursuit eye movements. Our procedurewas designed to mimic experiments that used a precisely timed changein the direction of target motion as the instructive signal for directionallearning in pursuit7, but here we replaced the change in target directionwith MT microstimulation.

The basic pursuit stimulus consisted of individual trials of step-ramptarget motion4. The target moved at 201 s–1 along the ‘pursuit axis,’which was chosen for each experiment as the cardinal direction closestto orthogonal to the preferred direction of the MT neurons recorded ata given stimulation site. We measured learning in the ‘learning axis,’which was defined as the cardinal axis closest to the preferred directionof the multiunit activity at the stimulation site and was always ortho-gonal to the pursuit axis. An example of a microstimulation-learningexperiment is illustrated in Figure 1 for a site where the MT neuronspreferred upwards motion. Each experiment began with a ‘baselineblock’ of pre-learning ‘probe trials’ that consisted only of target motionin the pursuit direction (here rightward) to verify that before learning,rightward target motion elicited very little vertical eye motion (Fig. 1b,top). Next we presented ‘learning trials’ in which microstimulation wasapplied in MT for 300 ms starting 200 ms after the onset of targetmotion (Fig. 1a, middle). Stimulation at this site during ongoingpursuit elicited a small smooth eye movement with a major upward


1Howard Hughes Medical Institute, W.M. Keck Foundation Center for Integrative Neuroscience, Neuroscience Graduate Program, and Department of Physiology, Universityof California, San Francisco, California 94143-0444, USA. 2Present address: Department of Neurobiology, Harvard Medical School, 220 Longwood Avenue, Boston,Massachusetts 02115, USA. Correspondence should be addressed to M.R.C. ([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

component15 (Fig. 1b, middle). After at least 100 learning trials, weassessed learning by measuring the vertical smooth eye movement onpost-learning probe trials of rightward target motion in which stimula-tion was not delivered in MT (Fig. 1a, bottom); infrequent probe trialswere interspersed among continuing learning trials.

The learned vertical eye movement had two components: an earlypeak followed by a later trough (Fig. 1b, bottom). The early componentof learning was upward, towards the preferred direction of the neuronsat the stimulation site, and peaked at around the time when micro-stimulation began in the learning trials. The later component of thelearned eye movement was in the opposite direction, comprising aseemingly wrong-way response that was not observed when theinstructive stimulus for learning was a change in target direction7.

At each individual stimulation site, we customized the directions oftarget motion for the preferred directions of neurons at that site. Forease of presentation, however, we have rotated and reflected the datafrom each experiment, and will present our findings as if the learningdirection represented by the site in MT were upward, and the pursuitdirection used for real target motion were rightward. To quantify theamount of learning, we subtracted the vertical eye velocity recordedduring probe trials in the baseline block from that recorded in identicalprobe trials toward the end of the learning block. Averaging theresponses from all experiments in each monkey shows that the smootheye movements evoked by stimulation on learning trials were unidi-rectional (Fig. 2a), whereas the learned eye movements were bidirec-tional (Fig. 2b). To represent variability across experiments, we also

have summarized the learned eye velocities by plotting the amplitudesof the peak and trough for each individual experiment as a function ofthe times at which they occurred and connecting the points for eachindividual experiment with a line (Fig. 2c). The amplitudes of the earlypeaks showed a wide range, from very small to as large as 11 s–1, but therelationship between the time and amplitudes of the peaks was quiteconsistent: all lines in Figure 2c had similar, negative slopes. Onaverage, the peaks in learned eye velocity occurred around the timecorresponding to the onset of microstimulation on learning trials,whereas the oppositely-directed troughs occurred later, within the300 ms interval that would have contained microstimulation onlearning trials.

We found considerable variability across stimulation sites in theamplitudes of the learning induced by MT microstimulation (Fig. 2c).The variability did not correlate with any of our measures of thereceptive field properties of each site, including direction tuning,preferred speed, spatial receptive field location and size, amplitude ofthe stimulation-evoked eye movement or whether the site responded

Pre-learning probe

StimLearning

0 200 400 600 800

Post-learning probe

Time from stimulation onset (ms)–200

Vertical eye velocity

0 200 400 600 800

3° s

–1

–200

a b

0 200 400 600–200

a

Learning

Monkey Q

MT stim

0 200 400 600–200

Monkey E

1° s

–1

MT stim

0 200 400 600–200

bProbe

Time from stimulation onset (ms)0 200 400 600–200

1° s

–1

250 ms

250 ms

0 200 400–200

Lear

ned

eye

velo

city

(1°

s–1

)

–1.5

–1.0

–0.5

0

0.5

1.0c

Time from stimulation onset (ms)0 200 400–200

Figure 1 A sample experiment using MT microstimulation to instruct

learning. (a) Schematic of the experimental design for a learning experiment.

From top to bottom, the three schematics show the probe trials in the

baseline block and the learning and probe trials in the learning block. The

bold arrow indicates the rightward motion of the pursuit target as a function

of time. The vertical dotted lines and the horizontal bar labeled ‘Stim’

indicate the interval when MT was stimulated in the learning trials.

(b) Averaged eye velocity along the learning axis, which was vertical in thisexperiment. In each panel, the black trace shows vertical eye velocity as a

function of time and the horizontal dashed line indicates zero eye velocity. In

the second and third panels, the gray trace reiterates the vertical eye velocity

in the pre-learning probe trials, also shown by the black trace in the top

panel. For this and all figures, learning trial data are from early in the learning

block, before learning has occurred, and learned eye movements on probe

trials are after at least 100 learning trials. Upward deflection of the velocity

traces corresponds to upward eye motion.

Figure 2 Summary of learned eye velocity when stimulation in MT provided

the instructive signal for learning. (a) Time course of averaged eye velocity

responses to MT stimulation in learning trials. The two vertical dotted

lines and the horizontal bar labeled ‘MT stim’ indicate the interval of

microstimulation. Data were taken from the first few learning trials before

measurable learning had occurred. (b) Time course of learned eye movements

(7 s.e.m.) along the learning axis, measured in infrequent probe trials.

Microstimulation started 200 ms after the onset of target motion in Monkey

Q and either 200 or 250 ms after the onset of target motion for Monkey E.

Data in a and b are averages across experimental days (n ¼ 30 for Monkey Q,

n ¼ 6 for each temporal interval for Monkey E). (c) Summary of the two

components of learned eye movements. Each point plots the positive or

negative peak of learned eye velocity as a function of the time of the peak.

Each pair of open symbols connected by a line shows data from an individualexperiment. Filled symbols connected by bold lines show averages across all

experiments. For Monkey E, circles and continuous lines show data obtained

with an interval of 200 ms; squares and dashed lines show data obtained

with an interval of 250 ms. Positive values of eye velocity indicate eye motion

in the learning direction.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

better to the pursuit target or to the ensuing image motion that resultedfrom the motion of the eye across a stationary background (Supple-mentary Fig. 1).

Origin of the oppositely directed component of learning

The relative timing and directions of the two components of learningsuggested that the first component was instructed by image motionsignals resulting directly from MT stimulation, whereas the later,oppositely directed component might be instructed by real imagemotion signals. Microstimulation in MT evoked an eye movementon learning trials that drove the eye away from the target, causing targetmotion relative to the eye. Thus, stimulation provided consistent realimage motion signals opposite to the direction of the stimulation-evoked eye movement. The real, target-related image motion signalswould be delayed relative to those mimicked by stimulation in MT,because of (i) the latency between stimulation and the evoked eyemovement and (ii) visual system processing delays. Thus, visualmotion signals resulting from MT microstimulation and the ensuingtarget motion relative to the eye would predict oppositely directed,temporally segregated components of learning like those we observed(Figs. 1 and 2).

To test the hypothesis that target-related image motion was respon-sible for the second, oppositely directed component of learning weobserved (Figs. 1 and 2), we stabilized the target on the eye andeliminated image motion during stimulation of MT by using therecorded eye position to drive target motion in real time. Microstimu-lation with target stabilization on learning trials evoked unidirectionaleye movements (Fig. 3a) that were similar to those evoked withoutstabilization (Fig. 2a). However, the learned eye velocity expressed inprobe trials was now also unidirectional and was always towards thepreferred direction of the stimulation site (Fig. 3b). Individual experi-ments and trials also showed the basic features of the averages(Supplementary Fig. 2). Thus, target stabilization eliminated thesecond, oppositely directed component of learning. We conclude thatvisual signals resulting from microstimulation in MT and fromreal target motion relative to the eye are both sufficient to instructdirectional learning in pursuit. Putting these signals in conflict causesa response that reflects the influence of both signals and theirrelative timing.

Microstimulation-induced versus visually induced learning

If MT provides instructive signals for pursuit learning under naturalconditions, then we should find excellent quantitative agreementbetween the learned eye movements induced by microstimulation inMT and those induced by natural visual stimuli. Our previous experi-ments using a change in target direction to induce learning7 providedonly unidirectional instructive signals. Therefore, we conducted beha-vioral experiments with a natural visual stimulus designed to mimic the

0 200 400 600–200

Learning

Monkey Q

MT stim

0 200 400 600–200

Monkey E

1° s

–1

MT stim

0 200 400 600–200

Probe

Time from stimulation onset (ms)

0 200 400 600–200

1° s

–1

a

b

0 200 400 600–200

Monkey Q

Target not stabilized

0 200 400 600–200

Monkey E

10° s

–1

0 200 400 600–200

Time from background motion onset (ms)

0 200 400 600–200

1° s

–1

10° s

–11°

s–1

0 200 400 600–200

Bkgd up

Learning trials

0 200 400 600–200

Probe trials

0 200 400 600–200

Monkey Q

Target stabilized

0 200 400 600–200

Monkey E

0 200 400 600–200

Time from background motion onset (ms)

0 200 400 600–200

a

b

c

d

Figure 4 Learning instructed by motion of a visual background during orthogonal target motion. The diagrams on the left summarize the visual stimuli used to

induce learning (top) and to evaluate the learned eye velocity (bottom). The interval indicated by the two vertical dotted lines and the horizontal bar labeled‘Bkgd up’ indicates the time when the background stimulus moved upwards, starting 200 ms after target motion onset. (a,b) Learning instructed by

background motion without target stabilization. (c,d) Learning instructed by background motion with target stabilization on the learning trials. (a,c) Averaged

eye velocity responses during the first few learning trials, before measurable learning had occurred. Data are averages across experimental days (n ¼ 7 for each

monkey) and are plotted as a function of time from onset of visual background motion (time 0). (b,d) Learned eye movements on probe trials after at least 100

learning trials. Data are plotted as mean 7 s.e.m. as a function of the time that visual background motion began on learning trials.

Figure 3 Summary of learned eye velocity when stimulation in MT combined

with target stabilization provided the instructive signal for learning. (a) Time

course of averaged eye velocity responses to MT stimulation in learning trials.

The two vertical dotted lines and the horizontal bar labeled ‘MT stim’ indicate

the interval of microstimulation and target stabilization. Data were taken

from the first few learning trials before measurable learning had occurred.

(b) Time course of learned eye movements (7 s.e.m.) along the learning axis,

measured in infrequent probe trials. Microstimulation started 200 ms afterthe onset of target motion in both monkeys. Data show averages across eight

and five experimental days for Monkeys Q and E, respectively.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

sequence of signals in MT and real image motion provided duringmicrostimulation; we replaced MT stimulation with the motion of avisual background.

As before, each trial began when the monkey fixated a pursuit target;shortly thereafter a texture of stationary random dots appeared (Fig. 4,left). On pre-learning and post-learning probe trials, the visual back-ground remained stationary as the pursuit target underwent rampmotion. On learning trials, the visual background remained stationaryfor the first 200 ms of pursuit target motion and then moved at 201 s–1

for 300 ms in a direction orthogonal to the direction of ongoingpursuit. In learning trials, the motion of the background evokedsmooth eye movement towards the direction of background motion18

(Fig. 4a). In subsequent probe trials, after at least 100 learning trials, thelearned eye velocity had two components, one towards the direction ofthe background motion and one in the opposite direction (Fig. 4b).When the target was stabilized relative to the eye, the response tobackground motion on learning trials with stabilization (Fig. 4c) wassimilar to that without target stabilization, but the learned eye move-ment was entirely in the direction of the background motion (Fig. 4d).The second, oppositely directed component of the learned eye velocityagain vanished. We conclude that both pursuit target-related andbackground motion are capable of instructing pursuit learning andthat microstimulation in MT is an effective mimic of these signals.

Comparison of the timing of the learned eye velocities in the presentstudy shows temporal differences between the learning instructed byMT microstimulation versus motion of a visual background. Weplotted the magnitudes of the peak and trough of learned eye velocityas a function of their times for both microstimulation and backgroundexperiments (Fig. 5), aligned at the onsets of MT stimulation orbackground motion (t ¼ 0). The general shape of the learned eye

movements was similar, but those instructed by MT stimulationoccurred earlier relative to the onset of the experimental manipulationthan did those instructed by motion of the visual background. How canwe understand these temporal differences?

If activity in MT were the instructive signal for both microstimula-tion and background-motion experiments, then the timing of thelearned eye movements for the two learning conditions should be offsetby an interval corresponding to the latency of MT neuronal responsesto visual stimuli. To evaluate this prediction, we varied the intervalbetween the onset of pursuit target motion and instructive backgroundmotion on different days. Aligning the learned eye velocities on theonset of pursuit target motion (Fig. 6a) reveals that changing thetemporal intervals affected both the shape and timing of the learned eyevelocity, as predicted from ref. 7. Lengthening the interval between theonset of horizontal target motion and the vertical instructive signalfrom 100 ms to 150 ms or to 200 ms caused a longer latency and a largerpeak in the learned eye movement towards the preferred direction ofthe stimulation site. There was a similar effect (Fig. 2b) when theinterval between the onset of target motion and MT stimulation wasvaried in Monkey E.

We obtained the same temporal profile of learned eye movementsonly when we compared the learning induced by electrical stimulationfor one interval with that induced by background motion that started50 to 75 ms closer to the onset of visual target motion (Fig. 6b). In

0 200 400–200

Lear

ned

eye

velo

city

(1°

s–1

)

0

0.5

1.0

–0.5

–1.0

–1.5

Time from stimulation onset (ms)

Monkey Q

0 200 400–200

Monkey E

MT meanBkgd mean

0 200 400 600 800

Lear

ned

eye

velo

city

(1°

s–1

)

0

0.5

–0.5

Bkgd 100Bkgd 150Bkgd 200

Monkey Q

0 200 400 600 800

Bkgd 150Bkgd 200

Monkey E

0 200 400 600 800

Lear

ned

eye

velo

city

(1°

s–1

)

0

0.5

–0.5

–1.0

1.0 Bkgd 100Bkgd 150Bkgd 200MT 200

Time from target motion onset (ms)

0 200 400 600 800

Bkgd 150Bkgd 200MT 200MT 250

a

b

Figure 5 Temporal relationship between learning instructed by MT

microstimulation and visual background motion. Each point plots the positive

or negative peak of learned eye velocity as a function of the time of the peak.

Each pair of open symbols connected by a line shows data from an individual

visual background motion experiment. Filled circles connected by bold lines

show averages across all background experiments. Filled squares and bold

dashed lines show averages of the learning instructed by MT stimulation.

Positive values of eye velocity indicate eye motion in the learning direction.Time zero indicates the onset of visual background motion or MT

microstimulation in learning trials, which was 200 ms after target motion

onset in both cases.

Figure 6 Comparison of learning instructed by MT stimulation and visual

background motion at different times after the onset of pursuit target motion.

(a) Average time courses of learned eye velocity: green, blue and red traces

show results for experiments in which learning was induced by visual

background motion that began 100, 150 or 200 ms after the onset of pursuit

target motion, respectively. The colored arrows indicate the times of visual

background motion in the learning trials for each experiment. (b) Each point

plots the average positive or negative peak of the learned eye velocity as a

function of the time of the peak. Each pair of symbols connected by a line

shows data from an individual learning condition. Bold circles connected by

dashed lines indicate the results of learning instructed by stimulation of MT.

Smaller circles connected by thin lines indicate the results of learning

instructed by background motion. For experiments with background motion,the colors have the same meaning as in a. For experiments with MT

stimulation, the color scheme has been shifted by 50 ms: blue and red

symbols indicate the MT stimulation occurred 200 or 250 ms after the onset

of pursuit target motion in learning trials. In both a and b, time zero indicates

the onset of pursuit target motion.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Monkey Q, the learned eye velocity for MTstimulation with an intervalof 200 ms had a temporal structure that fell between that for the learnedeye velocities from background motion with intervals of 100 ms and150 ms. In Monkey E, the best agreement in temporal structureoccurred when the interval between the onset of target motion andinstructive MT stimulation was more than 50 ms longer than theinterval between target motion onset and instructive backgroundmotion. These data would be expected if MT neurons were activatedby visual motion with a 50 to 75 ms latency19–22 and activity in MTprovided an instructive signal for the learning induced by backgroundmotion as well as by microstimulation.

