Nature Neuroscience February 2005

www.nature.com/natureneuroscience EDITORIAL OFFICE 345 Park Avenue South, New York, NY 10010-1707 Tel: (212) 726 9319, Fax: (212) 696 0978 Editor: Sandra Aamodt Associate Editors: I-han Chou, Annette Markus, Kalyani Narasimhan Assistant Editor: Cara Allen Copy Editor: Dorothy Moore Production Editor: Ivelisse Robles Assistant Production Editor: Elizabeth A. Melchor Cover Design: Erin Boyle Editorial Assistant: Jessica Chen MANAGEMENT OFFICES NPG New York 345 Park Avenue South, New York, NY 10010-1707 Tel: (212) 726 9200, Fax: (212) 696 9006 Publisher: Beatrice Renault Executive Editor: Linda Miller Chief Technology Officer: Howard Ratner Global Head of Advertising: Fabien Savenay Head of Nature Research & Reviews Marketing: Sara Girard Marketing Manager: Naomi Mulgrave Production Coordinator: Allyson Skinner Associate Director, New Technology: Timo Hannay Associate Director, Content Systems: Joe Landolfi Electronic Production Coordinator: Sarada Callison NPG London The Macmillan Building, 4 Crinan Street, London N1 9XW Tel: 44 207 833 4000, Fax: 44 207 843 4996 Managing Director: Annette Thomas Publishing Director: Peter Collins Editor-in-Chief, Nature Publications: Philip Campbell Marketing Director: Della Sar NPG Tokyo MG Ichigaya Bldg. 5F, 19-1 Haraikatamachi, Shinjuku-ku, Tokyo 162-0841 Tel: 81 3 3267 8751, Fax: 81 3 3267 8746 Asia-Pacific Publisher: Antoine E. Bocquet Manager: Koichi Nakamura Senior Marketing Manager: Peter Yoshihara Asia-Pacific Sales Director: Kate Yoneyama Asia-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 9481 Global Head of Display Advertising Sales: John Michael, Tel: 44 207 843 4960, Fax: 44 207 843 4996 Head of Display Advertising—Europe: Gerard Preston, Tel: 44 207 843 4960, Fax: 44 207 843 4996 Business Development Manager: Claire Hines, Tel: 44 207 843 4960, Fax: 44 207 843 4996 Asia-Pacific Sales Manager: Rinoko Asami, Tel: 81 3 3267 8751, Fax: 81 3 3267 8746 Western Region Sales Manager: George Lui, Tel: (415) 781 3804, Fax: (415) 781 3805 Sales Executives: New England: Sheila Reardon, Tel: (617) 399 4098, Fax: (617) 426 3717 New York, Mid-Atlantic, Southeast: Jim Breault, Tel: (212) 726 9334, Fax: (212) 696 9481 Midwest: Mike Rossi, Tel: (212) 726 9255, Fax: (212) 696 9481 Northwest: Mathieu DesJardins, Tel: (415) 781 6422, Fax: (415) 781 3805 Eastern England/Scotland/Italy/Spain/Israel: Matthew Wilkinson, Tel: 44 207 843 4960, Fax: 44 207 843 4749 South/West United Kingdom/Scandinavia/Holland: Marianne Boulakas, Tel: 44 207 843 4969, Fax: 44 207 843 4749 Northern Germany: Gerard Preston, Tel: 44 207 843 4960, Fax: 44 207 843 4749 Southern Germany/Austria/Switzerland/France/Belgium: Sabine Hugi-Fürst, Tel: 41 52761 3386, Fax: 41 52761 3419 [email protected] (US/Canada) [email protected] (Europe) [email protected] (Japan) Publisher: Ben Crowe, Tel: (212) 726 9245, Fax: (212) 696 9482 US Sales Manager: Peter Bless, Tel: (212) 726 9248, Fax: (212) 696 9482 Japan Sales Manager: Rinoko Asami, Tel: 81 3 3267 8751, Fax: 81 3 3267 8746 naturejobs Sales Director: Nevin Bayoumi, Tel: 44 207 843 4961, Fax: 44 207 843 4996 CUSTOMER SERVICE www.nature.com/help Senior Global Customer Service Manager: Gerald Coppin For all print and online assistance, please visit www.nature.com/help Purchase subscriptions: Americas: Nature Neuroscience, Subscription Dept., 303 Park Avenue South #1280, New York, NY 10010-3601. Tel: (866) 363 7860, Fax: (212) 689 9108 Europe/ROW: Nature Neuroscience, Subscription Dept., Macmillan Magazines Ltd., Brunel Road, Houndmills, Basingstoke RG21 6XS, United Kingdom. Tel: 44 1256 329 242, Fax: 44 1256 812 358 Japan: Nature Neuroscience, 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 8746 India: 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] Nature 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 4749 US Reprints Sales Executive: Sharda Tulsie, Tel: (212) 726 9631, Fax: (212) 679 0843 NATUREJOBS [email protected]

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

VOLUME 8 NUMBER 2 FEBRUARY 2005

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

E D I TO R I A L
123 Running the numbers

CO R R E S P ON D E N C E
Although the threshold for action potential initiation is lowest in the axon, the precise site of initiation is unknown. With multiple, simultaneous cell-attached recordings and immunofluorescent labeling, Häusser and colleagues find that action potentials initiate at the first node of Ranvier in cerebellar Purkinje neurons. This image shows the first node of Ranvier at the first axonal branchpoint of a Purkinje cell axon ensheathed in myelin. The axon is stained for calbindin (purple) and the myelin sheath for myelin basic protein (blue). (p 137)

125

Is the extrastriate body area involved in motor actions?

BOOK REVIEW
127 Smyelin for glia
Edited by Helmut Kettenmann and Bruce R Ransom

Reviewed by Ben A Barres

NEWS AND VIEWS
129 131 132 134 Craving cocaine pERKs up the amygdala Yarimar Carrasquillo & J David Sweatt ̈ see also p 212 Comm-ing across the midline Catherine E Krull ̈ see also p 156 The cochlear amplifier and Ca2+ current–driven active stereocilia motion Tianying Ren ̈ see also p 149 Channeling a ‘funny’ side of memory Daniel Johnston

B R I E F C O M M U N I C AT I O N S
137 140 The site of action potential initiation in cerebellar Purkinje neurons B A Clark, P Monsivais, T Branco, M London & M Häusser Large-scale changes in dendritic structure during reorganization of adult somatosensory cortex P W Hickmott & P A Steen Striate cortex (V1) activity gates awareness of motion J Silvanto, A Cowey, N Lavie & V Walsh The voices of wrath: brain responses to angry prosody in meaningless speech D Grandjean, D Sander, G Pourtois, S Schwartz, M L Seghier, K R Scherer & P Vuilleumier Pathological gambling is linked to reduced activation of the mesolimbic reward system J Reuter, T Raedler, M Rose, I Hand, J Gläscher & C Büchel

143 145
Detecting angry prosody in voices (p 145)

147

Nature Neuroscience (ISSN 1097-6256) is published monthly by Nature Publishing Group, a trading name of Nature America Inc. located at 345 Park Avenue South, New York, NY 10010-1707. Periodicals postage paid at New York, NY and additional mailing post offices. Editorial Office: 345 Park Avenue South, New York, NY 10010-1707. Tel: (212) 726 9321, Fax: (212) 696 0978. Annual subscription rates: USA/Canada: US$199 (personal), US$1,240 (institution). Canada add 7% GST #104911595RT001; Euro-zone: €289 (personal), €1,279 (institution); Rest of world (excluding China, Japan, Korea): £175 (personal), £775 (institution); Japan: Contact Nature Japan K.K., MG Ichigaya Building 5F, 19-1 Haraikatamachi, Shinjuku-ku, Tokyo 162-0841. Tel: 81 (03) 3267 8751, Fax: 81 (03) 3267 8746. POSTMASTER: Send address changes to Nature Neuroscience, Subscriptions Department, 303 Park Avenue South #1280, New York, NY 10010-3601. Authorization to photocopy material for internal or personal use, or internal or personal use of specific clients, is granted by Nature Publishing Group to libraries and others registered with the Copyright Clearance Center (CCC) Transactional Reporting Service, provided the relevant copyright fee is paid direct to CCC, 222 Rosewood Drive, Danvers, MA 01923, USA. Identification code for Nature Neuroscience: 1097-6256/04. Back issues: US$45, Canada add 7% for GST. CPC PUB AGREEMENT #40032744. Printed by Publishers Press, Inc., Lebanon Junction, KY, USA. Copyright © 2004 Nature Publishing Group. Printed in USA.

i

VOLUME 8 NUMBER 2 FEBRUARY 2005

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

ARTICLES
149 156 Ca2+ current–driven nonlinear amplification by the mammalian cochlea in vitro D K Chan & A J Hudspeth ̈ see also p 132 Comm function in commissural axon guidance: cell-autonomous sorting of Robo in vivo K Keleman, C Ribeiro & B J Dickson ̈ see also p 131 Cyclic AMP controls BDNF-induced TrkB phosphorylation and dendritic spine formation in mature hippocampal neurons Y Ji, P T Pang, L Feng & B Lu Activity-dependent liberation of synaptic neuropeptide vesicles D Shakiryanova, A Tully, R S Hewes, D L Deitcher & E S Levitan Endocytosis-dependent desensitization and protein synthesis–dependent resensitization in retinal growth cone adaptation M Piper, S Salih, C Weinl, C E Holt & W A Harris Coactivation and timing-dependent integration of synaptic potentiation and depression H-X Wang, R C Gerkin, D W Nauen & G-Q Bi Invariant computations in local cortical networks with balanced excitation and inhibition J Mariño, J Schummers, D C Lyon, L Schwabe, O Beck, P Wiesing, K Obermayer & M Sur Bistability of cerebellar Purkinje cells modulated by sensory stimulation Y Loewenstein, S Mahon, P Chadderton, K Kitamura, H Sompolinsky, Y Yarom & M Häusser Central amygdala ERK signaling pathway is critical to incubation of cocaine craving L Lu, B T Hope, J Dempsey, S Y Liu, J M Bossert & Y Shaham ̈ see also p 129 Dynamics of motion signaling by neurons in macaque area MT M A Smith, N J Majaj & J A Movshon Using visual direction in three-dimensional motion perception J M Harris & V F Drga A representation of the hazard rate of elapsed time in macaque area LIP P Janssen & M N Shadlen Prefrontal white matter volume is disproportionately larger in humans than in other primates P T Schoenemann, M J Sheehan & L D Glotzer

164
Comm regulates Robo transport in axon guidance (p 131 and 156)

173 179

187

194

202
Activity mobilizes neuropeptide vesicles (p 173)

212

220 229 234 242

Representation of probability over time in parietal cortex (p 234)

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.

NATURE NEUROSCIENCE

iii

E D I TO R I A L

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Running the numbers
rains are extremely sensitive pattern detectors—so sensitive, in fact, that they often detect patterns that do not actually exist1. Daniel Kahneman, professor of psychology at Princeton University, won the 2002 Nobel Prize in economics for his studies of the cognitive shortcuts that lead to such errors. Most scientists are aware that people do poorly at detecting biases in conclusions based on incomplete or unrepresentative data, and thus evaluate their hypotheses by formal statistical testing. Unfortunately, the same cognitive mechanisms that lead to pattern detection errors in everyday life, such as a tendency to interpret new information as supporting one’s current beliefs, can also lead to faulty intuitions about the correct application of statistical tests. To help assure readers that the conclusions of our papers do not reflect such biases, the Nature journals have developed a set of guidelines for the analysis and reporting of statistics in our pages, which can be found on our website at http://www.nature.com/neuro/authors/ index.html, along with a checklist for authors. Some of the guidelines simply aim to ensure that the statistical evidence for each finding is clearly described: what tests were used, how many samples were evaluated in each condition, which comparisons were done, and what significance level was found (reported as the actual P value, not merely “P < 0.05”). Graphs should include error bars, clearly labeled as standard error or standard deviation. The Methods section of all papers that include statistical testing should contain a subsection describing the analysis. We will make sure that all the required information is presented in the final version of the paper. We are also instituting a standard set of requirements for the statistical analysis itself that editors and referees will evaluate before a paper is accepted for publication. In particular, all data sets should be summarized with descriptive statistics, including a measure of center, such as the mean or median, and a measure of variability, before further analyses are done. Authors will be asked to justify their choice of analysis and the exclusion of any data points, and to confirm that their data conform to the assumptions underlying the tests that were used. We invite referees to point out any potential areas of statistical concern and let the editors know if they feel that a particular paper needs to be evaluated by a statistics expert. Following the new guidelines should help authors avoid several common statistical errors. One of the most widespread is the use of multiple comparisons, which increases the risk of false-positive results. For example, carrying out a series of pairwise comparisons by t-tests gives a higher chance of a falsely ‘significant’ result (because each test has a 1-in-20 risk of a false positive at P < 0.05, one would expect 1 false positive out of every 20 tests performed) than evaluating the same data with a single analysis of variance (ANOVA) at the same significance level. Along the same lines, analyses of functional imaging data should be corrected for multiple comparisons to account for

B

testing across multiple voxels. Another common error is the failure to recognize that most parametric tests require the data to be normally distributed. If this assumption is not valid for a particular data set, a nonparametric test should be used instead. Another point that is less widely recognized is that ANOVAs require approximately equal variance across the different groups or conditions examined. If this assumption is violated (as it frequently is, because variance often scales with the mean), a nonparametric test or appropriate transformation of the data is necessary. Finally, researchers should take care to choose the correct statistical tests for small data sets (roughly n < 10), for which ranges are a more appropriate measure of variability than standard deviations or standard errors. No set of general guidelines can protect against all possible sources of statistical error, of course. Needless to say, all scientists should understand the reasoning behind their analysis, rather than leaving the choice of test to the discretion of their favorite statistical software package. Segregating data into subgroups is another common source of error and bias. Researchers should select data to analyze as a subgroup with care, preferably based on an independent variable, rather than, for instance, sorting their data into ‘high-responding’ and ‘low-responding’ subject groups for further analysis. Negative findings should be stated with caution, and if critical to the conclusions, supported by a power analysis that indicates that the number of subjects would be adequate to detect an effect of the expected size. In many universities, statistical experts are available for consultation with researchers in other departments who need help in designing experiments and analyses, preferably before the data are collected. Even careful neuroscientists have a tough task, however, because they often work with data that are mathematically complex and thus difficult to analyze correctly. Spike rates tend not to be normally distributed and not to have equal variances across groups. They also violate an underlying assumption of cross-correlation analysis: that data are ‘stationary’—meaning that their stochastic properties do not change with time2. Another assumption often violated in neuroscience is that data points are independent of one another; because neurons are highly interconnected, physiological responses are often correlated. There is nothing shameful about cognitive shortcuts; during our evolution, it has often been more adaptive to be able to evaluate a situation and respond quickly than to get the precisely correct answer. Using rigorous analysis methodology allows scientists to bypass the potential bad consequences of this tendency and get the right answer most of the time. We hope that our new statistical guidelines will contribute to this effort, and welcome feedback on the new policy, which can be sent to the editors at [email protected]. í
1. Gilovich, T. How We Know What Isn’t So: The Fallibility of Human Reason in Everyday Life (Free Press, New York, 1993). 2. Brown, E.N., Kass, R.E. & Mitra, P.P. Nat. Neurosci. 7, 456–461 (2004).

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

123

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

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Is the extrastriate body area involved in motor actions?
To the Editor: Astafiev et al.1 report that unseen, visually guided motor acts activate the extrastriate body area (EBA)2. This finding has potential implications for understanding the interactions between motor and perceptual systems and suggests a mechanism by which the visual stimulation resulting from one’s own motor acts is distinguished from that produced by others3. We replicated the experiment of Astafiev et al. and found, in line with their findings, actionrelated modulation in the EBA. However, a closer look showed that the region involved in visually guided motor acts is distinct from the EBA and that action-related modulation and body selectivity are unrelated. We scanned 13 subjects with an fMRI localizer for the EBA (contrasting headless bodies with faces, scenes and tools). In the same session, we compared unseen visually guided finger movements with a perceptually matched control condition in an event-related design (see Supplementary Fig. 1 online). Replicating Astafiev et al., we found a significant effect of finger movements in left (t12 = 4.5, P < 0.001) and right (t12 = 4.0, P < 0.005) EBA. For each subject, a whole-brain contrast of finger movements versus control significantly (P < 0.00001, uncorrected) activated a bilateral temporal-occipital region (mean peak Talairach coordinates (x, y, z): left: –46, –65, –1; right: 53, –56, 0) that was close to the EBA (left: –45, –74, –3; right: 48, –68, 0). The peak of this actionrelated region (ARR), however, was significantly anterior to the EBA (left: t12 = 5.4, P < 0.001; right: t12 = 5.9, P < 0.001). Moreover, the spatial overlap4 of ARR with the EBA (at P < 0.0005, uncorrected) was only 14% (see Supplementary Fig. 2 online). Note that the partial overlap of ARR and the EBA does not necessarily mean that the same neurons are involved in both motor actions and body perception. If this were the case, we would expect a positive voxel-by-voxel correlation between selectivity for bodies and actionrelated modulation. To test this, we defined for each subject the intersection of ARR and the EBA and calculated the correlation between the strength (as expressed by T values) of action-related activity compared to control, and body selectivity. The average correlation between these two measures was not statistically different from zero (r = 0.00, P = 0.96). This suggests that the region shared by ARR and the EBA contains interleaved but functionally independent neural populations. To verify these findings, we scanned five subjects with an additional EBA localizer, using a contrast of body parts versus object parts. We replicated all of our key findings: a significant effect of pointing within the EBA (P < 0.05), a significantly anterior peak of ARR compared to EBA (P < 0.05), low spatial overlap of EBA and ARR (19%) and, if anything, a negative voxelby-voxel correlation between action-related activity and body selectivity (r = –0.14, P = 0.08). In contrast, the correlation between selectivity for whole bodies and for body parts was significantly positive (r = 0.42, P < 0.05), showing that the absence of a correlation between action-related activity and body selectivity was not due to insufficient statistical power. Thus, the temporal-occipital area that is involved in executing motor actions is distinct from the EBA. It may instead correspond to an area anterior to MT that is activated when subjects generate action-related words5. It also falls near the putative human homolog of MST, which represents visual motion in the periphery6,7. Further studies will be needed to determine the relationship between motor activity, action representation and visual motionselective regions in lateral temporal cortex.
Note: Supplementary information is available on the Nature Neuroscience website (http://www.nature.com/ natureneuroscience/).
Sprenger, M. & Scheltens, P. Magn. Reson. Imaging 16, 105–113 (1998). 5. Martin, A., Haxby, J.V., Lalonde, F.M., Wiggs, C.L. & Ungerleider, L.G. Science 270, 102–105 (1995). 6. Dukelow, S.P. et al. J. Neurophysiol. 86, 1991–2000 (2001). 7. Huk, A.C., Dougherty, R.F. & Heeger, D.J. J. Neurosci. 22, 7195–7205 (2002).

Marius V. Peelen and Paul E. Downing School of Psychology, University of Wales, Bangor, UK. e-mail: [email protected]
1. Astafiev, S.V., Stanley, C.M., Shulman, G.L. & Corbetta, M. Nat. Neurosci. 7, 542–548 (2004). 2. Downing, P.E., Jiang, Y., Shuman, M. & Kanwisher, N. Science 293, 2470–2473 (2001). 3. Jeannerod, M. Nat. Neurosci. 7, 422–423 (2004). 4. Rombouts, S.A., Barkhof, F., Hoogenraad, F.G.,

Astafiev et al. reply: Neural activity in visual cortex can be modulated by extraretinal signals involving attention, working memory, familiarity and multisensory integration. We have demonstrated that actions can also bias activity in visual cortex, including the early visual cortex (such as V1 and V2) and the extrastriate body area (EBA), which is specialized for the recognition of human bodies1,2. Downing and Peelen replicated our result for the EBA. At issue is whether action-related activity is centered within the EBA or in an adjacent cortical region. We reported that action-related activity was more common in the ventral-posterior part of the EBA but did not indicate whether this activity was coextensive with the EBA. Downing and Peelen report that the overlap between action-related activity (ARA) and the EBA is relatively small (14–19%). The overlap between two statistical maps defined by different task contrasts depends on the formula for computing overlap and on the control condition and number of observations for each contrast. The six subject maps in the authors’ Supplementary Figure 2 show that ARA was sometimes much smaller than the EBA (for example, in subjects 3 and 4). Even though ARA was largely contained within the EBA for these subjects, the overlap measure cited by the authors would necessarily yield a small number. The control condition must also be considered. Downing and Peelen’s paradigm involved a ‘preparatory’ period, in which subjects encoded a cue and either planned a hand movement to a target or just maintained fixation, and a ‘target/execution’ period, in which subjects either shifted attention and pointed to a target or just passively viewed it. Because fMRI signals in this study were apparently

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

125

CO R R E S P ON D E N C E
response, has a similar distribution within the EBA but a lower z-score. Figure 1d shows voxels in which activity is higher for action than for covert detection (ANOVA, interaction of condition (arm, foot, detection) × time; P < 0.005, uncorrected). The overlap with the EBA is about 14%. Therefore, although the bulk of action-related activity overlaps with the EBA if a low-level control is used, the overlap is considerably lower with a high-level control. This pattern would be consistent with Downing and Peelen’s results if their control was more similar to covert detection than to fixation, except that their ARA was anterior to the EBA. Downing and Peelen also claim that ARA and EBA reflect different and perhaps interleaved neuronal populations. Adaptation techniques3,4, rather than the authors’ correlational approach, are commonly used for this purpose. The authors do not explain why involvement of the same neurons in both tasks necessarily implies that variations in action-related activation and ‘body-related’ activation should be correlated. They appear to assume that the pattern of activity across different neurons is the same when a body part is viewed passively and when a limb is moved. Finally, we emphasize that even if the EBA and ARA are in adjacent regions, their topographical proximity suggests important functional relationships between visual body representations and observer actions1,5.
Sergei V Astafiev1, Christine M Stanley1. Gordon L Shulman2 & Maurizio Corbetta1–3 Departments of 1Radiology, 2Neurology and 3Anatomy and Neurobiology, Washington University School of Medicine, 4525 Scott Avenue, St. Louis, Missouri 63110, USA. e-mail: [email protected]
1. Astafiev, S.V., Stanley, C.M., Shulman, G.L. & Corbetta, M. Nat. Neurosci. 7, 542–548 (2004). 2. Downing, P.E., Jiang, Y., Shuman, M. & Kanwisher, N. Science 293, 2470–2473 (2001). 3. Grill-Spector, K. et al. Neuron 24, 187–203 (1999). 4. Buckner, R.L. et al. Neuron 20, 285–296 (1998). 5. Jeannerod, M. Nat. Neurosci. 7, 422–423 (2004).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 1 Overlap between group-averaged activity in ARA and EBA. Partial view of flattened representation of the left hemisphere. (a) EBA localization. (b) Action-related activation. (c) Activity during the covert detection control. (d) Voxels in which activity is higher for action than covert detection. Black outline in all panels shows the group-averaged EBA. STS=superior temporal sulcus. Color scale represents z–score.

averaged over the two periods, activity in ARA relative to the passive control task reflected motor planning, target detection and spatial orienting in addition to motor execution. In our study, action-related activity was limited to execution of a pointing response (preparatory activity was separately estimated) compared either to a fixation baseline or to a covert attention task that controlled sensory factors, target detection and spatial attention. Figure 1 shows the strong influence of the control condition on the overlap between ARA and the EBA. The black outline in all panels shows the group-averaged EBA, obtained from the contrast of body part versus object perception. Figure 1a shows the extent of EBA activa-

tion obtained by comparing passive viewing of body parts versus object parts in a group of subjects (n = 10). Figure 1b superimposes the group-averaged response for action-related activations (computed from the main effect of time during pointing with the right hand or foot in a random-effects ANOVA, with fixation as baseline). The overlap between ARA and the EBA is high ((number of ARA voxels within EBA)/(number of EBA voxels) = 92%), and activity is strongest in the posterior part of the EBA, just as for passive viewing of body parts (Fig. 1a). Figure 1c shows that activity during the covert detection control, in which subjects covertly shifted attention and detected the target without a motor

We welcome short letters on matters arising from previous papers in Nature Neuroscience or on other topics of widespread interest to the neuroscience community. Because space in this section of the journal is limited, priority is given to short (fewer than 500 words), well-written letters addressing the most topical issues. Typically, new data are not presented in this section, although they may occasionally be allowed at the discretion of the editors. Letters concerning material previously published in Nature Neuroscience are usually sent to the authors of the original piece for their comments and/or a formal reply. Letters may be edited for brevity and clarity.

126

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

NEWS AND VIEWS

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Craving cocaine pERKs up the amygdala
Yarimar Carrasquillo & J David Sweatt
In a rat model of cocaine craving, a new study shows that activation of the protein kinase ERK in the amygdala is necessary and sufficient for the growth of craving over time in withdrawal. The combination of experimental approaches used by these authors will serve as a model for future studies examining the molecular mechanisms underlying a cognitive or behavioral process. Most of us have experienced cravings at some point, whether for food, drugs, sex or something more idiosyncratic. What cellular and molecular mechanisms bring about this peculiar cognitive state, a preoccupation with the perceived need for a specific sensory stimulus or experience? In this issue, Lu et al. describe a series of experiments1 that begin to address these questions, using cocaineaddicted rats as a behavioral, cellular and molecular model of drug craving. The emotional distress of drug craving and its attendant vulnerability to relapse affect many recovering cocaine addicts. Cravingassociated relapse remains one of the main challenges for the effective treatment of cocaine addiction. Behavioral studies suggest that drug craving increases progressively over the withdrawal period and that it is strongly influenced by exposure to drug-associated environmental cues2–4. Thus, during withdrawal, ordinary events, places or situations that were previously associated with drug use promote a memory recall that triggers craving and provokes relapse. In addition, vulnerability to relapse among recovering cocaine addicts is higher after long periods of drug withdrawal than immediately after withdrawal begins. This behavioral phenomenon is referred to as ‘incubation’ of cocaine craving. Lu et al. have now systematically characterized its cellular and molecular underpinnings using their animal model of cocaine craving and relapse1. In the latest installment of their impressive drug addiction research program, the authors begin to unravel the molecular mechanisms of the incubation of cocaine craving by identifying for the first time a biochemical pathway in the amygdala that is crucial for this behavior. Using an exceptionally diverse experimental approach, this group demonstrated that activation of the extracellular signal–regulated kinase (ERK) in the central nucleus of the amygdala is both necessary and sufficient to induce cocaine craving during withdrawal. Their tripartite combination of experiments (what we will call ‘measure’, ‘block’ and ‘mimic’ types) makes this one of the few studies that convincingly link a specific molecular mechanism with a complex cognitive function. Here we focus on this

Yarimar Carrasquillo and J. David Sweatt are at the Department of Neuroscience, Baylor College of Medicine, One Baylor Plaza, Houston, Texas 77030, USA. e-mail: [email protected]

Figure 1 Model of hypothesis testing to study the molecular basis of complex behavior. (a) Animal model for the study of cocaine craving. See text for details. (b) The ‘measure’ experiment correlates a molecular change (ERK activation) with a specific behavior process (cocaine craving). (c) The ‘block’ experiment tests whether blocking ERK activation blocks cocaine craving. (d) The ‘mimic’ experiment asks whether directly activating ERK is sufficient to cause cocaine craving.

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

129

NEWS AND VIEWS
technical approach, because we feel it offers a model of hypothesis testing for future studies to elucidate the molecular basis of complex cognitive processes. For their study, Lu et al. used a rat model of cocaine craving and relapse that in many ways follows the basic principles of a classical pavlovian associative conditioning protocol4. In this behavioral model, cocaine craving is induced by exposure to previously learned drug-associated cues. The model has three different phases: training, withdrawal and extinction (Fig. 1a). During training, animals learn to press an active lever to receive an intravenous injection of cocaine, which is paired with a specific context and various sensory cues (such as a red house light, lever extension and tone-light signal). This pairing of cocaine with context and cues results in their subsequent association with the availability of the drug. At the end of the training period, the animals are withdrawn from cocaine and cocaine-associated cues for 1 or 30 d. After the withdrawal period, cocaine craving is evaluated by exposing the animals to the drug-associated context and cues, but without the cocaine reward. Drug-seeking behavior is evaluated by measuring the number of lever responses during this extinction test, which indicate that the animal is trying to get cocaine through the previously learned procedure. In characterizing this basic behavior, Lu et al. found that craving induced by exposure to the drug-associated environmental cues is much higher 30 d after withdrawal than 1 d after withdrawal, an indication that drug craving had increased with time in withdrawal. Lu et al. then started the identification of molecular substrates in the amygdala that could potentially mediate the incubation of drug craving by systematically evaluating their hypothesis that ERK signaling in the amygdala mediates this behavior1. To test this hypothesis, Lu et al. used their rat model in combination with an array of experimental approaches (Fig. 1). This combination provided an exceptional variety of independent lines of evidence that strongly supported their hypothesis. The three experimental approaches test different predictions of a specific hypothesis. As the name implies, the ‘measure’ experiment directly tests whether the process hypothesized to be involved actually occurs in association with the specific behavior. The ‘block’ experiment tests whether blocking the process, pharmacologically or otherwise, blocks the behavior. The ‘mimic’ experiment asks whether directly activating the molecular process is sufficient to cause the behavior. Measure and block experiments have been commonly used in neuroscience for many years to study behavior at the anatomical, molecular and cellular level. On the other hand, the mimic approach, although potentially fascinating, has rarely been used for behavioral studies because of the technical and theoretical challenges associated with the attempt to trigger complex behaviors with a single experimental manipulation. Lu and colleagues managed to effectively and convincingly use all three approaches in their study, to reach the novel conclusion that ERK activation in the amygdala mediates cocaine craving in rats. Starting with the measure experiment (Fig. 1b), the authors showed that ERK activation in the central nucleus of the amygdala correlates with cocaine-seeking behavior in rats, using antibodies specific for the phosphorylated, activated state (phospho-ERK (pERK)) of the kinase in western blots of microdissected regions of the amygdala. This observation correlates a molecular change (ERK activation) with a specific behavior process (cocaine craving). After characterizing the behavioral model of craving and relapse described above, the authors found that ERK activation in the amygdala was higher 30 d after withdrawal than 1 d after withdrawal, which nicely correlates with the increased cocaine craving behavior observed 30 d after withdrawal. These results suggest that ERK activation in the amygdala mediates the incubation of cocaine craving. But is this activation functionally relevant to the behavior? To answer this question, Lu et al. evaluated the effects of blocking amygdala ERK activation on cocaine craving behavior (Fig. 1c). Acute infusion into the central amygdala of a drug to block ERK activation decreased cocaine-seeking behavior, demonstrating that the observed ERK activation is necessary for the manifestation of the behavior. These results further support their hypothesis that ERK activation in the amygdala mediates this behavior. Lu et al. also used a combination of the block and measure experimental approaches to dissect out the components of the biochemical cascade. They found that amygdala ERK activation induced by cocaine craving is mediated via upstream activation of NMDA receptors. Blocking NMDA receptors greatly attenuated the behavioral manifestation of cocaine craving as well. These interesting observations tie these results concerning cocaine craving into a wide body of literature implicating ERK and NMDA receptors in ‘normal’ synaptic plasticity and learning5, strengthening the emerging idea that addictive processes may hijack normal mechanisms of plasticity and memory formation6. The last, most impressive and most challenging piece of evidence in this study comes from the mimic experiments (Fig. 1d). Here the authors triggered the activation of ERK in the amygdala by infusing NMDA, which results in activation of ERK. This ‘artificial’ amygdala ERK activation caused an increase in cocaine-seeking behavior 1 d after withdrawal, when animals normally show significantly less cocaine-seeking behavior than they will after 30 d. The effect is specific for ERK signaling because the effect of NMDA is blocked by an inhibitor of ERK activation. These results demonstrate that activation of ERK in the central nucleus of the amygdala is sufficient to induce complex cocaine-seeking behavior in rats. This is exceptional and impressive evidence for the involvement of a specific molecule and anatomical brain region in triggering a specific behavior. Using an impressive and carefully designed experimental approach, Lu et al. provide an exceptionally strong argument supporting a pivotal role for the ERK signaling pathway in the central nucleus of the amygdala in the incubation of cocaine craving. These results extend the hypothesis that specific cellular and molecular events in the amygdala underlie this behavior and, further, identify for the first time a specific biochemical pathway in the amygdala that modulates cocaine-seeking behavior. In addition, these results suggest a novel role of the amygdala in modulating the motivational state and emotional distress of cocaine craving and further confirm the general role of the amygdala in emotional cognitive processes. Finally, their use of a combination of measure, block and mimic experimental approaches in this study of a complex behavior offers a model of hypothesis testing for future studies to examine the molecular basis of complex cognitive processes.
1. Lu, L. et al. Nat. Neurosci. 8, 164–172 (2005). 2. Gawin, F.H. & Kleber, H.D. Arch. Gen. Psychiatry 43, 107–113 (1986). 3. O’Brien, C.P. Science 278, 66–70 (1997). 4. Grimm, J.W., Hope, B.T., Wise, R.A. & Shaham, Y. Nature 412, 141–142 (2001). 5. Sweatt, J.D. J. Neurochem. 76, 1–10 (2001). 6. Kelley, A.E. Neuron 44, 161–179 (2004).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

130

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

NEWS AND VIEWS

Comm-ing across the midline
Catherine E Krull
Comm downregulates the Slit receptor Robo on commissural axons, thereby enabling midline crossing. Elegant new experiments show that Comm functions by diverting Robo to the endosomal pathway, preventing its delivery to the growth cone. Axons in the developing CNS eventually arrive at a ‘fork in the road’ where they must make a tough choice: to cross or not to cross the CNS midline. Midline cells express Slit. Growth cones that express the Slit receptor Roundabout (Robo) are repelled and do not cross. In Drosophila, the transmembrane protein Commissureless (Comm), which is expressed in midline glia and some axons, downregulates Robo surface expression. Axons that express Comm can cross the midline. How exactly Comm controls Robo has been a matter of some controversy. Two models have been suggested (Fig. 1). According to the ‘sorting’ model, Robo would be sent to the endosomal pathway from the trans-Golgi network when Comm is present. Thus, Robo would be prevented from traveling down the axon to reach the growth cone and interact with Slit. In the competing ‘clearance’ model, Comm would control Robo by rapidly removing it from the growth cone’s plasma membrane. In this case, endocytic vesicles containing Comm and Robo would be found in the growth cone. Keleman et al. have now directly tested the two models in a set of elegant experiments that take advantage of Drosophila genetics, cell biology and imaging approaches1. Comm is expressed in two cell types: neurons that cross the midline and glial cells that lie at the midline. But where is it needed? Keleman et al. performed a genetic rescue trick in comm–/– fly embryos, adding comm back precisely in a group of neurons that normally express comm and cross the midline. They found that this enabled many of these neurons to extend axons across the midline again, even though comm was still absent in midline glial cells. Adding comm to midline glial cells, in addition to the neurons, did not improve the recovery of midline crossing. These results indicate that Comm in midline glia does not direct axons across the midline, arguing against previous studies that supported the clearance model and indicated that Comm from midline glia contributed to axon guidance2. One point scored for the sorting model. © 2005 Nature Publishing Group http://www.nature.com/natureneuroscience
Sorting Clearance
Slit (Comm) Slit (Comm)

Golgi Comm + Robo

Comm ON Comm + Robo Lysosome

Comm ON

Comm

Robo

Endocytosis

Midline

Midline

Figure 1 Two models for regulation of Robo by Comm in axons crossing the midline in Drosophila. In the ‘sorting’ model (left), Comm would divert Robo at the trans-Golgi network to the endosomal/ lysosomal pathway. No anterograde transport of vesicles with Robo or Comm would be evident in the axon, though other vesicles would be transported normally. In the ‘clearance’ model (right), vesicles containing Comm and Robo would be transported anterogradely through the axon to the growth cone, and Robo would be delivered to the plasma membrane. Comm would then cause removal of Robo from the membrane by endocytosis, and Robo and Comm would colocalize to endocytic vesicles.

Catherine E. Krull is at the Department of Cell and Developmental Biology at the University of Michigan, Ann Arbor, Michigan 48109, USA. e-mail: [email protected]

Next, the authors examined the distribution of Comm proteins that had mutations in their extracellular or transmembrane domains, in the presence or absence of Robo. Only mutations in the endosomal sorting signal or a region near the transmembrane domain disturbed the localization of Comm and Robo. Another point scored for the sorting model, where Comm is thought to function as a link between Robo and the endosomal sorting machinery. The superior test of the two models comes from the next experiments: imaging the trafficking of Robo protein, with or without Comm, in vivo. If the sorting model is correct, then little or no Robo should be moving down axons when Comm is present. If the clearance model is correct, Robo should travel down the axon to the growth cone in the presence of Comm, and both proteins should localize to endocytic vesicles in the growth cone. The authors used a Robo-GFP transgene that generated a Robo-GFP fusion protein, and expressed it in CNS and PNS neurons. The localization of Robo-GFP was very similar to the distribution of endogenous Robo in CNS neurons. The authors then made movies recording the movement of Robo-GFP in PNS neurons in live fly embryos. Why in the PNS? Adding or removing Comm in CNS neurons alters their axons’ route, making it difficult to know

whether the changes in Robo-GFP trafficking were a direct or indirect result of varying Comm expression. In addition, it is much easier to image PNS axons, as they lie just below the epidermis. Thus, analyzing the movement of Robo-GFP in PNS neurons avoids several complications. The results were striking. Without Comm, Robo-GFP vesicles in the axon moved quickly to the growth cone, presumably using anterograde transport motors. In the presence of Comm, Robo-GFP protein was mostly absent from plasma membrane and the axon and was instead found in static vesicles inside the cell body. Together, these data add a third point in support of the sorting model. Finally, the authors demonstrate that ubiquitination by Nedd4 is not required for Comm to function; nor is Nedd4 needed for midline crossing, as had been suggested previously3. The final score is three points in support of the sorting model and zero points for the clearance model, indicating that Comm rules Robo via sorting and not via clearance from the growth cone surface. These data establish that Comm regulates trafficking of the Robo receptor. Without Comm, Robo is transported down the axon to the growth cone. With Comm, the anterograde transport of Robo down the axon is terminated.

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

131

Annette Markus

NEWS AND VIEWS
Has the controversy been settled? For the most part, the answer is yes. Only a few lingering issues need to be clarified. First, we need to image in vivo the trafficking of Robo-GFP in CNS neurons that extend axons across the midline. With some genetic tweaking and technical modifications4, it should be possible to accomplish this task readily using confocal time-lapse microscopy. Second, we now want to know how exactly Comm reroutes Robo in the trans-Golgi network. And finally, these studies clarify how Comm works mechanistically to regulate Robo at the growth cone, but they do not explain how Robo is localized at a subcellular level to distal portions of commissural axons that have already crossed the midline, yet remains absent from mid-segments that are progressing across the midline. Indeed, different segments of an axon can have distinct roles in guidance and pathway selection5,6. In this scenario, there may be another molecular component that excludes Robo protein from the midline-crossing segments of commissural axons. Furthermore, it is interesting to speculate that this unknown component maybe manufactured by midline glia. Robo may also be synthesized locally in the distal tips of axons after crossing the midline. There is good evidence that local protein synthesis in the growth cone is important for cytoskeletal remodeling and motility7. These possibilities remain to be tested in the future. Collectively, these studies illustrate the real power of combining genetics, cell biology and imaging, allowing one to visualize genetic manipulations in real time as opposed to in still images. There are sure to be more surprises in store.
1. Keleman, K., Ribeiro, C. & Dickson, B.J. Nat. Neurosci. 8, 156–163 (2005). 2. Georgiou, M. & Tear, G. Development 129, 2947–2956 (2002). 3. Myat, A. et al. Neuron 35, 447–459 (2002). 4. Cabantous, S., Terwilliger, T.C. & Waldo, G.S. Nat. Biotechnol. 23, 102–107 (2005). 5. Keleman, K. et al. Cell 110, 415–427 (2002). 6. Eberhart, J. et al. J. Neurosci. 24, 1070–1078 (2004). 7. Steward, O. & Schuman, E.M. Neuron 40, 347–359 (2003).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

The cochlear amplifier and Ca2+ current–driven active stereocilia motion
Tianying Ren
Despite 20 years of study, the mechanism underlying the cochlear amplifier remains contentious. In this issue, a novel in vitro cochlear preparation implicates calcium current–driven active stereocilia motion in generating amplification.

Animals can hear all kinds of environmental sounds: predator calls, thunder, speech, music, laughter and even whispers. These sounds can vary a millionfold in intensity and a thousandfold in frequency, and arrival times of critical information may be segregated by less than a millisecond. This enormous real-time signal processing happens in the cochlea, a remarkably sensitive organ with exquisite frequency selectivity and time resolution capabilities. Auditory scientists have long believed that an active amplification mechanism called the cochlear amplifier enhances sensory tissue vibration in response to low-intensity sound, giving the cochlea its sensitivity, but its mechanism has been a topic of intense debate. In this issue, Chan and Hudspeth describe a new preparation that has allowed them to study the cochlear amplifier in vitro, demonstrating a Ca2+ current–dependent hair bundle motility mechanism1. In 1948, Thomas Gold applied radio engineering principles to study hearing and concluded that the cochlea must contain an active amplification mechanism2, later termed the ‘cochlear

amplifier’ by Davis in 1983 (ref. 3). The cochlear amplifier requires metabolic energy, and it seems to be a nonlinear feedback process that boosts the amplitude of sound-induced vibrations cycle by cycle. However, despite more than 20 years of intensive study, direct experimental support for theories behind the amplifier’s mechanism has been limited owing to the inherent difficulties of measuring acoustic power in a living cochlea. Many believe that the cochlear amplifier resides in the outer hair cells in the organ

of Corti, which sits on the vibratory basilar membrane4 (Fig. 1), but two very different mechanisms have been proposed to explain amplification. One theory proposes that voltage-dependent changes in the length of outer hair cell bodies are responsible. In mammals, changing membrane potential induces conformation changes in prestin, a protein embedded in the lateral wall of the outer hair cell, which in turn cause somatic motility5. Hair cell receptor potentials result

Recording electrode Stimulating electrode Tectorial membrane

Inner hair cell Outer hair cells Basilar membrane

Sound

Tianying Ren is at the Oregon Hearing Research Center, Oregon Health & Science University, Portland, Oregon 97239-3098, USA and at the School of Medicine, Xi’an Jiaotong University, Xi’an, P.R. China. e-mail: [email protected]

Figure 1 Schematic of the in vitro cochlear preparation used by Chan and Hudspeth. Sounds vibrate the air, fluid, basilar membrane and organ of Corti in the direction of the vertical white arrow. This induces shearing motion between the tectorial membrane and the organ of Corti, which deflects the hair bundles of inner and outer hair cells (white horizontal arrow). In this study, the authors recorded sound-induced hair bundle vibration of inner hair cells while applying a transepithelial electrical potential across the organ through two stimulating electrodes.