DISCUSSION

Sensory instructive signals for pursuit learning

Sensory error signals have been hypothesized to act as instructivesignals for a number of forms of learning, and yet there are relativelyfew examples of vertebrate behaviors for which activation of a specificclass of sensory neurons has been demonstrated to be sufficient toinstruct learned changes in behavior; examples include classical con-ditioning of the eyelid response23, auditory frequency discrimination24

and now smooth pursuit eye movements.We have shown that electrical microstimulation in area MT is

sufficient as an instructive signal for learning in smooth pursuit eyemovements. Further, there is excellent agreement in the expression oflearning following instruction by MT microstimulation and by anatural visual stimulus. In particular, the learned eye movementinstructed by MT stimulation was almost identical to that instructedby motion of a visual background when the comparison took accountof the latency of MT responses to visual motion. Relative to theonset of pursuit target motion in our learning paradigm, MT stimula-tion had to occur 50 to 75 ms later than background motion, exactly asexpected if visual signals took 50 to 75 ms longer to reach the sites ofinstruction than those produced by activating MT directly. Thesimilarity of the learning in the two situations implies that activity inMT represents the instructional signals for learning even when theinstructional stimulus is provided by real image motion. Althoughother brain regions have similar response latencies and also couldcontribute instructional signals for pursuit learning, our findingsestablish a clear role for MT.

For instructional signals that arise from both MT microstimulationand real image motion, the temporal relation between the instructivestimulus and the learned eye velocity was notable. We have previouslyshown7 that learned eye movements are timed to correspond to thetime of a change in direction of the pursuit target. In the present study,we found that the learned eye velocity followed the sequence of twoimage motion instructive signals: first in one direction and then in theopposite direction owing to target motion with respect to the eye. Wewere able to link the second component of learning to the oppositelydirected image motion by experiments in which the target wasstabilized on the eye in the learning trials, to eliminate oppositelydirected image motion.

The fact that the pursuit system learned to make a rapid sequence ofeye motion in one direction and then the other instead of averaging thetwo opposing instructional signals over time indicates that it isspecialized for learning temporal properties of the instructional signalon a time scale of 100 ms. Further, both background motion andmicrostimulation were effective at instructing pursuit learning even ifthey originated from parts of the visual field that were clearly separatefrom the foveal location of the pursuit target. Thus, learning in thepursuit system seems to operate more like a machine than like acognitive process25. Pursuit learning responds to the association of

ongoing pursuit and specific temporal sequences of image motion andis driven by spatially non-specific instructional signals.

Because MT microstimulation caused an eye movement in thelearning trials, we cannot rule out the possibility that the visual signalsfrom MT have been converted at least partially into motor coordinatesbefore reaching the site of learning. In future experiments, subthresholdstimulation of MT might help resolve this issue26. For now, severalaspects of our data imply that the instructional signals for pursuitlearning are not related directly to eye movements. First, in experimentsthat did not use target stabilization, the eye movement responses onlearning trials were unidirectional, but the learned eye movements hada component in the opposite direction. Second, target stabilizationdissociated these two eye movements: it had only minimal effects onthe eye movement on learning trials, but it profoundly altered the shapeof the learned eye movement. Finally, in microstimulation and beha-vioral experiments, the first peak of the learned eye velocity is timed tocoincide with the time of MT activation and not with the time of theevoked eye movement.

Implications for sites and mechanisms of learning

Our experiments do not speak directly to the site of learning for pursuitother than to indicate that the site of learning must receive majorinputs from MT. A previous study27 provided evidence that the site ofpursuit learning is downstream from the frontal pursuit area. Ourfinding that the site of pursuit learning is also downstream from areaMT pushes candidate loci into subcortical structures, including brain-stem precerebellar nuclei and the cerebellum itself28. In primates, MT isa major source of visual motion signals to the cerebellum, accessingboth cerebellar mossy fibers and climbing fibers via the dorsolateralpontine nucleus and the nucleus of the optic tract29–31.

One aspect of our data that provides insight into potential neuralmechanisms for pursuit learning is that the pursuit system compensatesfor the temporal delay in the instructive signal from MT. Learnedeye movements began, and in some experiments reached their peaks,before the time when stimulation was delivered on learning trials. Thisfinding provides a constraint for potential plasticity mechanisms thatmight mediate pursuit learning. Specifically, these plasticity mechan-isms would be expected to have asymmetrical temporal requirements,as has been proposed for some cerebellar-based learned behaviors32,33

and demonstrated for several forms of cerebellar plasticity34–37.In conclusion, we have demonstrated that visual signals arising in

visual area MT are a powerful instructive signal for pursuit learning,cementing MT in the pathway for signals that instruct learning. We arestruck that the organization of the pursuit circuit bears strong resem-blance to the organization of circuits for a wide range of motorbehaviors. All use inputs from sensory cortex, cortico-cortical pathwaysthrough the parietal sensory-motor areas and recurrent loops throughboth the cerebellum and the basal ganglia38,39. Many researchers haveproposed that the cerebellum may be responsible for error-correctinglearning for a wide range of behaviors40–42, and pursuit has become animportant model system for understanding error-correcting learning.We propose that the nature of the instructive signals and the sites andmechanisms of learning in pursuit may provide a framework that canbe generalized to a wide range of learned motor behaviors.

METHODSAnimals. All experimental procedures were approved in advance by the

Institutional Animal Care and Use Committee of the University of California,

San Francisco and conformed to federal guidelines. Two male rhesus monkeys

(Macaca mulatta, 7–12 kg) served as subjects, and their eye movements were re-

corded with the magnetic search coil method. Surgical and training procedures


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

to prepare monkeys for the study have been described previously43,44. Both

monkeys were well trained on pursuit tasks but naive to pursuit learning

experiments. Experiments were conducted approximately five times per week

and lasted 2–5 hours. All experiments were conducted in a nearly dark room.

Visual stimuli. Visual stimuli were presented on an analog oscilloscope

(Hewlett Packard 1304A) with a refresh rate of 250 Hz. The display was

positioned 30 cm from the monkey and subtended 411 horizontal � 341

vertical of visual angle. Target position and velocity were specified through the

user interface on a DEC Alpha UNIX workstation and were controlled by the

outputs of two 16-bit digital-to-analog converters on a digital signal processing

board in a PC.

A small bright spot was presented as a pursuit target in individual trials.

Each trial began with a fixation interval of a duration that was randomized

between 500 and 1,000 ms. The target then underwent ‘step-ramp’ motion to

minimize the occurrence of catch-up saccades. The target stepped away from

the fixation point by 2.51 (Monkey Q) or 3.01 (Monkey E) and simultaneously

began moving at 201 s–1 towards the fixation point. Monkeys were rewarded

with a drop of fluid if they kept eye position close to target position throughout

the trial (see below). They completed approximately 1,500 pursuit trials, plus

several hundred receptive field-mapping trials, in each daily experiment.

In some experiments, we also presented a visual background that consisted

of randomly placed dots at an average density of 0.09 dots/deg2. The dots were

one-sixteenth as bright as the pursuit target. To avoid spatial overlap of the

target and the background, dots were visible within two separate invisible

apertures of 381 � 141, one above and one below the horizontally-moving

pursuit target. Thus, the pursuit target was centered vertically on a 61-tall

horizontal stripe in which background dots never appeared. The stripe between

the two apertures was only slightly taller than the average separation between

dots, thereby minimizing the interruption of the background pattern while

preventing the background from introducing ambiguity as to the task require-

ments by overlapping the path traversed by the target. The placement of

individual dots within the background pattern was varied randomly on a trial-

by-trial basis. Motion of the visual background consisted of 100% coherent

vertical motion of the dots at 201 s–1 within the stationary, virtual apertures.

In a subset of our experiments we stabilized targets with respect to the

moving eye45 to eliminate image motion resulting from eye motion relative to

the target on learning trials. The position of the eye was sampled every milli-

second and fed back to the computer so that it could be used to drive target

motion in real time. We used the monkeys’ eye movements in response to small

position and velocity errors to verify the accuracy of the calibration of the eye

coil45 and ensure accurate image stabilization. Comparison of the eye move-

ments evoked by target motion with and without target stabilization showed

neither the saccades nor the smooth eye accelerations that would have been

predicted by systematic errors in stabilization. In a subset of our experiments

we selectively stabilized the target only along the (horizontal or vertical)

learning axis and not along the orthogonal pursuit axis. Our results did not

depend on whether the target was stabilized along both or just one of the axes.

MT recording and microstimulation. We recorded in and stimulated MT in

three hemispheres of two monkeys (left and right hemispheres of Monkey Q,

left only in Monkey E). We used tungsten microelectrodes with impedances

from 800 KO to 1.2 MO measured at 1 kHz (Frederick Haer). Area MT was

identified as described previously44 based on stereotaxic coordinates, direc-

tional response properties of MT neurons, receptive field sizes, retinotopic

organization, and activity recorded in surrounding cortical areas. We used a

vertical approach to MT, so that we usually were able to identify the middle

superior temporal area (MST) and the lumen of the superior temporal sulcus

before the electrode entered MT. Once we understood the topographical

organization of MT in each hemisphere, we attempted to find sites with

receptive field locations in the central 101 of the visual field.

Once we entered MT, we searched for a site where the multiunit activity

indicated that nearby neurons shared a common direction preference for the

motion (usually at 161 s–1) of dots that were placed randomly within a 381 �301 aperture in the visual hemifield contralateral to the recording site. Only

strongly direction-selective sites with reasonably Gaussian direction tuning

curves were selected for further experimentation. Once a potential stimulation

site had been identified we characterized its direction and speed preferences for

motion of patches of dots, as well as its spatial receptive field location and size,

its responses to motion of pursuit targets and background textures, and the

effects of microstimulation on eye movements during ongoing pursuit.

We delivered biphasic current pulses under the control of a Grass S88

stimulator. Pulses had intensities of 30–50 mA and durations of 0.2 ms, and

occurred at a rate of 200 Hz in trains of 300-ms duration. In agreement with

the previous finding46 that small changes in electrode position away from the

center of a direction column can reduce the effectiveness of microstimulation in

MT, we found that microstimulation at sites with weak or ambiguous direction

tuning profiles usually failed to elicit smooth eye movements even during

ongoing pursuit. However, stimulation nearly always evoked a directional eye

movement at the relatively low currents we used, if performed at sites with

strong multiunit responses to motion and clear direction tuning. In contrast to

previous studies in which stimulation coincided with the onset of pursuit15,47,

the stimulation-evoked eye movements we observed during pursuit mainte-

nance were consistently towards (and not away from) the preferred direction of

the stimulation site. Because we required the motion signal injected through

stimulation to be as directional as possible, we abandoned sites at which

microstimulation applied during ongoing pursuit failed to produce a smooth

eye movement. The presence of a stimulation-evoked eye movement also

allowed us to monitor the continued effectiveness of the stimulation site

during long experiments.

Behavioral control. The learning experiments were similar in design to previous

studies on pursuit learning6,7,27 and consisted of two blocks of trials: a baseline

block and a learning block. Each block consisted of one or more trial types

presented in varying ratios in pseudorandom order. The baseline block consisted

of probe trials in which the target moved along the pursuit axis without any

additional experimental manipulations. Before proceeding to a learning block of

trials, we confirmed the consistency and stability of learning axis eye velocities

on baseline probe trials. If the eye velocities were not stable across different

fractions of the baseline block, we terminated the experiment for that day to

avoid contamination of our data from fluctuations caused by poor pursuit

performance. In up to 10% of the trials in the baseline block, microstimulation

was applied in MT starting 200 or 250 ms after the onset of target motion. The

learning block contained at least 80% learning trials and 20% or fewer probe

trials. Learning trials always consisted of pursuit target motion identical to that

of probe trials. In addition, either microstimulation in MT or motion of a visual

background was delivered for 300 ms beginning at a fixed interval (100–250 ms,

usually 200 ms) after the onset of target motion. In some experiments, the target

was stabilized on learning trials during the 300 ms segment when microstimula-

tion or visual background motion was delivered. Probe trials in the learning

block were identical to probe trials in the baseline block.

Monkeys were generally required to keep their eyes within a 31 window

around the pursuit target to complete a trial and receive a fluid reward. Our

experimental manipulations caused deviations in eye position of less than 11

and therefore should not have affected the monkey’s likelihood of successfully

completing the trial. However, to be absolutely certain that microstimulation,

background motion or learning did not affect reward probabilities, we opened

the fixation window along the learning axis to 51 during the segment of the

learning trials in which microstimulation was applied or the visual background

moved and during the corresponding segments on probe trials. In an attempt

to eliminate potential confusion between the target and the dots in the visual

background experiment, we also conducted the background experiments with

the fixation window in only the learning axis narrowed to as little as 11.

Learning was unaffected by these changes in fixation requirements; we obtained

excellent task performance and clear, bidirectional learning with the fixation

window either widened or narrowed along the learning axis.

Data analysis. Details of our data acquisition have been published before7,44.

Data from any given trial were used for analysis only if the trial had been

completed successfully by the monkey. Saccades were marked by hand using an

interactive computer program, and the portions of the eye velocity traces

corresponding to saccades were treated as missing data. If a saccade occurred

during the segment corresponding to microstimulation or visual background

motion, then the entire trial was excluded from analysis. Due to the small size


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

of the learned eye movements, we aligned responses to identical stimuli on the

onset of target motion or the time of microstimulation or visual background

motion and computed time averages of eye velocity. Responses to MT

stimulation were isolated by subtracting the eye velocity along the learning

axis on trials in which microstimulation was not delivered from that when

stimulation occurred. Learning was assessed by computing the difference

between averaged eye velocity in the learning axis for probe trials near the

end of the learning block and those in the baseline block. No attempt was made

to assess the amount of learning until B100 learning trials had been completed.


ACKNOWLEDGMENTSWe thank M. Brainard, D. Copenhagen, M. Mauk, W. Newsome, M. Orger,R. Ramachandran, P. Sabes and D. Bodznick for helpful discussion; L. Bocskai,D. Kleinhesselink, K. McGary, S. Ruffner and D. Wolfgang-Kimball for excellenttechnical support and S. Tokiyama, E. Montgomery, B. St. Amant andK. MacLeod for superb animal care. This work was supported by the HowardHughes Medical Institute, US National Institutes of Health grant NS34835,and an Achievement Rewards for College Scientists fellowship to M.R.C.




1. Schultz, W. & Dickinson, A. Neuronal coding of prediction errors. Annu. Rev. Neurosci.23, 473–500 (2000).

2. Rescorla, R.A. & Wagner, A.R. A theory of pavlovian conditioning: variations in theeffectiveness of reinforcement and nonreinforcement. in Classical Conditioning II:Current Research and Theory (eds. Black, A.H. & Prokasy, W.F.) 64–99 (AppletonCentury Crofts, New York, 1972).

3. Ito, M. Neural design of the cerebellar motor control system. Brain Res. 40, 81–84(1972).

4. Rashbass, C. The relationship between saccadic and smooth tracking eye movements.J. Physiol. (Lond.) 159, 326–338 (1961).

5. Robinson, D.A. The mechanics of human smooth pursuit eye movement. J. Physiol.(Lond.) 180, 569–591 (1965).

6. Kahlon, M. & Lisberger, S.G. Coordinate system for learning in the smooth pursuit eyemovements of monkeys. J. Neurosci. 16, 7270–7283 (1996).

7. Medina, J.F., Carey, M.R. & Lisberger, S.G. The representation of time for motor learning.Neuron 45, 157–167 (2005).

8. Boman, D.K. & Hotson, J.R. Predictive smooth pursuit eye movements near abruptchanges in motion direction. Vision Res. 32, 675–689 (1992).

9. Maunsell, J.H. & Van Essen, D.C. Functional properties of neurons in middle temporalvisual area of the macaque monkey. I. Selectivity for stimulus direction, speed, andorientation. J. Neurophysiol. 49, 1127–1147 (1983).

10. Albright, T.D. Direction and orientation selectivity of neurons in visual area MT of themacaque. J. Neurophysiol. 52, 1106–1130 (1984).

11. Newsome, W.T., Wurtz, R.H., Dursteler, M.R. & Mikami, A. Deficits in visual motionprocessing following ibotenic acid lesions of the middle temporal visual area of themacaque monkey. J. Neurosci. 5, 825–840 (1985).

12. Newsome, W.T., Wurtz, R.H. & Komatsu, H. Relation of cortical areas MT and MST topursuit eye movements II. Differentiation of retinal from extraretinal inputs. J. Neuro-physiol. 60, 604–620 (1988).

13. Lisberger, S.G. & Movshon, J.A. Visual motion analysis for pursuit eye movements in areaMT of macaque monkeys. J. Neurosci. 19, 2224–2246 (1999).

14. Komatsu, H. & Wurtz, R.H. Modulation of pursuit eye movements by stimulation ofcortical areas MT and MST. J. Neurophysiol. 62, 31–47 (1989).

15. Groh, J.M., Born, R.T. & Newsome, W.T. How is a sensory map read out? Effectsof microstimulation in visual area MT on saccades and smooth pursuit eye movements.J. Neurosci. 17, 4312–4330 (1997).

16. Salzman, C.D., Murasugi, C.M., Britten, K.H. & Newsome, W.T. Microstimulationin visual area MT: effects on direction discrimination performance. J. Neurosci. 12,2331–2355 (1992).

17. Bisley, J.W., Zaksas, D. & Pasternak, T. Microstimulation of cortical area MT affectsperformance on a visual working memory task. J. Neurophysiol. 85, 187–196 (2001).

18. Kodaka, Y., Miura, K., Suehiro, K., Takemura, A. & Kawano, K. Ocular tracking of movingtargets: effects of perturbing the background. J. Neurophysiol. 91, 2474–2483 (2004).

19. Schmolesky, M.T. et al. Signal timing across the macaque visual system. J. Neurophy-siol. 79, 3272–3278 (1998).

20. Raiguel, S.E., Xiao, D.K., Marcar, V.L. & Orban, G.A. Response latency of macaquearea MT/V5 neurons and its relationship to stimulus parameters. J. Neurophysiol. 82,1944–1956 (1999).

21. Cook, E.P. & Maunsell, J.H. Dynamics of neuronal responses in macaque MT and VIPduring motion detection. Nat. Neurosci. 5, 985–994 (2002).

22. Bair, W., Cavanaugh, J.R., Smith, M.A. & Movshon, J.A. The timing of response onsetand offset in macaque visual neurons. J. Neurosci. 22, 3189–3205 (2002).

23. Mauk, M.D., Steinmetz, J.E. & Thompson, R.F. Classical conditioning using stimulationof the inferior olive as the unconditioned stimulus. Proc. Natl. Acad. Sci. USA 83,5349–5353 (1986).

24. Talwar, S.K. & Gerstein, G.L. Reorganization in awake rat auditory cortex by localmicrostimulation and its effect on frequency-discrimination behavior. J. Neurophysiol.86, 1555–1572 (2001).

25. Kowler, E. Cognitive expectations, not habits, control anticipatory smooth oculomotorpursuit. Vision Res. 29, 1049–1057 (1989).

26. Moore, T. & Armstrong, K.M. Selective gating of visual signals by microstimulation offrontal cortex. Nature 421, 370–373 (2003).

27. Chou, I.H. & Lisberger, S.G. The role of the frontal pursuit area in learning in smoothpursuit eye movements. J. Neurosci. 24, 4124–4133 (2004).

28. Kahlon, M. & Lisberger, S.G. Changes in the responses of Purkinje cells in the floccularcomplex of monkeys after motor learning in smooth pursuit eye movements.J. Neurophysiol. 84, 2945–2960 (2000).

29. Distler, C., Mustari, M.J. & Hoffmann, K.P. Cortical projections to the nucleus of theoptic tract and dorsal terminal nucleus and to the dorsolateral pontine nucleus inmacaques: a dual retrograde tracing study. J. Comp. Neurol. 444, 144–158 (2002).

30. Alley, K., Baker, R. & Simpson, J.I. Afferents to the vestibulo-cerebellum and the originof the visual climbing fibers in the rabbit. Brain Res. 98, 582–589 (1975).

31. Brodal, P. The pontocerebellar projection in the rhesus monkey: an experimental studywith retrograde axonal transport of horseradish peroxidase. Neuroscience 4, 193–208(1979).

32. Raymond, J.L. & Lisberger, S.G. Neural learning rules for the vestibulo-ocular reflex.J. Neurosci. 18, 9112–9129 (1998).

33. Medina, J.F., Nores, W.L., Ohyama, T. & Mauk, M.D. Mechanisms of cerebellar learningsuggested by eyelid conditioning. Curr. Opin. Neurobiol. 10, 717–724 (2000).

34. Bell, C.C., Han, V.Z., Sugawara, Y. & Grant, K. Synaptic plasticity in a cerebellum-likestructure depends on temporal order. Nature 387, 278–281 (1997).

35. Wang, S.S., Denk, W. & Hausser, M. Coincidence detection in single dendritic spinesmediated by calcium release. Nat. Neurosci. 3, 1266–1273 (2000).

36. Chen, C. & Thompson, R.F. Temporal specificity of long-term depression in parallel fiber–Purkinje synapses in rat cerebellar slice. Learn. Mem. 2, 185–198 (1995).

37. Brenowitz, S.D. & Regehr, W.G. Associative short-term synaptic plasticity mediated byendocannabinoids. Neuron 45, 419–431 (2005).

38. Middleton, F.A. & Strick, P.L. Basal ganglia and cerebellar loops: motor and cognitivecircuits. Brain Res. Brain Res. Rev. 31, 236–250 (2000).

39. Tian, J.R. & Lynch, J.C. Subcortical input to the smooth and saccadic eye movementsubregions of the frontal eye field in Cebus monkey. J. Neurosci. 17, 9233–9247(1997).

40. Doya, K. Complementary roles of basal ganglia and cerebellum in learning and motorcontrol. Curr. Opin. Neurobiol. 10, 732–739 (2000).

41. Hikosaka, O., Nakamura, K., Sakai, K. & Nakahara, H. Central mechanisms of motorskill learning. Curr. Opin. Neurobiol. 12, 217–222 (2002).

42. Jueptner, M., Frith, C.D., Brooks, D.J., Frackowiak, R.S. & Passingham, R.E. Anatomy ofmotor learning. II. Subcortical structures and learning by trial and error. J. Neurophysiol.77, 1325–1337 (1997).

43. Carey, M.R. & Lisberger, S.G. Signals that modulate gain control for smooth pursuit eyemovements in monkeys. J. Neurophysiol. 91, 623–631 (2004).

44. Churchland, M.M. & Lisberger, S.G. Shifts in the population response in the middletemporal visual area parallel perceptual and motor illusions produced by apparentmotion. J. Neurosci. 21, 9387–9402 (2001).

45. Morris, E.J. & Lisberger, S.G. Different responses to small visual errors during initiationand maintenance of smooth-pursuit eye movements in monkeys. J. Neurophysiol. 58,1351–1369 (1987).

46. Murasugi, C.M., Salzman, C.D. & Newsome, W.T. Microstimulation in visual areaMT: effects of varying pulse amplitude and frequency. J. Neurosci. 13, 1719–1729(1993).

47. Born, R.T., Groh, J.M., Zhao, R. & Lukasewycz, S.J. Segregation of object and back-ground motion in visual area MT: effects of microstimulation on eye movements. Neuron26, 725–734 (2000).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 6 | JULY 2005 969

E R R ATA

Erratum: Control of synaptic strength and timing by the release-site Ca2+ signalJohann H Bollmann & Bert SakmannNat. Neurosci. 8, 426–434 (2005).

In the published version of this article, the labels were missing from the scale bars in Figure 2a. Panels a-c are reproduced correctly below.

(a) Three ∆F/F transients and evoked EPSCs (Ipost), from three experiments (thin traces, thick traces and dotted traces, respectively). Arrow indicates time of UV pulse. To reduce postsynaptic receptor desensitization, cyclothiazide (CTZ) was added to the bath, which prolongs the rise time of EPSCs. The [Ca2+] transient did not evoke measurable presynaptic currents (Ipre). Electrical artifact in Ipost removed for clarity. (b) Same traces as in a, with EPSCs normalized to their peak amplitude. (c) Two ∆F/F transients of different amplitude and evoked EPSCs, from two different cell pairs.

Erratum: Instructive signals for motor learning from visual cortical area MTMegan R Carey, Javier F Medina & Stephen G LisbergerNat. Neurosci. 8, 813–819 (2005).

The upper trace in the right panel of Figure 2b on page 814 was mislabeled as ‘250 ms’, when it should have read ‘200 ms’. The corrected figure is below.

(b) Time course of learned eye movements (± s.e.m.) along the learning axis, measured in infrequent probe trials. Microstimulation started 200 ms after the onset of target motion in Monkey Q and either 200 or 250 ms after the onset of target motion for Monkey E. Data in a and b are averages across experimental days (n = 30 for Monkey Q, n = 6 for each temporal interval for Monkey E).

2 ms

∆F/F

Ipost, scaled

5 ms

∆F/F

Ipost

Ipre

a b

500 pA

2 ms

∆F/F

Ipost

c

50%

200 pA

500 pA

0 200 400 600–200

bProbe

Time from stimulation onset (ms)0 200 400 600–200

1° s

–1

250 ms

200 ms

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

3D shape perception from combined depth cues inhuman visual cortex

Andrew E Welchman1, Arne Deubelius1, Verena Conrad1, Heinrich H Bulthoff1 & Zoe Kourtzi1,2

Our perception of the world’s three-dimensional (3D) structure is critical for object recognition, navigation and planning actions.

To accomplish this, the brain combines different types of visual information about depth structure, but at present, the neural

architecture mediating this combination remains largely unknown. Here, we report neuroimaging correlates of human 3D shape

perception from the combination of two depth cues. We measured fMRI responses while observers judged the 3D structure of two

sequentially presented images of slanted planes defined by binocular disparity and perspective. We compared the behavioral and

fMRI responses evoked by changes in one or both of the depth cues. fMRI responses in extrastriate areas (hMT1/V5 and lateral

occipital complex), rather than responses in early retinotopic areas, reflected differences in perceived 3D shape, suggesting

‘combined-cue’ representations in higher visual areas. These findings provide insight into the neural circuits engaged when the

human brain combines different information sources for unified 3D visual perception.

Visual environments are defined by multiple cues to depth structure,such as binocular disparity, perspective, texture, shading and motion.The computations and neural mechanisms required to extract thisinformation from the retinal images are likely to be radically differ-ent, but despite this, we perceive coherent 3D structures; somehow,the information provided by different depth cues is combined bythe brain1. The stereogram in Figure 1a provides an illustration ofthe combination process: in this 3D shape defined by two planesslanted in depth, horizontal binocular disparity and perspective cuesprovide different information as to the object’s 3D structure. However,when observers view such stimuli, they combine the slant informationfrom each cue to perceive an intermediate 3D shape. This is not merelya laboratory curiosity produced by inducing cue conflict; rather, thebrain routinely combines different depth cues to obtain different typesof information and reduce the effects of sensor and processing noise1–3.

Recent studies have examined the neural mechanisms that mediateprocessing of individual depth cues: namely, disparity4–6 and perspec-tive7,8. This work has considered relatively isolated depth cues ormultiple correlated cues, providing insights into sites where informa-tion about individual depth cues may converge. However, the neuralsubstrates underlying the perception of shape based on the combi-nation of cues to depth structure have not been investigated.We addressed this question using concurrent psychophysical andfMRI measurements.

We used a 3D shape stimulus depicting a hinged plane receding indepth (Fig. 1) in which the angular slant specified by two cues,horizontal binocular disparity and linear perspective, was different(‘inconsistent-cue stimulus’). Observers judged the 3D shape of thistest stimulus through comparisons with reference stimuli in which the

cues defining 3D structure were consistent (‘consistent-cue stimuli’). Ineach trial, the test stimulus and a reference stimulus were presentedsequentially, and they differed in one or both of the depth cues.

The fMRI measurements employed an event-related adaptationprocedure9,10. The technique capitalizes on neural adaptation andrepetition suppression effects whereby neural activity is lower forstimuli that have been viewed recently than for stimuli that have not.By comparing the adapted response evoked by the test stimulus showntwice (‘test-test’) with responses evoked by the test stimulus paired witha reference stimulus (‘test-reference’), we could deduce sensitivity inthe neural population to differences between the test and reference.For example, a higher fMRI response (rebound effect) when thetest-reference pair differs only in disparity information from theadapted test-test response would indicate sensitivity to changes inthe disparity cue.

By comparing psychophysical behavior and fMRI rebound effects,we demonstrate that responses in extrastriate ventral and dorsal areas,rather than in early retinotopic areas, correspond to changes inperceived 3D shape based on cue combination. These findingsprovide evidence that higher visual areas are involved in processingperceived global shape that depends on the combination of individualdepth cues.

RESULTS

Experiment 1: cue-based and percept-based processing

Observers judged which of the two viewed stimuli in a trial had a largerdihedral angle (a: openness; Fig. 1b). By presenting the inconsistent-cue test stimulus (perspective angle, Sp, ¼ 1361; disparity angle,Sd, ¼ 1161) and a range of consistent-cue reference stimuli (Sd ¼ Sp) we


1Max-Planck Institute for Biological Cybernetics, Postfach 2169, 72012 Tubingen, Germany. 2School of Psychology, University of Birmingham, Edgbaston, BirminghamB15 2TT, UK. Correspondence should be addressed to Z.K. ([email protected]).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

obtained a psychometric function (Fig. 1c). Fitting these data yieldedthe point of subjective equality (PSE; 50% point on the curve), toprovide a measure of perceived angle in the inconsistent-cue teststimulus (Sc). Observers combined the cues, perceiving an anglebetween that specified by each cue. However, the PSE (130.8 7 0.41)was closer to Sp (1361) than Sd (1161), indicating that perception wasdetermined more by perspective than by disparity11 (cue weights:wp ¼ 0.73; wd ¼ 0.27). As expected1, there was some between-subjectsvariability in cue weights. However, all observers showed evidence ofcue combination (Supplementary Table 1).

The fMRI analysis considered responses in independently definedregions of interest (ROIs; Fig. 2) within the visual cortex. We studiedearly retinotopic areas (V1, V2, V3, V3a, Vp, V4) involved in processingvisual features. We also investigated two higher visual areas implicatedin analyzing 3D information: the lateral occipital complex (LOC),which is involved in processing object shape12

, and hMT+/V5, which isimplicated in the analysis of motion13,14, structure from motionand depth15,16.

To assess sensitivity to both individual cue changes and changes inperceived 3D shape, we used five experimental conditions (Fig. 1d): (i)the Identical Condition, in which the inconsistent-cue test stimulus wasshown twice; (ii) the Disparity Change condition, in which the test-reference pair specified the same perspective information but differentdisparities; (iii) the Perspective Change condition, in which theperspective cue differed between test and reference; (iv) the BalancedChange condition, in which the cues changed in opposite directionsand (v) the All Change condition, in which the cues changed in thesame direction. For each individual subject, we computed the fMRItime course from each condition and ROI and normalized it to theresponse from the fixation baseline (Fig. 3a). We compared averagedfMRI responses at the peak of the fMRI time course in each ROI acrossconditions and subjects (Fig. 3b). The Identical Condition evokedsignificantly lower fMRI responses than the other conditions acrossvisual areas (F28,280 ¼ 3.16, P o 0.05), providing evidence for fMRIadaptation9,10. Furthermore, we observed a significant interaction(F4,280 ¼ 2.29, P o 0.05) for condition and ROI.

Figure 1 Stimulus and design. (a) Stereogram

of a stimulus similar to those used. Cross-eyed

fusion of the images yields the impression of a

hinged plane receding in depth. The perspective

cue is provided by the trapezoidal projection of

the rectangular plane; the disparity cue by the

different positions of corresponding features in

the two eyes. Moving the figure closer and furtheraway manipulates the conflict between the cues.

When the figure is moved away, the perceived

angle between the planes gets smaller: the

disparity cue specifies a more acute angle as

distance increases, but the perspective-specified

angle remains constant. (b) Cartoon portraying

disparity-defined shape (Sd) as more acute than

perspective-defined shape (Sp). Aerial view at

right: the observer fixates the vertex, obtaining

perspective (Sp) and disparity (Sd) information.

Combining the cues results in perceived 3D

shape (Sc) between Sd and Sp. (c) Psychometric

functions (n ¼ 11) for consistent-cue (Sd ¼ Sp ¼1161) and inconsistent-cue (Sd ¼ 1161, Sp ¼1361) test stimuli. (d) Stimulus space. The

diagonal Sd ¼ Sp refers to consistent-cue stimuli.

The inconsistent-cue test stimulus (filled circle,

Sd ¼ 1161, Sp ¼ 1361) was paired with

consistent-cue reference stimuli (filled squares:Sd ¼ Sp ¼ 1361, 1261, 1161, and 1061). DC,

Disparity Change (test and reference differ in disparity: DSd ¼ 201); BC, Balanced Change (perspective and disparity cues differ in opposite directions:

DSd ¼ 101, DSp ¼ –101); PC, Perspective Change (test and reference differ in perspective: DSp ¼ –201); AC, All Change (both cues differ: DSd ¼ –101,

DSp ¼ –301). Open squares: consistent-cue reference stimuli used in psychophysical tests (but not in fMRI experiments).