132

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

NEWS AND VIEWS

Radial displacement

Ac

e tiv

ICa2+

s Pa

e siv

no ICa2+

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Stimulus Sound pressure level
Figure 2 Ca2+ current–dependent nonlinear hair bundle responses. Hair bundle vibration in response to acoustic stimuli is proportional to input intensity without Ca2+ current (no ICa2+, dashed blue line) and shows amplification with Ca2+ current (ICa2+, red line). The solid black trace on the right shows a stimulus tone burst at a low sound intensity. The hair bundle response with Ca2+ current intact (solid red line) is significantly greater than that when the Ca2+ current is blocked by amiloride (dashed blue line).

from sound-driven basilar membrane displacements, so the voltage-dependent change in hair cell length follows the basilar membrane motion. If the time delay caused by the organ architecture between the sensory hair cells and the basilar membrane is appropriate, hair cell–generated force can enhance the sound-induced vibration of the basilar membrane6. Mice with targeted deletion of prestin manifest hearing loss and poor frequency selectivity and lack outer hair cell somatic motility7. These findings have led some to conclude that prestin-based somatic motility alone is responsible for the cochlear amplification. However, this view has been criticized as incomplete, because any feedback system consisting of multiple parts will be compromised if only one component in the feedback path is removed8. Also, somatic motility cannot account for performance at the full frequency range of the system, because at high frequencies, the low-pass filter formed by the resistance and capacitance of the cell membrane should greatly attenuate the receptor potential and restrict the somatic response9. Finally, lower vertebrates such as lizards10 do not have differentiated outer hair cells in their hearing organs, and somatic electromotility has never been demonstrated in any of these organisms. For these species, the mechanism responsible for highly sensitive and sharply tuned hearing must reside elsewhere. This brings us to the second theory of the cochlear amplifier mechanism: that it involves motion of the stereocilia bundle, a hair-like organelle in auditory sensory cells. Martin and Hudspeth11 found that sensory hair cells from the bullfrog vestibular organ oscillated spontaneously with amplitudes as great as 50 nm at frequencies from 5 to 40 Hz. When hair bundles were stimulated at a frequency close to the hair cell’s natural oscillation frequency, vibrations as small as 5 nm in amplitude entrained

hair-bundle oscillations, and the bundle vibrated more efficiently for small stimuli than for large stimuli. Because the power needed to overcome the viscous drag force was more than that provided by the stimulus probe, the hair bundles were thought to produce power actively and therefore to amplify bundle vibrations11. This power is generated by myosin motor molecules, which mediate a slow form of adaptation, and by Ca2+ current–dependent closure of transduction channels, which is responsible for fast adaptation12. Fast adaptation is a rapid decrease of transduction current in less than one thousandth of a second during a stationary displacement of hair bundles in the excitatory direction—that is, towards the taller side of the stereocilia bundle. This current decrease is caused by the transduction channel closure, which occurs when Ca2+ ions enter the bundles and bind to the internal face of the channel. Subsequently, channel closure increases the tension of tip-links, pulling the bundle toward its short side13. If bundle motion is oscillatory in timing or occurs at a frequency that matches the speed of channel closure, these changes in tip-link tension will boost bundle vibration. This bundle-generated force could increase cochlear sensitivity and frequency selectivity in the same manner as prestin-driven somatic electromotility8. In fact, fast adaptation does exist in rat cochlear outer hair cells, occurring with a time constant of 150 µs in 1.5 mM Ca2+ in vitro, and it may be even faster in a living cochlea14. Although these findings indicate that a bundle-based amplification mechanism may operate in the cochlea, it remains unknown whether it is actually responsible for the cochlear amplifier in an active mammalian cochlea. In this issue, Chan and Hudspeth1 present an important technical contribution to the field: an active in vitro preparation of the mammalian cochlea (Fig. 1). It was long believed that active

cochlear mechanisms could not be studied in vitro, but Chan and Hudspeth achieved this significant progress by providing perilymph- and endolymph-like ionic environments, an endocochlear potential–like transepithelial potential, and normal acoustic stimuli. The authors used their new preparation to demonstrate compressive nonlinearity, a characteristic of the cochlear amplifier, in hair bundle responses to acoustic stimuli. The authors note that the nonlinearity in this study was substantially different from that seen in sensitive cochlea in vivo, which they attributed to a number of factors, including low temperature, mismatching of the stimulus frequency and native resonant frequency, and the base-to-apex gradient of the cochlear nonlinearity. Perhaps the lack of a cochlear traveling wave in the in vitro preparation could also have been responsible for decreased nonlinearity observed in this study. The most exciting conclusion of this study is that a Ca2+ current–based mechanism was responsible for observed nonlinear hair bundle responses to sound (‘active’ line in Fig. 2). This conclusion was based on the following observations. First, either applying the transduction channel blocker amiloride or removing the transepithelial potential severely decreased nonlinear bundle response to sound (‘passive’ line in Fig. 2), implying that the nonlinear responses depended on normal transduction currents. Second, replacing monovalent cations in the endolymph solution with the impermeant cation NMDG, which largely blocks the transduction current but leaves Ca2+ entry unaffected, did not have any significant effect on the nonlinear hair bundle responses. These findings demonstrated that a Ca2+ current–based mechanism was responsible for the nonlinearity observed in this study. There are two possible mechanisms for Ca2+ action on hair bundles: binding to myosin motors and regulation of channel closure. Because Ca2+ binding to myosin motors and their movement along actin filaments is much slower, the latter is likely to be responsible for high-frequency amplification. Hair bundle movements have a narrow dynamic range, and the authors hypothesize that other mechanisms, namely somatic electromotility and myosinbased slow adaptation, may work to overcome this limitation by positioning the bundles and the basilar membrane so as to maximize the Ca2+ current–mediated fast bundle vibration. By showing evidence that Ca2+ current–mediated active hair bundle motility provides force to the cochlear amplifier, this outstanding study significantly advances our understanding of how the cochlear amplifier

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

133

NEWS AND VIEWS
works, and the preparation developed in this study will provide a valuable new approach for studying active cochlear mechanisms in vitro. A number of important questions remain, however. Chan and Hudspeth proposed a supplemental role for hair cell somatic motility in the cochlear amplifier, but this is not entirely consistent with data showing that disrupting prestin completely eliminates nonlinear cochlear amplification7. Also, as the cochlear amplifier is a wholeorgan phenomenon, and as it depends upon a normal traveling wave15, the exciting findings from this study must be tested in sensitive living cochlea. Chan and Hudspeth’s study will surely inspire such in vivo studies.
1. Chan, D.K. & Hudspeth, A.J. Nat. Neurosci. 8, 149– 155 (2005). 2. Gold, T. Proc. R. Soc. Lond. B Biol. Sci. 135, 492–498 (1948). 3. Davis, H. Hear. Res. 9, 79–90 (1983). 4. Robles, L. & Ruggero, M.A. Physiol. Rev. 81, 1305– 1352 (2001). 5. Zheng, J. et al. Nature 405, 149–155 (2000). 6. Nilsen, K.E. & Russell, I.J. Nat. Neurosci. 2, 642–648 (1999). 7. Liberman, M.C. et al. Nature 419, 300–304 (2002). 8. Fettiplace, R. & Ricci, A.J. Curr. Opin. Neurobiol. 13, 446–451 (2003). 9. Fridberger, A. et al. J. Neurosci. 24, 10057–10063 (2004). 10. Manley, G.A., Kirk, D.L., Koppl, C. & Yates, G.K. Proc. Natl. Acad. Sci. USA 98, 2826–2831 (2001). 11. Martin, P. & Hudspeth, A.J. Proc. Natl. Acad. Sci. USA 96, 14306–14311 (1999). 12. Choe, Y., Magnasco, M.O. & Hudspeth, A.J. Proc. Natl. Acad. Sci. USA 95, 15321–15326 (1998). 13. Gillespie, P.G. & Walker, R.G. Nature 413, 194–202 (2001). 14. Kennedy, H.J., Evans, M.G., Crawford, A.C. & Fettiplace, R. Nat. Neurosci. 6, 832–836 (2003). 15. Hubbard, A. Science 259, 68–71 (1993).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Channeling a ‘funny’ side of memory
Daniel Johnston
Voltage-gated ion channels shape the integration of synaptic input in dendrites. Forebrain-restricted deletion of the hyperpolarizationactivated channel HCN1 enhances spatial learning, demonstrating a behavioral role for an active dendritic conductance. The role of synaptic potentiation and depression in various forms of learning and memory is well established. However, attention has recently been focused on other potential mechanisms, with the idea that synaptic plasticity may not be the whole story in the neurobiology of learning and memory1,2. In a paper published in Cell3, Nolan et al. take a major step forward in linking synaptic plasticity, dendritic integration and memory. In an impressive collaboration among different laboratories, this study reports a role for a particular ionic current (once called the ‘funny’ current) in certain forms of memory. The results strongly suggest that postsynaptic mechanisms beyond the synapse are a critical component of learning and memory in the behaving animal. A cation current active at hyperpolarized membrane potentials was first described in heart cells and was called the ‘funny’ current for its unusual properties4. A similar current was later described in neurons and more descriptively, but less concisely, called the hyperpolarizationactivated, cyclic nucleotide–gated, nonselective cation current (Ih). Four genes, HCN1–4, encode the channels underlying Ih, with HCN1 and HCN2 forming Ih channels in many neurons4. In their new paper3, Nolan et al. deleted the HCN1 gene from the forebrain of mice and analyzed the behavioral and physiological consequences of the loss of Ih from neurons in this region. Surprisingly, they find that a significant decrease in Ih actually enhances learning in a hippocampus-dependent spatial memory task and increases long-term potentiation (LTP) specifically at direct perforant path (temporoammonic) inputs from layer III of entorhinal cortex to CA1 neurons in the hippocampus. In a previous paper5, Nolan et al. found that deletion of the HCN1 gene from the entire mouse led to profound deficits in motor learning. For example, mice could not learn to navigate to a submerged platform in a water maze experiment, even if given visible cues such as a flag on the platform, but instead tended to swim in circles. Nolan et al. now show that, in contrast, mice with a forebrain-only deletion of HCN1 (HCN1f/
f,cre)

have no such deficits in motor learning and perform similarly to control mice on the visible platform version of the water maze experiment. Moreover, when required to find a hidden platform based on the location of spatial cues, these mice learn faster than control mice and also have reduced path lengths when swimming to the platform. An intriguing finding is that both contextual and cued fear conditioning were unaltered in the HCNf/f,cre mice. Because contextual fear conditioning is also thought to be a hippocampus-dependent spatial memory task, these results clearly show that the role of Ih in animal behavior is complex, with different contributions to different types of learning.

f/f HCN1 fEPSP slope

Schaffer collateral input

f/f,cre HCN1

Temporoammonic input fEPSP slope

Time
Figure 1 The loss of Ih enhances LTP specifically at distal inputs to CA1 pyramidal neurons. The density of Ih channels in CA1 dendrites increases with distance from the soma, as indicated by the blue (lower channel density) to red (higher channel density) gradient. When these channels are absent, LTP is unaffected at Schaffer collateral inputs to CA1 (top graph), which synapse onto proximal dendrites, but is enhanced at temporoammonic inputs (bottom graph), which synapse onto distal apical dendrites where Ih expression is normally high. Upper line on graphs refers to neuron at right, and lower line refers to neuron at left.

Daniel Johnston is at the Center for Learning and Memory at the University of Texas at Austin, Austin, Texas 78712, USA. e-mail: [email protected]

134

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

NEWS AND VIEWS
Without Ih study. For example, why does the loss of Ih specifically affect only certain forebrain memory tasks10, and why does it affect motor learning and spatial memory in opposite ways? Given that there are many other voltage-gated ion channels expressed in dendrites that affect synaptic integration11–13, do any of these interact with Ih and also affect learning and memory? Despite these and other lingering questions, the results of Nolan et al. are a watershed of sorts in the field for several reasons. First, they clearly ascribe a role for Ih in hippocampal-dependent learning and memory. Second, because Ih is so heavily expressed in dendrites and has substantial effects on dendritic integration of synaptic input, a new emphasis on the role of dendritic mechanisms and intrinsic excitability in learning and memory emerges from this study. Tens of thousands of synaptic inputs impinge on the dendritic tree, and this study and others1 suggest that the ways in which the dendrites and the intrinsic properties of the neuron modify these inputs are important in the memory storage process itself. Ih has also been shown recently to be ‘plastic’14,15, or subject to activity-dependent changes, suggesting that changes in Ih during learning might be a substrate for memory. Thus, the results of this study are not ‘funny’ at all, but instead serious and intriguing.
1. Zhang, W. & Linden, D.J. Nat. Rev. Neurosci. 4, 885– 900 (2003). 2. Frick, A., Magee, J. & Johnston, D. Nat. Neurosci. 7, 126–135 (2004). 3. Nolan, M.F. et al. Cell 119, 719–732 (2004). 4. Robinson, R.B. & Siegelbaum, S.A. Annu. Rev. Physiol. 65, 453–480 (2003). 5. Nolan, M.F. et al. Cell 115, 551–564 (2003). 6. Magee, J.C. J. Neurosci. 18, 7613–7624 (1998). 7. Magee, J.C. Nat. Neurosci. 2, 508–514 (1999). 8. Poolos, N.P., Migliore, M. & Johnston, D. Nat. Neurosci. 5, 767–774 (2002). 9. Shah, M.M., Anderson, A.E., Leung, V., Lin, X. & Johnston, D. Neuron 44, 495–508 (2004). 10. Remondes, M. & Schuman, E. Nature 431, 699–703 (2004). 11. Johnston, D., Magee, J.C., Colbert, C.M. & Cristie, B.R. Annu. Rev. Neurosci. 19, 165–186 (1996). 12. Magee, J.C. Nat. Rev. Neurosci. 1, 181–190 (2000). 13. Häusser, M., Spruston, N. & Stuart, G.J. Science 290, 739–744 (2000). 14. van Welie, I., van Hooft, J.A. & Wadman, W.J. Proc. Natl. Acad. Sci. USA 101, 5123–5128 (2004). 15. Wang, Z., Xu, N., Wu, C., Duan, S. & Poo, M. Neuron 37, 463–472 (2003).

EPSPs © 2005 Nature Publishing Group http://www.nature.com/natureneuroscience With Ih
Figure 2 The loss of Ih enhances temporal summation of excitatory synaptic input. The summation of synaptic input depolarizes the neuron, deactivates Ih and reduces the summation of the input by hyperpolarizing the cell (red). When Ih is blocked or removed genetically, the amount of temporal summation is increased significantly (blue). Modified from ref. 7. (Scale bars: 5 mV, upper trace; 4 mV, lower trace; 100 ms)

Ih has many unique and difficult-to-intuit properties. It is partially active at rest, further activated by hyperpolarization and deactivated by depolarization. As a nonselective cation current, however, it has a reversal potential of around –30 mV, so turning on the current with hyperpolarization tends to depolarize the cell, whereas turning it off with depolarization leads to a hyperpolarization. In CA1 pyramidal neurons, the density of Ih channels increases with distance from the soma and is about sevenfold higher in the distal apical dendrites6. Summated excitatory synaptic input, which depolarizes the neuron, deactivates Ih and thus suppresses temporal summation7. Because of the density gradient of Ih along the dendrites, however, temporal summation is more dampened for distal than for proximal inputs, with the net result that the temporal summation of all inputs reaching the soma is about equal (that is, normalized7). The effect of Ih in reducing temporal summation is also somewhat dependent on the frequency of synaptic input, being greatest at intermediate frequencies and lesser at low and high frequencies8. With this background on the properties of Ih and its effect on synaptic integration, how does the loss of Ih in the forebrain lead to enhanced spatial memory? Focusing on the neuronal mechanisms that might underlie the behavioral phenotype of HCNf/f,cre mice, Nolan et

al. found changes in hippocampal-dependent network oscillations. Both low- and high-frequency oscillations appeared unchanged in the knockout mice, but power in the intermediate range, or theta frequency band (4–9 Hz), was enhanced. This is particularly interesting in light of the frequency-dependent effects of Ih on temporal summation mentioned above. In whole-cell recordings, CA1 pyramidal neurons had more negative resting potentials, higher input resistances and longer membrane time constants, all characteristic features of a loss of Ih (ref. 9). Furthermore, Nolan et al. found enhanced LTP only at the temporoammonic input from layer III of entorhinal cortex to these neurons (Fig. 1). Because the synapses from this pathway terminate at the most distal regions of the apical dendrites of CA1 neurons, where the density of Ih is normally the highest, the authors argue that the enhanced LTP is due to a greater temporal summation at those synapses. In other words, Ih can be thought of as a partial brake that reduces dendritic depolarization. Take away the brake in the knockout animals, and greater depolarization can occur with a given synaptic input, leading to greater spread of synaptic input to the soma and possibly to more LTP (Fig. 2). The more distal the input, the greater this effect would be. Undoubtedly, this work raises many interesting questions and highlights areas for further

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

135

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

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

The site of action potential initiation in cerebellar Purkinje neurons
Beverley A Clark1,3, Pablo Monsivais1,3, Tiago Branco2, Michael London1 & Michael Häusser1 Knowledge of the site of action potential initiation is essential for understanding how synaptic input is converted into neuronal output. Previous studies have shown that the lowestthreshold site for initiation of action potentials is in the axon. Here we use recordings from visualized rat cerebellar Purkinje cell axons to localize the site of initiation to a well-defined anatomical structure: the first node of Ranvier, which normally forms at the first axonal branch point. Although recordings from the axon initial segment in various neuronal types have demonstrated that action potential initiation normally occurs in the axon1–5, the precise site of action potential initiation within the axon remains unknown. This is in part due to the difficulty of making direct axonal recordings at substantial distances from the

soma. To facilitate axonal recordings, we filled rat cerebellar Purkinje neurons with a fluorescent dye and imaged the axons using a CCD camera (Fig. 1a and Supplementary Methods online; all procedures were carried out according to U.K. Home Office regulations). This permitted cell-attached patch-clamp recordings to be made from the axon under direct visual control. To identify the site of action potential initiation in the axon, we made simultaneous multiple patch-clamp recordings of extracellular action currents from the soma and from locations at various distances down the axon (Fig. 1a–c) during spontaneous firing of Purkinje cells, which occurs at rates comparable to those found in vivo6. Previous studies have shown that action potentials in the initial segment of the axon precede those recorded at the soma1,2,4,5. We predicted that this negative time difference between the axonal and the somatic action potential should be greatest at the site of initiation of the action potential. Figure 1d shows that there is a V-shaped relationship between the axonal-somatic action potential latency and the distance of the axonal recording site. The site of initiation of the action potential was identified as the nadir of this relationship, corresponding to the largest axon-soma latency advance. This occurred at a distance of 75 ± 11 µm from the soma (measured from a bilinear fit to the data points; n = 68 recordings). The latencies were very similar for spontaneous action potentials recorded in cell-attached recordings and those evoked by depolarizing current pulses in somatic

a

b
Soma

60 µm

200 µm 50 µm 100 ms 40 pA 30 pA 20 pA

c
Soma 50 pA 20 pA

d
Axon-soma latency (µs)

600

400

200

60 µm 200 µm

0

10 pA –0.5 0.0 0.5 Time (ms) 1.0

–200 0 100 200 300 400 Distance from soma (µm) 500 600

Figure 1 Direct measurement of the initiation site during spontaneous action potential firing. (a) Fluorescence image of a Purkinje cell filled with Alexa 488, illustrating the recording configuration. (b) Simultaneous triple cellattached recording of spontaneous action potentials from the soma (top electrode) and at two locations along the axon (middle and bottom electrodes, 60 and 200 µm) of the same Purkinje neuron shown in a. (c) Latency differences between the somatic action potential (aligned to t = 0 at its first peak) and the action potential at the two axonal recording sites (averages of 50 action potentials are shown). (d) Latency differences of spontaneous action potentials for 68 simultaneous somatic and axonal recordings (colored data points correspond to the recording in c; open circles are latencies from the derivatives of somatic whole-cell voltage recordings where action potentials were evoked using 0.1–0.3 nA current pulses). A bilinear fit to the data is shown, giving a maximum axon-soma advance at 75 ± 11 µm; its slope indicates antidromic and orthodromic axonal propagation velocities of 0.56 ± 0.11 and 0.77 ± 0.05 m s–1, respectively.

1Wolfson

Institute for Biomedical Research, Department of Physiology, and 2MRC Laboratory for Molecular Cell Biology, University College London, Gower Street, London WC1E 6BT, UK. 3These authors contributed equally to this study. Correspondence should be addressed to M.H. ([email protected]).

Published online 23 January 2005; doi:10.1038/nn1390

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

137

B R I E F C O M M U N I C AT I O N S a
Figure 2 The action potential initiation site corresponds to the first branch point. (a) Confocal images of Purkinje cells filled with biocytin (green, left) and counterstained against MBP (blue, middle) to show the myelin and its relationship to the individual Purkinje cells (overlay, right). Arrowheads point to the starting point of the myelin (and thus the end of the initial segment). The inset shows the first node/branch point (MBP stain not in the plane of the labeled axon digitally removed for clarity). Scale bar: 15 µm for main image, 5 µm for inset. (b) Confocal images of Purkinje cells filled with biocytin (green, left) and counterstained against ankyrin-G (red, middle). The arrowhead indicates the initial segment, the top arrow the first node and the lower arrow the second node (lower panel; 312 µm from the soma; intervening axon cut). Scale bar: 10 µm. (c) Histogram showing distance to the end of the initial segment (gray bars: 2 µm bins) and the first branch point (white bars: 8 µm bins) measured from the point of origin of the axon from the soma. Gaussian fits to the data points are shown. Mean values were 21 ± 4 µm (n = 31) and 82 ± 14 µm (n = 22) from the soma, respectively.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

b

c
Count

12 10 8 6 4 2 0 0 20 40 60 80 100 120 140

Distance from soma (µm)

whole-cell recordings (Fig. 1d), and they were not affected by SR95531 (10 µM), an antagonist of inhibitory synaptic transmission (n = 4; P = 0.4). Increasing the action potential frequency to 180–200 Hz to match high-frequency firing rates observed in vivo6 also did not change the latency as compared to spontaneous firing (n = 9; P = 0.14). Does the physiologically determined site of action potential initiation correspond to a particular anatomical feature of the axon? To address this question, we examined the structure of the axon using confocal microscopy of fluorescently labeled Purkinje cells in the same preparation (see Supplementary Methods online). The physiological initiation site corresponded very well with the origin of the first axonal branch point, giving rise to an axonal collateral measured to be 82 ± 14 µm from the soma (n = 22; Fig. 2; ref. 7). To determine the relationship between the branch point and myelination, we stained for myelin using immunofluorescent labeling of myelin basic protein (MBP). This showed that the initial segment terminated at only 21 ± 4 µm from the soma (Fig. 2; n = 31), ruling out the initial segment as the site of initiation. Furthermore, the first node of Ranvier was observed to occur at the first branch point (Fig. 2c; n = 22 of 22 axons). We confirmed and extended these observations using immunofluorescent labeling of ankyrin-G,

which links voltage-gated Na+ channels to the cytoskeleton and is closely associated with axonal Na+ channels in many neuronal types8. Dense ankyrin-G immunofluorescence was detected at axon initial segments8 and at the first node of Ranvier, highlighting the Y-shaped bifurcation of the branch point (Fig. 2b). Ankyrin-G immunofluorescence also allowed reliable identification of the second node, which was located 346 ± 30 µm from the soma (n = 3; Fig. 2b), thus ruling out the second node as the site of initiation. Taken together, these findings indicate that the site of origin of the action potential corresponds to the first node, which is normally at the first axonal branch point. To investigate the geometric constraints on spike initiation in Purkinje cell axons, we designed a computer model based on the morphology of a Purkinje cell (Fig. 3 and Supplementary Videos 1 and 2 online), incorporating a detailed reconstruction of its axon. The model successfully reproduced our experimental data on action potential initiation, and demonstrated that during action potential initiation, the spatial spread of the membrane potential initially showed a single sharp maximum at the first node and then fell steeply, attenuating by more than two-thirds of its amplitude at the initial segment or second node. Indeed, such a spatial profile was required to produce latency differences that matched our experimental data. The model thus confirms that initiation does occur at a specific location, namely the first node of Ranvier, and therefore serves as an important ‘proof of concept’ supporting our experimental conclusions. We have provided the first direct localization of the site of initiation of action potentials in a mammalian CNS neuron. The correspondence of the initiation site with the first node resolves a longstanding controversy: both early work in motoneurons9 and more recent studies in Purkinje cells2 and pyramidal cells4,5,10 were unable to definitively resolve whether the action potential is initiated at the distal end of the initial segment, the first or second nodes or even further down the axon. It will be of great interest to explore how axonal geometry, channel properties and densities interact to target initiation to the first node, and whether similar rules hold in other neuronal types. The location of the initiation site at the first node may reflect a balancing of conflicting requirements. The initiation site should be at some distance from the soma, both to minimize the effect of dendritic capacitance on threshold10,11 and to reduce the impact of the large axonal action potential conductances on synaptic integration12. This provides the most efficient use of the available Na+ channels, with the initial section of myelin further minimizing the effective capacitance ‘visible’ to the node, thus reducing the number of Na+ channels required for spike initiation and the energetic demands on the neuron associated with action potential firing13. However, if the action potential is initiated too far down the axon, this weakens the strength and precision of the link between synaptic input

138

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

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

50 µm

20 mV
500 µs

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

50 ms

c

20
Vm (mV)

0 –20 –40 –60 0
1 2 3 4 5 6 7

200

400

600

800

1,000

Distance from soma (µm)

Figure 3 A model of action potential initiation in Purkinje neurons. (a) Reconstructed Purkinje neuron used for the model. The axon was extended artificially (dashed line; see Supplementary Methods). The inset shows spontaneous action potential firing in the model. (b) Capacitive currents (upper traces, inverted and normalized) and voltage traces of the action potential at the first node of Ranvier (red), soma (green) and the second node of Ranvier (blue). The latency difference between the peaks of the downward currents at the first node and the soma was 195 µs, consistent with our experimental data (Fig. 1d). (c) A schematic diagram of the axon, indicating the relative locations of nodes of Ranvier (top). The graph shows the voltage profile along the length of the axon at different time points during action potential initiation: 11.295 ms (1), 16.245 ms (2), 16.485 ms (3), 16.645 ms (4), 16.730 ms (5), 16.845 ms (6) and 17.145 ms (7) after the peak of the previous action potential.

and spike output, and propagation to the soma may be less secure and associated with significant delays. The targeting of inhibitory input onto the axon initial segment14, exemplified by the ‘pinceau’ arrangement of basket cell axon terminals on Purkinje cells15, could be an additional constraint such that the location of the initiation site at the first node may afford some protection from asynchronous inhibitory synaptic conductances. Understanding how these different constraints are balanced to result in the localization of the initiation site should provide fundamental insights into the regulation of neuronal excitability.
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS We thank J. T. Davie, J.J.B. Jack, C. Racca and A. Roth for helpful comments; V. Bennett for the ankyrin-G antibody; and A. Gidon, L. Ramakrishnan, E. Rancz, and A. Roth for help with reconstructions and videos. This work was supported by grants from the Wellcome Trust, European Commission, and the Gatsby Foundation. M.L. is a Long-Term Fellow of the Human Frontier Science Program, and T.B. was funded by the Wellcome Trust 4-year PhD Programme in Neuroscience.

COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 13 October; accepted 20 December 2004 Published online at http://www.nature.com/natureneuroscience/
1. Stuart, G.J. & Sakmann, B. Nature 367, 69–72 (1994). 2. Stuart, G. & Häusser, M. Neuron 13, 703–712 (1994). 3. Häusser, M., Stuart, G., Racca, C. & Sakmann, B. Neuron 15, 637–647 (1995). 4. Colbert, C.M. & Johnston, D. J. Neurosci. 16, 6676–6686 (1996). 5. Stuart, G., Schiller, J. & Sakmann, B. J. Physiol. (Lond.) 505, 617–632 (1997). 6. Armstrong, D.M. & Rawson, J.A. J. Physiol. (Lond.) 289, 425–448 (1979). 7. Gianola, S., Savio, T., Schwab, M.E. & Rossi, F. J. Neurosci. 23, 4613–4624 (2003). 8. Zhou, D. et al. J. Cell Biol. 143, 1295–1304 (1998). 9. Coombs, J.S., Curtis, D.R. & Eccles, J.C. J. Physiol. (Lond.) 139, 232–249 (1957). 10. Colbert, C.M. & Pan, E. Nat. Neurosci. 5, 533–538 (2002). 11. Mainen, Z.F., Joerges, J., Huguenard, J.R. & Sejnowski, T.J. Neuron 15, 1427–1439 (1995). 12. Häusser, M., Major, G. & Stuart, G.J. Science 291, 138–141 (2001). 13. Attwell, D. & Laughlin, S.B. J. Cereb. Blood Flow Metab. 21, 1133–1145 (2001). 14. Somogyi, P., Tamas, G., Lujan, R. & Buhl, E.H. Brain Res. Brain Res. Rev. 26, 113–135 (1998). 15. Eccles, J.C., Ito, M. & Szentágothai, J. The Cerebellum as a Neuronal Machine (Springer-Verlag, Berlin, 1967).

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

139

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

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Striate cortex (V1) activity gates awareness of motion
Juha Silvanto1,2, Alan Cowey3, Nilli Lavie1,2 & Vincent Walsh1,2 A key question in understanding visual awareness is whether any single cortical area is indispensable. In a transcranial magnetic stimulation experiment, we show that observers’ awareness of activity in extrastriate area V5 depends on the amount of activity in striate cortex (V1). From the timing and pattern of effects, we infer that back-projections from extrastriate cortex influence information content in V1, but it is V1 that determines whether that information reaches awareness. Of the many visual stimuli that impinge on the retina, few are consciously perceived at any one time. Several lines of evidence have identified V1 as the area most likely to play a central role in awareness: it is the area in which activity correlates most closely with awareness, even if the experience is faulty1; it receives back-projections from all the extrastriate visual areas2; damage to this area in humans3,4 and monkeys5 can abolish visual awareness of all stimulus attributes in the corresponding parts of the visual field; and transcranial magnetic stimulation (TMS) to this area interferes with perception of neural activity induced in extrastriate cortex6. These studies show that V1 activity correlates with conscious experience but do not establish that it is the recursive connections between extrastriate and striate cortex that determine the content or presence of awareness7,8. It is therefore essential to determine whether activity in V1 arising from backprojections from extrastriate cortex can produce visual awareness. It has never been shown that the attributes of awareness are dictated by these back-projections, yet it is a cornerstone of many current views of visual awareness7–10. We therefore examined, through direct stimulation of the occipital cortex (see Supplementary Methods online), whether the amount of V1 activity dictates awareness and, if so, whether the V5–V1 cortical back-projection determines the content of conscious awareness. All procedures were approved by the ethics committee of University College London and informed written consent was obtained from each subject. We applied single-pulse TMS over V5 and V1 at different onset asynchronies from –80 to +80 ms and asked subjects to report their induced perceptions (see Supplementary Methods). Using a modified binary search paradigm11, the intensity of TMS was determined individually for each subject according to their phosphene threshold12 (the intensity of TMS at which a phosphene is produced on about 75% of occasions).

When TMS was applied to V1 above the phosphene threshold, subjects reported the presence of a small, stationary phosphene located in the contralateral lower visual field within a few degrees of the vertical meridian4,6,12. Suprathreshold TMS over V5 also elicited the experience of a phosphene in the visual field contralateral to stimulation but with different features. As one would expect given the differences between V1 and V5 receptive field properties13, it was larger, moving and of a different shape from the V1 phosphene. When subthreshold TMS (that is, TMS producing no phosphene on its own; intensity 20% below phosphene threshold was used for all subthreshold stimulations) was applied over V5 followed by a subthreshold pulse to V1, subjects did not report any phosphene. Crucially, however, when a subthreshold pulse was applied over V5 followed 10–40 ms later by a suprathreshold pulse over V1, subjects reported a phosphene. Notably, this phosphene was not merely the suprathreshold V1 phosphene. Rather, it acquired features of a suprathreshold V5 phosphene: subjects now reported the perception of movement (Fig. 1a) and the shape and size of their percept was a mixture of V1 and V5 phosphenes (Fig. 1b). This shows that activity in V5 that, on its own, is insufficient to induce a moving percept can produce such a percept if the level of induced activity in V1 is high enough. Furthermore, this also shows that what reaches awareness via V1 is characterized by back-projections from extrastriate areas. Although the present study was the first to assess human awareness through direct manipulation of the levels of activity in V1 with TMS, and hence allowed us to determine the critical role of V1 activity in motion awareness, the narrow time window for V5–V1 interaction that we report (10–50 ms) is consistent with previous reports of extrastriate-striate feedback interactions in motion6,13. Notably, we showed that subthreshold TMS over V5 followed by suprathreshold V1 TMS produces awareness, but when suprathreshold TMS over V5 (which results in the perception of moving phosphene) is followed by subthreshold V1, TMS phosphene perception is suppressed6. This contrast emphasizes the importance of activity in V1 in determining the presence of awareness. However, the level of activity in V5 (either supra- or subthreshold) when V5 stimulation is followed by V1 stimulation does not dictate whether phosphenes are perceived in this context. Our findings that moving phosphenes are perceived only when suprathreshold V1 stimulation follows, but not precedes, subthreshold V5 stimulation, together with the gradual increase in motion perception from the 10–50 ms period, precludes a simple feed-forward summation account and points instead to a critical time of backprojection arrival in V1. A feed-forward summation account, in which V5 activity is primed with subthreshold TMS before being summed with a feed-forward input from a suprathreshold V1 stimulation, remains logically possible if the present finding is taken in isolation. However, recent findings that V1 stimulation has an effect on motion perception both before and after the critical V5 stimulation period,

1Institute of Cognitive Neuroscience, University College London, 17 Queen Square, London WC1N 3AR, UK. 2Department of Psychology, University College London, 26 Bedford Way, London WC1H 0AP, UK. 3Department of Experimental Psychology, University of Oxford, South Parks Road, Oxford, OX1 3UD, UK. Correspondence should be addressed to J.S. ([email protected]).

Published online 9 January 2005; doi: 10.1038/nn1379

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

143

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

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 1 Motion and size of reported phosphenes. (a) Motion judgments of the six subjects who perceived moving phosphenes when V5 was stimulated above threshold. Scale was adapted from a previous study9: 0 = phosphene was absent; 1 = phosphene was stationary; 2 = subject was uncertain whether phosphene was moving or stationary; 3 = the phosphene was moving. When subthreshold TMS was applied over V5 40 ms before V1 was stimulated above threshold, five of the six subjects reported the perception of moving phosphenes. In contrast, V1 suprathreshold stimulation by itself always induced stationary phosphenes in all subjects. This difference in subjects’ motion judgments is statistically significant (Z = 2.236; P < 0.025, Wilcoxon test). When both V1 and V5 were stimulated below threshold, only one subject reported the perception of a (stationary) phosphene. (b) Shape/size judgments (n = 8). 1 = percept was like a V1 phosphene; 2 = percept was like a mixture of V1 and V5 phosphenes; 3 = percept was like a V5 phosphene; 0 = phosphene was absent. Five of the eight subjects perceived a mixture of V1 and V5 phosphenes when TMS was applied over V5 30 ms before V1. Subjects’ percepts in this condition differed significantly from those induced by suprathreshold TMS over V1 by itself (Z = 2.236; P = 0.025; Wilcoxon).

whereas V5 stimulation affects motion perception only during one critical time period before a later V1 stimulation, make this account unlikely14 because it predicts that V5 stimulation effects should postdate the latest of V1 effects. Finally, we confirmed the specificity of the timing and direction of the V5–V1 interaction by stimulating the frontal eye fields 0–60 ms after V5 and also by stimulating both V5s 40 ms apart. Neither manipulation had the effects we describe here (Supplementary Note 1).
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS This work was supported by a scholarship from the University College London Graduate School and grants from the UK Medical Research Council (A.C., N.L., V.W.) and the Wellcome Trust (V.W., A.C.). V.W. is supported by the Royal Society. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.

Received 8 September; accepted 23 November 2004 Published online at http://www.nature.com/natureneuroscience/

1. Ress, D. & Heeger, D.J. Nat. Neurosci. 6, 414–420 (2003). 2. Angelucci, A. et al. J. Neurosci. 22, 8633–8646 (2002). 3. Weiskrantz, L. Blindsight—A Case Study and Implications (Oxford University Press, Oxford, UK, 1986). 4. Cowey, A. & Walsh, V. Neuroreport 11, 3269–3273 (2000). 5. Cowey, A. & Stoerig, P. Trends Neurosci. 14, 140–145 (1991). 6. Pascual-Leone, A. & Walsh, V. Science 292, 510–512 (2001). 7. Pollen, D.A. Cereb. Cortex 9, 4–19 (1999). 8. Pollen, D.A. Cereb. Cortex 13, 807–814 (2003). 9. Lamme, V.A., Super, H., Landman, R., Roelfsma, P.R. & Spekjreise, H. Vis. Res. 40, 1507–1521 (2000). 10. Hochstein S. & Ahissar M. Neuron 36,791–804 (2002). 11. Tyrell, R.A. & Owens, D.A. Behav. Res. Methods Instrum. Comput. 20, 137–141 (1988). 12. Stewart, L., Walsh, V. & Rothwell, J. Neuropsychologia 39, 415–419 (2001). 13. Hupe, J.M. et al. Neurophysiol. 85, 134–145 (2001). 14. Silvanto, J., Lavie, N. & Walsh, V. Cereb. Cortex (in the press).

144

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

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

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Large-scale changes in dendritic structure during reorganization of adult somatosensory cortex
Peter W Hickmott1 & Patricia A Steen1,2 In adult rat somatosensory cortex (S1), neurons are biased and have less dendritic arbor close to the border between the forepaw and lower jaw representations. Changes in sensory experience cause changes in the functional organization of the neocortex. Therefore, we examined the morphology of neurons in the reorganized region of S1 after forepaw denervation. We found that during reorganization dendritic arbors changed to reflect the new location of the border. Changes in cortical organization underlie a variety of important neurological phenomena, including ‘phantom’ pain and sensation after amputation1 and recovery of function after stroke2; these changes are thought to underlie improvements in performance induced by some forms of learning3. Thus, understanding the mechanisms that underlie representational plasticity is a central issue in neuroscience. The primary somatosensory cortex (S1) is organized into representations separated by borders. Changes in incoming activity patterns can induce reorganization of these cortical representations3,4. We have previously shown that denervation of the forepaw shifts the functional border between the forepaw and lower jaw representations into the area no longer receiving forepaw input. At the normal forepaw–lower jaw border we have observed a physiological bias in layer 2/3 such that responses that are evoked by layer 2/3 stimulation from within the representation are larger than those evoked from across the border5. This bias shifts during denervation to reflect the new location of the border6. We have also demonstrated an anatomical bias at the border: the dendritic trees of neurons close to the border are biased away from the border7. However, it is unclear whether the dendritic bias will change with a shift in the functional border. Changes in experience or sensory activity can affect the anatomical structure of neurons. In rodent S1, neonatal removal of a row of whiskers yields neurons that have symmetrical dendrites8. Less drastic manipulations, such as whisker trimming, can also influence developing dendrites9. In V1, monocular deprivation early in development yields cells that change their dendritic orientation toward ocular dominance columns of the open eye10. Significant reorganization of axons can also be observed11. The potential for such structural plasticity in the adult, however, is less clear. Changes in dendrite complexity have been observed in rat sensorimotor cortex in response to complex environments or maze learning12. Yet some recent in vivo studies have demonstrated

considerable stability of dendrites and dendritic spines in adult animals, although effects of deprivation were observed, particularly on spines13,14. These changes, however, are not directly linked to changes in functional organization. Here we show that a shift in the functional border between the forepaw and jaw is accompanied by a shift in the anatomical border, as evidenced by a change in dendritic bias. (Some of these data have been presented in abstract form, available online: P.W. Hickmott, program 650.3, Abstract Viewer/Itinerary Planner of the Society for Neuroscience 2002 meeting.) All procedures conformed to US National Institutes of Health guidelines and were approved by the Institutional Animal Care and Use Committee at the University of California Riverside. Forepawand lower-jaw-responsive cortex was located in vivo using carbon fiber electrodes to record multiunit responses and a glass rod to stimulate the periphery (Fig. 1a). The border was defined either as a dual response site or as the midpoint between closely spaced (<50 µm) paw and jaw responsive sites, and then marked with the carbocyanine dye DiI. A partial denervation of the forepaw was then performed by cutting and ligating the radial and median nerves in the forelimb. After 7, 14 or 28 d the reorganized border was remapped and marked with DiI; denervation caused the border between the deafferented cortex and lower jaw functional representations to shift medially (Fig. 1b). Supragranular neurons were examined in coronal slices containing marked borders by intracellular filling with biocytin (Fig. 2a and Supplementary Methods). We categorized neurons based on their position relative to the original and reorganized borders: lateral to the original border (‘l’), in between the two borders (‘d’) and medial

Figure 1 Forepaw denervation induces reorganization of S1. (a) Schematic of normal S1 in the region of the forepaw–lower jaw border with rostral to the right and lateral toward the top. HP, hindpaw; FL, forelimb; AZ, agranular zone; RV, rostral vibrissae; N, nose; FBP, frontal buccal pads. (b) The mean amount of shift in the forepaw–lower jaw border is shown for 7, 14 and 28 d of denervation. *, significantly different from 7- and 14-d values (one-way ANOVA, followed by Fisher’s protected least squares difference (PLSD), P < 0.05). Error bars, s.e.m.

1Department

of Psychology, University of California Riverside, Olmsted Hall 1344, Riverside, California 92521. 2Interdepartmental Neuroscience Program, University of California Riverside, Riverside, California 92521. Correspondence should be addressed to P.W.H. ([email protected]). Published online 16 January 2005; doi:10.1038/nn1384

140

ADVANCE ONLINE PUBLICATION NATURE NEUROSCIENCE

B R I E F C O M M U N I C AT I O N S
Figure 2 Changes in dendritic bias during cortical reorganization. (a) Examples of biocytin-filled neurons from supragranular S1, showing the original (O) and reorganized (R) borders labeled with DiI (red, dotted lines). The cortical surface is indicated by the thin line; medial is to the left, lateral to the right. Left: cell from an animal denervated for 28 d, category ‘d’. Right: cell from an animal denervated for 14 d, category, ‘l’. Scale bar, 100 µm. (b) Categories of neurons analyzed. In undenervated animals (left schematic), cells close to the border (‘n’) and >500 µm away from the border ('c') were analyzed. In denervated animals, the original border site is indicated by the open circle and the reorganized border by the filled circle. Cells close to the reorganized border ('m'), close to the original border ('l') and between the borders ('d') were analyzed. (c) Schematic illustrating the modified Scholl method for calculating overall bias. (d–f) b/nb ratios for Scholl bias (d), number of branch points (e) and total dendritic length (f) for the categories of neurons defined in (c) for 7-, 14- and 28-d denervation. Numbers in the bars represent the number of cells analyzed for each category. *, significantly different from the values for neurons close to the normal border (‘n’); **, significantly different from the values for control neurons far from the normal border (‘c’). Error bars, s.e.m. Significance values are from a one-way ANOVA followed by Fisher’s PLSD, P < 0.05.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

to the reorganized border (‘m’) (Fig. 2b). For neurons between the borders, we defined the border side as the side close to the original border site. The mean distance of the neurons from the original border was not significantly different among the three durations (7, 14 and 28 d) of denervation (Supplementary Fig. 1). We used a modified Scholl analysis to determine the overall dendritic bias of neurons with respect to the original and reorganized border sites. Concentric circles at 20 µm intervals were superimposed on the twodimensional projection of each neuron and the neuron was divided into two halves (close to the border ('b') and far from the border ('nb')) by a line perpendicular to the cortical surface (Fig. 2c). The number of intersections of labeled dendrites with each half-circle was determined for the border (b) and nonborder (nb) sides. The b/nb ratio was calculated and used as a measure of bias, referred to as the bias ratio (see Supplementary Methods). Bias ratio values from reorganized S1 were compared to those from neurons close to the border in normal animals (‘n’ in Fig. 2b, left) and to those from control neurons far from the border in normal animals (‘c’ in Fig. 2b, left). Data from non-denervated animals were collected previously7. After 7 d of denervation, only neurons lateral and close to the original border (‘l’, red) showed significant bias (Fig. 2d). Neurons

between the original and reorganized border (‘d’, gold) were biased slightly away from the original border site, and neurons medial and close to the reorganized border, distant from the original border site, were unbiased (‘m’, blue); (Fig. 2d). By 14 d of denervation, neurons that were lateral to the original border (‘l’, red) showed no significant bias, whereas those medial to the reorganized border (‘m’, blue) were biased away from the new border. Neurons between the border sites (‘d’, gold) also began to reverse their bias away from the original border (Fig. 2d). After 28 d of denervation, the changes in bias for all three categories reflected the new location of the border: neurons medial to the reorganized border (‘m’, blue) and between the borders (‘d’, gold) were biased away from the reorganized border, whereas those lateral to the original border (‘l’, red) were unbiased (Fig. 2d). Notably, for neurons between the borders (‘d’, gold), this reflects a complete reversal of bias. The b/nb ratios for the number of branch points (Fig. 2e) and for the total length of all dendrites (Fig. 2f) showed a similar pattern of changes after denervation. The changes in bias ratio were due to changes in dendritic arbors on both sides of the neurons (Supplementary Fig. 2). For neurons medial to the reorganized border (‘m’), bias was generated by an increase in complexity on the non-border side, and for neurons lateral to the original border (‘l’), bias was generated by an increase in complexity on the border side. For neurons between the borders (‘d’), we observed an initial decrease in complexity on the non-border side followed by an increase on the border side. Furthermore, we observed changes in dendrites both close to and far from the soma (Supplementary Fig. 3). We hypothesize that the major cues responsible for the structural changes are related to the changes in activity patterns induced by denervation and not due to degeneration within the cortex itself. Denervation is known to cause slow atrophy of somatosensory hindbrain and thalamus, with lesser effects on the cortex. However, there is little evidence for this denervation-induced degeneration over the sort of timescale (<28 d) used here15. Furthermore, such degeneration is unlikely to have caused the directed changes in dendrites observed here. Here we document large-scale, complex changes in dendritic structure during reorganization of adult rat S1 caused by peripheral denervation. By analyzing neurons in discrete, functionally defined areas, we demonstrate dendritic changes that reflect the new functional organization. Previous studies in the adult have demonstrated changes in fine structure (spines) or general changes in anatomical complexity during reorgani-

NATURE NEUROSCIENCE ADVANCE ONLINE PUBLICATION

141

B R I E F C O M M U N I C AT I O N S
zation. Our studies highlight the capacity of adult cortical neurons to specifically remodel their structure during functional reorganization.
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS We thank S. Burns for comments on versions of this manuscript. This work was supported by the National Institute of Neurological Disorders and Stroke (NS42241).
Flor, H. J. Rehabil. Med. 41 (suppl.), 66–72 (2003). Friel, K.M. & Nudo, R.J. Somatosens. Mot. Res. 15, 173–189 (1998). Buonomano, D.V. & Merzenich, M.M. Annu. Rev. Neurosci. 21, 149–186 (1998). Kaas, J.H. Annu. Rev. Neurosci. 14, 137–166 (1991). Hickmott, P.W. & Merzenich, M.M. J. Neurosci. 18, 4403–4416 (1998). Hickmott, P.W. & Merzenich, M.M. J. Neurophysiol. 88, 1288–1301 (2002). Hickmott, P.W. & Merzenich, M.M. J. Comp. Neurol. 409, 385–399 (1999). Steffan, H. & Van der Loos, H. Exp. Brain Res. 40, 419–431 (1980). Maravall, M., Koh, I.Y., Lindquist, W.B. & Svoboda, K. Cereb. Cortex 14, 655–664 (2004). 10. Kossel, A., Lowel, S. & Bolz, J. J. Neurosci. 15, 3913–3926 (1995). 11. Trachtenberg, J.T. & Stryker, M.P. J. Neurosci. 21, 3476–3482 (2001). 12. Black, J.E., Sirevaag, A.M., Wallace, C.S., Savin, M.H. & Greenough, W.T. Dev. Psychobiol. 22, 727–752 (1989). 13. Trachtenberg, J.T. et al. Nature 420, 788–794 (2002). 14. Grutzendler, J., Kasthuri, N. & Gan, W.B. Nature 420, 812–816 (2002). 15. Jones, E.G. Annu. Rev. Neurosci. 23, 1–37 (2000). 1. 2. 3. 4. 5. 6. 7. 8. 9.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 4 August; accepted 6 December 2004 Published online at http://www.nature.com/natureneuroscience/

142

ADVANCE ONLINE PUBLICATION NATURE NEUROSCIENCE

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

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

The voices of wrath: brain responses to angry prosody in meaningless speech
Didier Grandjean1,5, David Sander1,5, Gilles Pourtois2, Sophie Schwartz2, Mohamed L Seghier2,3, Klaus R Scherer1 & Patrik Vuilleumier2,4 We report two functional magnetic resonance imaging experiments showing enhanced responses in human middle superior temporal sulcus for angry relative to neutral prosody. This emotional enhancement was voice specific, unrelated to isolated acoustic amplitude or frequency cues in angry prosody, and distinct from any concomitant task-related attentional modulation. Attention and emotion seem to have separate effects on stimulus processing, reflecting a fundamental principle of human brain organization shared by voice and face perception. Detection of potential threats may occur even when they are not initially in the focus of attention, eliciting enhanced perceptual analysis and re-orienting of processing resources1,2. Modulation of sensory processing by affective signals has been found in the visual cortex of humans3,4 and monkeys5, but it is not known whether such effects exist in modalities other than vision. We used functional magnetic resonance imaging (fMRI) to investigate (i) whether emotional prosody enhances

neural activity in the auditory cortex, (ii) whether such enhancement involves voice-selective areas and (iii) whether any emotional effects of prosody depend on selective attention to the voice. In Experiment 1, 15 healthy right-handed adults who gave informed written consent (7 women, mean age = 24.4 ± 4.6 years) underwent fMRI scanning (see Supplementary Methods) while listening to meaningless but word-like utterances pronounced with either angry or neutral prosody. These stimuli were extracted from pseudosentences previously validated6 and matched for mean acoustic energy. To manipulate voluntary attention orthogonally to emotional prosody, we used a dichotic listening paradigm in which stimuli of 750-ms duration were presented simultaneously to either ear (angry and neutral (‘AN’), neutral and angry (‘NA’), or neutral and neutral (‘NN’), respectively, on the right and left sides) during each trial in pseudorandom order (see Supplementary Figure 1). In two successive counterbalanced blocks within the same scanning run, participants selectively attended to either the left or right ear and performed a gender decision on the voice heard on the target side (mean accuracy = 88%). Imaging data were analyzed by event-related random-effect statistics across the whole brain volume using the Statistical Parametric Mapping 2 software (see Supplementary Methods). To identify brain regions responding to emotional prosody irrespective of the target side of auditory attention, we compared activity elicited by pairs containing an angry voice (AN plus NA) relative to pairs containing only neutral voices (NN). Results showed selective increases in activity in the middle part of the right superior temporal sulcus (STS) (x, y, z = 62, –30, 6; t = 5.80; P < 0.05 corrected; see Fig. 1a) as well as in homologous areas of left STS (x, y, z = –60, –24, 0; t = 4.43; P < 0.05

Figure 1 Cortical activations elicited by spatial attention and emotional prosody. (a) Right hemisphere activations: increased responses for angry relative to neutral speech prosody were found in right STS, across all task conditions (red, P < 0.001, Experiment 1). An anterior region of right STS was also modulated by attention directed to the left relative to the right ear (green, P < 0.005, Experiment 1). These activations by emotion and attention occurred within voice-selective areas responding to speech more than to corresponding fundamental frequency or amplitude envelope cues presented in isolation (blue line, Experiment 2). (b) Mean parameter estimates of activity (percentage change relative to the global mean intensity of signal; ± s.e.m.) for right STS in Experiment 1. Blood oxygenation level– dependent (BOLD) responses were higher for angry (dark gray) versus neutral (light gray) speech, irrespective of the side of angry speech and of the side of selective listening. In addition, responses were also increased during attention to the left compared to the right ear, irrespective of prosody (N, neutral; L, left ear; R, right ear). (c) The same cluster in right STS in Experiment 2. Activation occurred only in response to vocal stimuli, not to synthetic sounds with matched fundamental frequency level or matched amplitude envelope heard in isolation.