860

0.2

0.4

0.6

0.8

1.0

96 106 116 126 136 146

86

86

96

96

106

106

116

116

126

126

136

136

146

146

Aerial viewFront view

z

x

y

x

Right eye's image Left eye's image

Sd

Sc

Sp α

Sd

Sp

Sc

Sd = Sp

S d = S p

Sd ≠ Sp

Reference stimulus (deg)

Pro

port

ion

'larg

er α

' res

pons

es

PSE

Sp

(deg

)

Sd (deg)

Teststimulus

AC

PC

BC

DC

a c

b d

Left hemisphereA P AP Right hemisphere

STS

STS

LO

pFsITS

OTS

Cos

ITS

OTSpFs

Cos

LOV3aV3a

V1V1

V1V1V2V2

VpVpV4V4

V2V2V3V3

V3aV3a

V4V4 V2V2 V1V1VpVp

V3 V2 V1V3 V2 V1

hMT+(V5)

hMT+(V5)

Figure 2 Regions of interest within the visual cortex. Functional activation

maps for one subject showing the early retinotopic regions, hMT+/V5 and the

LOC (comprising the regions labeled pFs and LO). The functional regions are

superimposed on flattened cortical surfaces of the right and left hemispheres.

Dark gray, sulci; light gray, gyri. A, anterior; P, posterior. STS: superior

temporal sulcus, ITS: inferior temporal sulcus, OTS: occipitotemporal

sulcus, CoS: collateral sulcus. The LOC was defined as the voxels in ventral

occipitotemporal cortex showing significantly stronger activation (P o 10�4,

corrected) to intact images than to scrambled images of objects. hMT+/V5

was defined as the contiguous voxels in the ascending limb of the inferior

temporal sulcus showing significantly stronger activation (P o 10�4,

corrected) to moving than to static low-contrast concentric rings. The borders

of early retinotopic regions (V1, V2, V3, V3a, Vp, V4) were localized using

rotating wedge stimuli, and eccentricity mapping was achieved usingconcentric rings.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

To determine whether cortical areas process cue-based information,we compared Table 1a fMRI responses in the Identical Condition withthose evoked when the presented stimuli differed in one cue (DisparityChange and Perspective Change conditions). The Disparity Changecondition evoked significantly higher fMRI responses than the IdenticalCondition in V1, V2, V3, V3a, Vp and V4. The Perspective Changecondition evoked significantly higher fMRI responses thanthe Identical Condition in V1, V2 and dorsal (V3, V3a) extrastriateareas; higher mean responses in ventral areas (Vp, V4) were notstatistically significant.

In higher areas (LOC and hMT+/V5), the Perspective Change, butnot the Disparity Change, condition evoked significantly higher activitythan the Identical Condition. The small numerical trend for higherresponses in the Disparity Change compared with the Identical con-dition (Fig. 3b) appears consistent with previous neurophysiological17

and fMRI adaptation18 studies showing sensitivity to disparity inhigher visual areas. An interesting possibility is that the lack of strongsensitivity to disparity changes in our study is consistent with thepsychophysical data, where perception was determined more byperspective than by disparity information.

To examine the relationship between perception and fMRIresponses, we made comparisons based on psychophysical perfor-mance (Fig. 3c, inset). First, we compared fMRI responses in‘high-discriminability’ (Perspective Change and All Change)and ‘low-discriminability’ (Balanced Change and Disparity Change)

conditions. We reasoned that areas whose activity reflects perceived 3Dshape based on cue combination should show stronger fMRI responsesin high-discriminability conditions than in low-discriminabilityconditions. This activation pattern was evident in areas LOC andhMT+/V5 but not in the retinotopic visual areas, with the exceptionof area V3a (Fig. 3c, Table 1a).

Second, we compared fMRI responses in conditions that had similarperceptual discriminability. We reasoned that when differences inperceived shape were similar, there should be no differences in fMRIresponse in areas that represent 3D shape based on the combination ofcues. Based on psychophysics, the Balanced Change and DisparityChange conditions were not differentially discriminable (repeatedmeasures ANOVA: F1,30 o 0.01, P¼ 0.99). Accordingly, fMRI responsesin the Balanced Change and Disparity Change conditions were notsignificantly different in the higher visual areas (LOC, hMT+/V5) and inarea V3a. In contrast, the Disparity Change condition evoked signifi-cantly higher fMRI responses than the Balanced Change condition inthe other retinotopic visual areas (V1, V2, V3, Vp, V4). Taken together,these dissociable results suggest sensitivity to cue changes in theretinotopic visual areas, excepting V3a, whereas responses in extrastriateventral (LOC) and dorsal (hMT+/V5) visual areas reflect differences inperceived 3D shape that are dependent on cue combination.

To quantify the relationship between psychophysical and fMRIresponses, we conducted regression analyses (Fig. 4), which providedevidence for a significant relationship between psychophysical and

Reb

ound

inde

x =

fMR

I n

fMR

I ID

Sig

nal c

hang

e fr

om b

asel

ine

(%)

hMT+/V5

IDBCDCPCAC

LOC

Time (s)

Sig

nal c

hang

e fr

om b

asel

ine

(%)

0

0.1

0.2

0.3

LOC hMT+/V5

0.8

0.9

1

1.1

1.2

1.3

1.4

1.5

1.6

V1 V2 V3 V3a Vp V4 LOC hMT+/V5

0

0.1

0.2

0.3

0.4

0.5

BC DC PC AC

Psychophysics

Res

pons

e in

dex

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

V1 V2 V3 V3a Vp V4

– 0.1

– 0.2

0

0.1

0.2

0.3

0.4

– 0.1

– 0.2

0

0.1

0.2

0.3

0.4

– 0.1

– 0.2

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0 1 2 3

ID BC DC PC AC

4 5 6 7 8 9 10

0 1 2 3 4 5 6 7 8 9 10

0 1 2 3 4 5 6 7 8 9 10

V1

a b

c

Figure 3 fMRI results from Experiment 1. (a) Representative average time course (n ¼ 11) of the fMRI response (areas V1, LOC, hMT+/V5). Error bars,

7 s.e.m. (b) Averaged fMRI response from the peak time points of the fMRI time course in LOC, hMT+/V5 and retinotopic areas (hemispheres combined).

Error bars, 7 s.e.m. No differences were found between sub-regions of the LOC12 or hMT+/V5. (c) fMRI responses expressed as a rebound index across

ROIs. Rebound index equals peak fMRI response in a condition (fMRIn) divided by the response to the Identical Condition (fMRIID); 1 indicates an adapted

response, and values 41 indicate sensitivity to changes in the stimulus. Error bars (normalized s.e.m.) incorporate the error estimates of both numerator and

denominator. Inset: between-subjects mean psychophysical data (n ¼ 11) for each experimental condition expressed as a psychophysical response index: 0

represents random behavior, and 0.5 represents perfect discrimination. Error bars, 7 s.e.m.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

fMRI responses in areas LOC and hMT+/V5, but not in the early visualareas (Table 1b). Furthermore, to ensure that this effect was not specificto the chosen inconsistent-cue test stimulus, we repeated the experi-ment with a different test stimulus in which the dihedral anglesspecified by the disparity and perspective cues were swapped(Sd ¼ 1361, Sp ¼ 1161). We observed the same relationships underthese conditions (Fig. 4), providing complementary evidence thatresponses in higher, rather than early, visual areas reflect perceivedshape based on cue combination.

Experiment 2: examining variations in cue weights

An interesting aspect of previous studies examining cue combination isthe inter-subject variability in cue weights. This suggests that observersdiffer in the reliability with which they encode cues, leading todifferences in perception11,19. Consistent with previous psychophysicalreports, we observed differences in fMRI response in areas LOC andhMT+/V5 that corresponded to between-subjects differences in per-ceived 3D shape (Supplementary Table 1, Supplementary Fig. 1).

To examine further the relationship in individual observers betweenfMRI response and the ‘combined-cue’ shape percept, we studied fMRIresponses to stimuli that were chosen based on an individual’s cueweights. Rather than ensuring that observers viewed identical stimuli(Experiment 1), we generated stimuli for individuals after psychophy-sical tests that measured disparity and perspective cue weights. Thisallowed us to test for fMRI adaptation to perceived shape using stimulithat differed in their component cues but were reported to have similar3D shape.

Specifically, before collecting fMRI data, we measured each obser-ver’s cue weights (wd, wp) inside the magnet. Observers (n ¼ 3) madejudgments about the dihedral angle of aninconsistent-cue test stimulus (Sd ¼ 1461;Sp ¼ 1261) by comparisons with consistent-cue reference stimuli. Fitting these psychophy-sical data provided a measure of perceivedangle for the test stimulus (Sc) that was usedto estimate the disparity and perspective cueweights. With this knowledge, we createda Balanced Change (BC) test stimulus inwhich the disparity (Sd ¼ d1) and perspective(Sp ¼ p1) cues changed in opposite directionsso that when combined, perceived 3D shape(Sc ¼ wdd + wpp) was very similar to that of aconsistent-cue reference stimulus (Sd ¼ Sp ¼1261). Two additional inconsistent-cue teststimuli were created (Fig. 5a) that corre-sponded to individual cue differences betweenthe Balanced Change test stimulus and theconsistent-cue reference: the Disparity Changestimulus (Sd ¼ d, Sp ¼ 1261) and PerspectiveChange stimulus (Sd ¼ 1261, Sp ¼ p). Themagnitude of angular slant change betweenthe Disparity and Perspective Change teststimuli and the consistent-cue reference wasdifferent (that is, Dd 4 Dp). However, psy-chophysical performance was predicted to besimilar in these conditions, as the magnitudeof cue change reflected each observer’s cueweights (wd o wp). Before collecting thefMRI data, we validated these predictionsabout the perceived shape of the three teststimuli (Balanced Changed, Disparity Change,

Perspective Change) through further psychophysical testing, by com-parison with consistent-cue stimuli, as above.

For the fMRI experiment, these tailored test stimuli were used in anevent-related adaptation procedure similar to Experiment 1. Fourexperimental conditions were tested. The Identical Condition, whichconsisted of a consistent-cue reference stimulus (Sp ¼ Sd ¼ 1261)shown twice, provided a baseline measure. The Balanced Change,Disparity Change and Perspective Change conditions consisted of thereference stimuli (Sd ¼ Sp ¼ 1261) paired with each of the tailoredinconsistent-cue test stimuli (Fig. 5a).

We reasoned that areas engaged in 3D shape perception on the basisof combined cues should show fMRI responses in the Balanced Changecondition similar to those in the Identical condition, owing to theperceptual similarity of the test and reference 3D shapes. However,these areas should show increased responses when the cue changesrequired for the Balanced Change stimulus are made in isolation(Disparity and Perspective Change conditions) because of differencesin perceived 3D shape. Furthermore, because the Disparity andPerspective Change conditions had similar perceptual discriminability,similar fMRI responses should be observed in these conditions. fMRIresponses in the higher, but not early, visual areas were consistent withthese predictions (Fig. 5b–d). Specifically, fMRI responses in LOC andhMT+/V5 for the Balanced Change condition were similar to those forthe Identical Condition (Table 2a). In contrast, isolated cue changes(Disparity and Perspective Change conditions) resulted in similar fMRIresponses that were significantly higher than responses for the BalancedChange condition, in which the same cue changes were madeconcurrently. However, fMRI responses in the early visual areas forthe Balanced Change and the individual cue change conditions were

Table 1 Statistical analyses for Experiment 1

(a) Statistical analyses (repeated measures ANOVA with Greenhouse-Geisser contrasts) for Experiment 1

IC: DC IC: PC BC+DC: AC+PC DC: BC

ROI F1,280 P F1,280 P F3,280 P F1,280 P

V1 9.39 0.014 4.90 0.030 0.40 0.388 5.53 0.027

V2 8.85 0.015 6.17 0.024 0.77 0.382 8.93 0.015

V3 13.78 0.007 4.42 0.033 1.15 0.160 14.86 0.008

V3a 4.96 0.030 7.21 0.020 4.91 0.049 0.93 0.176

Vp 4.29 0.034 2.31 0.110 3.71 0.078 2.97 0.047

V4 3.42 0.042 1.91 0.124 2.12 0.116 2.62 0.049

LOC 1.49 0.141 6.63 0.022 14.70 0.013 0.89 0.180

hMT+/V5 2.21 0.137 7.34 0.019 14.45 0.013 0.17 0.298

(b) Correlation coefficient and associated ANOVAs for regressions on behavioral and fMRI data for each ROI in

Experiment 1

ROI R F1,42 P

V1 0.01 0.02 0.90

V2 0.06 0.16 0.68

V3 0.01 0.01 0.98

V3a 0.12 1.04 0.31

Vp 0.14 0.89 0.35

V4 0.09 0.36 0.54

LOC 0.44 15.87 0.001

hMT+/V5 0.40 12.82 0.001

ROI, region of interest. IC, Identical Condition. DC, Disparity Change. PC, Perspective Change. BC, Balanced Change.AC, All Change.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

not significantly different (Fig. 5d, Table 2a). Consistent with Experi-ment 1, these data suggest that fMRI responses in the higher, but notearly, visual areas relate to the perception of 3D shape on the basis ofcue combination.

Controls for possible confounds

To ensure that our experimental results could not be confoundedby differences in eye movement or attentional demands across

conditions, we conducted control experiments and additional analyses.First, eye position measurements indicated that eye movements werevery small and were not systematically different across experi-mental conditions (Supplementary Fig. 2). Second, to help maintaincorrect eye vergence, we used nonius fixation targets for the first(Supplementary Fig. 2) and second experiments. Furthermore, inour setup, changes in horizontal disparity would be the main stimulusthat could drive vergence eye movements, as perspective cue signals arecounteracted by conflicting disparities20. We observed large changesin the fMRI response that were associated with no change in thehorizontal disparity signal (Perspective Change conditions), making itunlikely that vergence changes confounded our results. However, torule out a contribution of eye vergence, we investigated fMRI responseswhen observers were required to fixate targets across the range ofdisparities used in our experiments. Using the fMRI design employedin the main experiments, we did not observe significant sensitivity tothe small changes in vergence that would be evoked by these disparitychanges in any of the measured ROIs (Fig. 6a;Table 2b). Therefore, it isunlikely that our results were confounded by vergence differencesacross conditions.

It is also unlikely that differences in allocation of attention couldaccount for the observed pattern of fMRI responses. In contrast to ourfMRI results, an attentional load explanation would predict higher

Psychophysical response

fMR

I res

pons

e

0.25

0.30

0.35

0.40

0.45

0.50

0 0.1 0.2 0.3 0.4 0.50.25

0.30

0.35

0.40

0.45

0.50

0.55

0.60

0.65

0 0.1 0.2 0.3 0.4 0.5

LOC hMT+/V5

V3aV1

–0.1

0

0.1

0.2

0.3

–0.1

0

0.1

0.2

0.3

LOC hMT+/V5

IDBCDCPC

Sig

nal c

hang

e fr

omba

selin

e (%

)

Sig

nal c

hang

e fr

om b

asel

ine

(%)

0

0.1

0.2

LOC hMT+/V50

0.1

0.2

0.3

0.4

LOC hMT+/V50

0.1

0.2

0.3

0.4

LOC

ID BC DC PC

hMT+/V5

Observer 3Observer 2Observer 1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

V1 V2 V3 V3a Vp V40

0.1

0.2

0.3

LOC hMT+/V5

–0.10

0.10.20.30.40.50.6

V1

0 1 2 3 4 5 6 7 8 910 0 1 2 3 4 5 6 7 8 910 0 1 2 3 4 5 6 7 8 910

(126,126)

(126, π)

∆p

∆d(δ,126)

wd = 0.21wp = 0.79

wd = 0.42wp = 0.58

wd = 0.27wp = 0.73

PC BC

DC

Time (s)

Sp

(deg

)

Sd (deg)

S d = S p

(δ, π)

a

c

d

bFigure 5 Design and results from Experiment 2.

(a) Design of Experiment 2, in which inconsistent-

cue test stimuli (filled circles) reflected each

subject’s cue weights (wd, wp). The inconsistent-

cue stimulus for the Balanced Change (BC)

condition was constructed so that perceived shape

was very similar to the consistent-cue reference

(filled square, Sd ¼ Sp ¼ 1261). Values for the BCtest stimulus for each subject: S1, Sd ¼ 961,

Sp ¼ 1341, measured Sc ¼ 124.4 7 2.51; S2,

Sd ¼ 961, Sp ¼ 1481, measured Sc ¼ 127.3 72.21; S3, Sd ¼ 981, Sp ¼ 1361, measured

Sc ¼ 128.5 7 1.91. The inconsistent-cue

stimuli for the Disparity Change (DC) and

Perspective Change (PC) conditions consisted

of the individual cue changes required for the

Balanced Change inconsistent-cue stimulus.

(b) Representative average time course (n ¼ 3)

of the fMRI response (areas V1, LOC, hMT+/V5).

Error bars, 7 s.e.m. (c) Averaged peak fMRI

response for each of the three observers in LOC

and hMT+/V5. Error bars, 7 s.e.m. (d) Averaged

peak fMRI responses in LOC, hMT+/V5 and

retinotopic visual areas. Between-subjects data

presented for illustration (n ¼ 3). The same

pattern was observed in each observer’s data.Error bars, 7 s.e.m.

Figure 4 Regressions of fMRI responses on psychophysical responses: fMRI

data from Experiment 1 (asterisks, dashed lines) against corresponding

psychophysical response index. fMRI data were normalized across subjects

(n ¼ 11) and areas. Filled circles and solid lines indicate regressions on

data obtained from an additional experiment in which a different test

stimulus was used (six subjects). We did not observe a difference between

the b (slope) coefficient of the regressions for each data set based on the

95% confidence intervals.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

fMRI responses when the discrimination task was hardest, as difficultconditions require prolonged, focused attention, resulting in higherfMRI responses21. We observed the opposite effect: fMRI responseswere highest in the All Change condition, in which the discriminationwas easiest (Fig. 3c: highest mean response index) and subjectsresponded fastest (Fig. 6b: shortest response times). Further, it isunlikely that observers chose to attend selectively to particular condi-tions, as trials were presented in quick succession and were randomlyinterleaved. Finally, it could be argued that perceived change is moreinteresting than perceived lack of change and, as a result, responses inthe high-discriminability conditions were increased. However, suchincreases in general alertness or arousal would result in increases infMRI response across the visual areas21. This is not consistent with thedifferent patterns of fMRI responses that we observed in the early andhigher visual areas.

DISCUSSION

Our study provides neuroimaging evidence for depth cue combinationin the human visual cortex and demonstrates the following mainfindings. First, higher visual areas in both the ventral (LOC) and thedorsal (hMT+/V5) pathways appear to represent the perceived global3D shape based on a combination of cues to slant. Second, fMRIresponses in most of the early visual areas indicate sensitivity to changesin the depth cues that define 3D structure rather than to changes in the‘combined-cue’ shape percept. That is, we observed significant fMRIresponses to disparity changes across early visual areas and to perspec-tive changes in V1, V2 and dorsal (V3, V3a) extrastriate areas. Thisobserved sensitivity to changes in disparity in early retinotopic cortex isconsistent with previous physiological22–24 and imaging5,18 studies.Our experiments were not designed to delineate the precise nature ofthis sensitivity. However, it is likely that the responses we havemeasured reflect a functional hierarchy of disparity processing, rangingfrom the processing of local disparities at early stages25 to surface-basedslant processing at later stages. Previous work investigating the proces-sing of perspective information has focused on parietal areas7,26,27. Ourfindings raise the possibility that perspective cue processing involvesa dorsal network ascending into parietal cortex. Finally, previousstudies5,6,16,18,28 have implicated V3a in processing 3D information.Our findings suggest a possible role of area V3a in the combinationof cues for global 3D shape (Fig. 3). However, the activity in area V3adid not seem to be as tightly coupled to perception as in higher areas(Figs. 4 and 5).

Cue combination: the problem and the process

To date, almost all that we know about the mechanisms of visual cuecombination comes from theoretical considerations and psychophysi-cal data. Recent studies have suggested that under many circumstances,the brain combines cues using weighted linear combination1. Underthis scheme, individual cues are initially processed largely indepen-dently to yield an estimate of an environmental property. Estimatesfrom different depth cues are then combined by weighting each sourceof information in proportion to its statistical reliability19.

The neural mechanisms that mediate depth cue processing have beenstudied extensively. In particular, several studies provide evidence forretinotopic visual areas involved in processing disparity23, stereoscopicedges29, 3D orientation30 and surfaces5,6,28,31. Furthermore, monkey ITcortex17,32–34 and human LOC15,35 have been implicated in processingcues for 3D structure. Finally, MT has been shown to respond todisparity-defined surfaces36 even in the absence of motion15,37, to carryinformation about 3D surface orientation4 and to mediate perceptionof structure from motion36,38.

Previous work has often considered sensitivity to single depth cues.For instance, disparity information is often ‘isolated’ using random dotstereograms. However, such viewing conditions often create large cue

Sig

nal c

hang

e fr

om b

asel

ine

(%)

ID +10 –10 –20

Res

pons

e tim

e (s

)

Condition

0

0.1

0.2

0.3

0.4

V1 V2 V3 V3a Vp V4 LOC hMT+/V50

0.1

0.20.3

0.4

0.50.6

0.7

0.80.9

ID BC DC PC AC

a b

Table 2 Statistical analyses for Experiment 2 and vergence control

experiment

(a) Statistical analyses (repeated measures ANOVA with Greenhouse-Geisser

contrasts) for Experiment 2.

DC+PC: BC ID: BC

ROI F1,45 P F1,45 P

V1 1.09 0.302 3.37 0.039

V2 1.42 0.240 2.22 0.074

V3 1.51 0.226 1.91 0.085

V3a 1.69 0.201 4.95 0.017

Vp 0.72 0.359 2.10 0.079

V4 1.10 0.300 3.09 0.045

LOC 10.88 0.006 0.18 0.577

hMT+/V5 6.69 0.013 0.50 0.404

(b) Statistical analyses (repeated measures ANOVA) on data obtained from the

vergence control experiment.

All conditions

ROI F3,12 P

V1 0.035 0.991

V2 0.098 0.959

V3 0.138 0.935

V3a 0.071 0.974

Vp 0.185 0.904

V4 0.055 0.982

LOC 0.360 0.783

hMT+/V5 0.211 0.886

Figure 6 Vergence control experiment and

response time data. (a) Vergence control: fMRI

responses when observers (n ¼ 4) viewed two

sequentially presented fixation stimuli at different

depth planes representing the range of disparities

presented in Experiment 1 (5.5–9.6¢). Observers

were required to determine whether the first or

second viewed stimulus was closer. Error bars,7 s.e.m. (b) Between-subjects mean response

time in each condition (Experiment 1). Error bars,

7 s.e.m. The shortest mean response times were

observed in the All Change condition.


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

conflicts: random dot patterns specify a flat surface (that is, no texturegradient), in contrast to the disparity information. The perturbationapproach1 we have adopted is a useful technique for studying theneural basis of depth cue processing, as it reduces such conflicts.However, to ensure our results were not limited to experimentallyinduced cue-conflict stimuli (which might exceed natural bounds), weexamined the relationship between perceptual discriminability andfMRI responses using consistent-cue stimuli. Consistent with Experi-ments 1 and 2, we observed stronger fMRI responses for high-discriminability conditions than for low-discriminability conditionsin the higher visual areas (Supplementary Fig. 3).

Cue combination: neural selectivity and invariance

Our fMRI findings provide insights into the neural correlates of cuecombination at the scale of large neural populations and suggestcortical regions in which responses are consistent with the ‘com-bined-cue’ percept. Further physiological studies (J.D. Nguyenkim &G.C. DeAngelis, Soc. Neurosci. Abstr. 368.12, 2004) in these regions,with higher spatiotemporal resolution, are necessary to determinewhether the neural responses underlying weighted-cue combinationare carried by single neuron responses, population codes or both. Thesestudies could determine whether individual neurons within an arearespond on the basis of cue combination, or whether subpopulationswithin an area carry information about the two cues independently.Our findings provide evidence that even if disparity and perspectiveinformation is encoded by separate subpopulations, the responsemagnitude of these subpopulations in higher areas seems to vary inaccordance with ‘combined-cue’ 3D shape perception.

Complementary to feature selectivity, invariance to image changes isa neuronal property important for visual perception and recognition.Recent studies provide evidence for representations independent of thecues that define an object’s shape in monkey IT8,32,39 and humanLOC10,40. Similarly, neurons in MT have been suggested to compute 3Dsurfaces from multiple cues41,42. Finally, recent human27,43–45 andmonkey7,16,26,46 fMRI studies suggest a network of occipitotemporaland parietal regions involved in processing 3D objects that are definedby different cues. These studies propose that regions involved inprocessing different depth cues could also mediate cue combination.However, information about the behaviorally relevant cue weights isnecessary to infer cue combination. The perturbation technique weadopted breaks the correlation between cue and percept changes,allowing us to directly test areas involved in cue combination. Ourdata suggest that occipitotemporal areas previously discussed withreference to cue invariance are implicated in processing combinationsof cues. Future studies will address the role of parietal regions, asthe high spatiotemporal resolution required for the acquisition of thefMRI adaptation data has not allowed us to measure fMRI responses inthese regions.

In conclusion, an exciting challenge in neuroscience is to understandhow the same sensory cues can result in different percepts47 and howdifferent cues can result in the same percept48. Our study provides afirst step in understanding neural circuits in human visual cortex thatare fundamental for resolving the perceptual ambiguities inherent incombining multiple information sources. Further studies on the neuralprocessing of complex 3D objects49 defined by multiple cues within andacross modalities19 will provide important insights into the networksmediating our unified perception of the 3D world.

METHODSObservers. Observers gave informed written consent. All had normal or

corrected-to-normal visual acuity without stereoscopic or color deficits.

Stimuli and psychophysical methods. Stimuli were receding hinged planes

(5 � 6.61) defined by perspective and horizontal disparity (range: 4.3–13.3¢)cues to 3D structure. They had the same projection-screen width irrespective of

the slant (range: 17–471) and were surrounded by an irregular background

(peripheral extent: 9.81 horizontal, 7.71 vertical) of squares (0.51) in the plane

of the screen, which enhanced relative disparities and was designed to promote

accurate eye vergence50. Stimuli were rear projected and viewed through a

mirror 10 cm above the eyes (viewing distance, 78 cm). Red-green anaglyphs

were used (crosstalk was only 0.6%). Photometric measures of the red and

green signals from the NECGT950 video projector were used for gamma

correction and to equate image luminance for each eye. Stimuli for Experi-

ments 1 (replication; Supplementary Fig. 2) and 2 included a pair of nonius

lines on each side of the fixation circle to promote stable eye vergence.

To evaluate the perceived shape of the test stimuli observers made compar-

isons with a range (86–1461) of consistent-cue reference stimuli (method of

constant stimuli). Observers fixated a circle (diameter ¼ 0.41) at the stimulus’

center and judged which of two sequentially presented stimuli (test and

reference order randomized) had a larger dihedral angle. Responses were

collected using optical response buttons. The PSE was calculated by fitting

(Probit analysis) the proportion of ‘larger dihedral angle’ responses to the test

stimulus at each consistent-cue reference angle. Before collecting fMRI data we

obtained a full psychometric function inside the magnet (e.g. Fig. 1d). For

Experiment 1, psychophysical measurements made concurrently with fMRI

acquisition provided data for four points on the function (filled squares

Fig. 1d). PSEs obtained with and without the scanner running did not differ

(t10 ¼ 2.23, P ¼ 0.52). We calculated a psychophysical response index to

describe the data. This is the absolute difference between the ‘larger dihedral

angle’ response proportion in a condition and 0.5: 0 represents random

behavior and 0.5 perfect discrimination. A criterion of 0.25 was used to

distinguish high- and low-discriminability conditions.

Cue weights were calculated by assuming weighted linear combination

where weights sum to 1 (ref. 1): Sc ¼ wdSd + wpSp, wd + wp ¼ 1. When

tailoring stimuli to observers’ cue weights (Experiment 2), we calculated

stimulus values using these formulas so that perceived 3D shape in the Balanced

Change condition would be very similar to the consistent-cue reference (1261).

To ensure these calculations were correct, we evaluated each observer’s percep-

tion of the inconsistent-cue test stimuli (BC, DC, PC) prior to collecting fMRI

data (as above).

Imaging. Data were collected using a 3-T Siemens scanner with gradient echo

pulse sequence (TR ¼ 2 s, TE ¼ 40 ms, localizers; TR ¼ 1 s, TE ¼ 40 ms, event-

related scans) and a head coil. The high resolution of the event-related scans

limited us to 11 axial slices (5 mm thick; 3 � 3 mm in-plane resolution)

covering occipitotemporal regions. Subjects ran one session of nine scans: five

localizers (LOC, twice; hMT+/V5, once; retinotopic, twice) and four (Experi-

ment 1) or six (Experiment 2) event-related scans (order counterbalanced

across subjects). For event-related experiments, each scan consisted of one

experimental trial epoch and two 8-s fixation epochs (one at the start and one

at the end). For Experiment 1, each scan had 24 experimental trials per

condition (n ¼ 5) and 24 fixation trials (providing a measure of baseline

activity). For Experiment 2, each scan had 25 experimental trials per condition

(n ¼ 4) and 25 fixation trials. Presentation order was counterbalanced so that

trials from each condition (including fixation) were preceded equally often

by trials from other conditions. A new trial began every 3 s and consisted of

two stimuli, each presented for 400 ms (ISI ¼ 150 ms), followed by a blank

(2,050 ms). Subjects did not perceive apparent motion between the stimuli.

Note that changes in disparity and perspective cues result in changes in spatial

position. The magnitude of these low-level stimulus changes was deliberately

kept very small (largest difference in position of corners was 6¢).

Data analysis. fMRI data were processed using BrainVoyager (Brain Innova-

tion). For each individual subject, regions of interest (ROIs: V1, V2, V3, V3a,

VP, V4, hMT+/V5, LOC) were defined using standard techniques10,15 (Fig. 2

and Supplementary Fig. 4). We averaged the signal intensity across trials in

each condition at each time point and converted these to percentage signal

change relative to fixation for each ROI and subject10. Sample between-subject

mean fMRI time courses in areas V1, LOC and hMT+/V5 are provided


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

(Figs. 3a and 5b); time courses in other areas were similar. Fitting the time

courses with the hemodynamic response function indicated that peak fMRI

responses occurred between 4 and 5 s after trial onset (time-to-peak parameter

for every ROI shown in Supplementary Fig. 5). Statistical analyses were

performed on the average fMRI response at these time points (repeated-

measures ANOVAs, Greenhouse-Geisser contrasts).

Finally, we controlled for the possibility that differences between experi-

mental conditions were due to different fMRI responses evoked by the

consistent-cue reference stimuli themselves (Experiment 1). We tested fMRI

responses when each consistent-cue reference stimulus was presented alone in a

trial. There were no significant differences between fMRI responses for the

different stimuli in any ROI (Supplementary Fig. 6), ensuring that fMRI

responses in the adaptation experiments were due to differences between the

test and reference stimuli.


ACKNOWLEDGMENTSPreliminary reports of this work were presented at the VisionSciences Society’s2003 meeting and at the Society for Neuroscience’s 2003 meeting. Thanksto S. Maier and J. Lam for help with data collection, and N. Logothetis,N. Kanwisher, J. Harris, S. McDonald, M. Ernst and R. Fleming for helpfuldiscussions and comments. Thanks also to R. van Ee for advice on stimulusgeneration. Supported by an Alexander von Humboldt Fellowship to A.E.W.,the Max-Planck Society and DFG grant TH812/1-1.




1. Landy, M.S., Maloney, L.T., Johnston, E.B. & Young, M. Measurement and modeling ofdepth cue combination – in defense of weak fusion. Vision Res. 35, 389–412 (1995).

2. Clark, J.J. & Yuille, A.L. Data fusion for sensory information processing systems (KluwerAcademic, Boston, 1990).

3. Hillis, J.M., Ernst, M.O., Banks, M.S. & Landy, M.S. Combining sensory information:mandatory fusion within, but not between, senses. Science 298, 1627–1630 (2002).

4. Nguyenkim, J.D. & DeAngelis, G.C. Disparity-based coding of three-dimensional surfaceorientation by macaque middle temporal neurons. J. Neurosci. 23, 7117–7128 (2003).

5. Backus, B.T., Fleet, D.J., Parker, A.J. & Heeger, D.J. Human cortical activity correlateswith stereoscopic depth perception. J. Neurophysiol. 86, 2054–2068 (2001).

6. Tsao, D.Y. et al. Stereopsis activates V3A and caudal intraparietal areas in macaques andhumans. Neuron 39, 555–568 (2003).

7. Tsutsui, K., Sakata, H., Naganuma, T. & Taira, M. Neural correlates for perception of 3Dsurface orientation from texture gradient. Science 298, 409–412 (2002).

8. Liu, Y., Vogels, R. & Orban, G.A. Convergence of depth from texture and depth fromdisparity in macaque inferior temporal cortex. J. Neurosci. 24, 3795–3800 (2004).

9. Grill-Spector, K. & Malach, R. fMR-adaptation: a tool for studying the functionalproperties of human cortical neurons. Acta Psychol. (Amst.) 107, 293–321 (2001).

10. Kourtzi, Z. & Kanwisher, N. Representation of perceived object shape by the humanlateral occipital complex. Science 293, 1506–1509 (2001).

11. Knill, D.C. & Saunders, J.A. Do humans optimally integrate stereo and texture informa-tion for judgments of surface slant? Vision Res. 43, 2539–2558 (2003).

12. Grill-Spector, K., Kourtzi, Z. & Kanwisher, N. The lateral occipital complex and its role inobject recognition. Vision Res. 41, 1409–1422 (2001).

13. Zeki, S. et al. A direct demonstration of functional specialization in human visual cortex.J. Neurosci. 11, 641–649 (1991).

14. Tootell, R.B. et al. Functional analysis of human MT and related visual cortical areasusing magnetic resonance imaging. J. Neurosci. 15, 3215–3230 (1995).

15. Kourtzi, Z., Buelthoff, H.H., Erb, M. & Grodd, W. Object-selective responses in thehuman motion area MT/MST. Nat. Neurosci. 5, 17–18 (2002).