1Geneva

Emotion Research Group, Department of Psychology, University of Geneva, bd. du Pont d’Arve 40, CH-1205 Geneva, Switzerland. 2Laboratory for Neurology and Imaging of Cognition, Departments of Neurology and Neurosciences, Centre Médical Universitaire, University of Geneva, rue Michel-Servet 1, CH-1211 Geneva 4, Switzerland. 3Department of Radiology, Geneva University Hospital, Rue Micheli-du-Crest 24, CH-1211 Geneva 14, Switzerland. 4Department of Psychology, University of Geneva, bd. du Pont d’Arve 40, CH-1205 Geneva, Switzerland. 5These authors contributed equally to this work. Correspondence should be addressed to D.G. ([email protected]). Published online 23 January 2005; doi:10.1038/nn1392

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

145

B R I E F C O M M U N I C AT I O N S
corrected, see Supplementary Figure 2). These regions correspond to the location of one of several brain areas previously reported as voice selective7,8. Such activations occurred irrespective of which ear was the target of attention (Fig. 1b), indicating that the brain could still detect emotional prosody from voices that the participant was supposed to ignore (anger versus neutrality in relevant ear: t > 4.14; anger versus neutrality in irrelevant ear: t > 4.35; P < 0.001, for bilateral STS). We found no interaction between emotion and attention in the auditory regions (even at a low threshold, P < 0.05 uncorrected). Notably, however, attention did modulate activation in the auditory cortex as shown by a general increase in activity in the right STS when participants had to judge voices from the left rather than the right ear, across all stimulus pairs (main effect of attention left side > right side; x, y, z = 60, –12, –9; t = 3.45, P = 0.002; see Fig. 1a). The peak of activation occurred in the most anterior part of the right STS region activated by anger. These results further demonstrate that participants followed selective attention instructions and differentially processed voices from the task-relevant ear. No reliable attention effect was found in left STS (right side > left side; x, y, z = –54, –24, –6; t = 2.40, P = 0.015, uncorrected, see Supplementary Figure 2). This pattern of involuntary responses to anger independent of any concomitant modulation by voluntary attention was further supported by a repeated-measure ANOVA on the parameters of activity extracted from bilateral STS using attention (right or left ear), stimulus type (NN, AN, NA), and hemisphere (right or left) as factors. Results confirmed a main effect of stimulus type (F2,13 = 21.8, P < 0.0001) but, critically, no interaction of stimulus type with attention (F < 1) or with hemisphere (F < 1). Conversely, a 2×2 ANOVA on right STS responses to AN and NA stimuli during left side versus right side attention showed a significant effect of attended side (F1,14 = 4.86, P = 0.045) but no effect of anger side (F < 1) and no interaction (F = 1.16, n.s.). Previous work has shown an activation of the middle temporal gyrus when attention is directed toward emotional meaning of vocal prosody as compared with semantic meaning9, but, to our knowledge, our results provide the first demonstration that the middle right STS is modulated by angry prosody in voices and that such modulation occurs irrespective of current attentional relevance in a selective listening task. This supports recent proposals that this brain region might subserve high-level analysis of complex acoustic information in human voices8,10. To verify that activation by anger in STS was driven by vocal prosody rather than by related low-level acoustic features, we conducted a second fMRI experiment with the same participants. As fundamental frequency and distribution of energy through time have a critical role in conveying emotional information in voices6, three categories of binaural stimuli of 750-ms duration were used in Experiment 2: angry or neutral speech, similar to the stimuli used in Experiment 1 but not heard before (AA-sp/NN-sp); synthesized sinusoid sounds matched for the mean fundamental frequency (F0) of each of the respective vocal stimuli used in Experiment 1 (AA-fo/NN-fo); and sounds consisting of white noise matched for the amplitude envelope of each stimulus used in Experiment 1 (AA-en/NN-en). Participants now had to judge whether two successive stimuli from the same category separated by a 100 ms silence were identical or different. Results from Experiment 2 showed a selective activation of bilateral STS (right: x, y, z = 68, –20, –4; 258 voxels, t = 12.87; left: x, y, z = –59, –12, 1; 199 voxels, t = 10.76; both P < 0.05 corrected) in response to speech sounds as compared with both F0- and envelopematched sounds (Fig. 1a). The STS voxels activated by angry prosody in Experiment 1 clearly overlapped with these voice-selective regions. Moreover, in Experiment 2, they also showed greater responses to angry versus neutral speech sounds (AA-sp > NN-sp, t > 3.31, P < 0.002; see also Supplementary Table 1), but no differential increases for the angry versus neutral equivalents of F0-matched sounds (AA-fo > NN-fo, t = 0.13) or envelope-matched sounds (AA-en > NN-en, t = 0.51) (Fig. 1c). Therefore, STS responses to angry speech, as observed in Experiment 1 irrespective of attended ear, were not simply driven by a particular range of frequency or a specific amplitude contour, but rather reflected more specific emotional voice-related processes. Our data demonstrate that emotional signals from angry prosody may increase activity in the associative auditory cortex, an effect that occurs even when voice prosody and location are irrelevant to the listener’s task, being independent or additive to any concomitant modulation by the spatial distribution of auditory attention in right STS. A similar enhancement by emotion additive to spatial attention was found in the face-sensitive fusiform area for fearful relative to neutral faces11. This suggests that the right STS might have a function in the auditory domain comparable to that of the right fusiform in the visual domain11, finely tuned to extract socially and affectively salient signals from conspecifics12. Future research, requiring new acoustics and synthesis technology, should address more systematically the range of critical acoustic parameters involved in processing different voice qualities and expression of different emotions in STS. Enhanced sensory responses to emotional events may constitute a fundamental neural mechanism shared by voice and face recognition systems, enabling emotion and attention interactions that prioritize orienting towards significant stimuli even when these are not in the focus of attention.
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS We thank T. Bänziger for helpful discussions. Supported by grants from the Swiss National Foundations (K.S., P.V. and S.S.). COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 5 November; accepted 28 December 2004 Published online at http://www.nature.com/natureneuroscience/

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

1. Taylor, S.F. et al. Neuroimage 8, 188–197 (1998). 2. Armony, J.L. & Dolan, R.J. Neuropsychologia 40, 817–826 (2002). 3. Pourtois, G., Grandjean, D., Sander, D. & Vuilleumier, P. Cereb. Cortex 14, 619–633 (2004). 4. Hadjikhani, N. & de Gelder, B. Curr. Biol. 13, 2201–2205 (2003). 5. Sugase, Y., Yamane, S., Ueno, S. & Kawano, K. Nature 400, 869–873 (1999). 6. Banse, R. & Scherer, K.R. J. Pers. Soc. Psychol. 70, 614–636 (1996). 7. Belin, P., Zatorre, R.J., Lafaille, P., Ahad, P. & Pike, B. Nature 403, 309–312 (2000). 8. Belin, P., Fecteau, S. & Bédard, C. Trends Cogn. Sci. 8, 129–135 (2004). 9. Mitchell, R.L.C., Elliott, R., Barry, M., Cruttenden, A. & Woodruff, P.W.R. Neuropsychologia 41, 1410–1421 (2003). 10. Kriegstein, K.V. & Giraud, A. Neuroimage 22, 948–955 (2004). 11. Vuilleumier, P., Armony, J.L., Driver, J. & Dolan, R.J. Neuron 30, 829–841 (2001). 12. Gervais, H. et al. Nat. Neurosci. 7, 801–802 (2004).

146

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

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

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Pathological gambling is linked to reduced activation of the mesolimbic reward system
Jan Reuter1, Thomas Raedler2, Michael Rose1, Iver Hand3, Jan Gläscher1 & Christian Büchel1 By analogy to drug dependence, it has been speculated that the underlying pathology in pathological gambling is a reduction in the sensitivity of the reward system. Studying pathological gamblers and controls during a guessing game using functional magnetic resonance imaging, we observed a reduction of ventral striatal and ventromedial prefrontal activation in the pathological gamblers that was negatively correlated with gambling severity, linking hypoactivation of these areas to disease severity. Gambling that interferes with interpersonal relationships and negatively affects financial and socioeconomic status has been defined as pathological gambling1, a common disorder with a lifetime prevalence of 1.6% in adults1 and important social consequences. Pathological gambling shares many features with drug addiction, such as the development of euphoria (‘high’), craving and tolerance1,2. The mesolimbic reward system is thought to play a crucial role in the development and maintenance of drug addiction2,3, and several lines of evidence converge toward the hypothesis that drug addicts have a deficient reward system and that drug intake is an attempt to compensate for this deficit4. By analogy to drug addiction, it has been speculated that pathological gambling might also be related to a deficiency of the mesolimbic dopaminergic reward system4. We addressed this question by studying 12 pathological gamblers and 12 closely matched healthy controls using functional magnetic resonance imaging (fMRI) and a guessing task known to robustly activate the ventral striatum5,6 (Fig. 1a,b and Supplementary Methods online). The study was approved by the local ethics committee, and all participants gave written informed consent before participating in this study. The behavioral data (Supplementary Data online) showed that participants successfully completed all three runs. First, we verified that our task robustly activated the ventral striatum. We observed significantly greater activity during winning than during losing in the ventral striatum in both groups (Fig. 1c,d and Supplementary Table 1; activation maps of each individual subject can be found in Supplementary Figs. 1 and 2). At the same threshold, fewer voxels were activated in pathological gamblers than in the controls, and only the controls showed additional activation in the ventromedial and ventrolateral prefrontal cortex (VMPFC) (Fig. 1c). A direct comparison of both groups (Fig. 2) showed significantly lower activation of the right ventral striatum in pathological gamblers than in controls (peak x,y,z: 33, 12,

–6 mm; t18 = 4.9, P < 0.05, corrected) (Fig. 2b and Supplementary Table 2). The opposite comparison (testing for greater activation in pathological gamblers than in controls) revealed no significant signal differences. In addition, pathological gamblers showed significantly weaker activation in the VMPFC (peak x,y,z: –3, 54, –12 mm; t18 = 4.7, P < 0.05, corrected) (Fig. 2c). We also performed a regression analysis trying to correlate signal changes in the ventral striatum with the severity of gambling in each patient. The severity of gambling in pathological gamblers (as assessed with a gambling questionnaire) showed a significant negative correlation with the response in the right ventral striatum (r = –0.77; t10 = 3.8, P < 0.05) and the response in the VMPFC (r = –0.53; t10 = 2.0, P < 0.05) (Fig. 2a,d). The patterns in areas demonstrating negative correlation are shown in Supplementary Figure 3, and peak voxels are summarized in Supplementary Table 3. To ensure that this result was not related to depression in some pathological gamblers or to differences in smoking habits, we performed two additional analyses: (i) a categorical analysis comparing non-depressed pathological gamblers to a smaller control group perfectly matched for smoking, and (ii) a correlation analysis without the depressed patients.

a
21 €

b
22 20 18 16 14 12 10 0 50 100 150 200 250

Euros

Gambling trials

c

d

Figure 1 Layout of the guessing task and main effect of winning. (a) Volunteers had to chose a playing card (either right or left) by pressing a button. If the color of this card was red, they won €1.00; otherwise they lost €1.00. (b) Predetermined course of wins and losses. (c,d) Activation of the ventral striatum in the controls (c) and pathological gamblers (d) is superimposed on an averaged T1-weighted magnetic resonance image at a threshold of P < 0.001.

1NeuroImage Nord, Department of Neurology. 2NeuroImage Nord, Department of Psychiatry and 3Behavioral Therapy Unit, Department of Psychiatry, Hamburg University Hospital Eppendorf, Martinistr. 52, 20246 Hamburg, Germany. Correspondence should be addressed to C.B. ([email protected]).

Published online 9 January 2005; doi:10.1038/nn1378

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

147

B R I E F C O M M U N I C AT I O N S a
BOLD response (a.u.)
1 0 –1 –2 –3 20 r = – 0.77, P < 0.05

25

30

35

40

45

50

Gambling severity

Figure 2 Differences in activation between the controls and the pathological gamblers. (a) Gambling severity of individual pathological gamblers. (b) Lower activation in the right ventral striatum of pathological gamblers compared with controls at P < 0.001 masked with the contrast winning > losing at P < 0.001. Activity within this area was negatively correlated with gambling severity of individual pathological gamblers (a). (c) Pathological gamblers also showed less activation in the ventromedial prefrontal cortex. Again, activation was negatively correlated with gambling severity (d). The y axes in a and d represent parameter estimates from the single subject analysis and are directly related to BOLD signal change.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

b
of reward in win trials. This is in accordance with recent functional neuroimaging data showing that the nucleus accumbens is involved in coding unexpected arousing events13. Reward-related responses in the ventral striatum can also be augmented by saliency14, so a reduced activation of the ventral striatum in pathological gamblers might be the result of a lower saliency of rewards in that group. Our guessing task is played at a fast pace and thus can best be studied with rapid event-related fMRI, which ensures a highly efficient design15. However, the efficiency for main effects such as winning or losing is extremely low in these designs. It is thus possible that a smaller difference between winning- and losing-related blood oxygenation level–dependent (BOLD) signals in the ventral striatum of pathological gamblers could stem at least partially from a higher BOLD signal during loss trials in this group. Indeed, a previous study has shown that the ventral striatum responds not only to rewarding but also to punishing stimuli in healthy controls6. In summary, a decreased activation of the ventral striatum, which is a hallmark of drug addiction2, and decreased VMPFC activation, which is related to impaired impulse control8, favor the view that pathological gambling is a non-substance-related addiction.
Note: Supplementary information is available on the Nature Neuroscience website.

y = 12 mm

y = 15 mm

y = 18mm

c

d
BOLD response (a.u.)
2 0 –2 –4 –6 20 25 30 35 40 45 50 r = – 0.53, P < 0.05

y = 54 mm y= – 54 mm

Gambling severity

x = – 3 mm

The categorical analysis revealed the same pattern as described above (Supplementary Table 4 and Supplementary Fig. 4). The additional correlation analysis confirmed our initial data and showed an even stronger relationship between gambling severity and hypoactivation of the right ventral striatum (r = –0.88; t6 = 4.5, P < 0.05) and VMPFC (r = –0.67; t6 = 2.2, P < 0.05). Drug self-administration experiments have suggested that organisms try to maintain a homeostatic baseline level of dopamine in the ventral striatum3, which in normal volunteers can be maintained by weak reinforcers found in everyday life. In contrast, in pathological gamblers (who have reduced ventral striatal activation), natural reinforcers are not strong enough for dopamine to reach and maintain this homeostatic baseline level3. At this stage, organisms seek additional, stronger reinforcers, such as drugs of addiction or gambling, to compensate for the lack of activation. Maintenance of drug addiction is linked to diminished impulse control, which is the reason that the addict is unable to quit despite adverse consequences emerging from the addiction1,2. Several lesion and neuroimaging studies have identified the pivotal role of the VMPFC in impulse control7. It is thus not surprising that drug addicts perform similarly to patients with prefrontal lesions in tasks involving decision-making8,9. Similarly to what has been observed in drug addicts, our data show a reduced activation in VMPFC in pathological gamblers. This is in accord with previous data10 and might represent the neural basis for impaired impulse control in pathological gamblers. With reference to formal models11,12, the strong activation of the ventral striatum in our paradigm might be explained by the difference of predicted reward in each trial (50%) and the unexpected delivery

ACKNOWLEDGMENTS This work was supported by grants from Bundesministerium für Bildung und Forschung, Volkswagenstiftung (C.B.), Deutsche Forschungsgemeinschaft and Studienstiftung des Deutschen Volkes (J.G.). We thank the Physics and Methods group at NeuroImage Nord (Hamburg) for help with magnetic resonance scanning and A. Heinz, D. Braus and T. Sommer for comments on an earlier draft of this paper. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 13 October; accepted 23 November 2004 Published online at http://www.nature.com/natureneuroscience/
1. Potenza, M.N., Kosten, T.R. & Rounsaville, B.J. J. Am. Med. Assoc. 286, 141–144 (2001). 2. Volkow, N.D., Fowler, J.S., Wang, G.J. & Goldstein, R.Z. Neurobiol. Learn. Mem. 78, 610–624 (2002). 3. Robbins, T.W. & Everitt, B.J. Nature 398, 567–570 (1999). 4. Blum, K., Cull, J.C., Braverman, E.R. & Comings, D.E. Am. Sci. 84, 132–145 (1996). 5. Knutson, B., Adams, C.M., Fong, G.W. & Hommer, D. J. Neurosci. 21, RC159 (2001). 6. Delgado, M.R., Nystrom, L.E., Fissell, C., Noll, D.C. & Fiez, J.A. J. Neurophysiol. 84, 3072–3077 (2000). 7. Robbins, T.W. Exp. Brain Res. 133, 130–138 (2000). 8. Rogers, R.D. et al. Neuropsychopharmacology 20, 322–339 (1999). 9. Bechara, A. J. Gambl. Stud. 19, 23–51 (2003). 10. Potenza, M.N. et al. Am. J. Psychiatry 160, 1990–1994 (2003). 11. O’Doherty, J.P., Dayan, P., Friston, K., Critchley, H. & Dolan, R.J. Neuron 38, 329–337 (2003). 12. McClure, S.M., Berns, G.S. & Montague, P.R. Neuron 38, 339–346 (2003). 13. Zink, C.F., Pagnoni, G., Martin, M.E., Dhamala, M. & Berns, G.S. J. Neurosci. 23, 8092–8097 (2003). 14. Zink, C.F., Pagnoni, C., Martin-Skurski, M.E., Chappelow, J.C. & Berns, G.S. Neuron 42, 509–517 (2004). 15. Friston, K.J., Zarahn, E., Josephs, O., Henson, R.N. & Dale, A.M. Neuroimage 10, 607–619 (1999).

148

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Ca2+ current–driven nonlinear amplification by the mammalian cochlea in vitro
Dylan K Chan & A J Hudspeth
An active process in the inner ear expends energy to enhance the sensitivity and frequency selectivity of hearing. Two mechanisms have been proposed to underlie this process in the mammalian cochlea: receptor potential–based electromotility and Ca2+-driven active hair-bundle motility. To link the phenomenology of the cochlear amplifier with these cellular mechanisms, we developed an in vitro cochlear preparation from Meriones unguiculatus that affords optical access to the sensory epithelium while mimicking its in vivo environment. Acoustic and electrical stimulation elicited microphonic potentials and electrically evoked hair-bundle movement, demonstrating intact forward and reverse mechanotransduction. The mechanical responses of hair bundles from inner hair cells revealed a characteristic resonance and a compressive nonlinearity diagnostic of the active process. Blocking transduction with amiloride abolished nonlinear amplification, whereas eliminating all but the Ca2+ component of the transduction current did not. These results suggest that the Ca2+ current drives the cochlear active process, and they support the hypothesis that active hair-bundle motility underlies cochlear amplification.

The active process that optimizes the inner ear’s response to sound is defined by four principal characteristics1–6. First, the active process expends energy to amplify sound stimuli. Second, its activity is frequency specific, thus greatly sharpening the acoustic response. Third, the active process shows a compressive nonlinearity that condenses a wide range of stimulus intensities into a narrow gamut of responses. Finally, it underlies the production of sound by the ear, a phenomenon termed spontaneous otoacoustic emission. Present in all tetrapod vertebrates, the active process accounts for the ear’s exquisite sensitivity and acuity, as well as for its prodigious range of responsiveness in terms of frequency and intensity. In the mammalian cochlea, the active process resides in the outer hair cells of the organ of Corti, which sits atop an elastic basilar membrane whose vibration is maximally sensitive to a range of characteristic frequencies ordered tonotopically along its length2. Several specializations of the cochlea are critical for the operation of the active process. The micromechanical arrangement of cells within the organ of Corti is essential for sound, in the form of pressure differences across the basilar membrane, to stimulate the hair cells. In addition, the fluid environment of the cochlea is unique: the mechanosensitive hair bundles on the apical surface of the sensory epithelium are bathed in endolymph, a K+-rich and Na+-poor solution containing only ∼25 µM Ca2+, whereas the basolateral cellular surfaces are exposed to perilymph, which is similar to extracellular fluid in composition. Disruption of this specialized environment causes significant structural changes, especially to the tectorial membrane7, and leads to cellular intoxication with inappropriate ions. Finally, there is normally a standing endocochlear potential across the epithelium such that the apical compartment is ∼80 mV positive relative to the basolateral compartment; the active process falters in the absence of this potential8,9.

In vivo laser interferometry of basilar-membrane movement, which preserves these specializations, has permitted extensive characterization of the macroscopic phenomenology of the cochlear amplifier2–6. Because disruption of the cochlear microenvironment upon acute dissection has precluded the investigation of nonlinear amplification in vitro10,11, however, the cellular basis of the active process is poorly understood. Two mechanisms have been proposed to underlie the cochlear active process in mammals. The first, somatic electromotility, is effected by prestin, a protein abundant in the plasma membrane of the outer hair cell12–14. In response to changes in the membrane potential, prestin alters its conformation and, consequently, the length of the cell itself. Prestin is required for normal hearing: knocking out the gene encoding prestin yields mice with severely attenuated responses to sound15. The second candidate mechanism, active hair-bundle motility, is based on the bundle’s capacity to exert forces that amplify mechanical stimuli16–19. In this case, force arises from the myosin motors involved in adaptation20 and from the Ca2+-dependent reclosure of transduction channels21,22, both of which are driven by the Ca2+ current entering the hair cell as a consequence of mechanoelectrical transduction. Active hair-bundle motility has been shown to produce all of the characteristics of the active process in nonmammalian tetrapods1, but its significance in mammals remains unknown. In this paper, we describe an in vitro preparation of the cochlea from Meriones unguiculatus, the clawed jird or Mongolian ‘gerbil,’ that permits observation of cochlear amplification and evaluation of the competing hypotheses concerning its origin. We find that Ca2+ flow through the mechanoelectrical transduction channel is both necessary and sufficient for nonlinear amplification, suggesting that active hairbundle motility underlies the cochlear active process.

Laboratory of Sensory Neuroscience and Howard Hughes Medical Institute, The Rockefeller University, 1230 York Avenue, New York, New York 10021, USA. Correspondence should be addressed to A.J.H. ([email protected]). Published online 9 January 2005; doi:10.1038/nn1385

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

149

ARTICLES a
AE M IHC CP OHC SE AP RE

b
TM

AE

BM AP

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

c

M

OHC

Figure 1 In vitro cochlear preparation. (a) The excised middle cochlear turn, shown in black as transected through the modiolus and outer bony wall, separates the two compartments of the experimental chamber. The apical and basal aspects of the organ of Corti (white) are immersed in artificial endolymph (AE) and artificial perilymph (AP), respectively. Pairs of recording electrodes (RE) and stimulating electrodes (SE) measure microphonic potentials and provide transepithelial electrical stimuli. Acoustic stimuli from an earphone (orange arrow) are delivered to the basilar membrane (red arrow) through the air-and-fluid-filled lower compartment. (b) A schematic drawing of the cochlear partition as mounted in the in vitro recording chamber shows the organ of Corti (white) suspended between the bone (black) of the modiolus (M) and outer cochlear wall. The hair bundles of inner (IHC) and outer (OHC) hair cells are stimulated by shearing motions between the basilar membrane (BM) and tectorial membrane (TM). Radial movement of hair bundles of the inner hair cells (horizontal red arrow) is measured with a photodiode; vertical movement (vertical red arrow) is detected with laser interferometry using a glass bead (orange) atop the tectorial membrane. (c) A micrograph taken through a dissecting microscope shows the upper surface of the preparation. Three rows of outer hair cells spiral around the cochlear modiolus. (d) A video micrograph shows the hair bundles of inner hair cells, one of which is marked (red arrow) to indicate the axis of movements detected with a dual photodiode. (e) A similar micrograph shows two of the three rows of outer hair cells and their V-shaped hair bundles. Scale bars: (c), 100 µm; (d) and (e), 10 µm.

d

e

disturb the tight mechanical and hydrodynamic coupling among the basilar membrane, organ of Corti and tectorial membrane. Instead, we used the mechanical responses of inner hair cells—whose hair bundles produce microscopic images with sharp contrast—to investigate the nature of the active process in outer hair cells. Acoustic properties Acoustic stimuli were provided through the lower compartment to the basal aspect of the sensory epithelium. To determine the acoustic properties of the system, we sealed a pressure transducer onto the recording chamber in place of the cochlear preparation and recorded the pressure in response to acoustic frequency sweeps. We detected slight fluctuations in the pressure amplitude with frequency; calibration of the earphone’s output against these irregularities eliminated the fluctuations, thus permitting the delivery to the sensory epithelium of a known pressure stimulus that was flat across the frequency range of 0.3–3.0 kHz (Fig. 2a,b). When an acoustic frequency sweep was applied, hair-bundle movements showed a marked resonance that was fit well by the Lorentzian relation characteristic of a second-order harmonic oscillator (Fig. 2c). As the resonance was traversed, the displacement response acquired a phase lag of π radians relative to the pressure stimulus. The phase of movement was constant along the exposed cochlear segment, indicating that the experimental conditions suppressed the traveling wave that traverses the basilar membrane in vivo. In a dead and thus passive preparation, the peak’s magnitude scaled linearly with stimulus intensity over a range of 60 dB (Fig. 2d). The resonant frequency was insensitive to the volume of liquid in the upper compartment but highly dependent upon that in the lower, in which 2–5 µl of perilymph yielded resonant frequencies of 400–1,100 Hz, comparable to the range of 700–1,500 Hz expected for this cochlear segment23. The frequency scaled inversely with the square root of the lower liquid mass, allowing us to calculate the stiffness of the isolated segment of basilar membrane as ∼120 N m–1. This value increased by ∼65% after formaldehyde fixation and by ∼200% after exposure to glutaraldehyde, confirming that the basilar membrane determined the system’s stiffness (Fig. 2e). Microphonic potential A basic metric for the health of a hair cell is its ability to conduct mechanoelectrical transduction. To test the integrity of this mechanism,

RESULTS In vitro preparation of the mammalian cochlea To study cochlear amplification in vitro, we developed an experimental preparation that recapitulates the in vivo properties of the cochlea while providing optical and electrical access to the hair cells themselves. By mounting the excised middle turn of the cochlea in a two-compartment recording chamber that allows the delivery of acoustic stimuli, we isolated the apical and basal aspects of the sensory epithelium and bathed them in readily exchangeable K+-rich endolymph and Na+-based perilymph solutions, respectively (Fig. 1a,b). Both visual observation and the measured transepithelial resistance of ∼2 kΩ confirmed the absence of significant leakage between the two compartments. The tectorial membrane remained in place and extended beyond the third row of outer hair cells. Because the preparation generally began to show signs of morphological and physiological deterioration 30–60 min after dissection, we performed all of our experiments within 30 min. During this time, the preparation was judged to be viable using two criteria. First, both the inner and outer hair cells appeared healthy upon microscopic inspection, without swelling, bundle malformations, or brownian motion of organelles (Fig. 1c–e). Second, as described below, the preparation showed electrical and mechanical responses to stimulation. During stimulation with sound of a particular frequency, the pressure gradient across the basilar membrane sets this elastic structure into oscillation. Basilar-membrane movement is communicated to the overlying organ of Corti, where shearing movements of the tectorial membrane deflect the hair bundles of inner hair cells, the ultimate acoustic detectors of the inner ear. The tectorial membrane also stimulates outer hair cells, which implement the active process by increasing the amplitude of basilar-membrane oscillation through the mechanism under investigation. To keep the experimental preparation as intact as possible, we did not

150

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
Figure 2 Basilar-membrane resonance. (a) A computer-generated voltage stimulus in the form of a geometric frequency sweep (top) drove an earphone that produced a constant pressure at the basilar membrane (middle) but elicited a marked resonance in the displacement of a hair bundle of an inner hair cell (bottom). (b) The frequency spectrum of the applied pressure demonstrates the linearity of the stimulus over the range of frequencies of interest. (c) The spectrum of bundle-displacement magnitude reveals a resonance peak at 850 Hz that is well fit by a Lorentzian curve (gray) with a Q0 of 4.7 and a peak sensitivity of 1.8 µm Pa–1. (d) The response amplitude of a passive hair bundle was linear over a wide range of stimulus intensities. The measured slope in this doubly logarithmic plot (black line) is 1.03 (r 2 = 1.00). (e) A plot of the inverse square of the resonant frequency (ω0) against the fluid mass in the lower compartment can be fit linearly (solid line) to obtain the stiffness of the system, in this instance 120 N m–1 (r 2 = 0.98). Fixation of the same preparation with formaldehyde increased the system’s stiffness to 200 N m–1 (dashed line; r 2 = 0.99).

a
0.5 V

0.1 Pa

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

10 nm

b
Sound pressure level (dB)

c
80 Displacement (nm) 40 60

50 ms

we measured its extracellular manifestation, the microphonic potential (Fig. 3). In response to stimuli of 60–80 dB sound pressure level (SPL) at the resonant frequency of each preparation, we recorded microphonic potentials 10–50 µV in peak-to-peak magnitude. Simultaneous recording of the microphonic potential and vertical velocity of the tectorial membrane showed that the phase of positive polarization of the lower compartment, which correlates with current flowing into hair cells, lagged the velocity towards scala media by π/2 (Fig. 3a); this accords with the conventional model of cochlear mechanics in which displacement towards scala media causes mechanoelectrical transduction channels to open10,24. Consistent with its origin in the hair cells, the microphonic potential showed a resonance at a frequency that matched that of the mechanical responses and was saturated at high stimulus intensities. The microphonic potential was reversibly abolished by filling the upper compartment with endolymph containing amiloride (Fig. 3b). A concentration-response curve revealed a half-maximal inhibitory concentration (KI) of 96 µM (Fig. 3e), indicating that the inhibitory activity of amiloride results from its blockage of the mechanoelectrical transduction channel, not of other receptors25. Application of 50 µM streptomycin, a structurally dissimilar transduction-channel blocker, also abolished the microphonic response (data not shown), as did treatment with endolymph in which K+ was replaced as the primary monovalent cation by N-methyl-D-glucamine (NMDG), which does not traverse the transduction channel26 (Fig. 3c). Application of a transepithelial potential to mimic the in vivo endocochlear potential changed the driving force on ions entering the hair cell and accordingly had a strong effect on the microphonic potential (Fig. 3d). The saturation observed at high and low transepithelial potentials may reflect the activation of voltagedependent conductances at these extreme extracellular potentials and of adaptation in the transduction machinery. The reversal of response polarity at a transepithelial potential near –50 mV indicates that the resting potentials of the hair cells were near that value, demonstrating their electrical integrity. Taken together, these results show that the hair cells could be stimulated effectively and functioned normally in vitro. Electrically evoked hair-bundle movement Each of the proposed mechanisms for the cochlear active process involves some form of reverse electromechanical transduction. To detect such a transduction process, we applied electrical stimuli across the sensory epithelium and monitored hair-bundle displacement. Application of stimuli between 5 and 40 µA in the form of 500-Hz sinusoids or 50-ms pulses elicited bundle movements up to 40 nm in peak-to-peak magnitude. These responses were abolished reversibly by amiloride with a KI of 75 µM, confirming their dependence on mechanoelectrical transduction (Fig. 4a). Small transients remained at the beginning and end

20

40

20 0 1 2 Frequency (kHz) 3

0

0

1 2 Frequency (kHz)

3

d
Displacement (nm)

e
0.15 ω0–2 (ms2) 0.10

100

10

0.05

1 20 40 60 80 Sound pressure level (dB) 0.00 0 5 10 Fluid mass (mg) 15

of the response to the electrical pulse even in the presence of amiloride. Because these transients could not be abolished by treatment with BAPTA, which disrupts tip links27, or salicylate, a blocker of electromotility28, they are likely to represent stimulation artifacts that underlie the residual signal in the sinusoidal response as well. The electrically evoked responses of hair bundles were not affected by substitution of NMDGbased endolymph, which lacks the K+ that ordinarily carries most of the transduction current (Fig. 4b)29. This result implies that the entry of Ca2+, which persists in NMDG endolymph, is sufficient to mediate the full extent of electrically evoked hair-bundle movement. Nonlinear amplification The selective amplification of low-intensity stimuli by the cochlear amplifier is manifested in the compressive nonlinearity of the basilar membrane’s response to pressure stimuli2. To seek this nonlinearity in vitro, we measured the displacement of hair bundles in response to acoustic frequency sweeps over a ∼50-dB range of stimulus intensities. In the presence of normal, K+-based endolymph and an applied transepithelial potential of +80 mV, we observed a compressive nonlinearity in the stimulus-response function (Fig. 5a), especially at low levels of stimulation. In a doubly logarithmic plot of hair-bundle displacement as a function of sound pressure, the slope over the range of 30–60 dB SPL was 0.74 (r2 = 0.94); this value differs significantly from unity (corresponding to linear responsiveness). When the transepithelial potential was absent or amiloride was applied, the response became linear, with a slope of 1.06 (r2 = 0.99) or 1.17 (r2 = 0.99), respectively (Fig. 5a). By contrast, the compressive nonlinearity persisted with a slope of 0.74 (r2 = 0.99) in the presence of NMDGbased endolymph and a +80-mV transepithelial potential (Fig. 5b). Under

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

151

ARTICLES
Microphonic potential (µV p-p) Microphonic potential (µV p-p)

a

d

40 30 20 10 0 –10 –100 –50 0 +50 +100 Transepithelial potential (mV)

5 µV 1 ms

0.1 mm/s 0.2 Pa

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

b
K+endolymph

c
K+endolymph NMDG endolymph K+endolymph wash 40 µV 20 ms 40 µV 20 ms

e
20

1.4 mM amiloride K+endolymph wash

10

0 1

10

100

1000

Amiloride concentraton (mM)

Figure 3 Microphonic potential. (a) Tectorial-membrane velocity (gray) and microphonic potential (black) were recorded simultaneously (top traces) in response to a 67-dB SPL, 800-Hz acoustic stimulus (black, bottom trace) in the presence of a +80-mV transepithelial potential. In this and all subsequent traces, positive polarization of the lower compartment, which correlates with current flowing into hair cells, and velocity towards scala media are represented by upward deflections. (b) Treatment with amiloride reversibly abolished the microphonic potential in response to sinusoidal stimulation at a resonant frequency of 700 Hz. (c) Replacement of the K+ in endolymph by channel-impermeant NMDG greatly reduced the microphonic response to stimulation at a resonant frequency of 400 Hz. (d) The peak-to-peak (p-p) magnitude of the microphonic potential was dependent on the transepithelial potential, which mimicked the in vivo endocochlear potential. (e) The concentration-response relationship for amiloride’s block of the microphonic potential is fit with a Langmuir isotherm (black line), revealing a KI value of 96 µM, consistent with the drug's role as a blocker of mechanoelectrical transduction channels.

these conditions, the sensitivity of hair-bundle displacement was greatest at the lowest stimulus intensity (Fig. 5c). These results indicate that the Ca2+ component of the transduction current, and not the dominant K+ component, is necessary and sufficient to drive the active process that underlies cochlear amplification. Because Ca2+ is required to maintain the integrity of tip links27, however, it was not possible to test the consequences of the ion’s complete removal on amplification. To compare more directly the nonlinearity observed here with that traditionally measured in vivo, we used laser interferometry to measure the vertical motion of a bead placed atop the tectorial membrane above an inner hair cell. In the presence of a +80-mV transepithelial potential and NMDG-based endolymph, a compressive nonlinearity with a slope of 0.85 (r2 = 1.00) was observed in the response to stimuli at the resonant frequency (Fig. 5d). In the absence of the transepithelial potential, the response was linear, with a power-law slope of 1.07 (r2 = 0.99). Although it was difficult to compare the absolute magnitudes of responses across experiments, the power-law dependences of bundle displacement on sound pressure were highly consistent. In the presence of K+-rich control endolymph, the mean slope in the 30–60 dB SPL range was 0.81 ± 0.12 (mean ± s.d.; n = 15). Substitution of NMDG for K+ yielded a slope of 0.74 ± 0.15 (n = 16), which is not significantly different (P > 0.14). By contrast, K+ endolymph containing amiloride yielded a slope of 1.04 ± 0.14 (n = 17), which differs significantly from each of the other two values (P < 0.001). These nonlinearities reveal an amiloride-sensitive active amplifier that operates in the presence of either K+- or NMDG-based endolymph. Adjustment of the transepithelial potential provided a rapid and reversible means of altering the effectiveness of the active process without physically perturbing the preparation. When this potential was reduced from +80 mV to 0 mV, the response immediately diminished; reapplying the positive potential restored the original response. We compared the

bundle displacement at the resonant peak in the presence of a +80-mV transepithelial potential to the displacements obtained immediately before and after in the absence of the applied potential. In the presence of K+ endolymph, this ratio was 1.33 ± 0.53 (n = 23); substitution with NMDG endolymph yielded a ratio of 1.22 ± 0.37 (n = 54), which is not significantly different (P > 0.19). In contrast, this ratio in the presence of amiloride was 0.97 ± 0.10 (n = 27), which differs significantly from the values obtained in K+ endolymph (P < 0.003) and NMDG endolymph (P < 0.00003). The ∼1.3-fold amplification observed when K+ or NMDG endolymph, but not amiloride, bathed the apical surface corroborates the results of the nonlinear slope analysis. DISCUSSION An active in vitro preparation of the mammalian cochlea The operation of cochlear hair cells is inextricably linked with their ionic, electrical and mechanical environment. Any study of these cells’ functions, especially one meant to address their role in the cochlear active process, is limited by the extent to which this in vivo environment can be mimicked. By separating the ionic environments of the epithelium’s apical and basal surfaces, maintaining a transepithelial current that reproduces the in vivo endocochlear potential, and providing acoustic stimuli that drive basilar-membrane movement with pressure gradients, we were able to observe active responses from the organ of Corti in vitro. Although we measured the radial displacement of hair bundles in this study, most previous investigations have instead tracked the vertical displacement of the basilar membrane. Modeling suggests, however, that the structure of the cochlear partition imposes an approximately unity gain between basilar-membrane movement and hair-bundle displacement30, and our preliminary laser-interferometric measurements of basilarmembrane displacement in the present preparation support this notion (data not shown). Taking this into account, the resonant frequency, Q0,
VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

152

ARTICLES a
K endolymph
+

1.4 mM amiloride K+ endolymph wash

a gradient may also explain the relatively poor frequency selectivity of the nonlinearity seen in this preparation and at the cochlear apex33; this broad tuning may actually serve to minimize the effects of the resonance mismatch35. If this gradient were physiologically relevant, our in vitro nonlinearity, like the passive mechanical characteristics described above, would fall well within the range expected for this cochlear turn by interpolation of observations made at the cochlear base and apex2. A Ca2+ current–based mechanism for cochlear amplification Application of amiloride or removal of the transepithelial potential, by blocking mechanoelectrical transduction channels or by reducing the driving force on permeant ions, respectively, severely attenuated the microphonic potential and linearized the bundle’s response to acoustic stimuli. This finding provides strong evidence that the cochlear amplification observed here rests upon the mechanoelectrical transduction apparatus rather than a transduction-independent mechanism36. On the other hand, when the apical surface of the sensory epithelium was bathed in NMDG-based endolymph and a transepithelial potential was applied, little microphonic potential could be recorded, yet amplification was still observed. This dissociation of the receptor potential (as reflected in the microphonic potential) from the active process (as revealed by the compressive nonlinearity) argues against the participation of any membrane potential–based process such as somatic electromotility at the frequencies studied here. The possibility remains, however, that electromotility plays a role in higher-frequency amplification at the cochlear base. In the presence of NMDG-based endolymph, only the small transduction current carried by the 25 µM Ca2+ and 1 mM Na+ in the apical solution entered the hair cells. This current could not generate a substantial receptor potential, and thus could not support large-scale conformational changes in prestin. It has been demonstrated, however, that Ca2+ current alone is sufficient to drive active hair-bundle motility: a solution similar to the NMDG-based endolymph used here, in which the only transduction channel–permeant ions present were small amounts of Ca2+ and Na+, nonetheless supported spontaneous hair-bundle movement and active amplification by the hair cells of the bullfrog’s sacculus16. A similar Ca2+ current–based mechanism likely underlies the nonlinearity and amplification observed in this study. Correlates of the two mechanisms for active hair-bundle motility in nonmammalian tetrapods have been observed in mammalian hair cells37–39: a slow one based on myosin motors and a fast one based on Ca2+-dependent channel reclosure. In the first of these mechanisms, the time course of Ca2+ binding to myosin motors and the motors’ movement along the actin cytoskeleton is on the order of 10–20 ms, probably too slow to account for the cycle-by-cycle amplification of stimuli up to 100 kHz seen in some mammals. The characteristic time for Ca2+dependent channel reclosure in mammalian outer hair cells, however, is in the range of hundreds of microseconds37, which approaches the speed required for high-frequency amplification. Integration of motile processes in amplification Three motile processes operate in the mammalian cochlea: Ca2+dependent channel reclosure, with a submillisecond time constant; somatic electromotility, with a characteristic time of ∼10 ms set by the membrane time constant; and myosin-based adaptation, with a time constant of ∼20 ms. What are their roles in amplification? Cycle-by-cycle force generation is most likely mediated by the fastest of these processes, Ca2+-dependent channel reclosure. This mechanism, however, has only a limited dynamic range; its ability to amplify relies on the hair bundle being poised on the edge of an oscillatory instability characterized by a Hopf bifurcation18,40,41. The role of the other two processes could be to position the bundle to maximize the activity of the fast, cycle-by-cycle amplifier41–43.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

b
K+ endolymph

NMDG endolymph

Stimulus 10 nm 20 ms
Figure 4 Electrically evoked hair-bundle movement. The application of sinusoidal (left) or pulsatile (right) current stimuli (gray) across the sensory epithelium evoked hair-bundle movements. (a) Treatment with amiloride reversibly diminished both responses, leaving only small transients at transitions in the pulse stimulus. (b) In another preparation, substitution of NMDG-based endolymph affected neither the amplitude nor the shape of the responses to electrical stimuli.

peak sensitivity, and basilar-membrane stiffness determined here are all comparable to those obtained in vivo2,5,6,20,31. It is therefore likely that the cochlear partition in this preparation responds to sound much as it does in the intact ear. Furthermore, the presence of a microphonic potential sensitive to transduction-channel blockade and transepithelial potential, as well as a mechanical response to electrical stimuli, attests to the health of the hair cells and integrity of both forward and reverse mechanoelectrical transduction. Finally, the hair bundle’s response to sound pressure showed a compressive nonlinearity and was sensitive to the transepithelial potential, instantiating the cochlear active process in this preparation. This nonlinearity was comparable both in the radial component of hair-bundle response and in the vertical component of cochlear-partition motion, suggesting linear coupling between these two motions and permitting comparison between the present nonlinearity and those studied in the past in vivo. The nonlinearity observed here differs from that described for the cochlear base2. Although the response became linear at high levels of stimulation, as it does when measured in vivo, the limited sensitivity of our recording technique precluded observation of the complementary linearization at very low stimulus levels4. Within the nonlinear range, the power-law slope of 0.7–0.8 does not reach the value of 0.3–0.4 observed at the cochlear base5,6. This difference may be related to temperature: the illumination system maintained the recording location at 31 °C, a value below the optimal operating temperature of the amplifier32. Moreover, any mismatch between the experimental system’s resonant frequency and the natural frequency of the cochlear segment might have lowered the power-law exponent. Finally, the difference may reflect a diminution in amplification towards the cochlear apex, as has been noted in studies of both basilar-membrane motion and auditory-nerve activity33,34. Such

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

153

ARTICLES a
Radial displacement (nm) 100
+ + +

b
100 Radial displacement (nm)

10

10

1 20

1 30 40 50 60 Sound pressure level (dB) 100 70

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

c
3 Sensitivity (µm Pa –1) 2 1 0.5

30 40 50 60 70 Sound pressure level (dB)

80

d
Vertical displacement (nm)

30 dB 46 dB 60 dB 66 dB

10

Figure 5 Compressive nonlinearity and amplification. The amplitude of hairbundle displacement at the resonance peak was measured over a range of stimulus levels. (a) When a +80-mV transepithelial potential was applied in the presence of K+-based endolymph (filled circles), the power-law slope of the bundle’s response (solid line) diverged from linearity (dashed gray line) at low stimulus levels. In contrast, when no transepithelial potential was applied (open circles) or when amiloride was added to the apical solution (crosses), the response became linear. (b) When the apical surface was bathed in NMDG-based endolymph, the compressive nonlinearity persisted: the response at low stimulus levels (solid line) diverged from linearity (dashed gray line). (c) The nonlinearity is demonstrated in the increasing sensitivity of the response in (b) with decreasing sound pressure level. (d) The vertical displacement of the cochlear partition in the presence of a +80-mV transepithelial potential and NMDG-based endolymph (filled circles) also yielded a power-law slope (solid line) that diverged from linearity (dashed gray line) at low stimulus levels. Turning off the transepithelial potential immediately linearized the response in the same preparation (open circles).