16. Orban, G.A. et al. Similarities and differences in motion processing between thehuman and macaque brain: evidence from fMRI. Neuropsychologia 41, 1757–1768(2003).

17. Janssen, P., Vogels, R. & Orban, G.A. Selectivity for 3D shape that reveals distinct areaswithin macaque inferior temporal cortex. Science 288, 2054–2056 (2000).

18. Neri, P., Bridge, H. & Heeger, D.J. Stereoscopic processing of absolute and relativedisparity in human visual cortex. J. Neurophysiol. 92, 1880–1891 (2004).

19. Ernst, M.O. & Banks, M.S. Humans integrate visual and haptic information in astatistically optimal fashion. Nature 415, 429–433 (2002).

20. Enright, J.T. Art and the oculomotor system: perspective illustrations evoke vergencechanges. Perception 16, 731–746 (1987).

21. Ress, D., Backus, B.T. & Heeger, D.J. Activity in primary visual cortex predictsperformance in a visual detection task. Nat. Neurosci. 3, 940–945 (2000).

22. Thomas, O.M., Cumming, B.G. & Parker, A.J. A specialization for relative disparity in V2.Nat. Neurosci. 5, 472–478 (2002).

23. Cumming, B.G. & Parker, A.J. Responses of primary visual cortical neurons to binoculardisparity without depth perception. Nature 389, 280–283 (1997).

24. Watanabe, M., Tanaka, H., Uka, T. & Fujita, I. Disparity-selective neurons in area V4 ofmacaque monkeys. J. Neurophysiol. 87, 1960–1973 (2002).

25. Nienborg, H., Bridge, H., Parker, A.J. & Cumming, B.G. Receptive field size in V1neurons limits acuity for perceiving disparity modulation. J. Neurosci. 24, 2065–2076(2004).

26. Taira, M., Tsutsui, K.I., Jiang, M., Yara, K. & Sakata, H. Parietal neurons representsurface orientation from the gradient of binocular disparity. J. Neurophysiol. 83, 3140–3146 (2000).

27. Shikata, E. et al. Surface orientation discrimination activates caudal and anteriorintraparietal sulcus in humans: an event-related fMRI study. J. Neurophysiol. 85,1309–1314 (2001).

28. Poggio, G.F., Gonzalez, F. & Krause, F. Stereoscopic mechanisms in monkey visualcortex: binocular correlation and disparity selectivity. J. Neurosci. 8, 4531–4550(1988).

29. von der Heydt, R., Zhou, H. & Friedman, H.S. Representation of stereoscopic edges inmonkey visual cortex. Vision Res. 40, 1955–1967 (2000).

30. Hinkle, D.A. & Connor, C.E. Three-dimensional orientation tuning in macaque area V4.Nat. Neurosci. 5, 665–670 (2002).

31. Bakin, J.S., Nakayama, K. & Gilbert, C.D. Visual responses in monkey areas V1 and V2 tothree-dimensional surface configurations. J. Neurosci. 20, 8188–8198 (2000).

32. Tanaka, H., Uka, T., Yoshiyama, K., Kato, M. & Fujita, I. Processing of shape defined bydisparity in monkey inferior temporal cortex. J. Neurophysiol. 85, 735–744 (2001).

33. Uka, T., Tanaka, H., Yoshiyama, K., Kato, M. & Fujita, I. Disparity selectivity of neuronsin monkey inferior temporal cortex. J. Neurophysiol. 84, 120–132 (2000).

34. Janssen, P., Vogels, R., Liu, Y. & Orban, G.A. Macaque inferior temporal neurons areselective for three-dimensional boundaries and surfaces. J. Neurosci. 21, 9419–9429(2001).

35. Gilaie-Dotan, S., Ullman, S., Kushnir, T. & Malach, R. Shape-selective stereo processingin human object-related visual areas. Hum. Brain Mapp. 15, 67–79 (2002).

36. Bradley, D.C., Chang, G.C. & Andersen, R.A. Encoding of three-dimensional structure-from-motion by primate area MT neurons. Nature 392, 714–717 (1998).

37. Palanca, B.J. & DeAngelis, G.C. Macaque middle temporal neurons signal depth in theabsence of motion. J. Neurosci. 23, 7647–7658 (2003).

38. Dodd, J.V., Krug, K., Cumming, B.G. & Parker, A.J. Perceptually bistable three-dimensional figures evoke high choice probabilities in cortical area MT. J. Neurosci.21, 4809–4821 (2001).

39. Sary, G., Vogels, R. & Orban, G.A. Cue-invariant shape selectivity of macaque inferiortemporal neurons. Science 260, 995–997 (1993).

40. Grill-Spector, K., Kushnir, T., Edelman, S., Itzchak, Y. & Malach, R. Cue-invariantactivation in object-related areas of the human occipital lobe. Neuron 21, 191–202(1998).

41. Albright, T.D. Cortical processing of visual motion. Rev. Oculomot. Res. 5, 177–201(1993).

42. Treue, S. & Andersen, R.A. Neural responses to velocity gradients in macaque corticalarea MT. Vis. Neurosci. 13, 797–804 (1996).

43. Murray, S.O., Olshausen, B.A. & Woods, D.L. Processing shape, motion and three-dimensional shape-from-motion in the human cortex. Cereb. Cortex 13, 508–516(2003).

44. Paradis, A.L. et al. Visual perception of motion and 3-D structure from motion: an fMRIstudy. Cereb. Cortex 10, 772–783 (2000).

45. Taira, M., Nose, I., Inoue, K. & Tsutsui, K. Cortical areas related to attention to 3Dsurface structures based on shading: an fMRI study. Neuroimage 14, 959–966 (2001).

46. Sereno, M.E., Trinath, T., Augath, M. & Logothetis, N.K. Three-dimensional shaperepresentation in monkey cortex. Neuron 33, 635–652 (2002).

47. van Ee, R., van Dam, L.C. & Erkelens, C.J. Bi-stability in perceived slant when binoculardisparity and monocular perspective specify different slants. J. Vis.2, 597–607 (2002).

48. Backus, B.T. inAdvances in Neural Information Processing Systems 14 (eds. Dietterich,T.G., Becker, S. & Ghahramani, Z.) (MIT Press, Cambridge, Massachusetts, 2002).

49. Todd, J.T. The visual perception of 3D shape. Trends Cogn. Sci. 8, 115–121 (2004).50. DeAngelis, G.C. & Newsome, W.T. Organization of disparity-selective neurons in

macaque area MT. J. Neurosci. 19, 1398–1415 (1999).


ART ICLES©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

5-HTTLPR polymorphism impacts human cingulate-amygdala interactions: a genetic susceptibilitymechanism for depression

Lukas Pezawas1,3, Andreas Meyer-Lindenberg1,3, Emily M Drabant1, Beth A Verchinski1, Karen E Munoz1,Bhaskar S Kolachana1, Michael F Egan1, Venkata S Mattay1, Ahmad R Hariri2 & Daniel R Weinberger1

Carriers of the short allele of a functional 5¢ promoter polymorphism of the serotonin transporter gene have increased anxiety-

related temperamental traits, increased amygdala reactivity and elevated risk of depression. Here, we used multimodal

neuroimaging in a large sample of healthy human subjects to elucidate neural mechanisms underlying this complex genetic

association. Morphometrical analyses showed reduced gray matter volume in short-allele carriers in limbic regions critical for

processing of negative emotion, particularly perigenual cingulate and amygdala. Functional analysis of those regions during

perceptual processing of fearful stimuli demonstrated tight coupling as a feedback circuit implicated in the extinction of negative

affect. Short-allele carriers showed relative uncoupling of this circuit. Furthermore, the magnitude of coupling inversely predicted

almost 30% of variation in temperamental anxiety. These genotype-related alterations in anatomy and function of an amygdala-

cingulate feedback circuit critical for emotion regulation implicate a developmental, systems-level mechanism underlying normal

emotional reactivity and genetic susceptibility for depression.

Depression is among the four leading causes of disability and diseaseburden throughout the world and is associated with serious medicalconditions and mortality across the lifespan1,2. The importance ofserotonergic neurotransmission for the pathogenesis of depression issuggested clinically by the efficacy of serotonin re-uptake inhibitors(SSRIs), the first-line treatment of depression and most related anxietydisorders1 and by induction of depression by tryptophan depletion insusceptible individuals2. Post-mortem and in vivo studies of theserotonin transporter (5-HTT) and receptors support a role for thisneurotransmitter system in depression and anxiety disorders1. Further-more, serotonin (5-HT) appears to be critical for the development ofemotional circuitry in the brain, and even transient alterations in 5-HThomeostasis during early development modify neural connectionsimplicated in mood disorders and cause permanent elevations inanxiety-related behaviors during adulthood3–5. Importantly, 5-HThas broad developmental effects, promoting differentiation not onlyof serotonergic but also of glutamatergic neurons, which transientlyexpress 5-HTT in limbic regions such as cingulate cortex3.

Since heritability of depression approaches 70% (ref. 2), and anxioustemperament is related to risk for depression2, there has been intenseinterest in candidate genes related to serotonergic function and thesephenotypes. A variable number of tandem repeats in the 5¢ promoterregion (5-HTTLPR) of the human serotonin transporter gene(SLC6A4) has been shown both in vitro and in vivo to influence

transcriptional activity and subsequent availability of the 5-HTT6,7.Specifically, the 5-HTTLPR short allele (s) has reduced transcriptionalefficiency compared with the long allele (l), and individuals carryingthe s allele tend to have increased anxiety related temperamentaltraits8,9, which are related to increased risk for depression10. It hasfurther been shown that s carrier status elevates the risk of depression inthe context of environmental adversity11 (a unique gene-environmentinteraction that has been independently replicated12,13), and s allelesadversely affect outcome of SSRI treatment14. Thus, there is convergingevidence that 5-HTTLPR genotype is related to the biology and risk fordepression and anxiety-related temperamental traits. Previously, usingfunctional magnetic resonance imaging (fMRI) we found that healthy,non-depressed s allele carriers show an exaggerated amygdala responseto threatening visual stimuli compared with individuals with the l/lgenotype, suggesting a possible link between variation in the gene and abasic brain mechanism involved in processing of negative emotion15,16.This finding has been independently replicated17. However, it remainedunclear how this neurobiological association relates to clinical end-points, as variation in amygdala response did not account for indivi-dual differences in behavioral measures of emotional reactivity15,16.

Here, we used a multimodal neuroimaging strategy to identifymechanisms on the level of neural systems contributing to behavioraland, potentially, clinical effects associated with 5-HTTLPR. Becauseeven simple emotionally-charged stimuli (for example, masked fearful


1Genes, Cognition and Psychosis Program, National Institute of Mental Health, National Institutes of Health, 10 Center Drive 4S235, Bethesda, Maryland 20892-1379,USA. 2Department of Psychiatry, University of Pittsburgh School of Medicine, Western Psychiatric Institute and Clinic, 3811 O’Hara Street, E-729, Pittsburgh, Pennsylvania15213, USA. 3These authors contributed equally to this work. Correspondence should be addressed to D.R.W. ([email protected]).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

eye-whites) are processed in the human brain by distributed interactivenetworks18, we hypothesized that the phenotypic expression of the5-HTTLPR genotype would involve the structure and function ofneural circuitries involved in emotion processing. We used voxel-based morphometry (VBM), a neuroanatomical MRI technique, totest for genetic association with the morphology of limbic circuitry,

consistent with neurodevelopmental studies of animals with altered5-HT function3–5. We then explored the functional relevance of theobserved structural manifestations with an fMRI strategy, focusing oninteractions within distributed mood circuitry important for affectgeneration and regulation, to test the hypothesis that 5-HTTLPRgenotype affects development and patterns of neural activation withinthis circuitry. Assuming that neural mechanisms of disease suscept-ibility exist in clinically healthy individuals inheriting risk alleles, werestricted our study to a large sample of healthy Caucasian subjectswithout any lifetime psychiatric diagnosis or treatment, allowing us toexclude disease-related heterogeneity and environmental confounders.

We studied the effect of a functional variation in the serotonintransporter gene on limbic circuitry implicated in mood disorders.We found that s carriers had reduced gray matter volume inperigenual cingulate and amygdala. During processing of fearfulstimuli, these same regions showed strong functional interactions.In s carriers this circuit was relatively uncoupled, and the magnitudeof cingulate-amygdala interaction was a strong predictor of variationin temperamental anxiety, indicating genotype-related alterationsin anatomy and function of a limbic feedback circuit critical fornegative emotion.

RESULTS

Morphometry

First, we performed a structural imaging study using ‘optimized’VBM19,20, a sophisticated and automated method designed to measuregray matter volume changes with sufficient sensitivity to detectgenotype effects of single nucleotide polymorphisms21. In comparisonto l/l genotype subjects, s allele carriers showed significantly reducedvolume of the perigenual anterior cingulate cortex (pACC) andamygdala (Fig. 1, Table 1). Structural volume changes were morepronounced in the pACC than in amygdala (Fig. 2a), and only pACCand right amygdala remained significant (P o 0.05) after correction formultiple testing (however, the appropriateness of correcting for multi-ple tests with respect to the amygdala is debatable, considering priordata showing 5-HTTLPR effects on amygdala function15,16). It isnoteworthy that the rostral subgenual portion of the anterior cingulate

cortex (rACC), a structure implicated indepression22, was the punctum maximum ofobserved gray matter volume reductionsin s allele carriers within the whole brain(Supplementary Figure 1 online).

Covariance of gray matter structures

Given the prior anatomical evidence of inter-connection between amygdala and pACC23,24,we determined the degree to which amygdalavolume was related to volume of the pACCacross all subjects by calculating measures of‘structural covariance’ based on VBM data25:using the general linear model, we estimatedacross the brain the degree to which regional(amygdala) volume covaried with that of atarget region (pACC), putatively reflecting anaspect of neuronal ‘wiring’ with that region25.We found significant positive covariation ofamygdala and pACC volume, again with amaximum in the rACC and another localmaximum in the more caudal supragenualportion of the anterior cingulate cortex(cACC; Fig. 2b, Table 1).

4.2

3.8

3.4

3.1

2.7

2.3

1.9

3.0

2.5

2.0

1.5

1.0

0.5

0.0

a b

c

Figure 1 Thresholded (P o 0.05) statistical maps restricted to the limbic

cortex and amygdala illustrating gray matter volume reductions of s allele

carriers in comparison to l/l genotype (n ¼ 114). (a) Surface projectionsdisplay significant volume reductions of bilateral perigenual anterior cingulate

cortex and medial amygdala. Note that peak differences within the whole

brain are found within the subgenual cingulate, a region implicated in the

biology of depression and anxiety. (b) A coronal section through the amygdala

displays significant bilateral volume reductions. (c) Color scales used for

surface (left) and coronal (right) views represent t-scores.

Table 1 Cluster maxima of morphometric analyses (n = 114)

Region Subregion t z Pa xb y z

Morphometry

l/l genotype 4 s carrier

Subgenual anterior cingulate cortexc BA 24 4.01 3.86 o0.001* �3 33 �2

Supragenual anterior cingulate cortex BA 24 2.87 2.81 0.002 0 30 4


Left amygdala — 2.39 2.35 0.009 �31 3 �15

Right amygdala — 2.35 2.31 0.010* 18 �2 �22

Structural covariance

Main effect

Subgenual anterior cingulate cortex BA 32 3.16 3.08 0.001* 3 36 �9

Supragenual anterior cingulate cortex BA 32 3.30 3.22 0.001* 3 36 16

l/l genotype 4 s carrier

Subgenual anterior cingulate cortex BA 25 3.11 3.04 0.001* �1 18 �10


aUncorrected P-values. bCoordinates have been transformed from MNI space to that of Talairach and Tournoux; * P o 0.05 aftercorrection for multiple comparisons based on a ROI of the limbic cortex or amygdala; cRegion with the maximal volumedifference within the entire brain (post-hoc analysis); statistics have been thresholded with a t-score value corresponding touncorrected P ¼ 0.05.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Relationship between regional volume and BOLD signal

Since our earlier reports of increased amygdala activation in responseto threatening stimuli included individuals in this VBM analysis15,16,we tested whether those functional results might be confounded bythese structural changes (Supplementary Table 1). We did not observea significant correlation between amygdala or pACC volume and fMRIblood oxygen level–dependent (BOLD) activation, suggesting thatgenotype differences in this functional parameter of amygdala responseare not driven by local structural alterations.

Functional connectivity

The structural evidence that amygdala and pACC volumes werecorrelated suggests that they represent a functional circuitry modulatedby genetic variation of the serotonergic system. We thus measuredfunctional connectivity between these regions using fMRI to acquireBOLD signal during the perceptual processing of fearful and threaten-ing facial expressions15. ‘Functional connectivity’ is a measure ofcorrelated activity, derived from BOLD fMRI data, between a reference(amygdala) and target region (pACC) used widely in the imagingcommunity as a simple and robust characterization of aspects offunctional integration25–27. Converging lines of evidence suggest thatthis measure reflects anatomically and functionally relevant couplingwithin neuronal circuitries25; however, a finding of ‘functional con-nection’ should not be interpreted as proving the presence of structuralor causal connections, as this measure is correlative in nature. We foundin the entire dataset that amygdala and pACC were significantly

‘functionally connected’ (Fig. 3a, Table 2). Notably, amygdala con-nectivity distinguished two functionally divergent regions withinpACC: rACC, which was positively correlated with the amygdala, andcACC, which was negatively correlated with it (Fig. 3a). These distinctzones of functional connectivity within the cingulate cortex (rACC,cACC) also showed strong positive connectivity with each other,suggesting that they might form a feedback loop with amygdala(Supplementary Table 2).

Functional connectivity and structural covariance

These various findings suggest that disruption of amygdala-pACCfeedback circuitry could underlie the earlier observation of increasedamygdala activity in s carriers during the processing of fearful stimuli(using the same procedure as here)15,16. Therefore, we analyzed theeffect of genotype on functional coupling between amygdala, rACC andcACC. Short allele carriers showed a highly significant reduction ofamygdala-pACC connectivity in comparison to l/l genotype indivi-duals (Fig. 3b, Table 2), particularly prominently in rACC (Fig. 4a).Within the cingulate, rACC-cACC functional connectivity did notdiffer by genotype (Supplementary Table 2). A similar finding waspresent in structural covariance, where s carriers showed significantlylower structural covariance between amygdala and rACC than did l/lindividuals (Fig. 2c, Table 1).

Temperament correlates

As an important external validation, we reasoned that if functionaluncoupling of this mood circuit underlies reported associations of5-HTTLPR with emotional phenotypes, functional connectivitybetween amygdala and rACC should predict normal variation intemperamental trait measures related to anxiety and depression thatalso have been related to 5-HTTLPR8,9. Therefore, we performed acorrelation analysis based on temperament ratings evaluated by the

110

100

90

Per

cent

age

volu

me

rela

tive

to I

/I m

ean

± 1

s.e.

m.

80

70

I/I g

enot

ype

s car

rier

5-HTTLP

R

I/I g

enot

ype

s car

rier

5-HTTLP

R

Subgenual cingulate

Amygdala

Right Left

Subgenual SupragenualCingulate

Right Left

120

110

100

Per

cent

age

stru

ctur

al c

ovar

ianc

ere

lativ

e to

I/I

mea

n ±

1 s.

e.m

.

90

80

70

3.3

3.1

2.8

2.5

2.2

1.9

1.7

a b

c

a b

5.1 3.3

2.8

2.2

1.7

1.1

0.6

0

3.1

1.1

–0.9

–2.8

–4.8

–6.8

Figure 2 Structural data illustrating peak volume changes and results of

structural covariance analyses (n ¼ 114). Plots represent extracted peak

results normalized to volume measures relative to the l/l genotype group

mean. (a) Subgenual anterior cingulate cortex volume is markedly reduced

(by 425%) in s allele carriers in comparison to l/l individuals. (b) Statistical

map of structural covariance analysis displays different degrees of positive

correlation between bilateral amygdala volume and perigenual cingulate

volumes, with two local peaks located supra- and subgenually. (c) Plotdisplays reduced structural covariance in s carriers in comparison to l/l

genotype individuals, particularly in the subgenual anterior cingulate cortex.

Figure 3 Statistical functional connectivity maps between bilateral amygdala

and perigenual anterior cingulate cortex representing degree of functional

coupling between these structures (n ¼ 94). (a) Subgenual cortical regions in

left hemisphere (top) and right hemisphere (bottom) correlate positively with

amygdala activity during the perception of threatening faces, whereas

supragenual regions correlate negatively (color bar represents t-scores).

(b) 5-HTTLPR s allele carriers show significantly less functional coupling

between amygdala and perigenual anterior cingulate cortex than l/l

individuals, particularly in the subgenual region (color bar represents

absolute t-scores).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Tridimensional Personality Questionnaire (TPQ)28, choosing the totalharm avoidance subscale, a well-validated heritable measure linkedto anxiety and risk for depression that has been weakly related to5-HTTLPR status29. Almost 30% of the variance in harm avoidancescores was predicted by our measure of amygdala-rACC functionalconnectivity. In contrast, local functional or structural measures ofsingle regions were of no predictive value (Fig. 4b, SupplementaryTable 3), consistent with earlier results15,16.

Analysis of amygdala–medial prefrontal cortex relationship

A recent fMRI study17 reported that 5-HTTLPR genotype affectedfunctional coupling of the amygdala with ventromedial prefrontalcortex (Brodmann area (BA) 10) in a sample of 29 normal malesubjects analyzed with the same statistical approach employed herein.In contrast to our data in pACC, this study17 showed increasedcoupling of both amygdala and a region of ventral medial prefrontalcortex (BA 10) in s allele carriers compared with l/l genotype. Corticalregions analyzed in this and our study were largely non-overlapping(Supplementary Figure 2); our analysis was based on a region ofinterest defined a priori by a genotype effect on brain structure andcontained most of the ACC. In contrast, these authors17 created aspherical search region in medial prefrontal cortex, which did not covermajor parts of the ACC and specifically not the region where we report

maximal effects for genotype (rACC) (Sup-plementary Figure 2). To compare theseresults further, we constructed in our datasetthe same search region as reported17 andperformed a new functional connectivity ana-lysis based on this region and the amygdala.We found a strong main effect on functionalconnectivity with amygdala and BA10 in thesame direction observed previously17 (Supple-mentary Table 4). With respect to genotypewe found a weak effect of increased functionalconnectivity in s carriers in comparison to l/lgenotype (Supplementary Figure 3) thatbecame more pronounced if only males were

studied (Supplementary Table 4). This suggests that our functionalconnectivity data are consistent with the main effect of functionalconnectivity as recently reported17 and are, with respect to genotype, atleast qualitatively consistent across these studies.

DISCUSSION

Our multimodal imaging approach has identified an effect of5-HTTLPR genotype on structure, function and interconnections ofa neural circuit encompassing both amygdala and regions of the rACCand cACC. These structures have been prominently implicated instudies of depression and negative emotion22,30,31. Moreover, themeasure of amygdala-rACC connectivity impacted by this polymorph-ism predicted a notable degree of variation in a measure of anxioustemperament that also has been linked to risk for depression12.Convergent evidence from neuroimaging and neuropathological stu-dies suggests a key regulatory role for these regions in negative emotionprocessing, particularly the rACC (BA 32/25/24) but also the cACC(BA 32; ref. 23). Reduced activity of rACC is found in depression22 andinduced sadness31 and can be reversed by various antidepressanttherapies, including SSRIs32,33, sleep deprivation34 and even cingulo-tomy35. Notably, the region showing the greatest effect of 5-HTTLPRgenotype (pACC) is within the phylogenetically older archicorticalportion of the cingulate cortex36, a region that displays the highestdensity of 5-HTT terminals within the human cortex37 and that is atarget zone of dense projections from the amygdala24.

The functional connectivity analysis demonstrates that amygdala andpACC are significantly ‘functionally connected.’ It delineates twodistinct subregions that agree well with cytoarchitectonic and func-tional subdivisions (rACC, pACC) of this region as discussed above.This pattern of functional connectivity derived from fMRI data ismarkedly analogous to anatomical studies in primate brain showingmassive amygdala projections to rACC and efferent projections fromcACC back to the amygdala24. Convergent evidence strongly suggeststhat these interactions represent a functional feedback circuitry thatregulates amygdala processing of environmental adversity: stimulationof perilimbic prefrontal cortex inhibits amygdala function38, medialprefrontal cortex neurons also exert an inhibitory influence on theamygdala39 and lesions of this region markedly impair fear extinction40.

Given the evidence that rACC modulates amygdala activity byinhibition39,41, our finding of reduced amygdala connectivity in scarriers to rACC provides a potential mechanistic account for theobserved increase in amygdala activity in s carriers because reducedcoupling would translate into altered feedback regulation of amygdalaactivity15,16 (Supplementary Figure 4). It would be of interest toextend our observations on functional connectivity using ‘effectiveconnectivity’ methods that allow inferences about directionality in theinteractions observed here.

Table 2 Cluster maxima of functional connectivity analyses between bilateral amygdala and

perigenual cingulate cortex (n = 94).

Region Subregion t z Pa xb y z

Main effect

Subgenual anterior cingulate cortex BA 32 5.57 5.16 o0.001* 0 37 �2

Supragenual anterior cingulate cortex BA 32 �10.27 Inf o0.001* 0 34 26

l/l genotype > s carrier

Subgenual anterior cingulate cortex BA 32 3.27 3.18 0.001* 4 40 �6

Supragenual anterior cingulate cortex BA 32 �2.54 2.49 0.006* �4 38 17

aUncorrected P-values. bCoordinates have been transformed from MNI space to that of Talairach and Tournoux; * P o 0.05 aftercorrection for multiple comparisons based on ROIs derived from our structural analysis (perigenual cingulate). Inf, infinite.Statistics have been thresholded with a t-score value corresponding to uncorrected P ¼ 0.05.

Tot

al e

xpla

ined

var

ianc

e (R

2 ) of

tota

l har

m a

void

ance

sco

re (

HA

-TP

Q)

Amygdala

Functionalconnectivity

Gray mattervolume

BOLDsignal

*Subgenual cingulate

Supragenual cingulate

0%

5%

10%

15%

20%

25%

30%

l/l ge

notyp

e

s car

rier

5-HTTLP

R

Per

cent

age

func

tiona

l con

nect

ivity

rela

tive

to l/

l mea

n ±

1 s

.e.m

.

120

110

100

90

80

70

60

50

a b

Figure 4 Functional connectivity between the subgenual cingulate and

amygdala is dissociated by genotype and significantly explains harm

avoidance scores, whereas other functional or structural measures do not.

(a) Short allele carriers show significantly reduced functional connectivity

compared to l/l genotype (n ¼ 94). Plot represents extracted peak results

normalized to the mean absolute functional connectivity, relative to the l/l

genotype group. (b) Only functional connectivity between subgenual cingulateand amygdala explains (Bonferroni corrected, P o 0.05) harm avoidance

scores, in contrast to other structural and functional measures (n ¼ 26

subjects with both functional and structural data). Amygdala was used as

reference for connectivity measures. A small vertical line indicates values

close to zero. Error bars, s.e.m.


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

Taken together, our data show that 5-HTTLPR genotype affects thestructure and putative wiring of a core region within the limbic systemthought to be crucial for depression and anxiety-related temperamentaltraits. Since our morphometric analysis was age- and gender-independent and based on data from healthy Caucasian individualswithout any history of psychiatric illness or treatment, we conclude that5-HTTLPR affects brain development of these core regions and thefunctionality of related brain circuitries. Our finding that 5-HTTLPRimpacts cortical structure is consistent with longstanding evidence that5-HT plays an important role in cortical development, shapingneuronal circuitries by regulating synaptic plasticity and neuronalactivity patterns of serotonergic and non-serotonergic neurons3. Ser-otonergic neurons are among the first neurons to be generated, andeven non-serotonergic neurons may transiently express 5-HTT withincingulate cortex in animals3. This pattern of expression in non-serotonergic neurons within a specific temporal window has beenhypothesized to underlie the formation and fine-tuning of specificconnectivity patterns3, possibly through regulation of synaptogenesisand growth cone motility42. Reduced relative connectivity in s allelecarriers may be further compounded by the ability of 5-HT to alterdifferentiation of glutamatergic neurons, the major projection neuronsfor cortico-cortical interactions3.

Our data further demonstrate that 5-HTTLPR specifically affects afunctional interaction between amygdala and rACC, an effect thatpredicted a large degree of variance in normal subjects in a tempera-mental trait related to neuroticism and associated with greater risk fordepression. These findings echo those from studies suggesting a keyrole for rACC in neural interactions corresponding to affective person-ality styles, which are risk factors for depression43. As noted, asubstantial body of preclinical and clinical work shows the functionalrelevance of this circuitry in inhibitory modulation of the amygdala bythe rACC, with a specific role in fear extinction and depression. Theamygdala is densely directly connected to the rACC as mentionedabove. In contrast, amygdala connections to ventromedial prefrontalcortex (BA10), which may also participate in regulating amygdalaactivity17, if they exist at all44,45, are sparse46. Thus, the previouslyreported finding17 of increased functional connectivity in s allelecarriers in this region (supported in our own data set) is likely basedon indirect anatomical interconnections. It is, therefore, tempting tospeculate that ventromedial BA10 may function as an indirect regula-tory area, and the observation of increased ‘functional connectivity’ ins allele carriers may then represent a compensatory mechanism foralterations in the primary regulatory loop involving the cingulatedelineated here.

Two differences may have contributed to the effect in BA10 beingstronger in the previous study17 than in our own data. First, differencesin task design, which there17 required explicit processing of emotion-laden complex visual scenes, resulting in a more pronounced top-downregulation of limbic structures39 expected to engage more upstreammodulatory prefrontal regions. Second, the previous study17 examinedonly males, and both our reanalysis and a recent report on emotionalprocessing of negatively perceived verbal attributes indicate a possiblegender effect in BA10 (ref. 47). We therefore propose that decreasedcoupling in amygdala-rACC feedback circuitry leads to a relativelydysregulated amygdala response, and that prior observations of amyg-dala hyperreactivity and increased anxiety-related traits in s allelecarriers are based on changes in this circuit and probably lead tocompensatory overactivity in the ventromedial prefrontal cortex.

Our findings suggest a causal mechanism linking developmentalalterations in 5-HT–dependent neuronal pathways to impaired inter-actions in a regulatory network mediating emotional reactivity. The

effects of 5-HTTLPR genotype converge on the rACC in our structuraland functional connectivity analyses, which, coupled with evidence ofuniquely dense serotonergic innervation of this region, argues thatgenetically driven variation in 5-HTsignaling shapes the connectivity ofamygdala with this region. These alterations are manifested in anxiety-related temperamental traits, possibly reflecting inadequate regulationand integration of amygdala-mediated arousal, leading to an increasedvulnerability for persistent negative affect and eventually depression inthe context of accumulating environmental adversity. While investiga-tions of localized structural and functional abnormalities have pro-vided insights about depression, our data underscore the importance ofstudying genetic mechanisms of complex brain disorders at the level ofdynamically interacting neural systems. We suggest that such relation-ships better capture the functional consequences of neurodevelop-mental processes altering circuitry function implicated in humantemperament and psychiatric disorders.

METHODSAssessment of subject data. Demographics. Subjects were culled from a larger

population after careful screening48 to ensure they were free of any lifetime

history of psychiatric or neurological illness, psychiatric treatment or drug or

alcohol abuse (Supplementary Table 5). Only Caucasians of European ancestry

were studied to avoid stratification artifacts. All available scans of subjects

meeting these criteria were used. Subjects gave written informed consent and

participated in the study according to the guidelines of the National Institute of

Mental Health Institutional Review Board. Structural MRIs from 114 subjects

were used and customized templates were created from a larger sample of 214

healthy volunteers21. Functional MRIs of 94 subjects were studied. Twenty-six

(28%) subjects from the functional analyses were also part of our morpho-

metric analyses.

Mood and personality assessment. The harm avoidance subset of the TPQ was

administered, as it has been noted as a putative index of heritable behavior

traits reportedly related to anxiety, amygdala function and 5-HTTLPR status.

Harm avoidance scores were available for 109 subjects included in the

morphometric study and 79 subjects included in the functional study.

Genotyping. Genotyping was performed as described previously15. In addi-

tion, our sample was genotyped with a panel of 100 unlinked SNP loci to

survey for occult genetic stratification between 5-HTTLPR genotype groups

and showed no significant differences in frequency at any of these SNPs,

including several that have been associated with variation in brain function (for

example, COMT, BDNF, GRM3, GAD1, APOe4; available upon request).

Imaging. Functional task. During fMRI scanning, subjects completed a simple

perceptual task previously described to robustly engage the amygdala15,16.

During two blocks of an emotion task, subjects viewed a trio of faces, selecting

one of the two faces (bottom) that was identical to the target face (top). Per

block, six images were presented sequentially for 5 s, three of each gender and

target affect (angry or afraid) derived from a standard set of pictures of facial

affect. Emotion tasks alternated with three blocks of a sensorimotor control

task where faces were replaced with simple geometric shapes.