+

+

+

+

1

eter (Bailey Instruments). Animal procedures were performed with approval from the Institutional Animal Care and Use Committee of Rockefeller University.
30 40 50 60 Sound pressure level (dB) 70

0.5 2 1 Frequency (kHz)

3 4

Adaptation could act locally on the bundle to adjust the tension in the gating springs, whereas electromotility, with its large dynamic range, might modulate the resonant properties of the cochlear segment in which the hair cell sits. At low frequencies, the receptor potential of the outer hair cell includes an offset potential owing to the asymmetry in the transduction current44. In the absence of a cycle-by-cycle signal, this response may provide an error signal for the slower feedback processes, one that could be exploited to set the operating point of the fast amplifier. At high frequencies, however, outer hair cells do not produce such an offset potential43. In such cases, stimulation through efferent fibers of the olivocochlear bundle, which has been shown to linearize basilar-membrane responses45, could act through these slow processes to control amplification. The three motile mechanisms would thus work in concert over a wide range of temporal and spatial scales to provide the ear with a dynamic amplifier. METHODS
In vitro cochlear preparation. Cochleae were excised from 17–28-day-old jirds euthanized with 200 mg kg–1 pentobarbital (Abbott Laboratories) and were placed in dissecting solution containing 145 mM NaCl, 3 mM KCl, 250 µM CaCl2, 250 µM MgCl2, 2 mM sodium pyruvate, 5 mM D-glucose, and 10 mM Na2HPO4 at pH 7.35. Using a number 11 scalpel blade, we transected the cochlea perpendicular to its main axis between the basal and middle turns and affixed it by the cut bone surface atop a 1.5-mm hole in a plastic disk with cyanoacrylate glue (Iso-Dent, Ellman International). We created windows in the thin shelves of bone that form the floor and ceiling of the middle cochlear turn, providing a direct optical path through that turn’s sensory epithelium. To prevent leakage along the cochlear spiral, all entrances to the cochlear duct were sealed with cyanoacrylate glue. Reissner’s membrane was removed and the disk was mounted apical side up into a twocompartment recording chamber. This preparation was typically completed in less than 20 min. The basal surface was bathed in artificial perilymph consisting of dissecting solution with 1.3 mM CaCl2 and 0.9 mM MgCl2; the apical surface was immersed in artificial endolymph comprising 150 mM KCl, 25 µM CaCl2, 1 mM sodium pyruvate, 5 mM D-glucose, and 10 mM K2HPO4 at pH 7.35. Some experiments used NMDG-based endolymph containing 150 mM NMDG, 25 µM CaCl2, 1 mM sodium pyruvate, 5 mM D-glucose, and 10 mM H3PO4 at pH 7.35. In others, 1.4 mM amiloride was added to K+-based endolymph. To ensure complete washout, all solution exchanges were performed twice over 4–5 min. All experimental solutions were oxygenated and used at a room temperature of 20–24 °C. The temperature at the site of the experimental preparation, which was heated by the illumination system, was measured with a copper-constantan thermistor and BAT-8 thermom-

Acoustic stimulation. Acoustic stimuli were provided by an earphone (ER-1, Etymotic Research) coupled to a port leading to the lower compartment of the recording chamber and driven by a differential amplifier (AM 502, Tektronix) at unity gain. We determined the intensity and frequency profiles of the acoustic pressure stimulus delivered to the cochlear segment by placing a piezoresistive pressure transducer (8507C-1, Endevco) in place of the preparation and recording the sound pressure in the absence of liquid in the lower compartment. These profiles permitted calibration of the stimuli so that known sound-pressure levels with a flat frequency response from 0.3–3.0 kHz could be delivered to the sensory epithelium. There was a conduction delay of 1.3 ms between the onset of acoustic stimulation and the pressure response at the position of the specimen. All sound pressure levels are reported with reference to a root-mean-square pressure of 20 µPa. The sound pressure level (SPL) scale used in this report is a logarithmic representation of the root-mean-square sound pressure relative to this reference value; an increment of 20 dB corresponds to an increase in pressure by a factor of ten. Detection of hair-bundle motion. The displacement of hair bundles of inner hair cells was recorded using procedures analogous to those used to record responses in the bullfrog’s sacculus46. The hair bundles of inner hair cells were imaged with an upright microscope (MPS, Carl Zeiss) fitted with a 40× water-immersion objective of numerical aperture 0.8 and illuminated with a mercury lamp equipped with a heat filter. The image of the bundle was magnified 1,000× and projected onto a dual photodiode, which permitted detection of subnanometer displacements calibrated by imposing a 20-µm offset pulse to the photodiode before each recording. Direct microscopic observation of the organ of Corti confirmed that the hair bundles of inner hair cells were being deflected relative to the reticular lamina. Laser interferometry. The vertical velocity of the cochlear partition was measured using a laser interferometer (501 OFV, Polytec) coupled into the optical system described above. A 30-µm glass bead was placed on the tectorial membrane above an inner hair cell for signal acquisition. Displacement magnitudes were calculated offline from the interferometric velocity records. Determination of basilar-membrane stiffness. The passive basilar membrane behaved as a second-order resonant system whose angular resonant frequency ω0 was determined by its mass m and stiffness Κ: ω0 = K m

The mass of the liquid in the upper compartment, which was open to the atmosphere and possessed a wide meniscus, did not affect the resonant frequency and was disregarded in this analysis. We could determine the system’s stiffness by varying the mass of the liquid in the lower compartment and measuring the resonant frequency of hair-bundle motion. The mass of perilymph calculated from the

154

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
length and diameter of the liquid column extending from the sample chamber (mC) did not include the mass of the dead volume (m0) directly beneath the basilar membrane. We therefore applied the equation ω0–2 = 1 K mC + m0 K
12. Brownell, W.E., Bader, C.R., Bertrand, D. & de Ribaupierre, Y. Evoked mechanical responses of isolated cochlear outer hair cells. Science 227, 194–196 (1985). 13. Zheng, J. et al. Prestin is the motor protein of cochlear outer hair cells. Nature 405, 149–155 (2000). 14. Santos-Sacchi, J. New tunes from Corti’s organ: the outer hair cell boogie rules. Curr. Opin. Neurobiol. 13, 459–468 (2003). 15. Liberman, M.C. et al. Prestin is required for electromotility of the outer hair cell and for the cochlear amplifier. Nature 419, 300–304 (2002). 16. Martin, P. & Hudspeth, A.J. Active hair-bundle movements can amplify a hair cell’s response to oscillatory mechanical stimuli. Proc. Natl. Acad. Sci. USA 96, 14306–14311 (1999). 17. Hudspeth, A.J., Choe, Y., Mehta, A.D. & Martin, P. Putting ion channels to work: mechanoelectrical transduction, adaptation, and amplification by hair cells. Proc. Natl. Acad. Sci. USA 97, 11765–11772 (2000). 18. Fettiplace, R., Ricci, A.J. & Hackney, C.M. Clues to the cochlear amplifier from the turtle ear. Trends Neurosci. 24, 169–175 (2001). 19. Ricci, A. Active hair bundle movements and the cochlear amplifier. J. Am. Acad. Audiol. 14, 325–338 (2003). 20. Martin, P., Mehta, A.D. & Hudspeth, A.J. Negative hair-bundle stiffness betrays a mechanism for mechanical amplification by the hair cell. Proc. Natl. Acad. Sci. USA 97, 12026–12031 (2000). 21. Choe, Y., Magnasco, M.O. & Hudspeth, A.J. A model for amplification of hair-bundle motion by cyclical binding of Ca2+ to mechanoelectrical-transduction channels. Proc. Natl. Acad. Sci. USA 95, 15321–15326 (1998). 22. Howard, J. & Hudspeth, A.J. Compliance of the hair bundle associated with gating of mechanoelectrical transduction channels in the bullfrog’s saccular hair cell. Neuron 1, 189–199 (1988). 23. Müller, M. The cochlear place-frequency map of the adult and developing Mongolian gerbil. Hear. Res. 94, 148–156 (1996). 24. Hu, X., Evans, B.N. & Dallos, P. Direct visualization of organ of Corti kinematics in a hemicochlea. J. Neurophysiol. 82, 2798–2807 (1999). 25. Jørgensen, F. & Ohmori, H. Amiloride blocks the mechano-electrical transduction channel of hair cells of the chick. J. Physiol. (Lond.) 403, 577–588 (1988). 26. Lumpkin, E.A., Marquis, R.E. & Hudspeth, A.J. The selectivity of the hair cell’s mechanoelectrical-transduction channel promotes Ca2+ flux at low Ca2+ concentrations. Proc. Natl. Acad. Sci. USA 94, 10997–11002 (1997). 27. Assad, J.A., Shepherd, G.M. & Corey, D.P. Tip-link integrity and mechanical transduction in vertebrate hair cells. Neuron 7, 985–994 (1991). 28. Shehata, W.E., Brownell, W.E. & Dieler, R. Effects of salicylate on shape electromotility and membrane characteristics of isolated outer hair cells from guinea pig cochlea. Acta Otolaryngol. (Stockh.) 111, 707–718 (1991). 29. Corey, D.P. & Hudspeth, A.J. Ionic basis of the receptor potential in a vertebrate hair cell. Nature 281, 675–677 (1979). 30. Rhode, W.S. & Geisler, C.D. Model of the displacement between opposing points on the tectorial membrane and reticular lamina. J. Acoust. Soc. Am. 42, 185–190 (1967). 31. Naidu, R.C. & Mountain, D.C. Measurements of the stiffness map challenge a basic tenet of cochlear theories. Hear. Res. 124, 124–131 (1998). 32. Ohlemiller, K.K. & Siegel, J.H. The effects of moderate cooling on gross cochlear potentials in the gerbil: basal and apical differences. Hear. Res. 63, 79–89 (1992). 33. Cooper, N.P. & Rhode, W.S. Mechanical responses to two-tone distortion products in the apical and basal turns of the mammalian cochlea. J. Neurophysiol. 78, 261–270 (1997). 34. Cooper, N.P. & Yates, G.K. Nonlinear input-output functions derived from the responses of guinea-pig cochlear nerve fibres: variations with characteristic frequency. Hear. Res. 78, 221–234 (1994). 35. Eguíluz, V.M., Ospeck, M., Choe, Y., Hudspeth, A.J. & Magnasco, M.O. Essential nonlinearities in hearing. Phys. Rev. Lett. 84, 5232–5235 (2000). 36. Rybalchenko, V. & Santos-Sacchi, J. Cl– flux through a non-selective, stretch-sensitive conductance influences the outer hair cell motor of the guinea-pig. J. Physiol. (Lond.) 547, 873–891 (2003). 37. Kennedy, H.J., Evans, M.G., Crawford, A.C. & Fettiplace, R. Fast adaptation of mechanoelectrical transducer channels in mammalian cochlear hair cells. Nat. Neurosci. 6, 832–836 (2003). 38. Holt, J.R., Corey, D.P. & Eatock, R.A. Mechanoelectrical transduction and adaptation in hair cells of the mouse utricle, a low-frequency vestibular organ. J. Neurosci. 17, 8739–8748 (1997). 39. Fettiplace, R. & Ricci, A.J. Adaptation in auditory hair cells. Curr. Opin. Neurobiol. 13, 446–451 (2003). 40. Magnasco, M.O. A wave traveling over a Hopf instability shapes the cochlear tuning curve. Phys. Rev. Lett. 90, 058101 (2003). 41. Vilfan, A. & Duke, T. Two adaptation processes in auditory hair cells together can provide an active amplifier. Biophys. J. 85, 191–203 (2003). 42. Kim, D.O. Active and nonlinear cochlear biomechanics and the role of outer-hair-cell subsystem in the mammalian auditory system. Hear. Res. 22, 105–114 (1986). 43. Russell, I.J., Cody, A.R. & Richardson, G.P. The responses of inner and outer hair cells in the basal turn of the guinea pig cochlea and in the mouse cochlea grown in vitro. Hear. Res. 22, 199–216 (1986). 44. Dallos, P., Santos-Sacchi, J. & Flock, Å. Intracellular recordings from cochlear outer hair cells. Science 218, 582–584 (1982). 45. Murugasu, E. & Russell, I.J. The effect of efferent stimulation on basilar membrane displacement in the basal turn of the guinea pig cochlea. J. Neurosci. 16, 325–332 (1996). 46. Martin, P., Bozovic, D., Choe, Y. & Hudspeth, A.J. Spontaneous oscillation by hair bundles of the bullfrog’s sacculus. J. Neurosci. 23, 4533–4548 (2003). 47. Bozovic, D. & Hudspeth, A.J. Hair-bundle movements elicited by transepithelial electrical stimulation of hair cells in the sacculus of the bullfrog. Proc. Natl. Acad. Sci. USA 100, 958–963 (2003).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

in which the slope and intercept of the linear fit of ω0–2 to mC yielded the stiffness and mass of the dead volume, respectively. To confirm the origin of this measured stiffness in the basilar membrane, we measured the stiffness before and after overnight fixation at 4 °C in 4% formaldehyde in dissecting solution. Electrical stimulation and recording. Two pairs of silver–silver chloride electrodes were used to provide electrical currents and measure potentials. Each pair comprised one electrode immersed in the upper-compartment endolymph and a second embedded in agar contacting the lower-compartment perilymph. For electrical stimulation, signals were delivered across the sensory epithelium with a stimulus isolation unit47 (A395, World Precision Instruments). This unit also provided constant transepithelial currents that reproduced the in vivo +80-mV endocochlear potential. Microphonic and transepithelial potentials across the epithelium were monitored by the second pair of electrodes; the former signal was amplified 50,000× before analog-to-digital conversion. Data collection and analysis. Stimulation and recording were performed using LabVIEW 7.0 (National Instruments) on a computer (Precision 650, Dell) at digital output and sampling rates of 20 kHz. All analog input and output signals were subjected to low-pass filtering at 10 kHz with eight-pole Bessel filters. Data analysis was done using Excel 2003 (Microsoft) and Mathematica 5 (Wolfram Research). For determination of hair-bundle displacement at the resonance peak, each frequency spectrum was fit with a Lorentzian curve, from which the peak amplitude was extracted. Linear-regression fits of data points are reported with correlation coefficients; averaged values from repeated experiments are reported as means ± s.d. Statistical significance was assessed by Student's onetailed t-test; P values <0.05 were considered statistically significant.
ACKNOWLEDGMENTS The authors thank A. Hinterwirth for construction of the experimental chamber, B. Fabella for computer programming, and the members of our research group for comments on the manuscript. This work was supported by US National Institutes of Health grants DC00241 and GM07739. A.J.H. is an Investigator of Howard Hughes Medical Institute. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 17 November; accepted 13 December 2004 Published online at http://www.nature.com/natureneuroscience/
1. Manley, G.A. Evidence for an active process and a cochlear amplifier in nonmammals. J. Neurophysiol. 86, 541–549 (2001). 2. Robles, L. & Ruggero, M.A. Mechanics of the mammalian cochlea. Physiol. Rev. 81, 1305–1352 (2001). 3. Ruggero, M.A., Rich, N.C., Recio, A., Narayan, S.S. & Robles, L. Basilar-membrane responses to tones at the base of the chinchilla cochlea. J. Acoust. Soc. Am. 101, 2151–2163 (1997). 4. Sellick, P.M., Patuzzi, R.B. & Johnstone, B.M. Measurement of basilar membrane motion in the guinea pig using the Mössbauer technique. J. Acoust. Soc. Am. 72, 131–141 (1982). 5. Overstreet, E.H., Temchin, A.N. & Ruggero, M.A. Basilar membrane vibrations near the round window of the gerbil cochlea. J. Assoc. Res. Otolaryngol. 3, 351–361 (2002). 6. Ren, T. & Nuttall, A.L. Basilar membrane vibration in the basal turn of the sensitive gerbil cochlea. Hear. Res. 151, 48–60 (2001). 7. Freeman, D.M., Masaki, K., McAllister, A.R., Wei, J.L. & Weiss, T.F. Static material properties of the tectorial membrane: a summary. Hear. Res. 180, 11–27 (2003). 8. Sewell, W.F. The relation between the endocochlear potential and spontaneous activity in auditory nerve fibres of the cat. J. Physiol. (Lond.) 347, 685–696 (1984). 9. Ruggero, M.A. & Rich, N.C. Furosemide alters organ of Corti mechanics: evidence for feedback of outer hair cells upon the basilar membrane. J. Neurosci. 11, 1057–1067 (1991). 10. He, D.Z., Jia, S. & Dallos, P. Mechanoelectrical transduction of adult outer hair cells studied in a gerbil hemicochlea. Nature 429, 766–770 (2004). 11. Ulfendahl, M. & Flock, Å. In vitro studies of cochlear excitation. Curr. Opin. Neurobiol. 8, 475–479 (1998).

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

155

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Comm function in commissural axon guidance: cell-autonomous sorting of Robo in vivo
Krystyna Keleman, Carlos Ribeiro & Barry J Dickson
Commissureless (Comm) controls axon guidance across the Drosophila melanogaster midline by regulating surface levels of Robo, the receptor for the midline repellent Slit. Two different models have been proposed for how Comm regulates Robo: a ‘sorting’ model and a ‘clearance’ model, both based on studies using heterologous cells in vitro. Here, we test these two models in vivo. We establish a genetic rescue assay for Comm, and use this assay to show that midline crossing does not require the presence of Comm in midline cells, as proposed by the clearance model. Moreover, by monitoring the trafficking of a Robo–green fluorescent protein (GFP) fusion in living embryos, we demonstrate that Comm prevents the delivery of Robo-GFP to the growth cone, as predicted by the sorting model. It has also been suggested that Comm must be ubiquitinated by the Nedd4 ubiquitin ligase. We show here, however, that ubiquitination of Comm is not required for its function in vitro or in vivo, and that Nedd4 is unlikely to function in axon guidance at the midline.

The choice of axons to cross or not to cross the midline of the central nervous system is arguably the best understood of all axon pathfinding decisions1. Extensive studies of axon pathfinding at the midline of the D. melanogaster ventral nerve cord and of the vertebrate spinal cord have shown that the choice of a crossing (commissural) or noncrossing (longitudinal) pathway depends on the differential sensitivity of axons to the midline repellent Slit2,3. Axons that cross are relatively insensitive to Slit and respond instead to attractants from the midline. After crossing, these commissural axons often turn into longitudinal pathways, extending alongside the midline on the opposite (contralateral) side. These postcrossing commissural axons, like ipsilateral axons, are repelled by Slit and thus do not cross (or recross) the midline. In D. melanogaster, commissureless (comm) controls the choice between a commissural and an ipsilateral pathway. It is expressed in commissural, but not ipsilateral, neurons4,5 and is both necessary4,6,7 and sufficient8,9 for axon growth across the midline. Commissureless (Comm) is a short transmembrane protein10 that regulates surface levels of Roundabout (Robo), the receptor for Slit4,11. It is generally accepted that Comm controls Robo surface levels, but there is considerable controversy over how it does so. Two very different models for Comm function have been put forth (see Supplementary Fig. 1 online): a ‘sorting’ model4 and a ‘clearance’ model5,11,12. The sorting model proposes that Comm controls the sorting of Robo at the trans-Golgi network4. In ipsilateral neurons and postcrossing commissural neurons, which do not express active Comm, Robo is sorted into vesicles destined for the growth cone. In contrast, in commissural neurons, which express Comm, Robo is instead sorted to late endosomes and lysosomes so that little, if any, Robo reaches the growth cone. In this model, Comm acts strictly autonomously

to regulate Robo levels in the growth cone and hence the choice of a commissural versus a longitudinal pathway. The clearance (or internalization) model proposes instead that Comm controls Robo levels by acting at the plasma membrane. In this model, Comm does not block the delivery of Robo to the growth cone but instead rapidly removes it by endocytosis11. A homophilic interaction between Comm protein expressed on axons and Comm protein expressed on midline cells is proposed to restrict Comm to the midline segments of commissural axons, ensuring that Robo is selectively depleted from this segment of the axon5,12. These two models thus offer conflicting views for the regulation of Robo by Comm. This controversy is fueled mainly by two key issues. First, mechanistic studies have relied largely on heterologous cell culture systems: either mammalian COS-7 cells4 or D. melanogaster S2 cells11. It is not clear how readily data from these in vitro systems can be extrapolated to D. melanogaster commissural axons in vivo. The only indication that they have any relevance at all is that a specific cytoplasmic sequence motif in Comm is required both for midline crossing in vivo and for Robo regulation in vitro4,11. Second, comm is expressed not only in commissural neurons but also in midline cells. This suggests an additional nonautonomous role for Comm10 that cannot be recapitulated using these in vitro systems. A role for Comm in midline cells is incorporated into the clearance model but not into the sorting model. Resolving these issues is paramount to obtaining a deeper understanding of how axons choose between longitudinal and commissural pathways in the D. melanogaster nerve cord. This was our goal in the present study. Specifically, we aimed to investigate in detail the mechanism of Comm function in vivo using as our guide these two

Institute of Molecular Biotechnology of the Austrian Academy of Sciences (IMBA), Dr. Bohr-Gasse 3-5, A-1030 Vienna, Austria. Correspondence should be addressed to B.J.D. ([email protected]). Published online 16 January 2005; doi:10.1038/nn1388

156

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
poxn neurons all cross in wild-type embryos (Fig. 1a,e) but never in the commO72 null background (Fig. 1b,e). We then added a GAL4-responsive UAS-comm transgene to specifically restore comm function in the poxn neurons. This rescued the midline crossing of poxn axons in 67% of segments (n = 200; Fig. 1c,e). This result is consistent with our earlier transplantation studies4 (40% of comm+ clones in a comm– background included contralateral projections). Note that in both cases the percentage of individual axons crossing is somewhat lower. It is not possible to resolve individual axons in these experiments, so we scored as positive those segments or clones in which at least one axon crosses. Nevertheless, in both cases a significant number of commissural axons extend across the midline, despite the comFigure 1 Comm in midline cells does not contribute to midline crossing. (a–d) Stage 16 embryos plete absence of comm function from midline carrying poxn-GAL4 and UAS-τlacZ transgenes in various genetic backgrounds stained with antibodies cells. It is also important to note that neither to β-galactosidase to visualize projections of poxn neurons (white) and antibodies to HRP to visualize the poxn neurons nor those examined in the O72 O72 mutant (b), comm , UAS-comm (c), the axon scaffold (blue). Wild-type background (a), comm transplantation studies normally pioneer the and commO72, UAS-comm, slit-comm (d) are shown. (e) Percentage of segments in which one or commissural pathways. We find it remarkamore poxn axons cross the midline. n, Number of segments examined; + and –, presence or absence, ble that they can, however, do so in situations respectively, of the relevant transgene or endogenous (endog.) comm function. Each transgene was present as a single copy, and embryos were scored blind to their genotype. where no other axons cross. That some poxn axons cross in these experiments demonstrates the autonomous conflicting models derived from the in vitro studies. Our data support requirement for comm; that not all cross shows the nonautonomous the sorting model: we provide direct evidence that Comm regulates requirement. With this genetic assay established, we were now in a delivery of Robo to the growth cone in vivo and that midline cros- position to investigate the origin of this nonautonomous requirement. sing does not require either Comm in midline cells or a homophilic One possibility4 is that this reflects a ‘follower’ or ‘community’ effect in Comm-Comm interaction. axon pathfinding: namely, that the outcome of a particular guidance In addition, it has also been argued that Comm function in midline decision may depend on the genotype of not only the neuron in quescrossing requires its ubiquitination and that Nedd4 is the responsible tion but also the neurons with which it fasciculates and on which it ubiquitin ligase11. Although initially proposed in the context of the may rely for navigation. In this case, the nonautonomous requirement clearance model, these data are also consistent with the sorting model. for comm would simply reflect the autonomous role of comm in other However, in a set of genetic experiments designed to reproduce and commissural neurons. However, an alternative explanation is that this extend these findings11, we have found that ubiquitination cannot be nonautonomous requirement reflects a contribution from comm noran essential part of Comm function in vivo and have not found any mally provided by midline cells5,10. evidence in support of the proposed role for Nedd4. To distinguish these possibilities, we investigated whether restoring comm expression at the midline as well as in the poxn neurons would RESULTS substantially improve the degree of rescue. We prepared transgenes in which comm was directly expressed under the control of either of Comm in midline cells does not contribute to midline crossing In the initial characterization of the comm gene10, it was reported to two midline promoters (from the genes slit and single-minded (sim), be strongly expressed in midline cells but not in neurons. From this respectively14,15, and introduced them into the poxn partial rescue it was concluded that Comm must act exclusively in midline cells to background. Neither the slit-comm nor the sim-comm transgene regulate the growth of commissural axons across the midline. This view enhanced the frequency of midline crossing by poxn axons (Fig. 1d,e). was revised with the demonstration that comm is indeed expressed in Thus, the nonautonomous requirement for comm in the crossing of neurons and that midline crossing is not rescued in a comm mutant poxn axons cannot be explained by a requirement for comm in midline if comm expression is restored only at the midline4,5. These findings cells. We confirmed expression of Comm at the midline from each of suggested that comm has an essential role in neurons but left open the these transgenes (data not shown). Also, the comm cDNA fragment question of whether comm is required only in neurons, as predicted by used in these transgenes is exactly the same as the fragment used in the sorting model, or is required in both neurons and midline cells, as our active UAS-comm transgenes, and the same slit and sim promoters predicted by the clearance model. Initial attempts to resolve this critical have been successfully used in rescue assays for netrin and slit—genes issue came to different conclusions4,5. This confusion can be attributed that do act in midline cells during commissural axon guidance2,16 (M. in part to the lack of a phenotypic rescue assay for comm that would Brankatschk and B.J.D., unpublished data). We therefore conclude that provide a more definitive answer as to where comm function is (and is comm in midline cells does not contribute in any way to the regulation not) needed for midline crossing. of midline crossing, even in the sensitized genetic background of the To establish such a rescue assay, we focused on a set of commissural poxn partial rescue assay. This is in agreement with the sorting model neurons labeled by the poxn-GAL4 driver13 (poxn-GAL4-14). These but not the clearance model.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

157

ARTICLES
Minimal lumenal and transmembrane sequence requirements When we tested our lumenal and transmembrane domain mutants in Based on a small set of amino-terminal deletions and chimeric proteins, these two in vivo assays, we obtained results consistent with the COS cell it has been proposed12 that Comm function requires specific sequences data (Fig. 2). Almost all mutant Comm proteins were fully functional in both its extracellular and its transmembrane domains. In particular, in both in vivo assays, again with the sole exception of Comm108–131: Comm was found to interact homophilically when expressed in S2 cells, CD8. Notably, our transmembrane substitution, Comm132–160:CD8, requiring a region between amino acids 62 and 98. Because deletion is also as functional in vivo as it is in vitro, arguing against the notion mutants that fail to interact in S2 cells are also nonfunctional in vivo, that specific sequences in the transmembrane domain are essential for homodimerization was proposed to be essential for Comm function Comm localization and function12. in midline crossing12. Taking all Comm mutants together (those we have generated here as We re-examined these findings with a more extensive series of well as those from our initial characterization of Comm4) we have now deletions in the lumenal domain, including nested deletions from deleted or substituted amino acids 2–108, 132–160, 179–219 and 245–370, both the N terminus and the membrane-proximal region (Fig. 2). In all without impairing Comm localization or Robo relocalization in vitro, assessing sequence requirements in or near the transmembrane domain, or Comm function in vivo. Only mutations in the endosomal sorting sigwe were concerned that large deletions might simply disrupt the inser- nal (located between 220 and 244) or the lumenal juxtamembrane region tion or stability of Comm in the membrane. We therefore generated two (108–131) completely disrupt Comm localization, Robo relocalization small substitutions, replacing either the membrane-spanning region and Comm function. These data are consistent with the sorting model, (amino acids 132–160) or the membrane-proximal region (amino acids which predicts that the only essential function of Comm in midline cross108–131) by the corresponding regions of CD8. ing is to couple Robo to the endosomal sorting machinery. In particular, We first examined the localization of each of these proteins in tran- our data do not support the notion that Comm must homodimerize sient transfection experiments carried out in COS cells, both with and to function in vivo12. Specifically, although a ∆2–98 deletion mutant without cotransfection with Robo. Almost all of the mutant Comm (Comm∆2) was previously found to be unable to homodimerize12, proteins had the normal vesicular distribution, suggesting that they our identical mutant (Comm∆2–98) localized correctly in COS cells, were correctly targeted to late endosomes and lysosomes (Fig. 2). In each case where Comm was targeted correctly, Robo, if coexpressed, colocalized with Comm. The only exception was the mutant in which the membrane-proximal region was replaced by the corresponding region of CD8 (Comm108–131:CD8). This mutant had a diffuse localization throughout the cell (possibly in the endoplasmic reticulum). Robo, if coexpressed, was neither recruited to endosomes nor colocalized with Comm108–131:CD8, but rather showed a mixture of diffuse and plasma membrane staining. We infer that amino acids in this region are essential for the correct localization of Comm. By a process of elimination, we also suspect that this region is important for the ability of Comm to recruit Robo to endosomes. However, because Comm108–131:CD8 itself is mislocalized, it was not possible to determine if it could still relocalize Robo. We prepared UAS-comm transgenes for each of these mutants to assay their function in midline crossing in vivo. Previous structure-function studies of Comm have used a rather crude pan-neuronal gain-of-function assay4,11,12. The poxn rescue assay we have established (Fig. 1), however, provides a much more reliable and quantifiable assay for Comm function in vivo. In addition, we Figure 2 Structural requirements in the Comm lumenal and transmembrane domains. (a) Expression also used a single-cell gain-of-function assay9, of wild-type and mutant Comm proteins in COS-7 cells (top row), dorsal Ap neurons (middle row) using Ap-GAL4 to drive expression in the three and poxn neurons (bottom row). COS-7 cells were cotransfected and stained for Robo (antibody to Ap neurons per hemisegment and focusing HA, green) and Comm (antibody to myc, red). Stage 16 embryos carrying UAS-τlacZ, UAS-comm and our attention on the single, isolated, dorsal either Ap-GAL4 in a wild-type background or poxn-GAL4 in a commO72 background were stained with Ap neuron. The Ap neurons are all ipsilateral antibodies to β-galactosidase (white) and anti-HRP (blue). (b) Structure, localization and function of each of the mutant Comm proteins. Localization of Comm and Robo in COS-7 cells was scored as neurons and do not express comm. Yet when E (endosomal), PM (plasma membrane) or d (diffuse staining in cytoplasm, possibly endoplasmic forced to express a functional comm transgene, reticulum). Crossing of Ap and poxn axons was scored as the percentage of segments in which at least the Ap neurons almost always extend their one axon crosses (n ≥ 80 for each genotype). Black and gray bars represent independent transgene axons across the midline9 (Fig. 2). insertions, with the same insertion indicated by black bars for both Ap and poxn assays.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

158

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
Figure 3 Comm blocks Robo transport along the axon. Maximum intensity projections of living embryos expressing UAS-robo-GFP (green) and UAS-mRFP1-moe (red, highlighting the actin cytoskeleton) driven by poxn-GAL4, either without (a–d) or with (e–h) UAS-comm. Images in a, b, e and f have been deconvolved. (b and f) The green channel only for a and e, respectively. (c and g) Successive time points from two movies, taken at intervals of 2.2 s and 8 s, respectively (Supplementary Videos 1 and 4 online, respectively). These images have been false-colored to highlight Robo-GFP vesicles in the axons (arrows). (d and h) Kymographs of the two movies represented in c and g, respectively, as inverted gray-scale images, with pink shading highlighting the intervals shown in c and g. In all panels, the soma are to the left and the growth cones to the right. Embryos are mid stage 15 (a,b,e,f), late stage 15 (g) or stage 16 (c). Note that single vesicles sometimes appear as doublets due to their movement between successive scans at a single time point. Scale bars, 10 µm.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

relocalized Robo and was fully functional in both the rescue assay and the single-cell gain-of-function assay. Comm prevents Robo transport in axons These genetic studies have provided evidence in support of the sorting model and against the clearance model. The most stringent test of these two models, however, would be the direct visualization of Robo trafficking in the presence and absence of Comm. This has not been addressed, even in the in vitro studies, which have assessed Robo trafficking only indirectly and have failed to reach a consensus. Experiments in S2 cells have only examined the steady-state distributions of these two proteins11, from which internalization from the plasma membrane was inferred but not demonstrated. In contrast, surface-labeling experiments carried out using COS-7 cells4 suggested that Comm diverts Robo to endosomes even before it reaches the cell surface. However, as already noted, the trafficking of these two proteins may well be quite different in COS-7 cells in vitro and in D. melanogaster neurons in vivo. We therefore set out to observe directly the trafficking of Robo in vivo. The sorting model predicts that little if any Robo should be transported down axons in the presence of Comm. The clearance model predicts that Robo should still be delivered normally to the axon and growth cone. To test these predictions, we generated a UAS-robo-GFP transgene encoding a full-length Robo protein fused at its C terminus to GFP. In both fixed

and living embryos, we saw clear expression of Robo-GFP with the poxnGAL4 driver in the poxn neurons in the central nervous system (CNS) and in neurons of the polyinnervated sensory organs in the peripheral nervous system. In the CNS of fixed embryos, Robo-GFP recapitulated the distribution of endogenous Robo: high in the soma and post-crossing segments of the poxn commissural axons, but absent from the midline segments (data not shown). However, to define the mechanism by which Comm regulates Robo, we focused on neurons in the peripheral nervous system rather than in the CNS. Paradoxically, the peripheral nervous system is better suited to this than the CNS. In the CNS, adding or removing Comm alters an axon’s trajectory, so we would not know whether any change we might see in Robo-GFP trafficking is a direct consequence of Comm expression or an indirect consequence of the altered path. Simply comparing Robo-GFP trafficking in commissural axons before and after crossing is also not a solution, as we would not know whether any differences seen were because of changes in Comm activity or some other unrelated change that occurs upon midline crossing. Fortunately, the peripheral nervous system neurons that express poxn-GAL4 do not express comm, and ectopic expression of comm does not alter their axon projections toward the CNS. Their long axons also lie just below the epidermal surface, making them ideal for these imaging studies. Without coexpression of Comm (poxn-GAL4, UAS-robo-GFP; Fig. 3a–d), Robo-GFP in poxn sensory neurons localizes to the plasma

Figure 4 A Comm that cannot be ubiquitinated is still functional in vitro and in vivo. (a) Expression of wild-type and lysine-free Comm (CommKR) in COS-7 cells (top row), dorsal Ap neurons (middle row) and poxn neurons (bottom row). (b) Structure, localization and function of CommKR. Black, gray and white bars represent independent transgene insertions (n ≥ 80 for each genotype).

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

159

ARTICLES
membrane of the soma and axon but is also seen in vesicles in both soma and axon. Vesicles in the soma positive for Robo-GFP appear brighter and larger than those in the axon (Fig. 3a,b). To facilitate the dynamic analysis of Robo-GFP vesicles in the axon, we bleached a portion of the axon to eliminate the signal from the Robo-GFP associated with the plasma membrane. This allowed us to detect movement of RoboGFP vesicles along the axon. We recorded 16 movies from poxn-GAL4, UAS-robo-GFP embryos. A series of time points from one such movie is shown (Fig. 3c), and this and two other movies are available online (Supplementary Videos 1–3). Figure 3d shows a kymograph of the movie of Figure 3c in which the y and z axes have been collapsed and the data displayed on an x-t plot. Anterograde vesicles appear in the kymograph as a series of dots moving to the right (for example, arrows labeled ‘antero’), retrograde vesicles as dots moving to the left (for example, arrows labeled ‘retro’) and stationary vesicles as a vertical series of dots. Most of the Robo-GFP vesicles in the axon moved rapidly toward the growth cone, often stopping and restarting in a manner reminiscent of fast anterograde transport by kinesin motors17,18. Excluding their brief stationary periods, these vesicles moved with an average speed of 1.39 ± 0.60 µm s–1 (mean ± s.d., n = 21) and a maximum of 2.75 µm s–1, speeds that are consistent with kinesin-mediated anterograde transport. Only rarely did we detect Robo-GFP vesicles moving in the retrograde direction. These retrograde vesicles were slower than those moving in anterograde direction, with an average speed of 0.70 ± 0.30 µm s–1 (n = 6) and a maximum of 1.06 µm s–1. The distribution and dynamics of Robo-GFP are very different in embryos in which Comm is coexpressed (poxn-GAL4, UAS-robo-GFP, UAS-comm; Fig. 3e–h). Overall, the distribution of Robo-GFP in poxn sensory neurons changed in much the same way as seen in COS cells: Robo-GFP was largely depleted from the plasma membrane of both soma and axon and concentrated in intracellular vesicles in the soma (Fig. 3e,f). Vesicles positive for Robo-GFP were still observed in the axon but at a much lower frequency than in embryos without the UAScomm transgene. Where they were observed, these axonal GFP vesicles were also larger and brighter than those observed in the absence of Comm. We recorded six movies to examine the movement of RoboGFP vesicles in the axon (Fig. 3g,h and Supplementary Videos 4 and 5 online). In contrast to the rapid movement of Robo-GFP vesicles observed in the absence of Comm, those in the presence of Comm are essentially stationary; they do not behave as transport vesicles. We suspect they may be endosomes, which have previously been observed in the axons of cultured vertebrate neurons19 and D. melanogaster neurons in vivo (K. Wucherpfennig and M. Gonzalez-Gaitan, personal communication). This is also not a general block in axon transport, as the membrane markers CD8-GFP (data not shown) and mRFP-moe (Fig. 3e) are still transported normally, and these axons still extend normally toward the CNS. These data now establish that Comm regulates the intracellular trafficking of the Robo receptor. Without Comm, Robo is delivered by fast axonal transport toward the growth cone; with Comm, the anterograde trafficking of Robo ceases almost entirely. We consider this a direct and compelling in vivo confirmation of the sorting hypothesis. Comm function does not require its ubiquitination It has been proposed11 that Comm function in vivo requires its ubiquitination by the Nedd4 ubiquitin ligase. Although initially put forth in the context of the clearance model, ubiquitination by Nedd4 would also be consistent with the sorting model. Indeed, the yeast ortholog of Nedd4, Rsp5p, seems to have an analogous role in sorting the Gap1p permease from the Golgi to the vacuole, the yeast equivalent of the lysosome20,21. However, we did not detect ubiquitination of Comm in our in vitro assays (data not shown), and the published data were also not entirely

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 5 No evidence for Nedd4 function in midline crossing. (a–c) Stage 16 embryos carrying elav-GAL4, UAS-comm and either no Nedd4 transgene (a), UAS-Nedd4 (b), or UASNedd4CA (c), stained with monoclonal antibody 1D4 to FasII. All embryos show identical phenotypes (compare to Fig. 6b,f,g of Myat et al.11). (d–f) Stage 16 embryos of genotype slit 2 robo 1 / + +; elav-GAL4 / + (d), slit 2 robo 1 / + +; elav-GAL4 / UAS-Nedd4 (e), and slit 2 robo 1 / + +; elav-GAL4 / Df(3L)ED4688 (f), stained with antibody to FasII. In each of these images, FasII-positive axons cross the midline in two to three of the four segments shown. (g) Organization of the Nedd4 genomic region, showing P element insertions and the extent of the Df(3L)ED4688 deletion. For the EY00500 insertion, the arrow indicates the direction of GAL4-driven transcription. (h) Quantification of midline crossing defects, indicating the number of segments per embryo in which one or more FasII-positive longitudinal axon bundles aberrantly cross the midline. n, number of embryos examined. Error bars indicate s.e.m. ***P < 0.0001, Mann-Whitney test. P > 0.1 for all other genotypes compared with the corresponding 1407-GAL4 or elav-GAL4 control. All embryos were scored blind to their genotype. (i–l) Stage 16 wild-type (i,k) and Df(3L)ED4688 (j,l) embryos stained with monoclonal antibody 1D4 to FasII (i,j) or BP102 (k,l). Df(3L)ED4688 embryos are indistinguishable from wild type.

160

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
consistent with the sorting model. For example, Comm was shown to be polyubiquitinated in 293T cells, apparently targeting it for proteasomal rather than lysosomal degradation11. Although ubiquitination can target proteins to endosomes, this usually involves monoubiquitination rather than polyubiquitination22. In addition, ubiquitination of Comm in 293T cells required cotransfection of Nedd4, yet Comm is sorted correctly to endosomes in both mammalian COS-7 cells4 and D. melanogaster S2 cells11 without cotransfection of Nedd4. In light of these somewhat conflicting data from the in vitro studies, we decided to test whether ubiquitination is indeed necessary for Comm function in vivo. Ubiquitination is only possible on lysine residues or, rarely, on a protein’s N terminus22. Because Comm’s N terminus is lumenal, the only available sites for ubiquitination are the ten lysine residues in the cytoplasmic domain. A truncation that removes the last 126 amino acids of Comm, including six of the ten lysine residues, is still fully functional in vivo4. In the context of this deletion, we mutated each of the four remaining lysines to arginine and replaced the C-terminal c-myc epitope tags (which contain lysine residues) with hemagglutinin (HA) epitope tags (which do not). The resulting lysine-free Comm mutant, CommKR, was then tested in the COS-7 cell sorting assay in vitro and in both the Ap gain-of-function assay and the poxn rescue assay for midline crossing in vivo. In all of these tests, CommKR was as active as wild-type Comm (Fig. 4). We conclude that Comm does not need to be ubiquitinated. No evidence for Nedd4 function in midline crossing The full activity of the lysine-free CommKR mutant is a strong argument against the notion that Comm must be ubiquitinated to regulate midline crossing11. It does not, however, rule out a role for Nedd4, which may have some other substrate in midline crossing. Nevertheless, with the CommKR mutant casting some doubt on the role of ubiquitination, we thought that the genetic evidence for Nedd4 function in midline crossing needed bolstering before initiating any further mechanistic studies. With this objective in mind, we first attempted to reproduce and extend the genetic data11 that had initially linked Nedd4 to Comm function. One set of experiments first used11 to test for Nedd4 function in midline crossing was to determine whether coexpression of wild-type Nedd4 or a catalytically inactive Nedd4 would modify the pan-neuronal comm gain-of-function phenotype, in which FasII-positive longitudinal axons aberrantly cross the midline8. In this initial report11, wild-type Nedd4 (UAS-Nedd4) enhanced the frequency of aberrant midline crossing in elav-GAL4, UAS-comm embryos, whereas the catalytically inactive form of Nedd4 (UAS-Nedd4CA) suppressed this phenotype. Using the same transgenic lines and scoring embryos blind to their genotype, we could not reproduce these findings (Fig. 5a–c). Quantification of these phenotypes is problematic, because elav-GAL4, UAS-comm embryos have such a strong phenotype, with aberrant crossing of FasII-positive axons in almost every segment2,4,8,11. A second and more reliable assay used in the initial study of Nedd411 is the slit robo transheterozygous assay. In slit robo / + + embryos, FasII-positive axons cross the midline in three to five segments per embryo2 (Fig. 5d,h). This phenotype is highly sensitive to the dosage of several other genes, which can either increase or decrease the frequency of crossing23–26. Although this is often misinterpreted as proof that these genes act in the same pathway as Slit and Robo, it can more correctly be taken as an indication that these genes may act in the same process. In this regard, it is reasonable to expect that, if Nedd4 is required for Comm to regulate Robo trafficking, it should also test positive in this assay. Indeed, it was reported that pan-neuronal expression of Nedd4 (in elav-GAL4, UAS-Nedd4 embryos), which does not itself induce any aberrant crossing, nevertheless enhances the frequency of ectopic crossing in the slit robo / + + background11. We repeated these studies, again using the same transgenic lines as in the initial study11 and scoring embryos blind to their genotype. As a further control, we also tested overexpression of Nedd4 using another pan-neuronal driver, 1407-GAL4, as well as the same elav-GAL4 driver. In addition, we extended the predicted range of Nedd4 levels in both directions, using an EP insertion in the endogenous Nedd4 locus (EY00500) to obtain even higher Nedd4 levels (owing to the presence of 14 GAL4 binding sites, compared with the five sites in UAS-Nedd4), and a chromosomal deficiency (Df(3L)ED4688) to reduce Nedd4 levels (Fig. 5g). Despite testing what should be a large range in Nedd4 levels and trying two different pan-neuronal drivers, we did not detect any genetic interaction with Nedd4 (Fig. 5d–f,h). As a positive control, we included UAS-ena in these experiments, scoring these embryos blind together with the Nedd4 set. Ena is thought to act positively in Robo signal transduction23. Consistent with this, we found that overexpression of Ena suppresses the slit robo / + + phenotype (Fig. 5h). We also examined embryos homozygous for the deficiency Df(3L)ED4688. We generated this deficiency using the DROSDEL kit for the production of isogenic, molecularly defined chromosomal deletions27. Df(3L)ED4688 deletes 79,079 base pairs (bp), including the entire Nedd4 open reading frame and five other genes predicted in Release 3.2.1 of the Berkeley D. melanogaster genome annotation (Fig. 5g). The CNS axon scaffold seems completely normal in Df(3L)ED4688 homozygous embryos; in particular, we saw no defect at all in commissure formation (Fig. 5i–l). Thus, none of the genes in this region, including Nedd4, is required zygotically for midline crossing. In summary, zygotic Nedd4 null mutants have no commissural defect (Fig. 5i–l), the pan-neuronal comm gain-of-function phenotype is sensitive to neither misexpression of wild-type nor catalytically inactive Nedd4 (Fig. 5a–c), and the slit robo partial loss-of-function phenotype is not modified across a large range of Nedd4 expression levels (Figs. 5d–f,h). Thus, the key genetic findings from the initial study of Nedd411 were not reproducible and did not withstand more extensive testing. We also note that an interaction between Nedd4 and Comm was only observed in vitro or when both proteins were coexpressed in yeast or mammalian systems11. An interaction between the endogenous proteins has not yet been reported. These biochemical data might be explained simply by the promiscuous interactions of overexpressed proteins due to the high affinity of WW domains, such as those in Nedd4, for PPXY motifs, as occur in Comm28. Thus, while our data do not formally exclude a role for Nedd4 in midline crossing, we see no reason to reject the null hypothesis that it has no such role. DISCUSSION In the D. melanogaster ventral nerve cord, comm acts as a switch gene to regulate midline crossing; it is expressed in commissural but not ipsilateral neurons and is both necessary and sufficient for crossing4–6,8,9. Genetic data indicate that Comm acts antagonistically to Robo, the receptor for the midline repellent Slit6,8, and this can be attributed to the ability of Comm to regulate Robo surface levels4,11. The mechanism by which Comm regulates Robo surface levels has been controversial, however. Specifically, two distinct models have been proposed: the sorting model4 and the clearance (or internalization) model 5,11,12 (Supplementary Fig. 1). These conflicting models have been based largely on studies carried out in heterologous cells in vitro. The experiments reported here were aimed at determining the mechanism of Comm action in vivo, focusing specifically on points of contention between these two models. One critical issue on which the two models disagree is the requirement for comm in midline cells. We have tested this here using a rescue assay in

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

161

ARTICLES
which midline crossing of poxn commissural neurons is rescued in a comm mutant background by restoring comm function exclusively in the poxn neurons themselves. Similar results have also been obtained with eg-GAL4, which labels a different subset of commissural neurons (K.K. and B.J.D., unpublished data). These data demonstrate that the autonomous function of comm in commissural neurons is sufficient for midline crossing, a conclusion also consistent with our earlier transplantation experiments4. However, these genetic assays, like our original transplantation assays, also demonstrate a nonautonomous requirement for comm: not all comm+ neurons cross in these experiments. We have argued that this is likely to come from an autonomous function of comm in the guidance of other commissural axons4, with the nonautonomy simply reflecting a ‘follower’ or ‘community’ effect as is commonly observed in axon pathfinding. However, transgenic RNA interference experiments have suggested an alternative explanation5: that Comm provided by midline cells also contributes to axon guidance across the midline. The data presented here argue against this possibility. Specifically, the crossing of poxn neurons in a comm mutant background is no more frequent when comm function is restored both in the poxn neurons and in midline cells than when comm function is restored in the poxn neurons alone. Comm may well have some other function in the differentiation of midline cells, but it evidently does not contribute to the midline crossing of commissural axons in the manner proposed by the clearance model. The second critical issue on which the two models disagree is how Comm regulates the trafficking of Robo. To assess this, we have used a Robo-GFP fusion protein to observe directly the effect of Comm on Robo trafficking in D. melanogaster neurons in vivo. These studies provided confirmation of the sorting model. In the absence of Comm, Robo-GFP is transported along axons at a speed and in a discontinuous manner indicative of transport along microtubules by kinesin motors. In the presence of Comm, we still detected Robo-GFP in vesicles in the soma and, rarely, in axons. Yet, we no longer detected Robo-GFP at the plasma membrane or the transport of Robo-GFP vesicles along axons. We carried out these imaging studies on neurons of the peripheral nervous system rather than CNS neurons for biological reasons. Specifically, our aim was to define the effect of Comm on trafficking of Robo-GFP, so it was imperative to examine neurons whose axon trajectories are unchanged by the addition or removal of Comm. It was not our goal in these studies to understand how the sorting activity of Comm might be regulated during commissural axon guidance across the midline. However, with the role of Comm now clearly defined, it should be possible to address these questions by imaging Robo-GFP trafficking in the CNS. This poses some technical challenges, as CNS axons are considerably deeper and shorter than the peripheral nervous system axons we have examined here. There is, however, a strong incentive to overcome these difficulties, so that we can better understand how Comm activity and Robo trafficking are regulated during and after midline crossing. Together, these genetic and imaging studies have validated the sorting model and refuted the clearance model. Although we believe that this clarifies much of the confusion surrounding Comm function in vivo, it does leave us without an adequate explanation for how Robo comes to be highly enriched on the contralateral segments of commissural axons and excluded from their midline segments. This is readily explained by the clearance model, but not the sorting model. We have suggested that Robo may be restricted to the contralateral segments of commissural axons if Comm is inactivated after crossing and if Robo is inserted only at the growth cone and cannot spread along the axon shaft4. These assumptions remain to be tested. It is, however, also worth considering that the exclusion of Robo from commissures may in fact have nothing at all to do with Comm, except in the trivial sense that Comm is required for crossing. Perhaps some other factor, most likely provided by midline cells, is responsible for excluding Robo from commissures (both the trickle of Robo that escapes Comm during crossing and the surge of Robo that comes after crossing). Although this hypothesis challenges the dogma that Comm directly excludes Robo from commissures, we are not aware of any data that would refute it. The generation of a mutant form of Robo that is insensitive to Comm may help to resolve some of these issues. METHODS
Plasmids. We prepared the comm constructs in a pcDNA3.1+ vector (Invitrogen) for COS cell transfections or pUAST29 for D. melanogaster transformation. Comm proteins expressed in COS cells or D. melanogaster always included two C-terminal c-myc epitope tags or three HA tags in the case of CommKR. Mutations were generated using the overlap extension polymerase chain reaction (PCR) method and confirmed by sequencing. For slit-comm and simcomm transgenes, the comm-myc insert was subcloned from UAS-comm into pCaSpeR3-based vectors containing a 1.0-kb slit enhancer and noninducible hsp70 promoter14,16, and a 3.7-kb sim promoter15, respectively. UAS-robo-GFP encodes a full-length Robo protein with three N-terminal HA tags and one C-terminal GFP moiety. All other constructs have been described4. COS cell assays. COS-7 cell transfections and immunohistochemistry were carried out as described4. Slides were encoded so that protein localizations could be scored without knowledge of the particular comm mutant transfected. Comm (and, where appropriate, Robo) localization was scored in over 100 cells on each slide, binning into vesicular (presumably endosomal), plasma membrane, diffuse or mixed categories. All transfections were carried out at least in duplicate. For each experiment, Comm and Robo localizations were consistent both within and between transfections; this consensus localization is reported here. D. melanogaster stocks. The poxn-GAL4 line used here is a third-chromosome insertion of poxn-GAL4-1413. The commO72 allele30 bears a nonsense mutation in the 69th codon (S. Rajagopalan and B.J.D., unpublished data), as determined by PCR amplification and sequencing of the entire comm open reading frame from heterozygous flies. Df(3L)ED4688 was generated from the P element insertions CB-0732-3 and 5-HA-1908 as described27. To confirm the deletion in this chromosome, we prepared genomic DNA from single embryos from a y w; Df(3L)ED4688 / TM3, Ubx-lacZ stock and genotyped them using a PCR product length polymorphism associated with the single nucleotide polymorphism marker 3L18431. PCR was then carried out on homozygous and heterozygous Df(3L)ED4688 embryos to amplify the following: (A) 310 bp located 565 bp left of the CB-0732-3 insertion site; (B) 244 bp located 504 bp right of CB-0732-3; (C) 207 bp located 553 bp left of 5-HA-1908; and (D) 242 bp located 434 bp right of 5-HA-1908. Three of three Df(3L)ED4688 homozygous embryos gave the outside PCR products (A) and (D), but not the inside products (B) and (C), whereas five of five heterozygous embryos gave all four PCR products. UAS-mRFP1-moe was provided by M. Neumann and M. Affolter. Embryo staining and quantification. Embryos were fixed and stained with anti-β-galactosidase, anti-HRP, anti-FasII mAb 1D4 and/or BP102 as described32. Embryos were all genotyped using blue balancers, except the slit robo interactions with UAS-Nedd4, for which we strictly followed the procedure previously described11 in which embryos carrying one or two copies of UAS-Nedd4 are not distinguished. For quantification of phenotypes, embryos were encoded by K.K. and scored several days later by B.J.D. (for Nedd4 and ena interactions) or K.K. (for Ap and poxn assays) without knowledge of the genotype. Assays for all genotypes presented in a single figure were performed in parallel, and at least two replicate experiments were performed across a period of four months. The only exception to this was UAS-ena, which was only included in the later set. Slides from the first replicate were also rescored (still encoded) along with those of the later replicate. No significant difference was observed between scores for the same slide on different occasions or between scores for independent replicates of the same genotype. Confocal time-lapse microscopy. Living dechorionated embryos were glued with tape glue on coverslips, covered with 10S Voltalef oil and imaged on a Zeiss LSM 510 META Axiovert 200M confocal microscope using 40× or 63× objectives. To increase the signal, the pinhole was opened to 2–3 Airy units, and to

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

162

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
increase the scanning speed only the region of interest was scanned. Images were scanned at speed setting 12, with line averaging of two scans for each of three to six focal planes. Bleaching was performed by scanning 40 cycles at maximum laser intensity in the focal plane of the axon. Deconvolution was performed using Huygens Essential 2.6.0 (SVI). Images were processed with Imaris 4.0.6 (Bitplane). A Gaussian filter was applied to remove noise. To prepare kymographs, a maximum intensity projection of the z-axis was carried out for each time point, and this time series then loaded as a z-stack into Imaris. The kymograph was then obtained as a maximum intensity projection of the z-section. Measurements of vesicle speeds were performed using Volocity 3.0 (Improvision). Only vesicles with a high signal-to-noise ratio moving at a constant speed and direction were chosen. Speeds were measured across several time points, and an average speed was obtained for each vesicle.
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS We thank G. Tear, M. Noll, C. Klämbt, R. Hiesinger, and the Drosdel consortium for fly stocks; A. Graf and K. Paiha for technical assistance; and F. Schnorrer and G. Gilestro for discussions and comments on the manuscript. This work was funded in part by a grant from the Austrian Science Foundation (FWF). C.R. is supported by an EMBO long-term postdoctoral fellowship. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 28 October; accepted 23 December 2004 Published online at http://www.nature.com/natureneuroscience/
10. Tear, G. et al. commissureless controls growth cone guidance across the CNS midline in Drosophila and encodes a novel membrane protein. Neuron 16, 501–514 (1996). 11. Myat, A. et al. Drosophila Nedd4, a ubiquitin ligase, is recruited by Commissureless to control cell surface levels of the roundabout receptor. Neuron 35, 447–459 (2002). 12. Georgiou, M. & Tear, G. The N-terminal and transmembrane domains of Commissureless are necessary for its function and trafficking within neurons. Mech. Dev. 120, 1009– 1019 (2003). 13. Boll, W. & Noll, M. The Drosophila Pox neuro gene: control of male courtship behavior and fertility as revealed by a complete dissection of all enhancers. Development 129, 5667–5681 (2002). 14. Wharton, K.A., Jr. & Crews, S.T. CNS midline enhancers of the Drosophila slit and Toll genes. Mech. Dev. 40, 141–154 (1993). 15. Wharton, K.A., Jr., Franks, R.G., Kasai, Y. & Crews, S.T. Control of CNS midline transcription by asymmetric E-box-like elements: similarity to xenobiotic responsive regulation. Development 120, 3563–3569 (1994). 16. Mitchell, K.J. et al. Genetic analysis of Netrin genes in Drosophila: Netrins guide CNS commissural axons and peripheral motor axons. Neuron 17, 203–215 (1996). 17. Hirokawa, N. Kinesin and dynein superfamily proteins and the mechanism of organelle transport. Science 279, 519–526. (1998). 18. Goldstein, L.S. & Yang, Z. Microtubule-based transport systems in neurons: the roles of kinesins and dyneins. Annu. Rev. Neurosci. 23, 39–71 (2000). 19. Overly, C.C. & Hollenbeck, P.J. Dynamic organization of endocytic pathways in axons of cultured sympathetic neurons. J. Neurosci. 16, 6056–6064 (1996). 20. Helliwell, S.B., Losko, S. & Kaiser, C.A. Components of a ubiquitin ligase complex specify polyubiquitination and intracellular trafficking of the general amino acid permease. J. Cell Biol. 153, 649–662 (2001). 21. Soetens, O., De Craene, J.O. & Andre, B. Ubiquitin is required for sorting to the vacuole of the yeast general amino acid permease, Gap1. J. Biol. Chem. 276, 43949–43957 (2001). 22. Hicke, L. & Dunn, R. Regulation of membrane protein transport by ubiquitin and ubiquitin-binding proteins. Annu. Rev. Cell Dev. Biol. 19, 141–172 (2003). 23. Bashaw, G.J., Kidd, T., Murray, D., Pawson, T. & Goodman, C.S. Repulsive axon guidance: Abelson and Enabled play opposing roles downstream of the Roundabout receptor. Cell 101, 703–715 (2000). 24. Fan, X., Labrador, J.P., Hing, H. & Bashaw, G.J. Slit stimulation recruits Dock and Pak to the roundabout receptor and increases Rac activity to regulate axon repulsion at the CNS midline. Neuron 40, 113–127 (2003). 25. Lee, H. et al. The microtubule plus end tracking protein Orbit/MAST/CLASP acts downstream of the tyrosine kinase Abl in mediating axon guidance. Neuron 42, 913–926 (2004). 26. Hsouna, A., Kim, Y.S. & VanBerkum, M.F. Abelson tyrosine kinase is required to transduce midline repulsive cues. J. Neurobiol. 57, 15–30 (2003). 27. Ryder, E. et al. The DrosDel collection: a set of P-element insertions for generating custom chromosomal aberrations in Drosophila melanogaster. Genetics 167, 797–813 (2004). 28. Macias, M.J., Wiesner, S. & Sudol, M. WW and SH3 domains, two different scaffolds to recognize proline-rich ligands. FEBS Lett. 513, 30–37 (2002). 29. Brand, A.H. & Perrimon, N. Targeted gene expression as a means of altering cell fates and generating dominant phenotypes. Development 118, 401–415 (1993). 30. Hummel, T., Schimmelpfeng, K. & Klambt, C. Commissure formation in the embryonic CNS of Drosophila: I. Identification of the required gene functions. Dev. Biol. 209, 381–398 (1999). 31. Berger, J. et al. Genetic mapping with SNP markers in Drosophila. Nat. Genet. 29, 475–481 (2001). 32. Rajagopalan, S., Vivancos, V., Nicolas, E. & Dickson, B.J. Selecting a longitudinal pathway: Robo receptors specify the lateral position of axons in the Drosophila CNS. Cell 103, 1033–1045 (2000).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

1. Dickson, B.J. Molecular mechanisms of axon guidance. Science 298, 1959–1964 (2002). 2. Kidd, T., Bland, K.S. & Goodman, C.S. Slit is the midline repellent for the robo receptor in Drosophila. Cell 96, 785–794 (1999). 3. Long, H. et al. Conserved roles for Slit and Robo proteins in midline commissural axon guidance. Neuron 42, 213–223 (2004). 4. Keleman, K. et al. Comm sorts robo to control axon guidance at the Drosophila midline. Cell 110, 415–427 (2002). 5. Georgiou, M. & Tear, G. Commissureless is required both in commissural neurones and midline cells for axon guidance across the midline. Development 129, 2947–2956 (2002). 6. Seeger, M., Tear, G., Ferres-Marco, D. & Goodman, C.S. Mutations affecting growth cone guidance in Drosophila: genes necessary for guidance toward or away from the midline. Neuron 10, 409–426 (1993). 7. McGovern, V.L. & Seeger, M.A. Mosaic analysis reveals a cell-autonomous, neuronal requirement for Commissureless in the Drosophila CNS. Dev. Genes Evol. 213, 500–504 (2003). 8. Kidd, T., Russell, C., Goodman, C.S. & Tear, G. Dosage-sensitive and complementary functions of roundabout and commissureless control axon crossing of the CNS midline. Neuron 20, 25–33 (1998). 9. Bonkowsky, J.L., Yoshikawa, S., O’Keefe, D.D., Scully, A.L. & Thomas, J.B. Axon routing across the midline controlled by the Drosophila Derailed receptor. Nature 402, 540–544 (1999).

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

163

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Cyclic AMP controls BDNF-induced TrkB phosphorylation and dendritic spine formation in mature hippocampal neurons
Yuanyuan Ji1, Petti T Pang2,3, Linyin Feng1 & Bai Lu2
Synaptic actions of brain-derived neurotrophic factor (BDNF) are 'gated' by cyclic AMP (cAMP), but the underlying molecular mechanisms remain unclear. Here we report that cAMP regulates BDNF function in mature hippocampal neurons by modulating the signaling and trafficking of its receptor TrkB. cAMP gated the TrkB tyrosine kinase with three characteristic features: BDNFinduced TrkB phosphorylation was attenuated by inhibitors of cAMP signaling, it was potentiated by cAMP analogs, and activation of the cAMP pathway alone had no effect. In addition, cAMP facilitated trafficking of TrkB to dendritic spines, possibly by promoting its interaction with synaptic scaffolding protein PSD-95. Norepinephrinergic and dopaminergic agonists, which elevate intracellular cAMP concentration, also enhanced TrkB phosphorylation and its translocation to spines. cAMP gated long-term modulation by BDNF of spine density, but not the number of primary dendrites. These results reveal a specific role of cAMP in controlling BDNF actions in the brain, and provide new insights into the molecular mechanism underlying cAMP gating.

BDNF, a member of the neurotrophin family initially identified as a survival factor for peripheral neurons, has emerged as a critical factor that regulates synaptic development and plasticity in the central nervous system (CNS)1,2. Acute application of BDNF elicits a rapid potentiation of excitatory synaptic transmission in cultured hippocampal or cortical neurons3,4, and it facilitates long-term potentiation (LTP) as well as synaptic response to high-frequency stimulation (HFS) in neonatal hippocampal slices5. These relatively fast changes in synaptic efficacy may be translated into structural alterations when the synapses are exposed to BDNF for a longer period of time. These include axonal branching6,7, dendritic growth1,8 and activitydependent refinement of synapses9. A particularly interesting but less explored effect of BDNF is its regulation of dendritic spine formation. Long-term treatment of hippocampal slice cultures with BDNF increases the synapse number and spine density in apical dendrites of CA1 pyramidal neurons10. In cerebellar cultures, BDNF, together with cocultured granule cells, increases the spine density of Purkinje cells without affecting dendritic complexity11. All the functions of BDNF described above are mediated by the receptor tyrosine kinase TrkB, which is rapidly activated upon binding to BDNF, triggering multiple intracellular signaling pathways through protein-protein interactions12. cAMP seems to be important for the signaling and biological functions of BDNF. The survival effects of BDNF on retinal ganglion neurons requires cAMP13. The growth cones of embryonic spinal neurons show either attractive or repulsive turning responses towards a gradient of BDNF, depending on the

intracellular concentration of cAMP14. However, BDNF does not activate the cAMP signaling pathway directly15–17. Application of BDNF in Xenopus nerve-muscle cocultures induces a rapid potentiation of synaptic transmission at the neuromuscular synapses with the following unique features: (i) inhibitors of cAMP signaling block potentiation induced by high-dose BDNF; (ii) activators of cAMP signaling enhance the potentiating effects of low-dose BDNF; and (iii) cAMP analogs alone cannot mimic the effects of BDNF18. Based on these experiments, cAMP was proposed to act as a 'gate' that allows BDNF to achieve these effects. Is 'cAMP gating' a general mechanism involved in BDNF regulation of CNS neurons? In this study, we examined the effects of cAMP in BDNF regulation of dendritic growth and spine formation. In addition, we investigated the molecular mechanism underlying cAMP gating. We hypothesize that cAMP could gate BDNF function by modulating TrkB signaling. Using dissociated cultures of hippocampal neurons, we investigated whether BDNF-induced TrkB phosphorylation is gated by cAMP. Given that TrkB has been found in the postsynaptic density (PSD)19, we further examined whether cAMP modulates the trafficking of TrkB to the PSD. Our results suggest that 'cAMP gating' of TrkB phosphorylation and cAMP regulation of TrkB trafficking to the PSD may both contribute to the long-term regulation of dendritic spine formation by BDNF. These results may provide important insights into the mechanisms by which cAMP regulates the synaptic actions of BDNF in CNS neurons.

1Institute of Neuroscience, Shanghai Institutes of Biological Sciences, Chinese Academy of Sciences, Graduate School of the Chinese Academy of Sciences, 320 Yue Yang Road, Shanghai 200031, China. 2Section on Neural Development and Plasticity, National Institute of Child Heath and Human Development, National Institutes of Health, 35 Lincoln Drive, Bethesda, Maryland 20892-3714, USA. 3Department of Physiology, The Chinese University of Hong Kong, Shatin, Hong Kong. Correspondence should be addressed to B.L. ([email protected]).

Published online 23 January 2005; doi:10.1038/nn1381

164

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES a
Control BDNF 5 ng/ml

BDNF 5 ng/ml + Sp-cAMP

Sp-cAMP

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 1 cAMP facilitates the increase in dendritic spine density induced by BDNF. BDNF 5, 5 ng ml–1. BDNF 25, 25 ng ml–1. (a) Dendritic spines under various conditions. Hippocampal neurons (18–21 d) transfected with GFP at 7 d in vitro were fixed after the indicated treatments (Sp-cAMP; Rp-cAMP). Scale bar, 5 µm. (b,c) Sp-cAMP increases the effect of low dose BDNF (5 ng ml–1) on dendritic spines and filopodia. (d,e) Rp-cAMP blocks, but Sp-cAMP does not increase, the effect of high-dose BDNF (25 ng ml–1) on dendritic spines and filopodia. At least five independent cultures were examined. In this and all other graphs, error bars are s.e.m., and numbers associated with columns are the number of dendrites or neurons examined. **P < 0.01, ANOVA followed by post hoc test.

BDNF 25 ng/ml

BDNF 25 ng/ml + Sp-cAMP

BDNF 25 ng/ml + Rp-cAMP

Rp-cAMP

b
Protrusion density (spines/10 µm dendrite)

6 5 4 3 2 1

**

c
Protrusion density (filopodia/10 µm dendrite)

1.2 1.0 0.8 0.6 0.4 0.2

**

246 148 0 + BDNF 25 – – Sp-cAMP –

72 + +

68 – +

246 148 72 0 + + BDNF 25 – – + Sp-cAMP –

68 – +

d
6 Protrusion density (spines/10 µm dendrite) 5 4 3 2 1 79 + – – 15 + + – 54 + – + 19 – – + ** **

e
1.6 Protrusion density (filopodia/10 µm dendrite) 1.4 1.2 1.0 0.8 0.6 0.4 0.2 246 0 BDNF 25 – Sp-cAMP – Rp-cAMP – 79 + – – 15 + + – 54 + – + 19 – – + ** **

246 0 BDNF 25 – Sp-cAMP – Rp-cAMP –

RESULTS cAMP gating in the CNS Long-term treatment of hippocampal neurons with BDNF has been shown to increase the number of dendritic spines11. We tested whether cAMP gates BDNF regulation of spine formation. Confocal microscopy was used to measure the spine and filopodium densities separately (expressed as the number of spines or filopodia per 10-µm length of dendrites) of individual cultured hippocampal neurons (18–21 d), according to the criteria previously described20,21: spines, which include thin, mushroom- and branched-shaped types, and filopodia, which are thin, uniform-caliber, headless protrusions from the dendrites. Cultured hippocampal neurons treated with a low dose of BDNF (5 ng ml–1) for 24 h had a higher density of spines, but not filopodia, than control neurons (Fig. 1; control, 1.47 ± 0.11 for spine density, 0.51 ± 0.07 for filopodium density, n = 246; BDNF, 2.92 ± 0.11 for spine density, P < 0.01, 0.67 ± 0.06 for filopodium density, n = 148). A telling feature of cAMP gating is that cAMP should facilitate the function of low-dose BDNF. Indeed, when the cultures were pretreated for 15–20 min with 10 µM Sp-cAMP, a membrane-permeable activator of protein kinase A (PKA), and then treated with 5 ng ml–1 BDNF, the spine and filopodium densities increased to 4.95 ± 0.25 and 1.00 ± 0.07, respectively (Fig. 1b,c, n = 72; **, significantly different from BDNF-alone group, ANOVA,

P < 0.01). Furthermore, time-lapse imaging (Supplementary Videos 1–4) showed that new spines emerged, disappeared and re-emerged in the same location after treatment with BDNF alone. However, in cultures treated with Sp-cAMP plus BDNF, more spines appeared, remaining in the same location for 24 h, and most new spines showed rapid head morphing (Supplementary Videos 1–4). Thus, cultures pretreated with Sp-cAMP and BDNF developed more new spines than those treated with BDNF alone. Importantly, we found that treatment with Sp-cAMP alone for 24 h had no effect on spine or filopodium density (Fig. 1b,c: 1.84 ± 0.11 for spine density, 0.38 ± 0.05 for filopodium density, n = 68). Moreover, Sp-cAMP had no effect on the increase of spine or filopodium density induced by a higher dose of BDNF (25 ng ml–1) (Fig. 1d,e). The average spine densities with and without Sp-cAMP in the high-dose BDNF conditions were 4.40 ± 0.49 and 4.77 ± 0.17, and the average filopodium densities were 1.19 ± 0.13 and 1.30 ± 0.14, respectively. Another key feature of cAMP gating is that inhibition of cAMP signaling should prevent the function of high-dose BDNF. Indeed, application of Rp-cAMP (10 µM), a non-hydrolyzable potent inhibitor for PKA, had no effect on its own (1.62 ± 0.09 for spine density, 0.34 ± 0.05 for filopodium density, n = 19) but completely prevented the effect of high-dose BDNF (25 ng ml–1) on protrusion density (2.13 ± 0.12 for spine density, 0.59 ± 0.09 for filopodium density, n = 54, Fig. 1d,e). As the filopodia are thought to be precursors of more mature spines21,22, these results suggest that a certain threshold of cAMP is important for BDNF modulation of spine density. Activation of PKA facilitates spine formation induced by a low dose of BDNF, whereas inhibition of PKA attenuates spine formation induced by a high dose of BDNF. The role of cAMP in determining spine or filopodium density was further investigated using forskolin, an activator of adenylate cyclase. In the presence of 5 µM forskolin, the increase in spine or filopodium density induced by a low dose of BDNF (5 ng ml–1) was substantially magnified (Supplementary Fig. 1a and 1b, 3.87 ± 0.28 for spine density, 0.87 ± 0.12 for filopodium density, n = 41). Treatment with forskolin alone for 24 h was ineffective at increasing spine or filopodium density (1.51 ± 0.18 for spine density, 0.46 ± 0.13 for filopodium density, n = 12). In cultures treated with a high dose of BDNF (25 ng ml–1), forskolin had little effect on spine or filopodium density (Supplementary Fig. 1c and 1d). Taken together, these results indicate that cAMP may have purely a regulatory role, a mechanism referred to as gating23. BDNF has also been shown to be a potent regulator of dendritic growth in CNS neurons24. We next examined whether cAMP is important in gating BDNF regulation of dendritic growth of cultured hippocampal neurons (3 d). Exposure to BDNF at a high dose (25 ng ml–1) for 3 d resulted in robust growth of MAP2-positive primary dendrites (Fig. 2a). Quantitative analysis revealed that BDNF increased the number of primary dendrites by more than 80% (Fig. 2b; *, ANOVA followed by post hoc test, P < 0.001). Concurrent treatment with Rp-cAMP

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

165

ARTICLES a b
Number of primary dendrites/neuron 12 * 10 8 6 4 2 0 74 68 BDNF 5 ng/ml 58 45 66 84 56 BDNF 25 ng/ml 58 56 45 71 * * *

Control

BDNF 5 ng/ml

BDNF 25 ng/ml

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Sp-cAMP

Rp-cAMP

Control

Forskolin

BDNF 5 ng/ml

BDNF 25 ng/ml

Sp-cAMP

BDNF 5 ng/ml Sp-cAMP

BDNF 25 ng/ml Sp-cAMP

Rp-cAMP

BDNF 25 ng/ml Rp-cAMP

Figure 2 cAMP is not involved in the BDNF regulation of dendritic growth. (a) Representative images of MAP2-stained neurons. Hippocampal neurons (3 d) were incubated with various agents as indicated (Sp-cAMP; Rp-cAMP; forskolin) for 3 d and stained with anti-MAP2 antibody. Scale bar, 20 µm. (b) Summary of the effect of cAMP on BDNF modulation of dendritic growth. The number of primary dendrites per neuron was counted. At least three independent cultures were used in each condition. *P < 0.001, compared with control, ANOVA followed by post hoc test.

20th EPSP slope (% of the 1st)

EPSP slope (% of baseline)

(10 µM) did not block the BDNF effect. On the other hand, treatment with a low dose of BDNF (5 ng ml–1) alone had little effect on dendritic growth (Fig. 2b). Low-dose BDNF with forskolin or Sp-cAMP did not increase the number of primary dendrites, either (Fig. 2b). Thus, BDNF regulation of dendritic growth is not gated by cAMP. We next determined whether cAMP gates the acute modulation of CNS synapses by BDNF. BDNF elicits two acute effects at the CA1 synapses in the hippocampal slices. The first is to attenuate synaptic fatigue induced by a train of HFS (100 Hz, 100 pulses)5,25. Application of HFS induced a pronounced synaptic fatigue (43.36 ± 2.86%, n = 36) at Schaffer collatertal-CA1 synapses in neonatal hippocampal slices (p12–13), and treatment with exogenous BDNF (8 nM) for 2 h significantly attenuated the synaptic fatigue (52.92 ± 2.19%, n = 37) (Fig. 3a). However, treatment with Rp-cAMP did not prevent the effect of BDNF on HFS-induced synaptic fatigue (59.02 ± 2.34%, n = 33). Incubation with Rp-cAMP alone for 2 h had no effect, either (48.06 ± 3.84%, n = 23) (Fig. 3a). In a separate set of experiments, we found that perfusion of hippocampal slices with either a lower concentration of BDNF (2 nM) or the cAMP analog Sp-cAMP alone had no effect on
Figure 3 Role of cAMP in BDNF modulation of HFS-induced synaptic fatigue and LTP at hippocampal CA1 synapses. Neonatal hippocampal slices (p11–13) were treated with Rp-cAMP or Sp-cAMP, either with or without BDNF, for 2 h. (a) Rp-cAMP does not inhibit the modulation of synaptic fatigue by high-dose (8 nM) BDNF. The slope of the 20th EPSP in the train of HFS is presented as a percentage of the first EPSP slope. The number associated with each column represents the number of slices tested. Examples of EPSPs elicited by HFS recorded from a control (upper) and a BDNF-treated slice (lower) are shown on top of bar graph. *P < 0.05, **P < 0.01, compared with '–BDNF' groups, Student's t-test. (b) Sp-cAMP does not facilitate the modulation of synaptic fatigue by low-dose (2 nM) BDNF. (c) Rp-cAMP does not alter the potentiating effect of BDNF (8 nM) on LTP (Control, upper graph; Rp-cAMP, lower graph).

synaptic fatigue (Fig. 3b). More importantly, Sp-cAMP did not facilitate the effect of BDNF on synaptic responses (Fig. 3b). The second acute effect of BDNF is to facilitate HFS-induced LTP5. In neonatal hippocampus, HFS induced a synaptic potentiation that significantly decayed over the next 55–60 min (119.21 ± 5.48%, n = 12), and treatment of the slices with BDNF for 2–3 h significantly increased the magnitude of LTP (144.39 ± 10.08%, n = 10) (Fig. 3c). However,

a
Control BDNF 60 40 20 0 36 37 Control 23 33 Rp-cAMP * **

c 200
150

100

- BDNF (12 slices, 4 rats) + BDNF (10 slices, 4 rats)
50 0 20 40 60 80

200

b
20th EPSP slope (% of the 1st) 60 - BDNF + BDNF

150

40

100

20

50
0 5 17 Control 9 25 Sp-cAMP

Rp-cAMP (6 slices, 4 rats) Rp-cAMP + BDNF (8 slices, 5 rats)
0 20 40 Time (min) 60 80

166

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

Rp-cAMP

Sp-cAMP

Sp-cAMP

Forskolin

Forskolin

Forskolin

BDNF 5 ng/ml Forskolin

BDNF 25 ng/ml Forskolin

ARTICLES
a
3.5 TrkB phosphorylation (100% of BDNF 5 ng/ml) 3.0 2.5 2.0 1.5 1.0 0.5 0 BDNF 5 Sp-cAMP – – + – + + – + **

b
* 2.0 1.5 1.0 0.5 0 BDNF 5 SFK38393 isoproterenol – – – + – – + – + + + – – – + – + – TrkB phosphorylation (100% of BDNF 25 ng/ml) 2.5 TrkB phosphorylation (100% of BDNF 5 ng/ml) *

c
1.4 1.2 1.0 0.8 0.6 0.4 0.2 0 BDNF 25 Sp-cAMP Rp-cAMP – – – + – – + + – + – + – – + **

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

d
1.4 TrkB phosphorylation (100% of BDNF 5 ng/ml) 1.2 1.0 0.8 0.6 0.4 0.2 0 BDNF 5 Forskolin Sp-cAMP – – – + – – + – + + + – – – + – + –

e
TrkB phosphorylation (100% of BDNF 25 ng/ml) 1.2 1.0 0.8 0.6 0.4 0.2 0 BDNF 25 Rp-cAMP – – + – + + – –

f
1.6 TrkB phosphorylation (100% of BDNF 2 µM) 1.4

g
1.2 TrkB phosphorylation (100% of BDNF 8 µM) – – – + – – + – + + + – – – + – + – 1.0 0.8 0.6 0.4 0.2 0 BDNF 8nM Rp-cAMP – – + – + + – –

1.2 1.0 0.8 0.6 0.4 0.2 0

BDNF 2nM Forskolin Sp-cAMP

Figure 4 Regulation of TrkB phosphorylation at Tyr 490 by cAMP in mature hippocampal neurons. In this and other figures: BDNF 5, 5 ng ml–1; BDNF 25, 25 ng ml–1. Representative immunoblots for p-TrkB (p490, upper, detected by antibody from Cell Signaling) and total TrkB (lower, detected by antibody from Santa Cruz, Ab 794) are presented on top of each bar graph. (a) cAMP facilitates TrkB phosphorylation induced by low-dose BDNF. (b) Adrenoceptor or dopamine receptor activation potentiates TrkB phosphorylation induced by low-dose BDNF. (c) Blockade of cAMP signaling decreases, but activation of cAMP signaling does not affect, TrkB phosphorylation induced by high-dose BDNF. (d) Forskolin or Sp-cAMP does not increase the BDNF (5 ng ml–1)induced phosphorylation of TrkB in young neurons (cultured for 3 d). (e) Rp-cAMP does not decrease the BDNF (25 ng ml–1)-induced phosphorylation of TrkB in young neurons (cultured for 3 d). (f) Forskolin or Sp-cAMP does not increase the BDNF (2 nM)-induced phosphorylation of TrkB in neonatal hippocampal slices (p11–13). (g) Rp-cAMP does not decrease the BDNF (8 nM)-induced phosphorylation of TrkB in neonatal hippocampal slices (p11–13). n = 6 for all experiments. In this and all other western blot figures, the positions of molecular weight markers are indicated on the left of the figures. *P < 0.05, **P < 0.01, ANOVA followed by post hoc test.

inhibition of the cAMP pathway by Rp-cAMP did not attenuate the effect of BDNF on LTP (145.32 ± 15.22%, n = 8) (Fig. 3c). Slices treated with Rp-cAMP alone also had no effect on the magnitude of LTP (116.68 ± 9.60%, n = 6) (Fig. 3c). In addition, it is known that activation of cAMP alone does not potentiate synaptic efficacy26. Taken together, these results suggest that cAMP gating of BDNF regulation is specific: it controls the long-term regulation of spine formation in hippocampal neurons without affecting the acute modulation of hippocampal plasticity by BDNF. Gating of BDNF-induced TrkB phosphorylation by cAMP To understand the molecular mechanisms by which cAMP gates BDNF actions, we tested the possibility that cAMP signaling affects the tyrosine phosphorylation of the TrkB receptor itself. Western blot analysis was performed using an antibody that specifically recognizes TrkB phosphorylated on the tyrosine residue 490 (pY490). For quantitative analysis, we first measured dose-response relationship and time course of BDNF-induced TrkB tyrosine phosphorylation. Application of BDNF to cultured hippocampal neurons (12–14 d) for 15 min induced phosphorylation of p490 that was dependent on the concentration of BDNF. At 5 ng ml–1 BDNF, TrkB phosphorylation was submaximal (Supplementary Fig. 2a), and it took longer (15 min) to reach this

submaximal level (data not shown). TrkB phosphorylation reached a plateau when the concentration of BDNF was higher than 10 ng ml–1. Using 25 ng ml–1 of BDNF, TrkB was maximally phosphorylated within 5 min (Supplementary Fig. 2b). To determine whether cAMP facilitates BDNF-induced TrkB phosphorylation, we preincubated cultured neurons with 10 µM Sp-cAMP or 5 µM forskolin for 15 min. After BDNF treatment for 15 min, the cultures were harvested and TrkB phosphorylation was analyzed. Facilitation of TrkB phosphorylation induced by 5 ng ml–1, but not by 25 ng ml–1, BDNF was also observed using Sp-cAMP or forskolin. Treatment with Sp-cAMP or forskolin caused a 2.5- or 1.9fold increase, respectively, in TrkB phosphorylation upon exposure to a submaximal dose (5 ng ml–1) of BDNF (Fig. 4a, Supplementary Fig. 2c). Importantly, exposure to Sp-cAMP or forskolin alone had no significant effect on the base level of TrkB phosphorylation (P > 0.05, compared with controls). In the hippocampus, dopaminergic and norepinephrinergic afferents activate D1/D5 dopamine receptors and β-adrenoceptors respectively, leading to cAMP accumulation27,28. Application of the β-adrenoceptor agonist β-isoproterenol (10 µM) or the D1/D5 dopamine receptor agonist SKF38393 (100 µM) alone did not induce TrkB phosphorylation (P > 0.05, compared with controls). However, pretreatment with β-isoproterenol or SKF38393 followed by

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

167

ARTICLES a
TrkB phosphorylation (100% of BDNF 5 ng/ml) 2.5 2.0 1.5 1.0 0.5 0 – – – + – – + + – + – + – + – – – + ** **

b
TrkB phosphorylation (100% of BDNF 25 ng/ml) 1.2 1.0 0.8 0.6 0.4 0.2 0 – – – – + – – – + + – – + – + – + – – + – – – + **

c
TrkC phosphorylation (100% of NT3 5 ng/ml) 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0 – – – + – – + + – + – + – + – – – +

d
TrkC phosphorylation (100% of NT3 25 ng/ml) 1.2 1.0 0.8 0.6 0.4 0.2 0 NT-3 25 Forskolin Sp-cAMP Rp-cAMP – – – – + – – – + + – – + – + – + – – + – – – +

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

BDNF 5 Forskolin Sp-cAMP

BDNF 25 Forskolin Sp-cAMP Rp-cAMP

NT-3 5 Forskolin Sp-cAMP

Figure 5 Specificity of cAMP regulation of TrkB phosphorylation in mature hippocampal neurons. The experiments were done as described in Figure 4. Representative immunoblots are shown on top of each bar graph (n = 6 for all experiments). (a) Forskolin or Sp-cAMP facilitates TrkB phosphorylation at Tyr785 induced by low-dose BDNF (5 ng ml–1) in mature neurons. (b) Forskolin or Sp-cAMP does not affect, but Rp-cAMP inhibits, TrkB phosphorylation at Tyr785 induced by high-dose BDNF (25 ng ml–1). (c) Forskolin or Sp-cAMP does not facilitate TrkC phosphorylation induced by low-dose NT-3 (5 ng ml–1). TrkC phosphorylation is induced by NT-3, detected with anti-p490 (Cell Signaling) and normalized to total TrkC (detected by anti-TrkC, Ab 798 (Santa Cruz)). (d) Rp-cAMP does not inhibit TrkC phosphorylation induced by high-dose NT-3 (25 ng ml–1). **P < 0.01, ANOVA followed by post hoc test.

application of 5 ng ml–1 BDNF resulted in a 1.8-fold or 1.9-fold increase in the level of phosphorylated TrkB as compared with the application of 5 ng ml–1 BDNF alone (Fig. 4b). We next tested whether the effect of BDNF on TrkB phosphorylation could be attenuated by Rp-cAMP. Pretreatment with 10 µM RpcAMP had no effect on the base level of TrkB phosphorylation, but it largely abolished TrkB phosphorylation induced by a high dose (25 ng ml–1) of BDNF (Fig. 4c). Unlike TrkB phosphorylation induced by a submaximal dose of BDNF (which was enhanced by Sp-cAMP or forskolin), TrkB phosphorylation induced by a high dose of BDNF (25 ng ml–1) was not altered at all by Sp-cAMP or forskolin (P > 0.05, Fig. 4c and Supplementary Fig. 2d). These results show that cAMP does not directly activate TrkB, although TrkB activation in response to a low concentration of BDNF is enhanced when intracellular cAMP is increased. Given that more mature neurons were used for experiments testing cAMP gating of BDNF regulation of spine density, but immature cultures or slices were used to examine the effect of cAMP on dendrite growth or synaptic plasticity, it is possible that the cAMP gating mechanism may exist only in mature, but not in immature, hippocampal neurons. To test this idea, we measured the effect of cAMP on BDNF-induced TrkB phosphorylation in 3-d cultured neurons and p11–13 hippocampal slices. In contrast to our results using mature neurons, neither Sp-cAMP nor forskolin increased the BDNF-induced phosphorylation of TrkB (ANOVA, P > 0.6, compared with 5 ng ml–1 BDNF alone), and Rp-cAMP did not decrease the BDNF-induced phosphorylation of TrkB (ANOVA, P > 0.6, compared with 25 ng ml–1 BDNF alone) (Fig. 4d–g). Therefore, we suggest that the cAMP gating mechanism is associated with more mature neurons. At the stage in which the dendritic branching was

measured (3 d) or synaptic plasticity was recorded (p11–13 slices), BDNF-induced TrkB phosphorylation is not modulated by cAMP, and consequently BDNF/TrkB regulation of dendrite branching or synaptic plasticity is not gated by cAMP. Two experiments were performed to determine the specificity of cAMP gating mechanism in mature hippocampal neurons. First, we asked whether the cAMP regulation is restricted to the Shc binding site (pY490) of the TrkB receptor. We carried out experiments similar to those described earlier, but probing the western blots with an antibody that specifically recognizes TrkB phosphorylated at the PLC-γ binding site (tyrosine residue 785, pY785). Forskolin and SpcAMP facilitated TrkB phosphorylation at pY785 induced by a low dose of BDNF, and Rp-cAMP blocked high-dose BDNF-induced TrkB phosphorylation on that site (Fig. 5a,b). Thus, it appears that cAMP controls the tyrosine phosphorylation at both the Shc and PLC-γ binding sites of TrkB. Next, we examined whether the cAMP gating mechanism is also applicable to phosphorylation of TrkC, the receptor for neurotrophin 3 (NT-3). Application of NT-3 at either low (5 ng ml–1) or high (25 ng ml–1) concentration induced a reliable tyrosine phosphorylation of TrkC (Fig. 5c,d). Neither forskolin nor Sp-cAMP facilitated the TrkC phosphorylation by low-dose NT-3 (Fig. 5c). Furthermore, pretreatment of neurons with Rp-cAMP had no effect on TrkC phosphorylation by high-dose NT-3 (Fig. 5d). Thus, cAMP appears to specifically control BDNF-TrkB signaling without affecting NT-3-TrkC signaling.

a
Biotin TrkB Total TrkB

b
2.5 2.0 1.5 1.0
**

Figure 6 cAMP signaling does not alter cell surface expression of TrkB. Hippocampal neurons (12–14 d) were treated with various agents as indicated at 37 °C for 15 min, except for the positive control, in which the cells were treated with 50 ng ml–1 BDNF for 15 s at 21 °C. Cell surface proteins were biotinylated and then examined by western blot using anti-TrkB. (a) Representative western blots show that no treatment affects TrkB surface expression except a 15-s treatment with 50 ng ml–1 BDNF at 21 °C, which is known to increase surface TrkB (n = 4 for each experiment). (b) Summary of all experiments. **P < 0.01, ANOVA followed by post hoc test.