Structural image processing. Three-dimensional structural MRI scans were

acquired on a 1.5-T GE scanner using a T1-weighted SPGR sequence (TR/TE/

NEX 24/5/1, flip angle 451, matrix size 256 � 256, FOV 24 � 24 cm) with 124

sagittal slices (0.94 � 0.94 � 1.5 mm resolution) and pre-processed as

previously described21 followed by an optimized VBM protocol using custo-

mized templates19,20. Resulting gray matter images were smoothed with a

12-mm Gaussian kernel prior to statistics. Analysis was performed on Linux

workstation (RedHat) using MATLAB 6.52SP2 (MathWorks) within the Gen-

eral Linear Model49 in SPM2 (http://www.fil.ion.ucl.ac.uk/spm). The specifica-

tion of a design matrix identical to the one used in this study was described in

detail elsewhere21. Briefly, effects of 5-HTTLPR on gray matter volume were

examined by using an analysis of covariance model including the following

covariates of no interest: total gray matter volume, orthogonalized first- and

second-order polynomials of age, and gender. A hypothesis-driven regions of

interest (ROI) approach was used to investigate structural alterations induced

by genotype within the limbic cortex, including bilateral amygdala, hippocam-


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

pal formation, and cingulate cortex. Gray-matter volume changes were assessed

statistically using one-tailed t-contrasts after small-volume correction for the

limbic cortex or amygdala. False discovery rate estimations were used to correct

for multiple comparisons, and a probability of 0.05 was considered to be

significant. In addition, the t-statistic map representing decreased gray matter

volumes of s carriers in comparison to l/l genotype was used to define the

search region (ROI) for subsequent functional connectivity analyses by creating

a binary mask including voxels showing a significant reduction at P o 0.001

(uncorrected). Anatomical amygdala ROIs were based on Talairach labels using

the WFU Pick atlas (http://www.fmri.wfubmc.edu) software.

Functional image processing. BOLD fMRI was performed on a GE Signa 3-T

scanner using gradient echo EPI (24 axial slices, 4 mm thickness, 1 mm gap,

TR/TE ¼ 2000/28 ms, FOV ¼ 24 cm, matrix ¼ 64 � 64). Images were

processed as described previously15,16 using SPM99 (http://www.fil.ion.ucl.

ac.uk/spm). Briefly, images were realigned to the middle image of the scan

run, spatially normalized into a standard stereotactic space (MNI template)

using an affine and nonlinear (4 � 5 � 4 basis-functions) transformation,

smoothed with a 8-mm FWHM Gaussian filter and ratio normalized to the

whole-brain global mean. A statistical image for the contrast of the emotion

task versus the sensorimotor control was obtained for each subject and analyzed

in a second-level random effects model (ANOVA and one-tailed t-test) to

identify significant (P o 0.05, corrected for multiple comparisons) activations

within and between genotype groups.

Functional connectivity analyses. Our methods to measure ‘functional con-

nectivity’ have been described previously26,50. This measure examines the

covariation across the brain with the activation in a region (volume) of interest.

We used anatomical masks derived from VBM analyses to define these regions.

After mean and drift correction of the time series, median activity within this

region of interest was calculated (we prefer median as a robust estimator that

coincides with the mean under the assumption of normality) for each scan.

These values were then correlated across the brain with all voxel time series,

resulting in a map which contained, in each voxel, the correlation coefficient of

the time series in this voxel with that of the reference regions. These maps, one

per subject, were then analyzed in a random-effects model in SPM. To identify

regions significantly positively or negatively connected with the target region, a

one-sample t-test was used. To compare connectivity with the target region

across genotypes, a two-sample t-test was used. In each case, statistical

parametric maps reflected one-tailed significance (as is standard in the

employed software package, SPM) and were produced using appropriate

t-contrasts. Using the Gaussian random fields approach implemented in

SPM, correction for multiple comparisons was applied with inference restricted

to appropriate regions of interest as mentioned above.

Structural covariance analysis. ‘Structural covariance’ uses a similar measure

of coupling as functional connectivity, only this time not between functional

data (fMRI time series), but using voxel-wise gray matter volume maps derived

from VBM (see above). Voxels identified as significant in this approach will

have regional volume that is significantly positively or negatively correlated

with the target area across subjects. Summed volume in an anatomically

defined region of interest in standardized space was computed by adding the

local volume in all voxels comprising the ROI and used as a covariate of interest

in a random-effects general linear model in SPM that also included nuisance

covariates as described above for VBM. Statistical inference proceeded exactly

as described for VBM and functional connectivity by using appropriate

contrasts to derive SPMs for the significance of the volume covariate (to

identify regions positively or negatively covarying in volume with the target

region) or difference in correlation between the volume covariate split between

genotype groups (to identify regions differing in their dependence on target

area volume by genotype), again achieving multiple comparison correction

with the Gaussian random fields approach and restricting inference to appro-

priate regions of interest as detailed in the manuscript.

Creation of surface maps. SPM images files containing t-statistics have been

converted into AFNI format and warped from MNI into Talairach space using

AFNI software (http://afni.nimh.nih.gov/afni). C. Holmes’ pial brain surface in

Talairach space has been used as projection target, whereas its volume

representation in Talairach space served as grid parent (http://afni.nimh.nih.

gov/afni/suma). Only voxels with a maximal distance of 4 mm along the

segment of the cortical gray matter band were considered for mapping onto the

surface. A mapping function was chosen which maps only maximal statistical

values along the gray matter band segment in the brain volume onto the

surface. For surface maps displaying positive and negative statistics, maximal

absolute values have been mapped at the surface. Statistical maximal projection

maps were displayed within above-mentioned ROIs derived from VBM on

C. Holmes’ pial brain surface with SUMA, choosing Matlab’s standard color

bar where appropriate. A customized color bar has been generated for results of

the connectivity analysis irrespective of genotype in order to visualize positive

and negative ‘correlations’ in different subdivisions of the anterior cingulate.

Creation of plots. Plots represent extracted relative volumes, functional

connectivity or structural covariance b coefficients of SPMs at cluster peak

locations as reported in tables. Extracted values have been normalized to mean

values relative to the l/l genotype group using SPSS 11.0 (SPSS) for Mac OSX

(Apple Computers) resulting in percentage changes relative to l/l genotype.

Functional connectivity or structural covariance data have been transformed to

absolute measures before normalization in order to display connectivity/

covariance data on a positive scale.


ACKNOWLEDGMENTSWe thank A. Goldman and P. Fisher for technical assistance and H. S. Maybergfor comments on the manuscript. Furthermore, we thank T.E. Goldberg forproviding psychological data and J.H. Callicott for support of data acquisition.This work was supported by the US National Institute of Mental HealthIntramural Research Program.


Received 16 December 2004; accepted 14 April 2005


1. Nemeroff, C.B. & Owens, M.J. Treatment of mood disorders. Nat. Neurosci. 5 (suppl.),1068–1070 (2002).

2. Wong, M.L. & Licinio, J. Research and treatment approaches to depression. Nat. Rev.Neurosci. 2, 343–351 (2001).

3. Gaspar, P., Cases, O. & Maroteaux, L. The developmental role of serotonin: news frommouse molecular genetics. Nat. Rev. Neurosci. 4, 1002–1012 (2003).

4. Gross, C. & Hen, R. The developmental origins of anxiety. Nat. Rev. Neurosci. 5,545–552 (2004).

5. Ansorge, M.S., Zhou, M., Lira, A., Hen, R. & Gingrich, J.A. Early-life blockade of the5-HT transporter alters emotional behavior in adult mice. Science 306, 879–881(2004).

6. Lesch, K.P. et al. Association of anxiety-related traits with a polymorphism in theserotonin transporter gene regulatory region. Science 274, 1527–1531 (1996).

7. Murphy, D.L., Lerner, A., Rudnick, G. & Lesch, K.P. Serotonin transporter: gene, geneticdisorders, and pharmacogenetics. Mol. Interv. 4, 109–123 (2004).

8. Schinka, J.A., Busch, R.M. & Robichaux-Keene, N. A meta-analysis of the associationbetween the serotonin transporter gene polymorphism (5-HTTLPR) and trait anxiety.Mol. Psychiatry 9, 197–202 (2004).

9. Sen, S., Burmeister, M. & Ghosh, D. Meta-analysis of the association between aserotonin transporter promoter polymorphism (5-HTTLPR) and anxiety-related person-ality traits. Am. J. Med. Genet. B Neuropsychiatr. Genet. 127, 85–89 (2004).

10. Lotrich, F.E. & Pollock, B.G. Meta-analysis of serotonin transporter polymorphisms andaffective disorders. Psychiatr. Genet. 14, 121–129 (2004).

11. Caspi, A. et al. Influence of life stress on depression: moderation by a polymorphism inthe 5-HTT gene. Science 301, 386–389 (2003).

12. Kendler, K.S., Kuhn, J. & Prescott, C.A. The interrelationship of neuroticism, sex, andstressful life events in the prediction of episodes of major depression. Am. J. Psychiatry161, 631–636 (2004).

13. Eley, T.C. et al. Gene-environment interaction analysis of serotonin system markers withadolescent depression. Mol. Psychiatry 9, 908–915 (2004).

14. Murphy, G.M. Jr., Hollander, S.B., Rodrigues, H.E., Kremer, C. & Schatzberg, A.F.Effects of the serotonin transporter gene promoter polymorphism on mirtazapine andparoxetine efficacy and adverse events in geriatric major depression. Arch. Gen.Psychiatry 61, 1163–1169 (2004).

15. Hariri, A.R. et al. Serotonin transporter genetic variation and the response of the humanamygdala. Science 297, 400–403 (2002).

16. Hariri, A.R. et al. A susceptibility gene for affective disorders and the response of thehuman amygdala. Arch. Gen. Psychiatry 62, 146–152 (2005).

17. Heinz, A. et al. Amygdala-prefrontal coupling depends on a genetic variation of theserotonin transporter. Nat. Neurosci. 8, 20–21 (2005).

18. Whalen, P.J. et al. Human amygdala responsivity to masked fearful eye whites. Science306, 2061 (2004).

19. Good, C.D. et al. A voxel-based morphometric study of ageing in 465 normal adulthuman brains. Neuroimage 14, 21–36 (2001).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e

20. Ashburner, J. & Friston, K.J. Voxel-based morphometry–the methods. Neuroimage 11,805–821 (2000).

21. Pezawas, L. et al. The brain-derived neurotrophic factor val66met polymorphism andvariation in human cortical morphology. J. Neurosci. 24, 10099–10102 (2004).

22. Drevets, W.C. et al. Subgenual prefrontal cortex abnormalities in mood disorders. Nature386, 824–827 (1997).

23. Phillips, M.L., Drevets, W.C., Rauch, S.L. & Lane, R. Neurobiology of emotion percep-tion II: Implications for major psychiatric disorders. Biol. Psychiatry 54, 515–528(2003).

24. Paus, T. Primate anterior cingulate cortex: where motor control, drive and cognitioninterface. Nat. Rev. Neurosci. 2, 417–424 (2001).

25. Ramnani, N., Behrens, T.E., Penny, W. & Matthews, P.M. New approaches for exploringanatomical and functional connectivity in the human brain. Biol. Psychiatry 56,613–619 (2004).

26. Meyer-Lindenberg, A. et al. Evidence for abnormal cortical functional connec-tivity during working memory in schizophrenia. Am. J. Psychiatry 158, 1809–1817(2001).

27. Friston, K.J., Frith, C.D., Liddle, P.F. & Frackowiak, R.S. Functional connectivity: theprincipal-component analysis of large (PET) data sets. J. Cereb. Blood Flow Metab. 13,5–14 (1993).

28. Cloninger, C.R., Svrakic, D.M. & Przybeck, T.R. A psychobiological model of tempera-ment and character. Arch. Gen. Psychiatry 50, 975–990 (1993).

29. Reif, A. & Lesch, K.P. Toward a molecular architecture of personality. Behav. Brain Res.139, 1–20 (2003).

30. Drevets, W.C. et al. Glucose metabolism in the amygdala in depression: relationship todiagnostic subtype and plasma cortisol levels. Pharmacol. Biochem. Behav. 71, 431–447 (2002).

31. Mayberg, H.S. et al. Reciprocal limbic-cortical function and negative mood: convergingPET findings in depression and normal sadness. Am. J. Psychiatry 156, 675–682(1999).

32. Drevets, W.C., Bogers, W. & Raichle, M.E. Functional anatomical correlates of anti-depressant drug treatment assessed using PET measures of regional glucose metabo-lism. Eur. Neuropsychopharmacol. 12, 527–544 (2002).

33. Mayberg, H.S. et al. Regional metabolic effects of fluoxetine in major depression:serial changes and relationship to clinical response. Biol. Psychiatry 48, 830–843(2000).

34. Wu, J.C., Buchsbaum, M. & Bunney, W.E., Jr. Clinical neurochemical implications ofsleep deprivation’s effects on the anterior cingulate of depressed responders. Neuro-psychopharmacology 25, S74–S78 (2001).

35. Dougherty, D.D. et al. Cerebral metabolic correlates as potential predictors of response toanterior cingulotomy for treatment of major depression. J. Neurosurg. 99, 1010–1017(2003).

36. Mega, M.S., Cummings, J.L., Salloway, S. & Malloy, P. The limbic system: an anatomic,phylogenetic, and clinical perspective. J. Neuropsychiatry Clin. Neurosci. 9, 315–330(1997).

37. Varnas, K., Halldin, C. & Hall, H. Autoradiographic distribution of serotonin transportersand receptor subtypes in human brain. Hum. Brain Mapp. 22, 246–260 (2004).

38. Stefanacci, L. & Amaral, D.G. Some observations on cortical inputs to the macaquemonkey amygdala: an anterograde tracing study. J. Comp. Neurol. 451, 301–323(2002).

39. Maren, S. & Quirk, G.J. Neuronal signalling of fear memory. Nat. Rev. Neurosci. 5,844–852 (2004).

40. Sotres-Bayon, F., Bush, D.E. & LeDoux, J.E. Emotional perseveration: an update onprefrontal-amygdala interactions in fear extinction. Learn. Mem. 11, 525–535 (2004).

41. Rosenkranz, J.A., Moore, H. & Grace, A.A. The prefrontal cortex regulates lateralamygdala neuronal plasticity and responses to previously conditioned stimuli. J.Neurosci. 23, 11054–11064 (2003).

42. Haydon, P.G., McCobb, D.P. & Kater, S.B. Serotonin selectively inhibits growth conemotility and synaptogenesis of specific identified neurons. Science 226, 561–564(1984).

43. Keightley, M.L. et al. Personality influences limbic-cortical interactions during sad moodinduction. Neuroimage 20, 2031–2039 (2003).

44. Carmichael, S.T. & Price, J.L. Limbic connections of the orbital and medial prefrontalcortex in macaque monkeys. J. Comp. Neurol. 363, 615–641 (1995).

45. Ghashghaei, H.T. & Barbas, H. Pathways for emotion: interactions of prefrontal andanterior temporal pathways in the amygdala of the rhesus monkey. Neuroscience 115,1261–1279 (2002).

46. Amaral, D.G. & Price, J.L. Amygdalo-cortical projections in the monkey (Macacafascicularis). J. Comp. Neurol. 230, 465–496 (1984).

47. Shirao, N., Okamoto, Y., Mantani, T., Okamoto, Y. & Yamawaki, S. Gender differences inbrain activity generated by unpleasant word stimuli concerning body image: an fMRIstudy. Br. J. Psychiatry 186, 48–53 (2005).

48. Egan, M.F. et al. Effect of COMT Val108/158 Met genotype on frontal lobe function andrisk for schizophrenia. Proc. Natl. Acad. Sci. USA 98, 6917–6922 (2001).

49. Friston, K.J. et al. Statistical parametric maps in functional imaging: a general linearapproach. Hum. Brain Mapp. 2, 189–210 (1994).

50. Meyer-Lindenberg, A. et al. Neural basis of genetically determined visuospatial con-struction deficit in Williams syndrome. Neuron 43, 623–631 (2004).


A R T I C L E S©

2005

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

eneu

rosc

ienc

e


E R R ATA

Erratum: Cytoplasmic domain structures of Kir2.1 and Kir3.1 show sites for modulating gating and rectificationScott Pegan, Christine Arrabit, Wei Zhou, Witek Kwiatkowski, Anthony Collins, Paul A Slesinger & Senyon ChoeNat. Neurosci. 8, 279–287 (2005)

Several sentences in this paper contained an error. On p. 283 in the left column, and on p. 284 in the left column, “Arg255” and “Arg259” should have read “Asp255” and “Asp259” respectively, in all cases.

Erratum: The MAPK pathway and Egr-1 mediate stress-related behavioral effects of glucocorticoidsJean-Michel Revest, Francesco Di Blasi, Pierre Kitchener, Françoise Rougé-Pont, Aline Desmedt, Marc Turiault, François Tronche & Pier Vincenzo PiazzaNat. Neurosci. 8, 664–672 (2005)

A sentence in this paper contained an error. The fourth sentence of the abstract should have read as follows: “In the hippocampus of the wild-type mice after stress, as well as in the cell lines, activation of glucocorticoid receptors greatly increased the expression and enzymatic activity of proteins in the MAPK signaling pathway and led to an increase in the levels of both Egr-1 mRNA and protein.”

Erratum: Craving cocaine pERKs up the amygdalaYarimar Carrasquillo & J David SweattNat. Neurosci. 8, 129–130 (2005)

Reference 1 contained incorrect page numbers. The corrected version should read as follows:1. Lu, L. et al. Nat Neurosci. 8, 212–219 (2005).

©20

05 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

uren

euro

scie

nce

nature neuroscience june 2005

Documents