0.5 0 Rp-cAMP 10 µM BDNF 50 ng/ml 15 s Sp-cAMP 10 µM BDNF 5 ng/ml Sp-cAMP 10 µM BDNF 5 ng/ml BDNF 25 ng/ml Control

168

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES a
Control
PSD TrkB Overlay
Figure 7 cAMP facilitates the colocalization of TrkB with PSD-95. Hippocampal neuronal cultures (18–21 d) were fixed and stained with antibodies against TrkB (middle) and PSD-95 (left). (a) Images showing examples of neuronal dendrites in control, Sp-cAMP, β-isoproterenol, SKF38393 or Sp-cAMP + KT5720 cultures. Arrows denote prominent TrkB spots colocalized with PSD-95 on the spines of neurons. Scale bar, 10 µm. (b) Quantification of TrkB receptor colocalization with PSD-95 in response to the indicated treatment. Numbers inside bars are numbers of neuronal dendrites examined. **P < 0.01, ANOVA followed by post hoc test.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Sp-cAMP

b
No. colocalized/No. PSD

4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 10

*

*

*

10

10

10

10

Finally, to determine whether the level of endogenous BDNF interferes with cAMP gating of TrkB phosphorylation or spine density, we used TrkB-IgG (1 µg ml–1), a BDNF scavenger that neutralizes the effects of endogenous BDNF. Neither TrkB-IgG alone nor TrkB-IgG with Rp-cAMP decreased TrkB phosphorylation (Supplementary Fig. 3a), nor did it affect spine density (Supplementary Fig. 3b) as compared with control conditions. Thus, endogenous BDNF does not have an obvious effect on our experimental results. Effects of cAMP on TrkB subcellular distribution A previous study showed that treatment with cAMP resulted in a significant increase in the surface TrkB expression in retinal ganglion cells29.

We therefore tested whether cAMP may also potentiate TrkB signaling by enhancing the expression of cell surface TrkB in hippocampal neurons (12–14 d), using the biotinylation assay30. A 15-min Sp-cAMP or BDNF treatment alone did not increase the levels of cell surface TrkB as compared with the untreated control cells in these hippocampal cultures (Fig. 6a). Exposure of cells to Sp-cAMP for 15 min followed by a 15min BDNF treatment did not lead to an increase in surface TrkB, either (Fig. 6b). As a positive control, a 15-s BDNF treatment of cultured hippocampal neurons at 21 °C resulted in an increase in cell surface TrkB (Fig. 6a), similar to what has been previously reported31. The total protein levels of TrkB were not significantly altered (Fig. 6a), suggesting that the changes observed in the biotinylation assay represent events taking place on the cell surface only. Thus, cAMP did not modulate surface expression of TrkB receptors in hippocampal neurons. Previous studies have indicated that full-length TrkB seemed to be evenly distributed on the surface of hippocampal neurons and did not show any preferential localization in axons or dendrites32. However, TrkB was markedly increased in the postsynaptic density after transient cerebral ischemia33. It is quite possible that cAMP modulates the trafficking of TrkB receptors to the postsynaptic density, contributing to an enhanced response to BDNF in dendritic spines. To test this possibility, hippocampal neurons cultured for 18–21 d were treated with Sp-cAMP, β-isoproterenol or SKF38393 and immunostained for TrkB or PSD-95. In control cultures, TrkB showed a nonuniform subcellular localization in the soma and dendrites, with little distribution in the spines (Fig. 7a). In contrast to the relatively diffuse TrkB pattern, PSD-95 staining was present in distinct clusters on the dendritic shafts and spines with low levels of diffuse staining in the dendrites (Fig. 7a), as reported previously34. Treatment of the cultures with Sp-cAMP (10 µM), β-isoproterenol (10 µM) or SKF38393 (100 µM) for 15 min induced changes in TrkB staining, from a diffuse somatodendritic pattern to patches on both the dendritic shaft and spines (Fig. 7a). The TrkB fluorescence spots were colocalized mostly, but not entirely, with PSD-95 clusters, predominantly at dendritic spines. Moreover, treatment with 200 nM KT5720, a specific inhibitor of PKA, before exposure to Sp-cAMP abolished the Sp-cAMP-induced TrkB trafficking and colocalization with PSD-95 (Fig. 7b). These results raise the possibility that activation of the cAMP pathway may facilitate the mobilization of TrkB into PSD-95–containing spines and/or synapses. To investigate how Sp-cAMP could induce TrkB to mobilize into the PSD, we performed coimmunoprecipitation (co-IP) experiments using the lysates from 21-d-old cultured hippocampal neurons. An antibody to PSD-95 (anti–PSD-95) immunoprecipitated PSD-95 together with a significant amount of full-length TrkB receptor from cultures treated with Sp-cAMP or forskolin for 15 min, suggesting that cAMP facilitates the interaction between TrkB and PSD-95 (Fig. 8a). On the other hand, anti–PSD-95 did not coprecipitate a significant amount of TrkB in Rp-cMAP-treated cultures (Fig. 8a). Furthermore, pretreatment with 200 nM KT5720 prevented coimmunoprecipitation of TrkB and PSD95 induced by Sp-cAMP stimulation (Fig. 8a).

Sp-cAMP KT 5720

SKF 38393

β-isoproterenol

Sp-cAMP

β-isoproterenol

SKF 38393

Control

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

Sp-cAMP KT 5720

169

ARTICLES a
145 Total TrkB Anti-TrkB Anti-TrkC 118 66 Anti-PLCγ Anti-NR1 Anti-Shc Anti-Shc Anti-PSD-95 IgG Sp-cAMP Rp-cAMP Forskolin Control Sp-cAMP KT5720 Neg. Whole-cell lysate

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

52 93

Figure 8 cAMP enhances the association of TrkB with PSD-95. Hippocampal neurons cultured for 21 d were treated with the agents indicated for 15 min. (a) Immunoblot analysis of PSD-95 immunoprecipitates. The lysates of hippocampal neurons were precipitated by anti–PSD-95. The blot was probed with the indicated antibodies. Note that PSD-95 antibody immunoprecipitated TrkB only in cells treated with Sp-cAMP or forskolin. Top, total TrkB in the lysate (loading control). Neg, normal mouse IgG used as a negative control. The shadow in the anti-TrkC blot near the 'Forskolin' lane is nonspecific. (b) Immunoblot analysis of TrkB immunoprecipitates. The lysates of hippocampal neurons were precipitated by anti-TrkB. The blot was probed with anti-TrkB or anti–PSD-95. Note that Trk antibody immunoprecipitated PSD-95 only in cells treated with Sp-cAMP or forskolin. Top, total TrkB in the lysate (loading control). Neg, normal rabbit IgG used as a negative control.

IP: anti-PSD-95

b
93

Total PSD-95 Anti-PSD-95 Anti-Trk

IgG Control Rp-cAMP Sp-cAMP Sp-cAMP KT5720 Whole-cell lysate Forskolin Neg.

IP: anti-Trk

To determine the specificity of the co-IP experiments, we examined whether anti–PSD-95 can immunoprecipitate other proteins. After treatment with Sp-cAMP, anti–PSD-95 immunoprecipitated an increased amount of NR1, a subunit of the NMDA (N-methyl-Daspartate) receptor known to interact with PSD-95 (Fig. 8a). However, PSD-95 did not coprecipitate with TrkC, suggesting that the interaction between PSD-95 and TrkB is relatively specific (Fig. 8a). We also examined two proteins known to bind to the activated intracellular domain of TrkB. Quantitative analysis of the co-IP results indicated that Sp-cAMP increased the amount of PLC-γ, but not Shc, in the PSD95-TrkB complex (Fig. 8a). Rp-cAMP had no effect, and the effect of Sp-AMP was reversed when cells were cotreated with KT5720. Finally, we performed co-IP experiments in reverse order: the lysates from cultured neurons (21 d) were immunoprecipitated using anti-Trk and the western blot was probed with anti–PSD-95. Again, treatment with Sp-cAMP or forskolin resulted in association of TrkB with PSD95 (Fig. 8b). Taken together, these results suggest that the cAMP-PKA pathway may selectively enhance the translocation of TrkB into the PSD by facilitating the association of TrkB with the PSD-95 complex. DISCUSSION Commonly known as a second messenger, cAMP also has another important function: to serve as a gate that controls signal flow through other pathways23,26,35. The targets of cAMP include Raf and protein phosphatase I35,36. In the case of gating neurotrophic regulation of synapses, a characteristic feature is that cAMP is required for the full function of a particular neurotrophin (such as BDNF), but cAMP itself

cannot mimic the neurotrophin effects18,37. In this study, we show that cAMP modulates TrkB signaling in two ways. First, cAMP gates BDNFinduced TrkB phosphorylation, showing all three characteristic features: BDNF-induced TrkB phosphorylation is attenuated by inhibitors of cAMP signaling, it is potentiated by cAMP analogs, and activation of the cAMP pathway alone has no effect. Second, cAMP facilitates the distribution of TrkB into the PSD in the spines of hippocampal neurons. Thus, cAMP gating occurs at the TrkB receptor level, rather than at intracellular signaling molecules. There are several advantages for cAMP gating to function at the receptor level. First, because receptors are at the very first step of signal transduction, gating at the receptor level is more powerful and efficient. Second, although intracellular signaling molecules are shared by multiple signaling pathways, TrkB is the specific receptor for BDNF. Thus, cAMP could selectively control BDNF signaling without affecting other systems. For example, NT3-induced activation of TrkC is not affected by cAMP. We further show that norepinephrinergic and dopaminergic agonists mimic the effects of cAMP. Thus, an elevation of intracellular cAMP concentration triggered by afferent inputs may 'gate' TrkB phosphorylation as well as facilitate TrkB trafficking to the PSD, both of which may contribute to the long-term regulation of dendritic spine formation by BDNF. These results may provide important insights into the mechanisms by which cAMP regulates the synaptic actions of BDNF in CNS neurons. It is tempting to consider that pairing of cAMP and BDNF could potentially be used for coincidence detection, analogous to presynaptic activity and postsynaptic membrane depolarization leading to NMDA receptor activation, to mediate activity-dependent, long-term regulation of dendritic spines. An important finding of the present study is that cAMP gates BDNF regulation of dendritic spine formation in the CNS. It is important to note the specificity of the cAMP gating effect. First, cAMP does not gate the acute regulation of LTP and synaptic responses to tetanus by BDNF in neonatal hippocampal synapses. Second, cAMP does not control the long-term regulation of the number of primary dendrites by BDNF in immature hippocampal neurons. A possible interpretation of the results is that the cAMP gating mechanism may exist in mature, but not in immature, hippocampal neurons. Our previous work5 suggests that there is a critical time window to reveal the regulatory effects of exogenous BDNF on synaptic plasticity (synaptic response to tetanus, LTP) in immature hippocampal slices (p11–13). Consistent with these, we found that BDNF-induced TrkB phosphorylation was not modulated by cAMP in younger cultures in which the dendritic branching was measured (3 d) or in neonatal slices (p11–13) in which synaptic plasticity was recorded. Using subcellular fractionation, TrkB was shown to be present in the PSD fraction but only moderately enriched compared with CaM kinase II, GluR1, or GluR2/3 in the adult rat brain32, suggesting that the PSD fraction is a recipient of signals transmitted by BDNF. However,

170

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
full-length TrkB in the rat hippocampus is present both at and outside hippocampal synapses38,39. Rapid gain and loss of receptors at synapses are accounted for by lateral diffusion in the plane of the plasma membrane and by endocytosis and exocytosis38. To elicit synapsespecific modulation, it is crucial that TrkB be localized at the synapses, and preferably controlled by synaptic activity. We showed that TrkB is readily distributed to the dendritic spines in response to cAMP. Within the spines, cAMP facilitates the colocalization, and perhaps association, of TrkB with PSD-95, a key component in the PSD. Such translocation of TrkB into the PSD may represent a mechanism for relatively rapid enhancement of postsynaptic sensitivity to incoming BDNF signaling in the mature hippocampal synapses. Consistent with our observation, intense synaptic activity associated with limbic seizure has been shown to induce a marked accumulation of TrkB in the PSD in the hippocampus40. A brief ischemic episode, which might enhance cAMP concentration, also results in a marked increase of TrkB within PSD33. Thus, it is conceivable that a local increase in cAMP concentration in response to elevated synaptic activity may recruit TrkB into the PSD in the spines, leading to a selective modulation of active synapses. Taken together, our results suggest that two mechanisms, cAMP gating of TrkB phosphorylation and cAMP control of TrkB trafficking to the PSD, may both contribute to the long-term regulation of dendritic spine formation by BDNF. The precise mechanisms by which cAMP recruits TrkB to the PSD remain unknown. Although there is a putative PKA phosphorylation site, ArgProArgThr, containing Thr776 near the C terminus of TrkB, there is no evidence that this threonine is phosphorylated by PKA in cells. We found that upon treatment with cAMP, some TrkB was colocalized with PSD-95, a PSD protein known to control the clustering and anchoring of postsynaptic receptors and ion channels in the dendritic spines41,42. Furthermore, TrkB and PSD-95 could be coimmunoprecipitated in hippocampal neurons treated with cAMP. These results raise the possibility that cAMP induces TrkB translocation into spines by facilitating its interaction with PSD proteins. We do not know, however, whether TrkB directly interacts with PSD-95, or whether additional factors are required to facilitate the formation of a complex containing TrkB and PSD-95. To cluster at the PSD, proteins have to interact with the PSD-95 through their short PDZ-binding motif or consensus sequence (Thr/SerXVal) at the C terminus43. Although TrkB has a SerProVal motif, the valine residue is located in the seventh position, rather than at the end, of the C terminus. Extensive studies have been carried out to identify the proteins interacting with the cytoplasmic domain of TrkB as well as those interacting with PSD-95. So far there is no report of a direct interaction between TrkB and PSD-95. Thus, it is more likely that cAMP and PKA facilitate the translocation of TrkB into spines by more complex mechanisms than simple interaction with PSD-95. METHODS
Cell culture. Primary cultures of hippocampal neurons were plated on poly-Dlysine coated 12-well plates at 500,000 per well for biochemical experiments or on coverslips at 5,000 per coverslip for morphology experiments, as described30. For some experiments, cultures were treated with cAMP agonists forskolin (5 µM, Sigma), Sp-cAMP (10 µM, Biomol Research Laboratories), cAMP antagonist Rp-cAMP (10 µM, Sigma), PKA inhibitor KT5720 (200 nM, Sigma), β-adrenergic agonist β-isoproterenol (10 µM, Sigma) or dopaminergic agonist SKF38393 (100 µM, Sigma) for 15 min before BDNF (gift from Regeneron Pharmaceuticals) treatment. All experimental procedures were under the approval of the Animal Experiment Committee of Chinese Academy of Sciences. Immunoprecipitation and immunoblotting. The dissociated embryonic hippocampal neurons cultured in vitro or hippocampal slices of neonatal rat (p11–13) were lysed in an ice-cold RIPA lysis buffer, as described elsewhere30. The lysates were either mixed with sample buffer for SDS-PAGE, or used for further immunoprecipitation. Antibodies were from Santa Cruz Biotechnology unless indicated otherwise. Lysates were incubated overnight at 4°C with mouse anti–PSD-95 (4 µg, Upstate Biotechnology), rabbit anti-Trk (Ab C14) (4 µg), normal mouse IgG or normal rabbit IgG.. Protein A/G–agarose beads (50 µl) were then added for 2 h at 4 °C, precipitated, washed and resuspended in sample buffer. Samples were boiled for 5 min, subjected to 8% SDS-PAGE, and immunoblotted using rabbit anti–phospho-TrkA (Tyr490) (1:1,000, Cell Signaling), rabbit anti-TrkB (Ab 794, 1:500), rabbit anti–phospho-TrkB (Tyr785) (1:5,000, a gift from B. Sun, Shanghai Institutes of Biological Sciences, Shanghai, China), rabbit anti-TrkC (Ab 798, 1:200), mouse anti–PSD-95 family (PDZ domain) (1:1,000), rabbit anti-NR1 (1:1,000, Chemicon International), mouse anti-Shc (1:1,000, BD Transduction Laboratories), mouse anti–PLC-γ-1 (1:1,000, Upstate Biotechnology) or rabbit anti-Trk (Ab C14) (1:1,000). Horseradish peroxidase– conjugated anti-rabbit or anti-mouse secondary antibodies (1:10,000, Pierce) were used as secondary antibodies. Immunoreactive bands were visualized by enhanced chemiluminescence (ECL, Pierce). Densitometric analysis was conducted using Molecular Analysis software (Bio-Rad). The same experiments (with multiple experiment conditions) were repeated at least six times (n = 6). Surface biotinylation assay. Surface TrkB receptors were measured by biotinylation followed by western blotting using antibodies to TrkB, as described elsewhere29. In brief, various treatments were performed in a 37 °C incubator. For a positive control, a 15-s BDNF treatment was performed with the cell plate on ice with BDNF diluted to 50 ng ml–1 in cell culture medium pre-equilibrated to 21 °C. The cell cultures were quickly rinsed in ice-cold PBS-Ca-Mg (PBS, pH 7.4, containing 1 mM CaCl2 and 0.5 mM MgCl2). Cell surface proteins were biotinylated for 30 min with Sulfo-NHS-LC-Biotin (0.25 mg ml–1, Pierce) diluted in PBSCa-Mg. Biotinylation was stopped by removing the above solution and incubating the cells in 10 mM ice-cold glycine in PBS-Ca-Mg for 20 min. Neurons were then washed with cold PBS-Ca-Mg and lysed with RIPA buffer. Biotinylated proteins (160 µg) were precipitated with ImmunoPure Immobilized Streptavidin (25 µl, Pierce) overnight at 4 °C by constant mixing. The biotinylated protein precipitates were washed with RIPA buffer and processed for western blot analysis. DiI labeling, GFP transfection and spine density analysis. Protrusion (spine and filopodium) density was assessed in fixed neurons post-labeled with DiI, or individual spines were observed in neurons expressing green fluorescent protein (GFP). For the DiI labeling experiment, the hippocampal neurons at 18–21 d in vitro were fixed and labeled with DiI (Molecular Probes) as described22. For the GFP transfection experiment, hippocampal neurons grown in culture for 7 d were transfected with GFP by the calcium phosphate method. The medium-size (10–20 µm) neurons were observed 14 d after the transfection. Confocal images were obtained using a Zeiss confocal microscope (40×, NA 1.30, 488 nm laser, LSM 510). Optical serial sections of 0.5 µm were taken through the cells and reconstructed to yield complete 'three-dimensional' images of individual cells in focus. The densities of filopodia and of spines were measured.. Dendritic protrusions with lengths between 1–5 µm were divided into two categories: spines—stubby, thin, mushroom- and branchedshaped spines previously described in the literature20,21; and filopodia—thin, uniform-caliber, headless protrusions from the dendrites. Spine or filopodium density (number per 10-µm segment) was calculated by dividing the number of spines or filopodia by the length of the segment in micrometers and multiplying by 10. Each segment was counted as an individual observation (n = 1). Immunocytochemistry. The hippocampal neurons at 18–21 d in vitro were fixed as described44, and incubated with rabbit anti-TrkB (1:200, Chemicon), or mouse anti–PSD-95 (1:100, Upstate Biotechnology) in PBS containing 0.2% gelatin overnight at 4 °C. The cultures were washed, and incubated with rhodamine (Jackson Laboratories)- or Alexa Fluor 488 (Molecular Probes)conjugated secondary antibodies diluted in PBS containing 0.2% gelatin for 1 h at room temperature. Images were acquired with a Zeiss 63× (NA 1.40) objective of confocal laser-scanning microscope. Morphometric measurements were performed using Metamorph software (Universal Imaging) by setting a threshold of fluorescence intensity and were automatically counted and logged into Excel. For Figure 7, the yellow spots showing the colocalization for TrkB and PSD-95 were generated using the 'Multiply' function. The number of yellow fluorescence spots was divided by the number of red fluorescence spots indicative of PSD-95 localization. Units of colocalization were measured as a yellow/red fluorescence number normalized to untreated controls.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

171

ARTICLES
Slice preparation and electrophysiology. Hippocampal slices from neonatal rat (p11–13), were bathed in artificial cerebrospinal fluid (ACSF) at 34 °C at 15 ml h–1, and field excitatory postsynaptic potentials (EPSPs) were recorded using an Axoclamp-2B amplifier (Axon Instruments) as described5. All slices were treated with Rp-cAMP (100 µM) or Sp-cAMP (50 µM) with or without BDNF for 2 h before recording. A low concentration of BDNF (2 nM) was used with Sp-cAMP, whereas a higher concentration of BDNF (8 nM) was used with Rp-cAMP experiments. Assay for dendritic growth. Cultured hippocampal neurons at 3 d in vitro were grown in neurobasal medium containing B27 supplements and treated with the indicated agents for 3 d. Cells were stained with mouse anti-MAP2 (1:1,000, Chemicon), followed by DAPI staining. Images were acquired with a Leica microscope (40×). After taking images of MAP2-positive neurons with cell body for diameter (15–20 µm), the number of the primary dendrites per neuron was counted.
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS The authors thank Regeneron Pharmaceuticals for providing recombinant BDNF. This work was supported by funds from the National Institute of Child Heath and Human Development (NICHD) intramural program, Major State Basic Research Program of China (No. G2000077800) and National Natural Science Foundation of China (No. 30228020 and No. 30470533). COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 14 October; accepted 15 November 2004 Published online at http://www.nature.com/natureneuroscience/
16. Brewer, G.J. Serum-free B27/neurobasal medium supports differentiated growth of neurons from the striatum, substantia nigra, septum, cerebral cortex, cerebellum, and dentate gyrus. J. Neurosci. Res. 42, 674–683 (1995). 17. Tartaglia, N. et al. Protein synthesis-dependent and -independent regulation of hippocampal synapses by brain-derived neurotrophic factor. J. Biol. Chem. 276, 37585–37593 (2001). 18. Boulanger, L. & Poo, M. Gating of BDNF-induced synaptic potentiation by cAMP. Science 284, 1982–1984 (1999). 19. Wu, K. et al. Functional trkB neurotrophin receptors are intrinsic components of the adult brain postsynaptic density. Brain Res. Mol. Brain Res. 43, 286–290 (1996). 20. Harris, K.M., Jensen, F.E. & Tsao, B. Three-dimensional structure of dendritic spines and synapses in rat hippocampus (CA1) at postnatal day 15 and adult ages: implications for the maturation of synaptic physiology and long-term potentiation. J. Neurosci. 12, 2685–2705 (1992). 21. Ziv, N.E. & Smith, S.J. Evidence for a role of dendritic filopodia in synaptogenesis and spine formation. Neuron 17, 91–102 (1996). 22. Papa, M., Bundman, M.C., Greenberger, V. & Segal, M. Morphological analysis of dendritic spine development in primary cultures of hippocampal neurons. J. Neurosci. 15, 1–11 (1995). 23. Iyengar, R. Gating by cyclic AMP: expanded role for an old signaling pathway. Science 271, 461–463 (1996). 24. McAllister, A.K., Lo, D.C. & Katz, L.C. Neurotrophins regulate dendritic growth in developing visual cortex. Neuron 15, 791–803 (1995). 25. Gottschalk, W., Pozzo-Miller, L.D., Figurov, A. & Lu, B. Presynaptic modulation of synaptic transmission and plasticity by brain-derived neurotrophic factor in the developing hippocampus. J. Neurosci. 18, 6830–6839 (1998). 26. Blitzer, R.D., Wong, T., Nouranifar, R., Iyengar, R. & Landau, E.M. Postsynaptic cAMP pathway gates early LTP in hippocampal CA1 region. Neuron 15, 1403–1414 (1995). 27. Lezcano, N. & Bergson, C. D1/D5 dopamine receptors stimulate intracellular calcium release in primary cultures of neocortical and hippocampal neurons. J. Neurophysiol. 87, 2167–2175 (2002). 28. Brinton, R.D., Thompson, R.H. & Brownson, E.A. Spatial, cellular and temporal basis of vasopressin potentiation of norepinephrine-induced cAMP formation. Eur. J. Pharmacol. 405, 73–88 (2000). 29. Meyer-Franke, A. et al. Depolarization and cAMP elevation rapidly recruit TrkB to the plasma membrane of CNS neurons. Neuron 21, 681–693 (1998). 30. Du, J., Feng, L., Yang, F. & Lu, B. Activity- and Ca2+-dependent modulation of surface expression of brain-derived neurotrophic factor receptors in hippocampal neurons. J. Cell Biol. 150, 1423–1434 (2000). 31. Haapasalo, A. et al. Regulation of TRKB surface expression by brain-derived neurotrophic factor and truncated TRKB isoforms. J. Biol. Chem. 277, 43160–43167 (2002). 32. Kryl, D. et al. Subcellular localization of full-length and truncated Trk receptor isoforms in polarized neurons and epithelial cells. J. Neurosci. 19, 5823–5833 (1999). 33. Hu, B.R. et al. Assembly of proteins to postsynaptic densities after transient cerebral ischemia. J. Neurosci. 18, 625–633 (1998). 34. Kornau, H.C., Schenker, L.T., Kennedy, M.B. & Seeburg, P.H. Domain interaction between NMDA receptor subunits and the postsynaptic density protein PSD-95. Science 269, 1737–1740 (1995). 35. Chen, J. & Iyengar, R. Suppression of Ras-induced transformation of NIH 3T3 cells by activated G alpha s. Science 263, 1278–1281 (1994). 36. Blitzer, R.D. et al. Gating of CaMKII by cAMP-regulated protein phosphatase activity during LTP. Science 280, 1940–1942 (1998). 37. Tartaglia, N. et al. Protein synthesis dependent and independent regulation of hippocampal synapses by brain-derived neurotrophic factor. J. Biol. Chem. 276, 37585– 37593 (2001). 38. Drake, C.T., Milner, T.A. & Patterson, S.L. Ultrastructural localization of full-length trkB immunoreactivity in rat hippocampus suggests multiple roles in modulating activity-dependent synaptic plasticity. J. Neurosci. 19, 8009–8026 (1999). 39. Patterson, S.L. et al. Some forms of cAMP-mediated long-lasting potentiation are associated with release of BDNF and nuclear translocation of phospho-MAP kinase. Neuron 32, 123–140 (2001). 40. Wyneken, U. et al. Kainate-induced seizures alter protein composition and N-methylD-aspartate receptor function of rat forebrain postsynaptic densities. Neuroscience 102, 65–74 (2001). 41. Scannevin, R.H. & Huganir, R.L. Postsynaptic organization and regulation of excitatory synapses. Nat. Rev. Neurosci. 1, 133–141 (2000). 42. Sheng, M. Molecular organization of the postsynaptic specialization. Proc. Natl. Acad. Sci. USA 98, 7058–7061 (2001). 43. O'Brien, R.J., Lau, L.F. & Huganir, R.L. Molecular mechanisms of glutamate receptor clustering at excitatory synapses. Curr. Opin. Neurobiol. 8, 364–369 (1998). 44. Sala, C. et al. Regulation of dendritic spine morphology and synaptic function by Shank and Homer. Neuron 31, 115–130 (2001).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

1. Lu, B. Acute and long-term regulation of synapses by neurotrophins. Prog. Brain Res. 146, 137–150 (2004). 2. Poo, M. M. Neurotrophins as synaptic modulators. Nat. Rev. Neurosci. 2, 24–32 (2001). 3. Lessmann, V., Gottmann, K. & Heumann, R. BDNF and NT-4/5 enhance glutamatergic synaptic transmission in cultured hippocampal neurons. Neuroreport 6, 21–25 (1994). 4. Takei, N. et al. Brain-derived neurotrophic factor increases the stimulation-evoked release of glutamate and the levels of exocytosis-associated proteins in cultured cortical neurons from embryonic rats. J. Neurochem. 68, 370–375 (1997). 5. Figurov, A., Pozzo-Miller, L., Olafsson, P., Wang, T. & Lu, B. Regulation of synaptic responses to high-frequency stimulation and LTP by neurotrophins in the hippocampus. Nature 381, 706–709 (1996). 6. Cohen-Cory, S. & Fraser, S. E. Effects of brain-derived neurotrophic factor on optic axon branching and remodelling in vivo. Nature 378, 192–196 (1995). 7. Gallo, G. & Letourneau, P.C. Localized sources of neurotrophins initiate axon collateral sprouting. J. Neurosci. 18, 5403–5414 (1998). 8. McAllister, A.M., Katz, L.C. & Lo, D.C. Neurotrophins and synaptic plasticity. Annu. Rev. Neurosci. 22, 295–318 (1999). 9. Cabelli, R.J., Horn, A. & Shatz, C.J. Inhibition of ocular dominance column formation by infusion of NT-4/5 or BDNF. Science 267, 1662–1666 (1995). 10. Tyler, W.J. & Pozzo-Miller, L.D. BDNF enhances quantal neurotransmitter release and increases the number of docked vesicles at the active zones of hippocampal excitatory synapses. J. Neurosci. 21, 4249–4258 (2001). 11. Shimada, A., Mason, C.A. & Morrison, M.E. TrkB signaling modulates spine density and morphology independent of dendrite structure in cultured neonatal Purkinje cells. J. Neurosci. 18, 8559–8570 (1998). 12. Huang, E. J. & Reichardt, L. F. Trk receptors: roles in neuronal signal transduction. Annu. Rev. Biochem. 72, 609–642 (2003). 13. Meyer-Franke, A., Kaplan, M.R., Pfrieger, F.W. & Barres, B.A. Characterization of the signaling interactions that promote the survival and growth of developing retinal ganglion cells in culture. Neuron 15, 805–819 (1995). 14. Song, H.J., Ming, G.L. & Poo, M.M. cAMP-induced switching in turning direction of nerve growth cones. Nature 388, 275–279 (1997). 15. Gaiddon, C., Loeffler, J.P. & Larmet, Y. Brain-derived neurotrophic factor stimulates AP-1 and cyclic AMP-responsive element dependent transcriptional activity in central nervous system neurons. J. Neurochem. 66, 2279–2286 (1996).

172

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Activity-dependent liberation of synaptic neuropeptide vesicles
Dinara Shakiryanova1,4, Arvonn Tully1,4, Randall S Hewes2, David L Deitcher3 & Edwin S Levitan1
Despite the importance of neuropeptide release, which is evoked by long bouts of action potential activity and which regulates behavior, peptidergic vesicle movement has not been examined in living nerve terminals. Previous in vitro studies have found that secretory vesicle motion at many sites of release is constitutive: Ca2+ does not affect the movement of small synaptic vesicles in nerve terminals or the movement of large dense core vesicles in growth cones and endocrine cells. However, in vivo imaging of a neuropeptide, atrial natriuretic factor, tagged with green fluorescent protein in larval Drosophila melanogaster neuromuscular junctions shows that peptidergic vesicle behavior in nerve terminals is sensitive to activity-induced Ca2+ influx. Specifically, peptidergic vesicles are immobile in resting synaptic boutons but become mobile after seconds of stimulation. Vesicle movement is undirected, occurs without the use of axonal transport motors or F-actin, and aids in the depletion of undocked neuropeptide vesicles. Peptidergic vesicle mobilization and post-tetanic potentiation of neuropeptide release are sustained for minutes.

Neuropeptides are synthesized in the neuronal cell body, packaged in large dense core vesicles (LDCVs) and transported to nerve terminals. There they accumulate with little docking until they are released away from active zones, affecting mood, behavior, development and peripheral tissues. This release typically occurs slowly over long periods in response to repetitive firing. In contrast, classical transmitters are packaged in small synaptic vesicles (SSVs) at the nerve terminal and undergo fast phasic release at active zones in response to single action potentials. Furthermore, the supply of SSVs is unlimited because they are locally produced by endocytosis and refilled by transporters. Thus, peptidergic and classical neurotransmission are distinct types of neurosecretion that occur within single nerve terminals1–4. Despite these differences, the release of small-molecule transmitters and the release of peptide transmitters are based on conserved features. For example, the release of both neuropeptides and classical transmitters are triggered by Ca2+ influx through voltage-gated Ca2+ channels. Likewise, both fast and peptidergic transmissions rely on the same SNARE proteins5. Furthermore, secretory vesicle motion is constitutive at many sites of release: Ca2+ affects neither the motion of SSVs in retinal, hippocampal and neuromuscular junction nerve terminals, nor the motion of LDCVs in growth cones and endocrine cells6–11. Finally, both fast and peptidergic neurotransmission are sensitive to patterned electrical activity. For release of classical transmitters, multiple types of synaptic plasticity have been described12. Likewise, the activity dependence of neuropeptide release has been attributed to preparatory, priming and mobilization processes13–15. However, the cell biological mechanisms underlying these regulatory effects have not been defined.

Recently, neuropeptide release has been studied by imaging LDCVs directly in living cells. In vitro experiments have revealed diverse strategies for recruiting secretory vesicles. For example, predocked LDCVs on the surface of endocrine cells dominate the first few minutes of hormone release, whereas cytoplasmic vesicles tend to be depleted much later16–18. In contrast, during peptide release by growth cones, mobile undocked LDCVs are more rapidly depleted19–21. Thus far, the behavior of LDCVs in living synapses has not been examined. In vivo imaging of neuropeptide vesicles is now possible with transgenic animals that express green fluorescent protein (GFP)-tagged peptides22. In D. melanogaster, fluorescence microscopy has been used to detect peptidergic vesicles containing GFP-tagged atrial natriuretic factor (ANF-GFP) in axons and neuromuscular junctions, and Ba2+evoked release of the fluorescent peptide from identified type Ib and III synaptic boutons23 that differ markedly in their accumulation of LDCVs and SSVs24,25. ANF-GFP fluorescence has also been used to assay neuropeptide release evoked by native behaviors26,27. Here we further examine synaptic LDCV behavior and peptide release in D. melanogaster nerve terminals. We report robust depletion of synaptic neuropeptide stores based on Ca2+ influx-triggered liberation of restrained undocked LDCVs. Peptidergic vesicle motion in nerve terminals is undirected and persists for minutes after seconds of activity. RESULTS Activity-induced synaptic neuropeptide vesicle motion Synaptic LDCV motion, measured by fluorescence recovery after photobleaching (FRAP) of the fluorescent neuropeptide ANF-GFP, is not constitutive in predominantly peptidergic D. melanogaster type

1Department of Pharmacology, University of Pittsburgh, Pittsburgh, Pennsylvania 15261, USA. 2Departments of Zoology and Cell Biology, University of Oklahoma, Norman, Oklahoma 73019, USA. 3Department of Neurobiology and Behavior, Cornell University, Ithaca, New York 14853, USA. 4These authors contributed equally to this work. Correspondence should be addressed to E.S.L. ([email protected]).

Published online 9 January 2004; doi:10.1038/nn1377

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

173

ARTICLES
mobilization was abolished by removing extracellular Ca2+ or by adding the Ca2+ channel blocker Cd2+ to the bathing medium (Fig. 2b). Thus, Ca2+ influx evoked by stimulated activity increases synaptic peptidergic vesicle mobility. Physiological motor neuron activity also stimulates synaptic LDCV motion. Glutamatergic larval neuromuscular junctions in vivo are driven by repetitive bursts of motor neuron activity with intraburst frequencies reaching 100 Hz30,31. To preserve this central pattern generator (cpg)-driven motor neuron activity, the nervous system was left intact during the dissection in Ca2+-free medium. Under these conditions, application of Ca2+-containing standard saline re-established cpg-elicited rhythmic muscle contractions. Imaging type Ib boutons between contractions showed that synaptic neuropeptide vesicle mobility increased markedly (Fig. 2c). Thus, bursting activity endogenously generated by the central nervous system induced mobilization of synaptic peptidergic vesicles. Mobilization by liberation of peptidergic vesicles Stimulated neuropeptide vesicle motion seems to be the result of liberation of restrained vesicles. First, the extremely limited extent and slow time course of FRAP suggest that LDCVs are immobilized in resting synaptic boutons. Second, Ca2+-activated peptidergic vesicle movement in boutons is not polarized. The FRAP results mentioned earlier (Fig. 1) show that after stimulation, neuropeptide vesicles move along the lengthwise axonal axis of type III boutons. To test whether these vesicles also move across type III boutons, we used lengthwise edge photobleach profiles (Fig. 3a, left). Complete FRAP occurred upon Ca2+ influx, to again show that essentially all neuropeptide vesicles became mobile (Fig. 3a). More importantly, upon stimulation LDCVs moved radially toward the bouton surface. We also detected with wide-field microscopy vesicle redistribution in type Ib boutons that have central regions nearly devoid of neuropeptide (Fig. 3b). Confocal microscopy of large type Ib boutons verified that neuropeptide vesicles could move into the ‘dark’ central regions upon stimulation. For example, neuropeptide fluorescence in the highlighted central region in the bouton in Figure 3c increased by 216% even though total neuropeptide fluorescence was reduced by 33%. This implies that mobilized peptidergic vesicles moved away from the bouton surface where exocytosis occurred. As previously mobilized neuropeptide vesicles move toward the center and the periphery and along the length of boutons, their motion

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 1 FRAP reveals Ca2+-dependent synaptic neuropeptide vesicle motion in peptidergic type III synapses. (a) FRAP in type III boutons depolarized for 2 min in the presence (left) or absence (right) of Ca2+. Pre, before photobleaching; Post, after photobleaching; 2 min stim., depolarization for 2 min starting 12 min after the photobleach. Other times reflect time elapsed after photobleaching. Size bar, 2 µm. (b) Time course of FRAP obtained with depolarization (indicated by bar) in the presence (b) or absence (ć) of Ca2+ in type III boutons (n = 4). The ratio of fluorescence in the photobleached region to a distal unbleached region of the bouton was calculated and then normalized to the ratio before photobleaching to quantify FRAP. Error bars show s.e.m. in this and all following figures.

III boutons. In resting type III boutons, little FRAP occurred within 10 min (Fig. 1a), indicating that the mobility of synaptic LDCVs is limited. However, stimulation with a depolarizing version of standard saline28 (see Methods) induced an increase in neuropeptide in the photobleached region even though total peptide dropped owing to release (Fig. 1a). The near completeness of FRAP implies that almost all synaptic peptidergic vesicles were mobilized (Fig. 1b). Furthermore, this effect was blocked by removing extracellular Ca2+ (Fig. 1a,b). Thus, LDCV motion in peptidergic synaptic boutons is Ca2+-dependent. This mobilization can be detected without the use of photobleaching in predominantly glutamatergic type Ib boutons that possess some peptidergic vesicles23–25. Figure 2a shows a wide-field image of a resting type Ib bouton along with a pseudocolor image of the change in fluorescence in 3 s (labeled ∆F). Because measured total neuropeptide fluorescence did not change in this short interval, the ∆F panel indicates intrabouton peptidergic vesicle motion. A comparison of the control F and ∆F images indicates that few puncta are mobile. However, after electrical stimulation (70 Hz for 15 s) in the high-Mg2+/Ca2+ saline HL3 (ref. 29), synaptic neuropeptide vesicle motion was enhanced (Fig. 2a, ∆F, Stim.). Again, these data were obtained during a 3-s interval when there was no substantial change in total GFP-tagged neuropeptide fluorescence, indicating that the alteration in ∆F could not be attributed to release. Figure 2 Activity-dependent neuropeptide vesicle motion in type Ib boutons. (a) Top (F), wide-field The mobility illustrated in the ∆F images was neuropeptide vesicle fluorescence images of a type Ib bouton before (Con.) and after a 15-s 70-Hz quantified as a mobility index based on the stimulus (Stim.). Lower (∆F), the change in the top images that occurred in 3 s. Note that release, pixel-by-pixel correlation coefficient between measured as the change in total neuropeptide fluorescence, in the 3-s interval is negligible and so sequential images (see Methods). The mobility the ∆F images reflect vesicle motion. Size bar, 2 µm. (b) Either depolarization or bursting stimulation 70 Hz peak frequency) for 2 min increases neuropeptide vesicle mobility index was increased by depolarization for 2 min (35 Hz mean frequency, in the presence of Ca2+, but this effect is blocked by removal of Ca2+ or addition of 0.5 mM Cd2+ (Fig. 2b). A similar effect was elicited by stimu(n = 6). Open bars, before stimulation; filled bars, after stimulation. (c) Mobilization is induced by lating motor nerves to produce repetitive bursts intrinsic activity. Mobilization is compared before (0 Ca2+) and within 2 min after addition of Ca2+(intraburst frequency 70 Hz, burst duration containing standard saline to the intact nervous system to enable central pattern generator–induced 2 s, mean frequency 35 Hz). However, this rhythmic activity (Ca2+ (cpg)) (n = 4).

174

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
Figure 3 Neuropeptide vesicle movement is undirected after liberation. (a) Depolarization (Stim., indicated by bar in right graph) increases FRAP after tangential lengthwise photobleaching (n = 5) of type III boutons. (b,c) Redistribution of neuropeptide vesicles in type Ib synaptic boutons evoked by 2 min of stimulation (Stim.). (b) Wide-field images before and after bursting electrical stimulation. (c) Confocal images before and after depolarization of a large posterior segment type Ib bouton. Equatorial optical sections of confocal stacks are shown. Note that peptidergic vesicles moved into the central region of the bouton indicated by the white outline. Size bars, 2 µm.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

appears to be undirected as if LDCVs were freed to diffuse within synaptic boutons. This conclusion is supported by pharmacological experiments. First, we examined the effects of N-ethylmaleimide (NEM), a sulfhydryl reagent that inhibits many myosins and the axonal transport motors kinesin and dynein32–34. Earlier work has shown that peptidergic vesicles undergo both anterograde and retrograde transport in larval motor nerves23. We found that NEM irreversibly abolished all axonal transport of peptidergic vesicles (Fig. 4a). However, Ca2+ influx–induced neuropeptide vesicle mobilization and redistribution in type Ib boutons occurred after NEM treatment (Fig. 4b,c). The undirected synaptic neuropeptide vesicle movement that can be induced even after inhibition of many motors is consistent with liberation of restrained LDCVs. Second, we examined the effect of F-actin–depolymerizing drugs. Cytochalasin D treatment, which is known to be effective in type Ib boutons35, did not affect basal or electrically stimulated LDCV motion (Fig. 4d). Basal mobility was also unaffected by mycalolide B, a structurally and mechanistically distinct F-actin–depolymerizing drug that we found effectively blocks muscle contraction (data not shown). In addition, F-actin was difficult to detect in type Ib and III boutons with TRITC-phalloidin labeling (data not shown); its scarcity further suggests the lack of a role for polymerized actin. Because myosins require F-actin for translocation, these results further suggest that mobilization is based on untethering and diffusion rather than motor activation. Activity dependence and persistence of vesicle liberation Synaptic LDCV mobilization is activity dependent, sustained and reversible. Experiments with type Ib boutons showed that substantial activity is required to initiate peptidergic vesicle movement: 1 s of 70-Hz

stimulation did not affect LDCV mobility (Fig. 5a, circles). However, a 5-s 70–Hz stimulus induced an increase in synaptic peptidergic vesicle motion that was evident for minutes (Fig. 5a, triangles). Mobilization seemed to be maximal with a 15-s 70–Hz stimulus (Fig. 5b). Furthermore, LDCV motion in synaptic boutons stimulated by electrical activity (Fig. 5c) or depolarization (data not shown) reversed over a period of many minutes. This explains why mobilization induced by ongoing intrinsic activity (Fig. 2c) was not evident ∼10 min after dissecting animals in Ca2+ free medium. More importantly, this establishes that mobilization is not an irreversible consequence of excitotoxicity, but a persistent physiological activity–dependent response. Peptidergic vesicle mobilization in synapses is not a direct consequence of release. First, although NEM spared LDCV liberation (Fig. 4b), it blocked release as measured by loss of peptide fluorescence from type Ib boutons. Specifically, in contrast to controls (Fig. 6), after NEM treatment, depolarization for 2 min did not decrease synaptic peptide content (∆F equaled –1.9 ± 2.3%, a value statistically indistinguishable from zero). Thus, mobilization must not depend on or require exocytosis. Second, exocytosis is more phase locked
Figure 4 LDCV mobilization persists without axonal transport motor activity and F-actin polymerization. (a) NEM inhibits axonal transport of peptidergic vesicles. Sequential images of a motor nerve, acquired every 3 s, were color-coded red, green and blue, and then superimposed. Note that LDCV images are white in NEM-treated preparation, indicating that the LDCVs were immobile in axons. Size bar, 5 µm. (b) After NEM treatment, neuropeptide vesicle motion in type Ib synaptic boutons was increased by depolarization for 2 min (n = 9). (c) Panels show F and ∆F data from an NEM-treated type Ib bouton. See Figure 2 for explanation and comparison. Size bar, 2 µm. (d) Exposure to 10 mM cytochalasin D for 10 min (Cyto.) did not affect mobility in resting type Ib boutons (Con.) or the effect of 15 s of 70-Hz electrical stimulation (Stim.) (n = 5).

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

175

ARTICLES
Figure 5 Persistence and activity dependence of neuropeptide vesicle liberation in type Ib boutons. (a) Time course of neuropeptide vesicle motion after 1-s (circles) or 5-s (triangles) 70-Hz stimulations (n = 6). (b) Dependence of neuropeptide vesicle motion on duration of 70-Hz stimulation (n = 6). (c) Induction and reversal of mobilization with 2 min of bursting activity (indicated by bar; bursts: 2 s of 70-Hz stimulation, interbursts: 2 s) (n = 6).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

to electrical activity than LDCV mobility, because mobilization lasted for minutes (Fig. 5a,c) whereas release ceased within seconds (Fig. 6a). Therefore, mobilization is mechanistically distinct from exocytosis. Depletion of liberated synaptic neuropeptide vesicles Two independent types of experiments suggest that mobilized undocked synaptic LDCVs are depleted within minutes. First, measurement of Ca2+-dependent neuropeptide release indicates participation of freed peptidergic vesicles. If mobilized cytoplasmic LDCVs do not participate in release, then the paucity of docked vesicles in D. melanogaster type Ib and III synaptic boutons24,25 should be reflected in meager Ca2+-dependent neuropeptide secretion. However, substantial neuropeptide release was triggered within minutes by electrical stimulation of type Ib boutons in HL3 (Fig. 6b,c), and even faster release was induced with the stronger stimulus of depolarizing standard saline (Fig. 6d). The large extent of neuropeptide release evoked within minutes could not have been produced by a small population of predocked LDCVs, indicating avid participation by liberated synaptic neuropeptide vesicles. Second, optical sectioning by confocal microscopy revealed neuropeptide depletion away from the nerve terminal surface. For example, neuropeptide levels in the centers of large type III boutons drop markedly with a 3-min depolarization (Fig. 7a, n = 4). Although light microscopy cannot distinguish between a docked vesicle and a vesicle that is close to but not touching the membrane, the resolution of our optics is sufficient to conclude that any fluorescence more than 290 nm from the bouton surface comes from undocked LDCVs. Thus, the loss of peptide fluorescence in the center of large type III boutons must reflect depletion of undocked peptidergic vesicles. Similar results were obtained in another larger peripheral peptidergic bouton (data not shown, n = 5). Likewise, confocal microscopy showed that robust release is accompanied by loss of neuropeptide from throughout large posterior segment

type Ib boutons (Figs. 7b and 3c) that cannot be accounted for by the modest movement of peptidergic vesicles into the centers of these boutons. Thus, confocal microscopy reveals robust depletion away from the bouton surface that could only occur if LDCVs were mobile. As very few peptidergic vesicles are mobile before stimulation and nearly all LDCVs are very mobile after depolarization (Figs. 1 and 3a), liberated cytoplasmic neuropeptide vesicles must be depleted in synaptic boutons. Post-tetanic potentiation of neuropeptide release Our demonstration that mobilization persists for minutes led us to test for synaptic plasticity on this time scale. First, we identified type Ib boutons that showed little release in response to a 30-s bout of 3 Hz stimulation. Repeating this weak stimulation evoked reproducible release (Fig. 8a). We then examined the effect of a strong stimulation (70 Hz for 15 s) that induces mobilization (Fig. 5b). Release in response to the second bout of 3 Hz stimulation was markedly enhanced 2.5 min after the intense bout of activity (Fig. 8b). This is significant because it demonstrates a change in release associated with long-term mobilization. Typically, measurements of release have guided the study of regulatory mechanisms. However, in this case, monitoring the inner workings of nerve terminals in vivo led to the discovery that peptidergic neurotransmission is subject to minutes-long post-tetanic potentiation. DISCUSSION Neurotransmitter release varies among identified neurons, between classical and peptide cotransmitters, and with patterned activity. Yet the mechanisms that determine the time course and activity dependence of synaptic transmission are not understood. Our experiments with multiple methods and distinct synapse types show that the LDCV motion that aids in robust depletion of synaptic neuropeptide stores is unpolarized and independent of conventional motors. This is reminiscent of SSV behavior in ribbon synapse–containing nerve

Figure 6 Robust neuropeptide release by electrical activity or depolarization. Release was measured as the loss of peptide fluorescence (see Methods). (a) Release by type Ib boutons in response to a 15-s 70-Hz stimulus (n = 5). (b) First minute of release by type Ib boutons induced by 70 Hz stimulation in HL3 (n = 5). (c) Time course of release by type Ib boutons evoked by bursting stimulation (mean frequency 35 Hz) in HL3 (n = 6). (d) Release evoked by type Ib (n = 6) and III (n = 5) boutons in 3 min of depolarizing standard saline. Period of stimulation indicated by bar.

176

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 7 Depletion of synaptic neuropeptide content detected with confocal microscopy. (a) Horizontal (xy) and axial (yz) confocal sections of a large type III bouton showing neuropeptide fluorescence before and after a 3-min treatment with depolarizing standard saline. (b) Horizontal confocal sections of a large type Ib bouton before and after a 3-min treatment with depolarizing standard saline. Size bars, 2 µm.

Figure 8 Post-tetanic potentiation of neuropeptide release by type Ib boutons. (a) Release responses to two trials of 30 s of 3-Hz stimulation (n = 7). The interval between trials was 5 min. (b) Identical stimulation as a except that a 15-s 70-Hz stimulus was delivered 2.5 min before the second trial (n = 8). Note that the tetanus potentiated the response to 3-Hz stimulation. Release was measured as loss of peptide fluorescence.

terminals and LDCVs in growth cones6,7,19,21. However, peptidergic vesicle motion in synaptic boutons is not constitutive. Rather, Ca2+ influx induced by depolarization, electrical stimulation or endogenous central pattern generator–induced rhythmic activity increases LDCV movement in nerve terminals. We found this effect dramatic, because FRAP experiments show that nearly all synaptic LDCVs can be mobilized. Furthermore, confocal microscopy and secretion measurements establish that undocked neuropeptide vesicles are efficiently depleted in the first minutes of release. This could not have occurred if synaptic LDCVs remained immobilized and hence unavailable. Therefore, the simplest explanation for our results is that activity-dependent Ca2+ influx liberates neuropeptide vesicles to diffuse in synaptic terminals to be recruited for exocytosis. The substantial activity requirement for inducing peptidergic vesicle mobilization, the depletion of mobile LDCVs, and the post-tetanic potentiation that accompanies sustained LDCV mobility all indicate that liberation of restrained synaptic neuropeptide vesicles influences the time course of peptidergic transmission. Indeed, mobilization seems to be a type of activity-dependent priming for the undocked reserve neuropeptide vesicles abundant in nerve terminals. Mobilization may also be relevant for other aspects of the cell biology of peptidergic synapses. For example, liberation may facilitate replacement of old synaptic LDCVs destined for retrograde transport with newly synthesized LDCVs undergoing anterograde transport. It is also conceivable that a similar regulation occurs with SSVs. Although activity-dependent SSV motion has not been detected, it has never been measured within the small clusters of undocked SSVs that surround active zones in canonical fast synapses. Rather, previous FRAP studies with hippocampal

neurons and neuromuscular junctions8,9 had only enough resolution to detect lateral motion between these clusters. Recently, it was proposed that SSVs in the readily releasable pool that supports physiological fast transmission are undocked and mobile within the active zone–associated cloud of SSVs36. Thus, it will be important to determine whether the mobilization demonstrated here is unique to synaptic neuropeptide vesicles or operates within the clusters of SSVs adjacent to active zones to potentiate fast transmission. METHODS
Transgenic animals. All experiments used transgenic D. melanogaster (UASAnfGFP, FlyBase ID FBti0026990)23 in which Emerald GFP-tagged atrial natriuretic factor expression was driven by the yeast transcription factor Gal4. To induce expression of the GFP-tagged peptide, these flies were crossed with other transgenic lines that express Gal4 driven by endogenous sequences. For this study, we used the pan-neuronal promoter elav (elav-gal4, FBti0002575) as was done previously23 or a sequence upstream of a prohormone convertase gene (386-gal4, FBti0020938)37. Homozygous female wandering third instar larvae were filleted, and muscle 6, 7 and 12 neuromuscular junction boutons were studied. When necessary for confocal sectioning, the largest type Ib and III boutons were identified based on their GFP fluorescence. Initial experiments were performed in anterior segments A2–A5, but optical sectioning of type Ib boutons was only possible in the more posterior A5 segment where we discovered that these boutons are larger. Solutions and stimulation. Standard saline28 included 128 mM NaCl, 2 mM KCl, 1.8 mM CaCl2, 4 MgCl2, 35.5 mM sucrose, 5 mM sodium HEPES, pH 7.2. The high Mg2+/Ca2+ saline HL329 contained 70 mM NaCl, 5 mM KCl, 1.5 mM CaCl2, 20 mM MgCl2, 10 mM NaHCO3, 5 mM trehalose, 115 mM sucrose, 5 mM sodium HEPES, pH 7.2. Dissections were performed in the presence of 0-Ca saline in which Ca2+ was excluded or substituted with

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

177

ARTICLES
0.5 mM EGTA. Boutons were electrically stimulated via a suction electrode on the motor nerve after reintroduction of Ca2+. To minimize spontaneous muscle contractions that disrupt imaging during electrical stimulation experiments, the ventral ganglion was cut to prevent input from a central pattern generator31, HL3 was used, and 10 mM glutamate was included in the saline to desensitize postsynaptic receptors. Control experiments showed that glutamate did not affect synaptic LDCV movement and was not required for induction of motion. Indeed, data in Figures 1, 3a and 4c were obtained without glutamate. These measures were necessary because electrically evoked neuropeptide release was compromised by cutting nerves. Bursting stimulation was generated by alternating between 2-s trains (for example, 5 V for 0.5 ms at 70 Hz) and 2-s silent periods. In most depolarization experiments, the preparation was switched directly from a 0 Ca saline into ‘depolarizing standard saline’ in which 85 mM NaCl was replaced with KCl. However, the K+/Na+ substitution was performed with HL3 (to generate ‘depolarizing HL3’) for the direct comparison of depolarization to electrical stimulation (Fig. 2b). Because depolarization normally elicits muscle contraction, images were acquired after removing the excess bath KCl except after treatment with 1 mM NEM for 15 min, which irreversibly inhibits muscle contraction. Imaging. Neuromuscular junctions were imaged at room temperature on upright microscopes with direct water immersion objectives. Wide-field epifluorescence data were collected via objectives with numerical apertures (NAs) ranging from 0.9 to 1.1 with cooled CCD cameras with 6.7-µm-wide pixels. FRAP experiments and optical sectioning were performed with a Zeiss Pascal scanning confocal microscope equipped with a 0.95-NA objective. Release was quantified as the loss in bouton neuropeptide fluorescence. This was possible because photobleaching was minimal in these studies. Furthermore, because we imaged axons and multiple en passant boutons in each experiment, our time-lapse imaging revealed that depletion of peptide content occurred without redistribution of peptidergic vesicles into axons, in accordance with previous experiments23. For ease of presentation, only representative single boutons are shown in the figures. Furthermore, some figures use pseudocolor scales generated by Image J or Zeiss Pascal confocal software. Units in these figures are arbitrary, but the vertical dimension is linear from minimal to maximal values in the figure. We initially discovered the effect of activity on peptidergic vesicle motion in type Ib boutons by examining time-lapse movies. However, for analysis, it was necessary to develop an unbiased indicator of vesicle dynamics. Therefore, we made use of the pixel-by-pixel correlation coefficient (CC) between neighboring images, acquired at 3-s intervals, to quantify the degree of vesicle movement. Quantification of such fluctuations is utilized in fluorescence correlation spectroscopy (FCS), a method that is analogous to noise analysis for the study of channels. However, deducing diffusion coefficients by FCS requires rigorous geometrical boundary conditions that are not apparent for type Ib boutons. Therefore, we characterized LDCV motion by calculating a mobility index equal to 1-CC. As this is a statistical value, we acquired a series of images under one experimental condition and used the mean value of the mobility index determined from the series of images. A number of internal controls indicate that the mobility index accurately reflects vesicle motion, and not release. First, relative values agreed with qualitative estimates from inspection of time-lapse movies. Second, the stimulus induced change in the mobility index was unaffected by inhibiting release with NEM. Third, the time course for the change in the mobility index did not parallel the time course of release.
ACKNOWLEDGMENTS This research was supported by US National Institutes of Health grant NS32385 (to E.S.L.) and Oklahoma Center for Science and Technology grant HR03-048S (to R.S.H.). We thank C. Ziegler for technical assistance. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 23 September; accepted 26 October 2004 Published online at http://www.nature.com/natureneuroscience/
4. Whim, M.D. & Lloyd, P.E. Frequency-dependent release of peptide cotransmitters from identified cholinergic motor neurons of Aplysia. Proc. Natl. Acad. Sci. USA 86, 9034–9038 (1989). 5. Martin, T.F. The molecular machinery for fast and slow neurosecretion. Curr. Opin. Neurobiol. 4, 626–632 (1994). 6. Holt, M., Cooke, A., Neef, A. & Lagnado, L. High mobility of vesicles supports continuous exocytosis at a ribbon synapse. Curr. Biol. 14, 173–183 (2004). 7. Rea, R. et al. Streamlined synaptic vesicle cycle in cone photoreceptor terminals. Neuron 41, 755–766 (2004). 8. Henkel, A.W., Simpson, L.L., Ridge, R.M. & Betz, W.J. Synaptic vesicle movements monitored by fluorescence recovery after photobleaching in nerve terminals stained with FM1-43. J. Neurosci. 16, 3960–3967 (1996). 9. Kraszewski, K., Daniell, L., Mundigl, O. & DeCamilli, P. Mobility of synaptic vesicles in nerve endings monitored by recovery from photobleaching of synaptic vesicle-associated fluorescence. J. Neurosci. 16, 5905–5913 (1996). 10. Ng, Y.K., Lu, X. & Levitan, E.S. Physical mobilization of secretory vesicles facilitates neuropeptide release by nerve growth factor-differentiated PC12 cells. J. Physiol. 542, 395–402 (2002). 11. Becherer, U., Moser, T., Stuhmer, W. & Oheim, M. Calcium regulates exocytosis at the level of single vesicles. Nat. Neurosci. 6, 846–853 (2003). 12. Zucker, R.S. & Regehr, W.G. Short-term synaptic plasticity. Annu. Rev. Physiol. 64, 355–405 (2002). 13. Seward, E.P., Chernevskaya, N.I. & Nowycky, M.C. Exocytosis in peptidergic nerve terminals exhibits two calcium-sensitive phases during pulsatile calcium entry. J. Neurosci. 15, 3390–3399 (1995). 14. Brezina, V., Church, P.J. & Weiss, K.R. Temporal pattern dependence of neuronal peptide transmitter release: models and experiments. J. Neurosci. 20, 6760–6772 (2000). 15. Ludwig, M. et al. Intracellular calcium stores regulate activity-dependent neuropeptide release from dendrites. Nature 418, 85–89 (2002). 16. Steyer, J.A., Horstmann, H. & Almers, W. Transport, docking and exocytosis of single secretory granules in live chromaffin cells. Nature 388, 474–478 (1997). 17. Olofsson, C.S. et al. Fast insulin secretion reflects exocytosis of docked granules in mouse pancreatic B-cells. Pflugers Arch. 444, 43–51 (2002). 18. Duncan, R.R. et al. Functional and spatial segregation of secretory vesicle pools according to vesicle age. Nature 422, 176–180 (2003). 19. Han, W., Ng, Y.K., Axelrod, D. & Levitan, E.S. Neuropeptide release by efficient recruitment of diffusing cytoplasmic secretory vesicles. Proc. Natl. Acad. Sci. USA 96, 14577–14582 (1999). 20. Ng, Y.K. et al. Unexpected mobility variation among individual secretory vesicles produces an apparent refractory neuropeptide pool. Biophys. J. 84, 4127–4134 (2003). 21. Burke, N.V. et al. Neuronal peptide release is limited by secretory granule mobility. Neuron 19, 1095–1102 (1997). 22. Levitan, E.S. Using GFP to image peptide hormone and neuropeptide release in vitro and in vivo. Methods 33, 281–286 (2004). 23. Rao, S., Lang, C., Levitan, E.S. & Deitcher, D.L. Visualization of neuropeptide expression, transport, and exocytosis in Drosophila melanogaster. J. Neurobiol. 49, 159–172 (2001). 24. Atwood, H.L., Govind, C.K. & Wu, C.F. Differential ultrastructure of synaptic terminals on ventral longitudinal abdominal muscles in Drosophila larvae. J. Neurobiol. 24, 1008–1024 (1993). 25. Jia, X.X., Gorczyca, M. & Budnik, V. Ultrastructure of neuromuscular junctions in Drosophila: comparison of wild type and mutants with increased excitability. J. Neurobiol. 24, 1025–1044 (1993). 26. Husain, Q.M. & Ewer, J. Use of targetable GFP-tagged neuropeptide for visualizing neuropeptide release following execution of a behavior. J. Neurobiol. 59, 181–191 (2004). 27. Heifetz, Y. & Wolfner, M.F. Mating, seminal fluid components, and sperm cause changes in vesicle release in the Drosophila female reproductive tract. Proc. Natl. Acad. Sci. USA 101, 6261–6266 (2004). 28. Jan, L.Y. & Jan, Y.N. Properties of the larval neuromuscular junction in Drosophila melanogaster. J. Physiol. 262, 189–214 (1976). 29. Stewart, B.A., Atwood, H.L., Renger, J.J., Wang, J. & Wu, C.F. Improved stability of Drosophila larval neuromuscular preparations in haemolymph-like physiological solutions. J. Comp. Physiol. A 175, 179–191 (1994). 30. Barclay, J.W., Atwood, H.L. & Robertson, R.M. Impairment of central pattern generation in Drosophila cysteine string protein mutants. J. Comp. Physiol A. 188, 71–78 (2002). 31. Cattaert, D. & Birman, S. Blockade of the central generator of locomotor rhythm by noncompetitive NMDA receptor antagonists in Drosophila larvae. J. Neurobiol. 48, 58–73 (2001). 32. Pfister, K.K., Wagner, M.C., Bloom, G.S. & Brady, S.T. Modification of the microtubulebinding and ATPase activities of kinesin by N-ethylmaleimide (NEM) suggests a role for sulfhydryls in fast axonal transport. Biochemistry 28, 9006–9012 (1989). 33. Phelps, K.K. & Walker, R.A. N-ethylmaleimide inhibits Ncd motor function by modification of a cysteine in the stalk domain. Biochemistry 38, 10750–10757 (1999). 34. Perry, S.V. & Cotterill, J. The action of thiol reagents on the adenosine-triphosphatase activities of heavy meromyosin and L-myosin. Biochem. J. 96, 224–230 (1965). 35. Delgado, R., Maureira, C., Oliva, C., Kidokoro, Y. & Labarca, P. Size of vesicle pools, rates of mobilization, and recycling at neuromuscular synapses of a Drosophila mutant, shibire. Neuron 28, 941–953 (2000). 36. Rizzoli, S.O. & Betz, W.J. The structural organization of the readily releasable pool of synaptic vesicles. Science 303, 2037–2039 (2004). 37. Bantignies, F., Goodman, R.H. & Smolik, S.M. Functional interaction between the coactivator Drosophila CREB-binding protein and ASH1, a member of the trithorax group of chromatin modifiers. Mol. Cell Biol. 20, 9317–9330 (2000).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

1. Zupanc, G.K. Peptidergic transmission: from morphological correlates to functional implications. Micron. 27, 35–91 (1996). 2. Hokfelt, T., Broberger, C., Xu, Z.Q., Sergeyev, V., Ubink, R. & Diez, M. Neuropeptides— an overview. Neuropharmacology 39, 1337–1356 (2000). 3. Taghert, P.H. & Veenstra, J.A. Drosophila neuropeptide signaling. Adv. Genet. 49, 1–65 (2003).

178

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Endocytosis-dependent desensitization and protein synthesis–dependent resensitization in retinal growth cone adaptation
Michael Piper1,2, Saif Salih1,2, Christine Weinl1, Christine E Holt1 & William A Harris1
It has been proposed that growth cones navigating through gradients adapt to baseline concentrations of guidance cues. This adaptation process is poorly understood. Using the collapse assay, we show that adaptation in Xenopus laevis retinal growth cones to the guidance cues Sema3A or netrin-1 involves two processes: a fast, ligand-specific desensitization that occurs within 2 min of exposure and is dependent on endocytosis, and a slower, ligand-specific resensitization, which occurs within 5 min and is dependent upon protein synthesis. These two phases of adaptation allow retinal axons to adjust their range of sensitivity to specific guidance cues.

The continuous receptor-mediated signaling that occurs when a cell interacts with its environment can be regulated by adaptation. From the immune system to the nervous system, the result of the adaptation process is usually a resetting of sensitivity1,2. Adaptation seems to be especially crucial for the chemotropic responses of cells, including bacteria3 and macrophages4,5, in gradients of attractants or repellents. It seems reasonable to expect that adaptation also has a role in the chemotropic responses of growth cones. Previous work has shown that retinal growth cones launched on a platform of a repulsive guidance factor could grow further up an increasing gradient of the factor than could growth cones that were not launched on the platform, demonstrating that adaptation can extend the sensitivity of axons to guidance cues6. This adaptation process could be used for axonal orientation in gradients of guidance molecules, such as the ones retinal axons encounter in the tectum. Others have argued that adaptation may also be important along the pathway where axons meet various guidance cues and might need to adjust their sensitivity to navigate correctly7. Adaptation is often associated with a fast desensitization response and a slower resensitization response8. The phenomenon of growth cone desensitization was first demonstrated in chick retinal growth cones that failed to respond to a collapse-inducing signal after it was repeatedly presented9,10. Desensitization has also been shown to be involved in enabling growth cones to move on from attractive intermediate targets. Netrin-1 attracts commissural axons to the midline of the spinal cord, yet once these axons are exposed to netrin-1 they lose responsiveness to it11. Desensitization and resensitization have also been described in embryonic X. laevis spinal growth cones exposed to attractants12. Exposure to low levels of an attractant (either BDNF or netrin-1) for 30 min caused growth cones to fail to turn towards a

gradient of this factor, but if the exposure was continued for another 30–60 min, the growth cones recovered their ability to respond. Are the sequential processes of desensitization and resensitization part of a homeostatic reset mechanism that allows a growth cone to transiently turn its response off, then on again, so that it can sense the same guidance cue in the same way if it is presented a second time? Or do these processes allow a growth cone to adjust its sensitivity appropriately to particular background levels of a guidance cue? Here, we report on the cellular mechanisms underlying desensitization and resensitization in retinal growth cones, which allow them to adjust their sensitivity down and up as a function of exposure to a specific ligand. RESULTS Rapid desensitization and resensitization in growth cones We investigated the time course of desensitization and resensitization of retinal growth cones to guidance cues using collapse assays. We chose collapse assays because they are rapid, enabling the repulsive chemotropic responses of large numbers of growth cones to be monitored within minutes. X. laevis retinal growth cones undergo collapse in response to Sema3A, detectable within 2 min of exposure and maximal at 10 min13. Collapse is probably related to repulsive turning responses, as guidance cues that elicit repulsive turning when presented in a gradient usually cause collapse when added uniformly. The difference between a repulsive turn and collapse may, therefore, simply be the difference between a polarized and a global response. Such assays allowed us to measure the time course of desensitization and resensitization accurately (as shown in Fig. 1a). We exposed growth cones to a low dose of a guidance molecule that produced a minimal amount of growth cone collapse for various periods of time. This was

1Department of Anatomy, University of Cambridge, Downing Street, Cambridge CB2 3DY, United Kingdom. 2These authors contributed equally to this work. Correspondence should be addressed to W.H. ([email protected]).

Published online 9 January 2005; doi: 10.1038/nn1380

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

179

ARTICLES a
Dose of chemotropic agent

LD

HD

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

5 4 3 2 1 Pretreatment (min)

0

10 Collapse assay (min) Fix and determine total growth cone collapse
HD Sema3A HD netrin-1

b
Growth cone collapse (%)

75 70 65 60 55 50 45 40 35 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0

Figure 1 Adaptation time course for retinal growth cones. (a) To investigate the time course of desensitization and resensitization in retinal growth cones, explant cultures were pretreated with a low dose (LD) of either Sema3A or netrin-1 for 0.5–5 min, before a high dose (HD) of the same molecule was added for 10 min. Cultures were then fixed and the percentage of axons with collapsed growth cones determined. The LD was selected because it caused minimal collapse compared to the basal level of collapse shown by cultures treated with control medium (see Methods). (b) Pretreatment with a LD of Sema3A for 2 min or a LD of netrin-1 for 1 min caused a rapid desensitization to a HD, resulting in significantly lower growth cone collapse. After 5 min, growth cones had fully resensitized to a LD of either Sema3A or netrin-1, with collapse rates equivalent to cultures treated with a HD of either cue alone. Asterisks indicate a significant difference (*P < 0.05, **P < 0.005) between cultures that received both a LD pretreatment and a HD and those that received a HD alone; Mann-Whitney U-test. Error bars, s.e.m.

*

** ** **

** **

**

4.5

5.0

LD pretreatment (min)

immediately followed by exposure to a high dose that consistently produced a maximal collapse response. A pretreatment time of 2 min to a low dose of either Sema3A or netrin-1 caused a marked decrease in collapse in response to the high dose of either molecule, demonstrating that growth cones become rapidly desensitized (Fig. 1b). To investigate the time course of this effect more fully, pretreatment times were varied from 30 s to 30 min. Pretreatment with a low dose of Sema3A resulted in desensitization within 90 s, with maximal desensitization at 2 min. Growth cones then showed steady resensitization, showing a maximal response to a high dose after 4 min (Fig. 1b). Netrin-1 produced even faster desensitization, with growth cones being maximally insensitive after only 1 min of low-dose pretreatment and becoming resensitized within 5 min (Fig. 1b). Pretreatments for times ranging from 10–30 min did not show any further significant changes, suggesting that the resensitization process is complete by 5 min. Recalibration or homeostatic reset? What is the function of growth cone desensitization and resensitization? Is it a homeostatic reset mechanism, whereby a growth cone that

collapses in response to a dose of a guidance factor is able to recover from this encounter by first disengaging from the collapse-inducing factor and then re-engaging its response elements so that the same dose of a collapse factor can subsequently lead to the same response? Or are the processes of desensitization and resensitization used to adjust the sensitivity of a growth cone so that when a growth cone is exposed to a particular background level of a guidance cue, it adjusts or recalibrates its sensitivity? This latter recalibration mechanism is what is known, in the field of bacterial research, as adaptation, as it allows bacteria to climb up concentration gradients of attractants. If adaptation in retinal growth cones is a recalibration process, then a growth cone exposed to a background level of a repellent such as Sema3A should need more Sema3A than a naive growth cone to cause the same collapse response, and such exposed growth cones should respond differentially to larger amounts of Sema3A that unexposed growth cones could not differentiate: that is, the dose-response curve should be shifted and extended to higher doses in adapted as compared to naive growth cones. To test this idea directly, we compared the responses to a variety of doses of Sema3A in collapse assay of (i) naive growth cones and (ii) growth cones exposed to a low-dose pretreatment with Sema3A for 5 min (allowing resensitization). Naive growth cones showed more collapse than growth cones exposed to the low-dose pretreatment at all doses of Sema3A except the high dose (Fig. 2). Furthermore, naive growth cones also seemed to saturate their response at lower doses than exposed

Figure 2 Adaptation adjusts sensitivity. Growth cones exposed to a low dose (LD) of Sema3A for 5 min (filled circles), allowing both desensitization and resensitization, were compared to naive growth cones (open squares) on a range of Sema3A doses. The level of Sema3A in the high dose (HD) is normalized to 1.0 and the other doses are expressed as proportions of the normalized HD, with the LD being equivalent to 0.4. Adapted growth cones showed less collapse than naive growth cones at all doses except the HD. The response of naive growth cones saturated at lower doses than the adapted growth cones. At high doses of Sema3A (0.86–1.0), adapted growth cones showed a differential response, but the responses of naive growth cones appeared to saturate at 0.86. Asterisks indicate significant difference (*P < 0.05, **P < 0.01), Mann-Whitney U-test. Error bars, s.e.m.

180

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES a
Growth cone collapse (%) 70 65 60 55 50 45 40 35 1 1.5 2 2.5 3 3.5 4 * Figure 3 Resensitization is blocked by protein synthesis inhibitors. (a,b) Cultured retinal explants were pretreated with a low dose (LD) of either Sema3A or netrin-1 in the presence of cycloheximide (CHX), an inhibitor of ribosomal translation. Cultures were rinsed after pretreatment to ensure no pharmacological reagents were present during the collapse assay. Inhibition of mRNA translation during pretreatment with a LD of Sema3A (a) or a LD of netrin-1 (b) prevented resensitization, but not desensitization, to a high dose (HD) of the respective molecule. (c) Similarly, pretreatment with anisomycin (Aniso), another protein synthesis inhibitor, and a LD of either Sema3A or netrin-1 for 5 min, prevented resensitization of growth cones. Protein synthesis may be essential to the reacquisition of responsiveness to guidance cues after desensitization. Asterisks indicate significant difference at P < 0.05, Mann-Whitney U-test. Error bars, s.e.m.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Pretreatment (min) LD Sema3A CHX LD Sema3A + CHX

b
Growth cone collapse (%)

75 70 65 60 55 50 45 40 35 0.5 1 2 3 Pretreatment (min) CHX 4 5 *

LD netrin-1

LD netrin-1 + CHX

c
Growth cone collapse (%)
70 65 60 55 50 45 40 35 30

or netrin-1 was added in the absence of cycloheximide. The presence of cycloheximide during low-dose pretreatment did not alter desensitization of growth cones to either guidance cue (Fig. 3a,b). However, resensitization of growth cones did not occur when protein synthesis was inhibited during pretreatment (Fig. 3a,b). This implies that retinal growth cone resensitization, but not desensitization, is dependent on local protein synthesis. Inhibition of growth cone resensitization was also observed when the reversible inhibitor of peptidyltransferase activity on the ribosomes, anisomycin, was added during the pretreatment period (Fig. 3c). The rapid reversibility of cycloheximide and anisomycin is critical for these experiments, and is shown by the fact that axons pretreated with either cycloheximide or anisomycin alone, for various periods, show full collapse when exposed to a high dose of either netrin-1 or Sema3A, whereas exposure of growth cones to these guidance cues in the presence of cycloheximide or anisomycin prevents collapse14. The findings support the hypothesis that protein synthesis is required for growth cone resensitization12. Desensitization is dependent on endocytosis As desensitization of retinal growth cones was not dependent on protein synthesis, a different cellular mechanism must underlie this aspect of adaptation. Sema3A-induced collapse of neuronal growth cones is accompanied by increased endocytosis15,16, and the Sema3A receptors neuropilin-1 and plexin colocalize with endocytic vacuoles after Sema3A stimulation16. To test whether desensitization by Sema3A or netrin-1 is dependent on endocytosis, growth cones were pretreated with a low dose of either Sema3A or netrin-1 in the presence of phenylarsine oxide (PAO), an inhibitor of receptor-mediated endocytosis5,17. Pretreatment of cultures with a low dose of Sema3A or netrin-1 and PAO clearly inhibited desensitization (Fig. 4a,b), illustrating that receptor-mediated endocytosis may be an essential step in growth cone desensitization. Multiple endocytic pathways are used in biological systems. Clathrinmediated endocytosis targets proteins to the endosomal pathway, whereas lipid raft and caveolar pathways act as alternative means of trafficking18. As PAO nonspecifically inhibits endocytosis, we used a more selective reagent. MDC is a competitive inhibitor of transglutaminase, an enzyme essential for the formation of clathrin-coated vesicles19,20. Pretreatment with MDC and a low dose of either molecule also suppressed growth cone desensitization (Fig. 4c), suggesting that endocytosis induced by Sema3A and netrin-1 occurs in a clathrin-dependent fashion. Endocytosis of receptors and desensitization Neuropilin-1, a transmembrane glycoprotein that interacts with Sema3A21,22, forms a receptor complex with another transmembrane protein, plexin A1 (ref. 23). Binding of Sema3A to neuropilin-1 causes the dissociation of plexin from the receptor complex and initiates

*

* HD Sema3A HD netrin-1

LD Sema3A

Aniso + LD Sema3A

Pretreatment (5 min)

growth cones, showing that exposed growth cones can distinguish these higher doses, and suggesting that exposure to a low dose shifts the dose-response curve towards higher levels of the collapsing factor (Fig. 2). These results support a model in which desensitization and resensitization are components of an adaptation process that is involved in adjusting the sensitivity of growth cones in a way that could enable them navigate in gradients of guidance cues6,7. Resensitization requires protein synthesis To determine whether protein synthesis is involved in desensitization or resensitization, we blocked protein synthesis during the low-dose exposure by adding cycloheximide, a reversible inhibitor of the protein translocation reaction on ribosomes. At the end of all pretreatments involving pharmacological reagents, cultures were rinsed in inhibitor-free medium (see Methods). Then a high dose of Sema3A

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

LD netrin-1

Aniso + LD netrin-1

Aniso

Aniso

181

ARTICLES a
75 Growth cone collapse (%) 70 65 60 55 50 45 40 35 1 1.5 2 2.5 3 3.5 4 * * Figure 4 Desensitization is blocked by endocytosis inhibitors. (a,b) Cultured retinal explants were pretreated with a low dose (LD) of Sema3A (a) or a LD of netrin-1 (b) in the presence of PAO, an inhibitor of endocytosis. Cultures were rinsed after pretreatment. Inhibition of endocytosis during pretreatment prevented desensitization of growth cones to a high dose (HD) of the respective molecule, suggesting that endocytosis is necessary for this aspect of adaptation. (c) MDC also inhibited desensitization induced by a LD of either netrin-1 (1 min) or Sema3A (2 min) during pretreatment, demonstrating that clathrin-mediated endocytosis may underlie desensitization. Asterisks indicate significant difference (*P < 0.05, **P < 0.01), Mann-Whitney U-test. Error bars, s.e.m.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Pretreatment (min) LD Sema3A PAO LD Sema3A + PAO

b
Growth cone collapse (%)

75 70 65 60 55 50 45 40 35 0.5 1 2 3 Pretreatment (min) PAO 4 5 * *

domain of DCC after exposure to a low dose of netrin-1. Again, the removal and replacement of DCC from the surface of the growth cone mirrored the time course for desensitization and resensitization (Fig. 6a–e). We next assessed whether surface receptor loss was dependent on endocytosis by applying inhibitors. Indeed, PAO abolished Sema3Ainduced neuropilin-1 depletion at the cell surface (Fig. 5f,i–k). Similar results for the netrin-1 receptor, DCC (Fig. 6f,i–k), suggest that receptor endocytosis in growth cones may mediate desensitization. Replacement of receptors and protein synthesis Because desensitization was correlated with the endocytosis of receptors at the cell surface, we wondered whether the reappearance of receptors at the cell surface was dependent on protein synthesis during resensitization. We therefore exposed growth cones to a low dose of Sema3A + cycloheximide. A 2-min exposure showed that neuropilin-1 removal from the cell surface (Fig. 5g,k) was unaffected, which means that protein synthesis is not involved in receptor removal. What was more interesting was the reappearance of at least some of the receptor after 5 min of treatment with a low dose of Sema3A + cycloheximide (Fig. 5h,k). Similar experiments were performed after growth cone exposure to a low dose of netrin-1 + cycloheximide. Removal of DCC from the growth cone surface was not dependent on protein synthesis (Fig. 6g,k). In this case, too, we observed a partial recovery of DCC to the growth cone surface after 5 min of treatment with a low dose of netrin-1 + cycloheximide (Fig. 6h,k). These results suggest that protein synthesis may be required for the reappearance of receptors on the growth cone surface. However, the partial recovery of cell surface receptor in the absence of protein synthesis suggests that some of the reappearance is due to recycling from the endosomal compartment back to the plasma membrane. Desensitization and resensitization are ligand specific To test whether the adaptation process was ligand specific, we carried out collapse assays in which cultures were pretreated with a low dose of Sema3A and then treated with a high dose of netrin-1, or vice versa. Pretreatment with a low dose of Sema3A for 2 min did not cause desensitization to a high dose of netrin-1, and similarly, pretreatment with a low dose of netrin-1 (2 min) did not desensitize growth cones to a high dose of Sema3A (Fig. 7a). These data suggest that adaptation is ligand specific and are consistent with the idea that the loss of receptor from the surface mediates desensitization. Resensitization is protein synthesis dependent, whereas receptor replacement is only partially dependent on protein synthesis. Therefore, resensitization may rely, at least in part, on the synthesis of molecules that are common to both the Sema3A and netrin-1 pathways. For example, it could be that increased translation of cytoskeletal proteins such as β-actin could be involved in resensitization. If so, then resensitization should not be ligand specific. To test this, we performed a 5-min low-dose pretreatment with one molecule to allow complete resensitization

LD netrin-1

LD netrin-1 + PAO HD Sema3A HD netrin-1

c
Growth cone collapse (%) 70 65 60 55 50 45 40
MDC + LD Sema 3A LD Sema3A MDC + LD netrin-1 MDC LD netrin-1 MDC

*

**

35

2 min

Pretreatment

1 min

second-messenger pathways24,25. We therefore wanted to know if the endocytic removal of neuropilin-1 from the cell surface mediated desensitization to Sema3A. To investigate this, we exposed cultures to a low dose of Sema3A for times ranging from 2 to 10 min and subsequently analyzed the abundance of cell surface neuropilin-1 on growth cones by quantitative immunohistochemistry, using an antibody against the extracellular domain of neuropilin-1 in nonpermeabilized conditions (Fig. 5). Addition of a low dose of Sema3A for 2 min significantly diminished the amount of cell surface neuropilin-1 present on axonal growth cones (Fig. 5a–e). After 5 min treatment, cell surface neuropilin-1 reactivity had returned to levels seen before stimulation. Thus, the time course for removal and replacement of neuropilin-1 matches that course for desensitization and resensitization. Similar experiments were performed using an antibody to the extracellular

182

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
Fluorescent intensity (% control)

a

b

c

d

e

120 100 90 80 70 60 50 Control 2 5 10 LD Sema3A treatment (min)

**

0 min 2 min 5 min 10 min Neuropilin-1 immunoreactivity after LD Sema3A treatment

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

f

g

h

i

j

k

Fluorescent intensity (% control)

120 100 80 60 40

0 min

2 min

5 min

2 min

5 min

+ CHX + PA O Neuropilin-1 immunoreactivity after LD Sema3A treatment

**
Control 2 5 + CHX 2 + PAO 5

Figure 5 Neuropilin-1 depletion and replacement in adaptation. (a–e) Retinal explant cultures were treated with a low dose (LD) of Sema3A and the amount of neuropilin-1 localized on the surface of the growth cone was assessed by quantitative immunohistochemistry. After 2 min the level of cell surface neuropilin-1 was significantly lower than for the control, but after 5 min it had fully recovered. (f–k) Treatment in the presence of cycloheximide (CHX) did not affect the rapid removal of neuropilin-1 from the cell surface after 2 min, but seemed to partially limit its replacement into the membrane after 5 min (f–h,k). However, treatment in the presence of PAO abolished the reduction in neuropilin-1 surface staining in response to a LD of Sema3A (f,i–k). **P < 0.001 in comparison to controls, Student’s t-test. Error bars, s.e.m. Scale bar, 10 µm.

LD Sema3A treatment (min)

and then tested with the other molecule. If there were cross-resensitization, we would expect a higher than normal collapse rate. We found no evidence of cross-resensitization (Fig. 7b), which suggests that resensitization, like desensitization, may also be ligand specific (Fig. 7b). In this experiment, however, it is possible that the collapse response was saturated. To address this potential difficulty, we used an intermediate dose rather than a high dose of Sema3A after conditioning the growth cone to a low dose of netrin-1. The intermediate dose of Sema3A caused submaximal collapse in naive neurons, and this was not significantly altered in growth cones adapted to a low dose of netrin-1: that is, there is no indication of cross-resensitization (Fig. 7c). That both phases of adaptation are ligand specific implies that growth cones can fully adapt to one ligand, or family of related ligands, without changing their sensitivity to other ligands. DISCUSSION We suggest that desensitization and resensitization are two phases of an adaptational mechanism that can be used to reset the sensitivity of growth cones as they grow through increasing concentration of ligands, such as the gradients of ephrins in the tectum, or along the pathway as they navigate a b in an epithelium that exposes them to a multitude of different guidance factors

at different concentrations. This suggestion is based on findings that retinal growth cones pretreated with a low dose of a repellent showed a rapid, endocytosis-dependent desensitization in response to a high dose in a collapse assay. Resensitization was also very rapid, with growth cones regaining full responsiveness to collapse within 5 min of pretreatment. Protein synthesis was essential for resensitization. We also found that neuropilin-1 and DCC, the Sema3A and netrin-1 receptors, undergo an endocytosis-dependent depletion from the growth cone surface within minutes of exposure to their respective guidance cue, followed by a partially protein synthesis–independent reinsertion into the plasma membrane of the growth cone within 5 min of exposure. Finally, we found that these responses are ligand specific and that exposure to a low dose of Sema3A shifts the sensitivity of growth cones to higher doses of Sema3A. By exposing developing spinal neurons to a ligand such as netrin1 for 30 min, earlier researchers12 showed that these growth cones attenuate their capacity to turn in response to guidance cues. They also showed that these growth cones recover their sensitivity in a protein synthesis-dependent manner after 60–90 min. Clearly, the time
Fluorescent intensity (% control)

c

d

e

120 100 90 80 70 60 50 40 Control 1 5 10 LD netrin-1 treatment (min)

Figure 6 DCC depletion and replacement in adaptation. (a–e) Retinal explant cultures were treated with a low dose (LD) of netrin-1 and quantitative immunohistochemistry was used to assess the amount of DCC localized on the surface of the growth cone. After 1 min the level of cell surface DCC was significantly lower than the control, whereas after 5 min it had fully recovered. (f–k) Treatment in the presence of CHX did not affect the rapid removal of DCC from the cell surface after 1 min; however, inhibition of protein synthesis clearly limited the reinsertion of DCC into the surface membrane (f–h,k). PAO abolished the reduction in DCC surface staining in response to a LD of netrin-1 (f,i–k). **P < 0.001, *P < 0.05 in comparison to controls, Student’s t-test. Error bars, s.e.m. Scale bar, 10 µm.

0 min

1 min

5 min

10 min

**

DCC immunoreactivity after LD netrin-1 treatment

f

g

h

i

j

k
Fluorescent intensity (% control) 120 100 90 80 70 60 50 Control

* **
1 5 1 5 + CHX + PAO LD netrin-1 treatment (min)

0 min

1 min

5 min + CHX

1 min

5 min + PAO

DCC immunoreactivity after LD netrin-1 treatment

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

183

ARTICLES
Figure 7 Adaptation is ligand specific. (a) Cultured retinal explants pretreated with a low dose (LD) of Sema3A for 2 min, washed, then treated with a high dose (HD) of netrin-1 showed a full collapse response. Similarly, cultures pretreated with a LD of netrin-1 (2 min), rinsed, then treated with a HD of Sema3A showed a full collapse response. This suggests that desensitization is ligand specific. (b,c) A 5-min pretreatment with a LD of the heterologous cue did not elicit greater sensitivity to the other cue either at a HD (b), nor did a LD of netrin-1 elicit increased sensitivity to a submaximal intermediate dose (ID; 1:2 supernatant/culture medium) of Sema3A (c). Asterisks indicate significant difference (*P < 0.05, **P < 0.01), Mann-Whitney U-test. Error bars, s.e.m.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

courses of these effects they reported are much longer than the time courses observed in the present experiments. The reasons for this difference are not yet known. It might be because of basic differences in the experimental protocols, such as turning assays versus collapse assays, or the use of attractants by these researchers versus our use of repellents. Furthermore, they used substantially younger spinal neurons that survive in culture on their own yolk supplies12, whereas we used retinal neurons from older embryos for which L15 medium is essential for survival. Although the time course may be different, it is nevertheless clear that both studies revealed the presence of desensitization followed by resensitization, raising the basic question of the role of these phenomena in growth cone responsiveness. Desensitization may simply reduce the sensitivity of growth cones to a guidance cue and resensitization may return it to the previous level, as in a simple on-or-off situation. This ‘homeostatic reset’ mechanism might suggest that directional movements should be intermittent. Indeed, the zigzag tracks of spinal axons that have sometimes been seen in a concentration gradient of a guidance cue have been interpreted as evidence for this12. Another possibility is that growth cones adapt to the background levels, making them able to respond appropriately to the higher levels of guidance cues riding on top of these background

levels. The present study favors such an interpretation, as it shows that the adaptation process allows retinal growth cones exposed to background levels of Sema3A to respond differentially to higher levels of Sema3A than unexposed growth cones. It seems to us that adaptation in a gradient may be an ongoing process comprised of temporally overlapping components, with endocytosis being involved in decreasing the response and translation being involved in increasing the response, and with growth cones using these two mechanisms to adjust their sensitivity to a given background level of ligand. Although it is easy to understand an adaptation process that allows chemotaxis up a gradient of an attractant, it is more difficult to understand why it might be useful to adapt to a repellent. Yet adaptation to repellents is also seen in bacterial chemotaxis26,27. Our findings are also consistent with another set of earlier findings6 showing that exposure of retinal axons to a certain level of ephrins allowed these axons to crawl further up an ephrin gradient. In the retinotectal system, a shallow gradient of repellents covers the anterior-posterior axis of the tectum, and axons have to navigate appropriately within this gradient28. A recent study29 showed that up to a threshold level, ephrinA2 promotes the growth of retinal axons, but above this level it inhibits growth. The threshold level is graded: it is higher for nasal axons than for temporal ones. It is at present unclear how the adaptation process that we have described here relates to these threshold concentrations of ephrinA2 and the endpoints beyond which retinal axons will not grow. This is especially true given that in many animals, retinal axons grow past their most appropriate termination zones, and then form terminal arbors in the correct place by back-branching along the axon shaft while withdrawing their overshooting process30,31. The first phase of adaptation, desensitization, is dependent on rapid endocytosis. The speed of endocytosis observed here is not uncommon; the somatostatin receptor undergoes internalization and desensitization within 20 s of exposure to somatostatin32, and Sema3A has been reported to elicit a significant increase in fluid-phase endocytosis in axonal growth cones within 5 min16. The latter report was also the first to correlate Sema3A stimulation of growth cones with endocytosis, and also included a proposed model whereby rearrangement of neuropilin-1 and plexin receptors to F-actin–rich structures within the growth cone was a consequence of Sema3A stimulation16. The data shown here support this model, with Sema3A stimulating a rapid, endocytosis-dependent depletion of neuropilin-1 from the surface of retinal growth cones. Receptor-mediated endocytosis is preceded by clustering of receptor molecules, followed by internalization through either clathrin- or non-clathrin-mediated pathways18. Receptor-specific endocytosis as a mechanism of desensitization is consistent with the unchanged sensitivity of growth cones to Sema3A when pretreated with netrin-1, and vice versa. Previous work12 also showed that X. laevis embryonic spinal neurons show homologous, but not heterologous, desensitization. The data presented here strongly support the hypothesis that desensitization occurs at the level of membrane receptors, as

184

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
the loss of receptors from the growth cone surface via endocytosis is temporally correlated with the desensitization process, and blocking endocytosis prevents desensitization. Endocytosis of surface receptor may be a cause of desensitization. However, endocytosis may have a much greater role in growth cone navigation. It may be a necessary aspect of receptor signaling. For example, TGF-β receptor internalization to the endosomal compartment has been shown to promote SMAD2 signalling18,33. Furthermore, although loss of surface receptor can explain desensitization, it is also possible that the receptor remaining on the membrane becomes inactive during desensitization. Indeed, desensitization need not be a receptor-based phenomenon, as components of the signaling pathway downstream of receptor internalization could also be desensitized. The partial replacement of surface receptor when protein synthesis is blocked without a parallel increase in sensitivity would suggest that other components are involved. In this respect members of the MAP kinase (MAPK) family may be of particularly interest. Activation of ERK1/2 is suppressed in some endocytosis-defective cells34, MAPKs such as ERK1/2 have been localized to the endosomal compartment35–37 and receptor stimulation can enhance localization of JNK3 to intracellular vesicles38. In growth cones, both Sema3A and netrin-1 signal through various MAPK molecules to induce turning in a gradient or collapse12,39. Thus, it will be necessary to examine the full transduction cascade, from receptor activation to motor response, to find out precisely which components are regulated during the desensitization process. Our findings may also clarify the role(s) of protein synthesis during growth cone behavior. Our data suggest that resensitization is probably not simply due to the appearance of new receptors on the growth cone surface, as there is a partial replacement of receptor when protein synthesis is blocked, even though there is no recovery of sensitivity in this case. Perhaps, however, the recycled receptors are initially inactive and it is newly synthesized receptors that mediate the resensitization. Because these must represent only a fraction of the total surface receptor to the adapted ligand, a resensitized growth cone would have fewer active receptors than a naive one. This could explain why adapted growth cones show extended sensitivity. If some of the receptors expressed on the surface of resensitized growth cones were newly synthesized, one would expect to find the mRNA for these receptors in the growth cone. However, in preliminary studies, we have been unable to identify either DCC or neuropilin-1 from a retinal growth cone cDNA library (data not shown), which casts some doubt on this model. A growing body of data suggests that growth cones do harbor a subset of mRNA species, including those encoding β-actin40,41, neurofilament42, the microtubule-associated protein tau43 and actin-depolymerizing factor44. Ligand-induced collapse involves rearrangement of the cytoskeleton, such as a reduction in filamentous actin within the growth cone45. The need for protein synthesis during resensitization may reflect production of nascent cytoskeletal elements to be used during collapse and subsequent turning of growth cones. However, such an explanation would not be consistent with a ligand-specific process. Therefore, we suggest that the proteins that are synthesized to mediate resensitization may be involved in replenishing cytosolic components of specific signal transduction pathways. It has previously been demonstrated that protein synthesis is involved in retinal growth cone turning in response to gradients of Sema3A and netrin-1 and growth cone collapse14. Recently, it was shown, using collapse assays, that rapid protein synthesis is necessary for mouse DRG growth cones to maintain normal sensitivity to Sema3A46. These studies indicate that protein synthesis has a role in collapse and turning that may be independent of its role in adaptation. However, as the time course of observable collapse and turning is relatively long compared to the time course of resensitization shown here, interfering with temporal dynamics of growth cone sensitivity could explain all of these effects. Further experiments will be necessary to test this possibility. METHODS
Retinal cultures. Eye primordia from stage 35/36 X. laevis embryos47 were cultured in L15 medium without any added growth factors or serum as described previously13. Cultures for collapse assays and quantitative immunohistochemistry were grown for 24 h at 20 °C on glass cover slips precoated with 10 µg/ml polyL-lysine (Sigma) and 10 µg/ml laminin (Sigma). Collapse assays. Collapse assays were performed as described previously14,48 with minor modifications. The basal level of collapse (36.3% ± 3.3 (mean ± s.e.m.) for Sema3A controls, n = 1,106: 34.9% ± 2.4 for netrin-1 controls, n = 1,056) was similar to that described previously in cultured X. laevis retinal neurons14. A dose of 1:1 cell supernatant to fresh culture medium (supernatant was from Sema3A-expressing COS-7 cells transiently transfected with Sema3A plasmid, collected after 72 h) gave a collapse rate of 63.0% ± 1.3 (n = 2,004), which did not appreciably increase even at higher concentrations of Sema3A. This was defined as the high dose. A 1:4 ratio of supernatant to culture medium consistently gave a collapse rate just above the control value (that is, 43.6% ± 2.2 collapse, n = 615), and was defined as the low dose. The intermediate dose of Sema3A used to assess cross-adaptation in Figure 7c was a 1:2 ratio of supernatant to culture medium, which gave a collapse rate of 59.6% ± 1.49 (n = 745). For netrin-1 assays, the low dose was 150 ng ml–1 (42.4% ± 3.1 collapse, n = 921) and the high dose was 300 ng ml–1 (66.1% ± 1.0 collapse, n = 961). Other doses used are described in the legend to Figure 2. The data presented represent at least four independent experiments, with ≥400 growth cones for each individual data point. Values represent mean percent collapse ± s.e.m. Pharmacological reagents. Pharmacological reagents were bath-applied to cultures immediately before the application of the low dose in the collapse or quantitative immunohistochemistry assays, and were 40 µM anisomycin (Sigma), 25 µM cycloheximide (Sigma), 50 µM phenylarsine oxide (Sigma) and 10 µM monodansyl cadaverine (Sigma). After pretreatment with pharmacological reagents in collapse assays, cultures were washed twice with 900 µl of culture medium to remove the reagent before exposure to the high dose. Antibodies and digital quantification of fluorescence intensity. Neuropilin-1 was detected using an anti–neuropilin-1 antibody (Zymed Laboratories Inc.). DCC was detected on growth cones using an anti-DCC antibody (Oncogene Research Products). Growth cones were not permeabilized so as to detect only cell surface receptors. Growth cones were viewed on a Nikon inverted fluorescence microscope with a 100× Plan Apo objective. For quantification, growth cones were randomly selected with phase optics and fluorescent images were captured. The outline of a growth cone was traced digitally and the amount of fluorescence within the area of the growth cone was calculated digitally (Openlab, Improvision; IP Lab, Scanalytics Inc.). The background fluorescence for the same area was similarly calculated and subtracted from the growth cone value. This final value represented the mean fluorescence intensity per unit area. This was repeated to obtain the fluorescence intensities of 30–40 growth cones for each condition per experiment. Values were then normalized against controls. The data presented represent at least four independent experiments. Fluorescent intensities are presented as a percentage of treated growth cones compared to control treated growth cones ± s.e.m. Statistical analysis. Statistical analyses were carried out using the Mann-Whitney U-test for collapse assays and the Student’s t-test for the quantitative immunofluorescence assays.
ACKNOWLEDGMENTS We thank D. O’Connor, A. Dwivedy, D. Pask, S. Diamantakis, S. Shipway and E. Miranda for technical assistance, and K. Ohta and M. Tessier-Lavigne for the Sema3A and netrin-1 plasmids respectively. This work was supported by the Wellcome Trust and the Medical Research Council. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 3 August; accepted 10 November 2004 Published online at http://www.nature.com/natureneuroscience/

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

185

ARTICLES
1. Smythies, J. What is the function of receptor and membrane endocytosis at the postsynaptic neuron? Proc. R. Soc. Lond. B 267, 1363–1367 (2000). 2. Bredt, D.S. & Nicoll, R.A. AMPA receptor trafficking at excitatory synapses. Neuron 40, 361–379 (2003). 3. Falke, J.J., Bass, R.B., Butler, S.L., Chervitz, S.A. & Danielson, M.A. The two-component signaling pathway of bacterial chemotaxis: a molecular view of signal transduction by receptors, kinases, and adaptation enzymes. Annu. Rev. Cell Dev. Biol. 13, 457–512 (1997). 4. Samanta, A.K., Oppenheim, J.J. & Matsushima, K. Interleukin 8 (monocyte-derived neutrophil chemotactic factor) dynamically regulates its own receptor expression on human neutrophils. J. Biol. Chem. 265, 183–189 (1990). 5. Bourke, E. et al. IL-1β scavenging by the type II IL-1 decoy receptor in human neutrophils. J. Immunol. 170, 5999–6005 (2003). 6. Rosentreter, S.M. et al. Response of retinal ganglion cell axons to striped linear gradients of repellent guidance molecules. J. Neurobiol. 37, 541–562 (1998). 7. Loschinger, J., Weth, F. & Bonhoeffer, F. Reading of concentration gradients by axonal growth cones. Phil. Trans. R. Soc. Lond. B 355, 971–982 (2000). 8. Ferguson, S.S. & Caron, M.G. G protein–coupled receptor adaptation mechanisms. Semin. Cell Dev. Biol. 9, 119–127 (1998). 9. Kapfhammer, J.P. & Raper, J.A. Interactions between growth cones and neurites growing from different neural tissues in culture. J. Neurosci. 7, 1595–1600 (1987). 10. Kapfhammer, J.P. & Raper, J.A. Collapse of growth cone structure on contact with specific neurites in culture. J. Neurosci. 7, 201–212 (1987). 11. Shirasaki, R., Katsumata, R. & Murakami, F. Change in chemoattractant responsiveness of developing axons at an intermediate target. Science 279, 105–107 (1998). 12. Ming, G.L. et al. Adaptation in the chemotactic guidance of nerve growth cones. Nature 417, 411–418 (2002). 13. Campbell, D.S. et al. Semaphorin 3A elicits stage-dependent collapse, turning, and branching in Xenopus retinal growth cones. J. Neurosci. 21, 8538–8547 (2001). 14. Campbell, D.S. & Holt, C.E. Chemotropic responses of retinal growth cones mediated by rapid local protein synthesis and degradation. Neuron 32, 1013–1026 (2001). 15. Jurney, W.M., Gallo, G., Letourneau, P.C. & McLoon, S.C. Rac1-mediated endocytosis during ephrin-A2- and semaphorin 3A-induced growth cone collapse. J. Neurosci. 22, 6019–6028 (2002). 16. Fournier, A.E. et al. Semaphorin3A enhances endocytosis at sites of receptor-F-actin colocalization during growth cone collapse. J. Cell Biol. 149, 411–422 (2000). 17. Hertel, C., Coulter, S.J. & Perkins, J.P. A comparison of catecholamine-induced internalization of β-adrenergic receptors and receptor-mediated endocytosis of epidermal growth factor in human astrocytoma cells. Inhibition by phenylarsine oxide. J. Biol. Chem. 260, 12547–12553 (1985). 18. Di Guglielmo, G.M., Le Roy, C., Goodfellow, A.F. & Wrana, J.L. Distinct endocytic pathways regulate TGF-β receptor signalling and turnover. Nat. Cell Biol. 5, 410–421 (2003). 19. Schutze, S. et al. Inhibition of receptor internalization by monodansylcadaverine selectively blocks p55 tumor necrosis factor receptor death domain signaling. J. Biol. Chem. 274, 10203–10212 (1999). 20. Ray, E. & Samanta, A.K. Dansyl cadaverine regulates ligand induced endocytosis of interleukin-8 receptor in human polymorphonuclear neutrophils. FEBS Lett. 378, 235–239 (1996). 21. He, Z. & Tessier-Lavigne, M. Neuropilin is a receptor for the axonal chemorepellent Semaphorin III. Cell 90, 739–751 (1997). 22. Kolodkin, A.L. et al. Neuropilin is a semaphorin III receptor. Cell 90, 753–762 (1997). 23. Winberg, M.L. et al. Plexin A is a neuronal semaphorin receptor that controls axon guidance. Cell 95, 903–916 (1998). 24. Takahashi, T. et al. Plexin-neuropilin-1 complexes form functional semaphorin-3A receptors. Cell 99, 59–69 (1999). 25. Tamagnone, L. et al. Plexins are a large family of receptors for transmembrane, secreted, and GPI-anchored semaphorins in vertebrates. Cell 99, 71–80 (1999). 26. Taylor, B.L. An alternative strategy for adaptation in bacterial behavior. J. Bacteriol. 186, 3671–3673 (2004). 27. Bibikov, S.I., Miller, A.C., Gosink, K.K. & Parkinson, J.S. Methylation-independent aerotaxis mediated by the Escherichia coli Aer protein. J. Bacteriol. 186, 3730–3737 (2004). 28. Flanagan, J.G. & Vanderhaeghen, P. The ephrins and Eph receptors in neural development. Annu. Rev. Neurosci. 21, 309–345 (1998). 29. Hansen, M.J., Dallal, G.E. & Flanagan, J.G. Retinal axon response to ephrin-as shows a graded, concentration-dependent transition from growth promotion to inhibition. Neuron 42, 717–730 (2004). 30. Yates, P.A., Roskies, A.L., McLaughlin, T. & O’Leary, D.D. Topographic-specific axon branching controlled by ephrin-As is the critical event in retinotectal map development. J. Neurosci. 21, 8548–8563 (2001). 31. Sakurai, T., Wong, E., Drescher, U., Tanaka, H. & Jay, D.G. Ephrin-A5 restricts topographically specific arborization in the chick retinotectal projection in vivo. Proc. Natl. Acad. Sci. USA 99, 10795–10800 (2002). 32. Beaumont, V., Hepworth, M.B., Luty, J.S., Kelly, E. & Henderson, G. Somatostatin receptor desensitization in NG108-15 cells. A consequence of receptor sequestration. J. Biol. Chem. 273, 33174–33183 (1998). 33. Hayes, S., Chawla, A. & Corvera, S. TGFβ receptor internalization into EEA1-enriched early endosomes: role in signaling to Smad2. J. Cell Biol. 158, 1239–1249 (2002). 34. Vieira, A.V., Lamaze, C. & Schmid, S.L. Control of EGF receptor signaling by clathrinmediated endocytosis. Science 274, 2086–2089 (1996). 35. Sorkin, A. & Von Zastrow, M. Signal transduction and endocytosis: close encounters of many kinds. Nat. Rev. Mol. Cell Biol. 3, 600–614 (2002). 36. Pol, A., Calvo, M. & Enrich, C. Isolated endosomes from quiescent rat liver contain the signal transduction machinery. Differential distribution of activated Raf-1 and Mek in the endocytic compartment. FEBS Lett. 441, 34–38 (1998). 37. Rizzo, M.A., Shome, K., Watkins, S.C. & Romero, G. The recruitment of Raf-1 to membranes is mediated by direct interaction with phosphatidic acid and is independent of association with Ras. J. Biol. Chem. 275, 23911–23918 (2000). 38. McDonald, P.H. et al. β-arrestin 2: a receptor-regulated MAPK scaffold for the activation of JNK3. Science 290, 1574–1577 (2000). 39. Campbell, D.S. & Holt, C.E. Apoptotic pathway and MAPKs differentially regulate chemotropic responses of retinal growth cones. Neuron 37, 939–952 (2003). 40. Olink-Coux, M. & Hollenbeck, P.J. Localization and active transport of mRNA in axons of sympathetic neurons in culture. J. Neurosci. 16, 1346–1358 (1996). 41. Bassell, G.J. et al. Sorting of β-actin mRNA and protein to neurites and growth cones in culture. J. Neurosci. 18, 251–265 (1998). 42. Sotelo-Silveira, J.R. et al. Neurofilament mRNAs are present and translated in the normal and severed sciatic nerve. J. Neurosci. Res. 62, 65–74 (2000). 43. Litman, P., Barg, J., Rindzoonski, L. & Ginzburg, I. Subcellular localization of tau mRNA in differentiating neuronal cell culture: implications for neuronal polarity. Neuron 10, 627–638 (1993). 44. Lee, S.K. & Hollenbeck, P.J. Organization and translation of mRNA in sympathetic axons. J. Cell Sci. 116, 4467–4478 (2003). 45. Fan, J., Mansfield, S.G., Redmond, T., Gordon-Weeks, P.R. & Raper, J.A. The organization of F-actin and microtubules in growth cones exposed to a brain-derived collapsing factor. J. Cell Biol. 121, 867–878 (1993). 46. Li, C. et al. Correlation between semaphorin3A-induced facilitation of axonal transport and local activation of a translation initiation factor eukaryotic translation initiation factor 4E. J. Neurosci. 24, 6161–6170 (2004). 47. Cornel, E. & Holt, C. Precocious pathfinding: retinal axons can navigate in an axonless brain. Neuron 9, 1001–1011 (1992). 48. Shewan, D., Dwivedy, A., Anderson, R. & Holt, C.E. Age-related changes underlie switch in netrin-1 responsiveness as growth cones advance along visual pathway. Nat. Neurosci. 5, 955–962 (2002).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

186

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Coactivation and timing-dependent integration of synaptic potentiation and depression
Huai-Xing Wang1, Richard C Gerkin2,3,5, David W Nauen2–5 & Guo-Qiang Bi1–3
Neuronal synaptic connections can be potentiated or depressed by paired pre- and postsynaptic spikes, depending on the spike timing. We show that in cultured rat hippocampal neurons a calcium/calmodulin-dependent protein kinase II (CaMKII)mediated potentiation process and a calcineurin-mediated depression process can be activated concomitantly by spike triplets or quadruplets. The integration of the two processes critically depends on their activation timing. Depression can cancel previously activated potentiation, whereas potentiation tends to override previously activated depression. The time window for potentiation to dominate is about 70 ms, beyond which the two processes cancel. These results indicate that the signaling machinery underlying spike timing–dependent plasticity (STDP) may be separated into functional modules that are sensitive to the spatiotemporal dynamics (rather than the amount) of calcium influx. The timing dependence of modular interaction provides a quantitative framework for understanding the temporal integration of STDP.

In the mammalian brain, patterned neuronal activity leads to modification of specific synaptic connections1, a process that is believed to be central to the development of precisely organized neuronal circuits as well as for learning and memory2–4. A hallmark of such activity-induced synaptic plasticity is that different activity patterns can lead to either strengthening or weakening of synapses, commonly known as longterm synaptic potentiation (LTP) or depression (LTD), respectively4–7. In traditional studies of homosynaptically induced plasticity, LTP is generally induced by bursts of high-frequency stimulation of input axons, whereas LTD is induced by low-frequency stimulation5–7. In the recently characterized STDP, the polarity of synaptic modification depends on the precise timing of individual pre- and postsynaptic spikes4,8,9: potentiation is induced if a postsynaptic spike repetitively follows a presynaptic spike by a few milliseconds, whereas depression is induced if the temporal order of the spike pairing is reversed10–19. Such sensitivity to activity patterns is crucial for the formation of specific engrams in neuronal circuits20,21. To what extent can a synapse precisely interpret incoming spike patterns? In vivo, a synapse experiences complex patterns of ongoing activity. Will the cellular processes involved in LTP and those involved in LTD both be activated when multiple pre- and/or postsynaptic spikes occur within milliseconds of one another? If so, how do these opposing processes interact to result in the final synaptic modification? Using cultured hippocampal neurons, we found that spike triplets or quadruplets could activate concomitantly a CaMKII-mediated potentiation process and a calcineurin-mediated depression process. The two modular processes integrated nonlinearly in a timing-dependent manner: depression canceled previously activated potentiation, whereas

potentiation tended to override previously activated depression. The time window for potentiation to dominate is about 70 ms. In addition, blocking L-type calcium channels selectively suppressed depression, resulting in net potentiation. These results provide new quantitative rules for STDP integration. They also reflect the dynamic nature of the underlying cellular signaling. RESULTS Asymmetric temporal integration of STDP Dual perforated whole-cell patch-clamp recordings were made on pairs of cultured hippocampal neurons, in which STDP has been demonstrated by repeated pairing of pre- and postsynaptic spiking at 1 Hz13. Based on the previously characterized spike timing window of STDP in these neurons, a potentiation (P) process can be triggered by a ‘pre-post’ spike pair (spike timing t > 0), whereas a depression (D) process can be induced by a ‘post-pre’ spike pair (t < 0). Here, P and D refer to the initial induction processes in the signaling pathways that lead to the final expression of STDP. In most studies and models of STDP, it is assumed that the spike pairing events in more complex spike patterns induce the P and D processes independently18,19,22–27. Under such an assumption, the simplest cases that could involve both potentiation and depression processes are triplet spiking paradigms (Fig. 1). For a triplet with a spiking sequence of pre-post-pre, P is presumably activated by the first spike-pairing event (pre-post) with spike timing t1, and this is followed by D, which is activated by the second pairing (post-pre) with spike timing t2. In a triplet case with post-pre-post spikes, the two processes are activated in the opposite temporal order. Hereafter, we use the two spike timing values {t1, t2} to denote a triplet

1Department

of Neurobiology, 2Center for Neuroscience, 3Center for the Neural Basis of Cognition and 4Medical Scientist Training Program, University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania 15261, USA. 5These authors contributed equally to this work. Correspondence should be addressed to G.Q.B. ([email protected]). Published online 16 January 2005; doi:10.1038/nn1387

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

187

ARTICLES a
EPSC amplitude (pA)
300 200 100 −20 −10 0 10 20 30
pre +10 –10 post
20 ms 100 pA

b
200 150 100 50 0
20 ms

−20

−10

0

10

20

c
EPSC amplitude (pA)
–10 +10
100 pA

d
300 200 100 0
20 ms

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

500 400 300 200 100

20 ms

200 pA

–5

+5

40 pA

+5

–5

30

Figure 1 Triplet experiments showing asymmetric temporal integration of STDP. (a,b) Results from typical stimulation experiments with pre-postpre triplets with (a) 10-ms and (b) 5-ms intervals. Data points show the peak amplitudes of monosynaptic EPSCs elicited by test stimuli (0.05 Hz) before and after repetitive triplet stimulation (arrow, 60 repetitions at 1 Hz during which the postsynaptic cell was under current clamp). Insets show traces of EPSCs (average of five consecutive events) 5 min before (left) and 20 min after (right) triplet stimulation. (c,d) Results from typical stimulation experiments with post-pre-post triplets with (c) 10-ms and (d) 5-ms intervals. (e) Summary of triplet experiments pre-post-pre {+10, –10} (n = 12) and post-pre-post {–10, +10} (n = 7). (f) Summary of triplet experiments pre-post-pre {+5, –5} (n = 6) and post-pre-post {–5, +5} (n = 7). Error bars, s.e.m.

e
1.6

−20

−10

0

10

20

30

f

0 −20 1.6 1.4 1.2 1.0

−10

0

10

20

30

Normalized EPSC

1.4 1.2 1.0 0.8

tri{+10,–10} tri{–10,+10}

tri{+5,–5} tri{–5,+5}

to be due to active integration, because in the same experimental setting, post-pre spike pairs did result in LTD (Supplementary Fig. 1). Therefore, rather than summing linearly, the integration of potentiation and depression processes is temporally asymmetric: the two processes cancel when potentiation is triggered first, whereas potentiation dominates when it is triggered second. Coactivation of kinase and phosphatase signaling modules It is well established that protein kinases, especially CaMKII, are key to the induction of LTP by conventional tetanic stimulation paradigms, whereas phosphatases, including calcineurin and protein phophatase1, are involved in LTD28–33. These pathways are likely to be involved in the induction of STDP by spike pairing with either positive (protein kinases) or negative (phosphatases) spike timing. They also may be coactivated by triplets or more-complex spike patterns. If this is the case, these pathways could be the molecular substrates for the proposed P and D processes that activate independently and then

−10

0 10 Time (min)

20

30

0.8

−10

0 10 Time (min)

20

30

paradigm; thus t1 > 0 and t2 < 0 for pre-post-pre triplets, whereas t1 < 0 and t2 > 0 for post-pre-post triplets. We first examined the effects of triplet stimulation with spike timing intervals of ±10 ms. As in previous studies that used spike pairs13, each triplet was repeated at 1 Hz during the 60-s induction period. In cultured hippocampal neurons, paired spiking with +10 ms spike timing induces significant LTP, whereas paired spiking with –10 ms timing induces significant LTD13. If P and D integrate linearly, as assumed in many STDP models22–27, potentiation and depression in both pre-post-pre {+10, –10} and post-pre-post {–10, +10} paradigms should offset to result in the same outcome. Whereas no significant synaptic change was induced by the pre-post-pre triplet {+10, –10} (Fig. 1a,e), the postpre-post triplet {–10, +10} did lead to marked LTP (Fig. 1c,e). We found similar results in experiments using triplets with intervals of ±5 ms (Fig. 1b,d,f). The amount of LTP that was induced by post-pre-post triplets (–10, +10; STDP ratio ± s.e.m., 1.27 ± 0.05) or (–5, +5; 1.28 ± 0.05) was similar to that induced by pre-post spike pairs alone (1.25 ± 0.05; see Supplementary Fig. 1 online). It should be noted that the amount of LTP that was induced by pre-post spike pairs under these experimental conditions was lower than that reported in the previous study13. This is accounted for by the difference in initial synaptic strength, which significantly influences the degree of LTP induction13: on average, older cultures and stronger synapses were used in the current studies. Further, failure to induce significant change by the pre-post-pre triplets seemed

200 pA

EPSC amplitude (pA)

200 150

+5

–5 20 ms

+8

–8 20 ms

400 300

100 50 –20 –10 0 10 20 30
100 pA

200 100

–10

0

10

20
200 pA 20 ms

c
EPSC amplitude (pA)

d
500 400 300 200 100 *
–8 +8

250 200 150 100 –20 –10

–5

+5 20 ms

0

10

20

30
100 pA

0

–10

0

10

20

EPSC amplitude (pA)

+5 20 ms

300 200

+8 20 ms

200 150 100 –20 –10 0 10 20 30 100 –10
KN-62 6 6 7 5 3

Figure 2 Effects of CaMKII inhibition on the induction and temporal integration of STDP. (a–f) The extracellular perfusion solution contained (a,c,e) the CaMKII antagonist KN-62 (10 µM; 5-ms intervals) or (b,d,f) the control analog KN-92 (10 µM; 8-ms intervals). (a,b) Results from typical experiments using pre-post-pre triplets. (c,d) Results from typical experiments using post-pre-post triplets. The synaptic response in d contained a GABAergic polysynaptic component (*). In this example, after the induction of LTP, a second time of stimulation (arrowhead) with prepost-pre triplets did not induce any pronounced change. (e,f) Results from typical experiments using pre-post spike pair stimulation. (g) Summary of all experiments using triplet and pair stimulation in the presence of KN62. (h) Summary of all experiments with KN-62 or KN-92. Number of experiments in each data set is shown above each column. Error bars, s.e.m.

g 1.2
STDP ratio

1.0 0.9 0.8 –10 0 10 20 Time (min) 30

STDP ratio

1.1

KN62 {+5,–5} KN62 {–5,+5} KN62 {+5}

h
30 20 10 0 –10 –20 –30
Pr e

0

10
KN-92

20

188

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

Pr ost epo st -p Po re st -p re -p os t Pr ep Pr ost epo st -p re Po st -p re -p os t

-p

200 pA 4

e250

f

200 pA

a

b

30

ARTICLES
Figure 3 Effects of calcineurin inhibition on the induction and temporal integration of STDP. (a–f) Experiments were performed in the presence of (a,c,e) the calcineurin antagonist CsA (10 µM; 5-ms intervals) or (b,d,f) FK-520 (2.5 µM; 8-ms intervals). (a,b) Results from typical experiments using pre-post-pre triplets. (c,d) Results from typical experiments using post-pre-post triplets. (e,f) Results from typical experiments using post-pre spike pair stimulation. (g) Summary of all experiments using triplet and pair stimulation in the presence of CsA. (h) Summary of all experiments in CsA or FK-520. Number of experiments in each data set is shown above each column. Error bars, s.e.m.
200 pA 200 pA 20 ms

EPSC amplitude (pA)

a500
400 300 200 –20 –10

b
700 600 500 400
+8 −8

+5

−5

20 ms

0

10

20

30
200 pA

–10

0

10

20
200 pA 20 ms

c
EPSC amplitude (pA)
250 200 150 100 50 –20 –10
−5 20 ms −5 +5 20 ms

d
500 400 300
−8 +8

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

integrate to result in net synaptic modification, as assumed in theoretical as well as experimental studies18,19,22–26. Indeed, in the presence of KN-62, a specific blocker of CaMKII34, both triplets (pre-post-pre and post-pre-post) resulted in substantial depression (Fig. 2a,c,g,h). In contrast, in the presence of KN-92, an analog of KN-62 that does not inhibit CaMKII but has nonspecific inhibitory effects on voltagegated K+ channels similar to those of KN-62, the triplets led to either cancellation or LTP as in control conditions (Fig. 2b,d,h). As we had expected, KN-62, but not KN-92, also eliminated LTP that was induced by pre-post spike pairing (Fig. 2e–h). Notably, no significant LTD was observed in pre-post spike pairing experiments with the blockade of CaMKII. In the presence of cyclosporin A (CsA) or FK-520, blockers of calcineurin32, both triplets induced pronounced potentiation (Fig. 3a–d,g,h). On the other hand, post-pre spike pairing in the presence of CsA or FK-520 resulted in neither LTP nor LTD (Fig. 3e–h). Therefore, the kinase and phosphatase pathways seem to be coactivated by either triplet paradigm, because blocking one pathway can unmask the effect of the other. In addition, the activation of each pathway may be caused only by specific pairing; that is, pairing with positive timing activates the kinase-dependent P process, whereas pairing with negative timing activates the phosphatase-dependent D process. Such specificity may arise from the requirements for highly local and dynamic calcium signaling during STDP induction. Dominance of potentiation in STDP integration The amount of STDP that is induced by spike pairs depends on the spike timing interval10,13. This offered us an opportunity to investigate the role of induction strength in the integration of P and D using triplets with different spike timing intervals (Fig. 4). In this set of experiments, we selected spike timing of ±5 ms for the activation of strong P or D and ±15 ms for the activation of weak P or D. Of the four triplet paradigms, pre-post-pre {+15, –5} and {+5, –15} and post-prepost {–5, +15} and {–15, +5}, three resulted in potentiation (Fig. 4). The potentiation induced by the post-pre-post triplet {–5, +15} indicated that weak P could overcome strong D when P was activated later (Fig. 4c,e). On the other hand, even when D was activated later, weak D could not negate strong P (Fig. 4b,f). These results indicate that P is dominant in the integration. It is also notable that triplet {+15, –5} did not produce pronounced depression but resulted in cancellation of potentiation and depression that was quantitatively similar to triplets {+10, –10} and {+5, –5} (see also Fig. 6c). Therefore, even in situations where P was not dominant, P and D did not linearly sum but resulted in cancellation. Asymmetric integration with quadruplet stimuli Because the pre-post-pre triplet involves two presynaptic spikes, shortterm synaptic plasticity could in principle influence the outcome of STDP integration. At the culture stages used in these studies, however, most synapses in these hippocampal neurons showed moderate paired-pulse depression, and the paired-pulse ratio was uncorrelated

0

10

20

30
100 pA

–10

0

10

20
200 pA 20 ms

EPSC amplitude (pA)

e
400 300 200

f
300 200 100

−8

100 –20

–10

0

10

20

30

-10

0
CsA 7 7

10
FK-520 7

20

Normalized STDP (%)

g1.5
STDP ratio
1.4 1.3 1.2 1.1 1.0 –10 0

CsA {+5,–5} CsA {–5,+5} CsA {–5}

h
30 20 10 0 –10
7

7

8

10 20 Time (min)

30

with observed changes in synaptic strength (Supplementary Fig. 2). If the integration of P and D is assumed to be linear, paired-pulse depression should lead to more potentiation by pre-post-pre triplets than by post-pre-post triplets, the opposite of our experimental observation. Therefore, our results probably reflect nonlinearity in the interaction. To further evaluate this, we used quadruplet configurations that consist of two spike pairs (with spike timing of ±5 ms) separated by an interval T (defined as the time interval from the midpoint of the D-inducing spike pair to the midpoint of the P-inducing pair; T is positive when P follows D and negative when P precedes D; Fig. 5). Like the triplet configurations, quadruplet pre-post-post-pre induced little long-term synaptic change, whereas quadruplet post-pre-prepost induced pronounced LTP (Fig. 5). Therefore, the integration of P and D in quadruplet paradigms follows the same rule as in triplets: P and D cancel when P is induced first, whereas P dominates when it is induced second. Dependence of integration on initial synaptic strength In our culture system, there is notable heterogeneity in the strength of synaptic connections. Previous studies in the same system have shown that the amount of spike-timing-dependent LTP, but not LTD, depends on the initial synaptic strength13. To evaluate how initial strength influences the outcome of STDP integration, we plotted the measured STDP ratio in individual triplet and quadruplet experiments against initial excitatory postsynaptic current (EPSC) size. A negative correlation between the STDP ratio and the initial EPSC amplitude was found in experiments using LTP-inducing triplet or quadruplet paradigms: considerably more potentiation was induced in initially weaker synapses (Fig. 6a). It is also notable that no correlation was found between STDP ratio and initial strength in experiments using paradigms that were ineffective at inducing LTP (Fig. 6a). Because LTP (but not LTD) that was induced by paired

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

tPr pre epo st -p Po re st -p re -p os t Po st -p Pr re epo st -p re Po st -p re -p os t

Po s

189

ARTICLES a
80 pA

b
+15

EPSC amplitude (pA)

–5
20 ms

250 200 150 100

100 75 50 25

+5

–15
20 ms

Figure 4 STDP integration after biased triplets with unequal positive and negative timings suggests dominant role of potentiation. (a–d) Results from typical experiments with (a) pre-post-pre triplet {+15, –5}, (b) pre-post-pre triplet {+5, –15}, (c) post-pre-post triplet {–5, +15} and (d) post-pre-post triplet {–15, +5}. (e) Summary of triplet experiments pre-post-pre {+15, –5) (n = 7) and post-pre-post {–5, +15} (n = 6). (f) Summary of triplet experiments pre-post-pre {+5, –15} (n = 6) and post-pre-post {–15, +5} (n = 5). Error bars, s.e.m.
30

–20

–10

0

10

20

30

–20

–10

0

10

20
150 pA

200 pA

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

c
EPSC amplitude (pA)
500 400 300 200 100 –20

d
500 400 300 200

–5

+15
20 ms

–15

+5
20 ms

50 pA

–10

0

10

20

30

–20

–10

0

10

20

30

Normalized EPSC

e 1.3
1.2 1.1 1.0 0.9 0.8
tri{+15,–5} tri{–5,+15}

f
1.3 1.2 1.1 1.0

of LTP (Fig. 6c). The case of pre-post-pre triplet {+5, –5} is particularly telling, because specific blockade of calcium influx through L-type channels (thus less total calcium influx) resulted in potentiation, in contrast to the cancellation observed in the absence of drug. This suggests that specific calcium sources, and thus the spatial and temporal patterns of calcium transients at the synapse, are of critical importance in the induction and integration of STDP. Dependence of integration on P and D activation timing Comparing the amount of LTP induced by quadruplet post-pre-prepost {+20} to that induced by the two post-pre-post triplets {–10, +10} and {–5, +5} (Fig. 6b,c), it is apparent that the quadruplet induced less LTP (P = 0.016, t-test). One possibility is that nonlinear integration of P and D depends on the time interval between P and D; the 25-ms interval (T) used in the quadruplet was longer than that in the triplet cases. To obtain a more complete picture of the timing dependence of this integration, we carried out more quadruplet experiments with spike timing of ±5 ms for the P- or D-inducing spike pairs and T varying from –105 to +105 ms. P and D canceled for all negative T values (pre-post-post-pre quadruplets) regardless of the interval (average synaptic change, 1.2 ± 1.8%; n = 29). This is similar to the triplet results (1.4 ± 1.8%; n = 18; Fig. 7). For positive T (post-pre-pre-post quadruplets), however, the integration depended on the time interval. When T was small, marked LTP (18.1 ± 2.7% for T < 50 ms; n = 22) resulted from a dominating P as in the triplet situations (27.4 ± 3.3%; n = 14). P and D seemed to cancel when T was longer, resulting in no change (3.6 ± 3.0% for T > 70 ms, n = 6). The most notable feature in this summary (Fig. 7) is that P dominates the interaction when it follows D (hereafter referred to as D→P interaction) within a window of 70 ms. At a phenomenological level, this can be formulated as supralinear summation, with a ‘history-dependent’ term of extra potentiation that depends on the strength of P and D (denoted as P and D, respectively) and decays with increasing D→P interval T: δ(D→P, T) = α × P × D × exp(–T/τ). Thus, the unitary change in synaptic weight (w) that is due to the integration is ∆w = D + P + δ(D→ P, T). Here, exponential decay was chosen for simplicity, with the time

0.9 0.8

tri{+5,–15} tri{–15,+5}

–10

0 10 Time (min)

20

30

–10

0 10 Time (min)

20

30

spiking depended on initial synaptic strength, the cancellation of potentiation and depression that was independent of initial strength and that occurred during the ineffective triplet or quadruplet paradigms further indicates that the two processes do not integrate by linear summation. Based on the dependence of potentiation on initial strength (Fig. 6a), we normalized the measured STDP ratio for experiments with potentiation-inducing paradigms (Fig. 6b,c). The normalized STDP ratio reduced the bias in the measurement that was due to the initial strength variability. Comparing the quantitative results from triplet and quadruplet experiments with opposite orders of effective P or D induction—triplet {+10, –10} versus triplet {–10, +10}, triplet {+5, –5} versus triplet {–5, +5} and quadruplet {T = +25} versus quadruplet {T = –25; Fig. 6c}—it is clear that temporal asymmetry is a general feature of STDP integration. The L-type channel is needed for spike-timing-dependent LTD but not LTP13. Thus, we examined whether L-type channels are involved in the integration of P and D. In the presence of nimodipine, an L-type calcium channel antagonist, both the pre-post-pre triplet {+5, –5} and the post-pre-post triplet {–5, +5} resulted in similar amounts

a
EPSC amplitude (pA)

b
200 150 100 50 0
–20 –10 0 10 Time (min) 20 30

c
T=+25
30 pA

+5 T=–25 –5
20 ms

50 pA

–5
EPSC amplitude (pA)

+5
20 ms

80 70 60 50 40
–20 –10 0 10 Time (min) 20

1.3

Quad{+20} Quad{–20}

Normalized EPSC
30

1.2 1.1 1.0 0.9 –10 0 10 Time (min) 20 30

Figure 5 Asymmetric integration follows quadruplet stimuli. A quadruplet paradigm consisted of two spike pairs with ±5-ms spike timing that were separated by a longer interpair interval, T. (a) Result from a typical quadruplet experiment with T = –25 ms. (b) Result from a typical quadruplet experiment with T = +25 ms. (c) Summary of all quadruplet experiments with T = +20 to +25 ms (Quad{+20}; n = 9) and T = –20 to –25ms (Quad{–20}; n = 10). Error bars, s.e.m.

190

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES a
1.4
STDP ratio
Tri+ Tri– Quad+ Quad–

b
Cumulative fraction

c
1.0
Normalized STDP (%)
Pre-post-pre Post-pre-post 7 6 6 7 6 5 6 9

0.8 0.6 0.4 0.2 0.0
0.8 1.0 1.2 1.4
Tri{+10,–10} Tri{–10,+10} Tri{+5,–5} Tri{–5,+5} Quad{–20} Quad{+20}

40 30 20 10 0
12 6 7 10

1.2 1.0 0.8
0 100 200 300 400 500

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Initial EPSC (pA)

Normalized STDP ratio

1.6

–10

Figure 6 Normalization of STDP integration. (a) Dependence on initial strength in integration. Tri+ included data from experiments using post-pre-post triplets {–10, +10}, {–5, +5} and {–15, +5}. Tri– included data from experiments using pre-post-pre triplets {+10, –10}, {+5, –5} and {+15, –5}. Quad+ included quadruplet experiments with 0 < T < 25 ms. Quad– included quadruplet experiments with T < 0. All Tri+ and Quad+ data were used to fit a single exponential curve R = 1 + R1 × e–A/A1 (R, STDP ratio; A, initial EPSC amplitude). Fitting result: R1 = 0.33 ± 0.05; A1 = 428 ± 196 pA (χ2/degrees of freedom = 0.015; R2 = 0.14). (b) Cumulative histogram of normalized STDP ratio for quadruplet and triplet experiments. Based on the actual initial EPSC amplitude (A) and the curve fit parameters (R1 = 0.33, A1 = 428), a correction value ε = R1 × (e–A/A1 – e–100/A1) was calculated as the predicted deviation of the STDP ratio from that for a fixed initial strength of 100 pA. Normalized STDP ratio was the measured STDP ratio after subtracting ε. Quad{+20} and Quad{–20} included the same dataset as in Figure 5c. (c) Summary of mean normalized STDP (in percent change) under different triplet and quadruplet conditions. Also included are results from experiments performed in the presence of nimodipine (nimo, 20 µM) using triplets {–5, +5} and {+5, –5}. Number of experiments in each data set is shown above each column. Error bars, s.e.m.

constant τ (∼70 ms) derived from the right half of the timing window of the P and D interaction (Fig. 7). A scaling factor, α, can be chosen so that when P and D are both maximally activated, δ cancels D, resulting in ∆w = P. For P→D integration, the left half of the timing window (Fig. 7) indicates that the two modules tend to cancel. In principle, this nonlinearity could be formulated as a set of nonlinear filters that allow previously activated P to influence subsequent D. Alternatively, the cancellation may reflect intrinsic instability of P and D modules (or subsequent signals), which can be implemented by additional terms that gate the effects of the modules. DISCUSSION Rules of STDP integration Experimental studies of STDP have shown characteristic spike timing dependence that to a large extent is consistent across systems and preparations10–19,35,36. This property, in the form of various spike timing windows, can be regarded as a first-order rule because it describes the effects of simple paradigms in which only a single event of pre- and postsynaptic spike pairing could occur within the timescale of tens of milliseconds. To encompass more realistic activity patterns involvFigure 7 Asymmetric Pre time window of +5 –5 +5 –5 T<0 T>0 STDP integration. Post 1.5 Only quadruplet and Quad triplet experiments 1.4 Tri with unbiased timing 1.3 were included. STDP ratio and T were 1.2 defined as in previous 1.1 figures. Results 1.0 from quadruplet experiments with 0.9 smaller T were 0.8 consistent with 0.7 those from triplet –100 100 –50 0 50 experiments. For Time (ms) experiments with larger T, quadruplet results showed a 70-ms window within which potentiation dominates. Outside this window, potentiation and depression appeared to cancel each other.
STDP ratio

ing multiple pre- and postsynaptic spikes, most STDP models assume simple second-order rules for the integration of STDP, such as linear summation22–26,37. Our results, however, indicate that nonlinear second-order rules of STDP integration may occur, a consideration of which is likely to improve predictions of synaptic modifications that result from naturalistic spike trains. Nonetheless, these rules are likely to represent only part of the complexity of STDP integration. Other issues, for example, whether only near-neighbor spike interactions count, remain to be addressed38. Further, it is not clear whether morecomplex spike patterns may engage additional (higher-order) rules. Ultimately, these issues will be resolved through better understanding of the underlying molecular signaling mechanisms. The domination of P in D→P integration that we observed in hippocampal neurons is in opposition to the ‘first-spike-dominating’ rule— the spike that occurs first suppresses the efficacies of later spikes—that is found in layer 2/3 of the visual cortex19. One explanation is that layer 2/3 neurons may use additional mechanisms, such as strong short-term depression in both synaptic transmission and the back-propagation of action potentials, which in turn function as gates to limit downstream interaction between P and D from occurring in these synapses and to assure the priority of the earlier spike pairs. In other words, the firstspike-dominating rule may describe the interaction between adjacent spikes, whereas the timing-dependent integration of P and D reflects the properties of downstream signaling processes. Notably, in layer 5 of the visual cortex, STDP integration seems to follow an LTP-dominating rule18. Intuitively, our observations in cultured hippocampal neurons predict that with naturalistic spike train stimuli, LTP would also dominate when the spike rate is high. In the present study, however, the dominance of LTP is at least in part due to the detection of a specific temporal structure (such as D→P triplets and quadruplets) in the spike trains. Therefore, in contrast to the generality of the first-order rule of STDP that emphasizes causality detection as a fundamental neural function, multiple forms of the second-order rule may be ‘tuned’ to serve specific functions in different circuits. Calcium signaling in STDP At the cellular level, timing-dependent P and D integration underscores the dynamic nature of the intracellular signaling that underlies STDP. In accord with classical studies of LTP and LTD5–7, our results indicate that

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

Tri{ +1 0,– 10 Tri{ } +5 ,–5 } Tri{ +1 5, Tri{ –5} +5 ,–1 nim 5 o{+ } 5,– 5} Qu ad Tri{ {–20} –1 0 Tri{ ,+10 } –5 ,+5 } Tri{ –5 ,+ Tri{ 15} –1 5 nim ,+5} o{– 5 Qu ,+5} ad {+2 0}

191

ARTICLES
calcium and subsequent kinase and phosphatase pathways are crucial for STDP induction and integration. Several aspects of P and D integration that we observed, however, contradict the classical picture that the postsynaptic calcium increase dictates the outcome of synaptic potentiation or depression28,31,39. For example, whereas pre-post spike pairs induced LTP, most pre-post-pre triplets and pre-post-post-pre quadruplets that were tested did not result in substantial potentiation, as if the extra calcium influx that was triggered by the second presynaptic (and postsynaptic) stimulation had a negative impact. In addition, blocking L-type calcium channels allowed the pre-post-pre triplet to induce LTP, indicating a negative role for calcium influx from these channels. Therefore, in agreement with findings that classical LTP and LTD may depend on the duration of calcium elevation40 and on the activation of specific receptor subunits41, our results underscore the notion that when and where calcium influx occurs may be more critical than the amount of calcium increase for STDP. In other words, if postsynaptic calcium signals the induction of STDP, it must do so with its spatiotemporal dynamics at millisecond and submicron resolution. Such specificity may arise from the highly organized complexity of the postsynaptic density and nearby structures42–44. Ca2+ influxes from different channels seem to preferentially activate different kinase signaling pathways45–47. This could be due to the physical proximity of specific kinases to specific channels and/or distinct contributions of different channels to the temporal dynamics of intracellular Ca2+ transients. Indeed, with incorporation of multiple enzymatic pathways to detect the temporal profiles of calcium influx, a conductance-based biophysical model can reproduce the spike timing window and the asymmetric integration of STDP48. Modular activation and dynamic interaction of P and D Our results demonstrate that, downstream of calcium, the CaMKIIdependent P signal and the calcineurin-dependent D signal were coactivated by multiple spike paradigms because pharmacological inhibition of either signal led to the expression of the other. Meanwhile, pairing paradigms with either positive or negative spike timing activated only the P signal or the D signal, respectively. Therefore, these early signaling processes can be regarded as relatively independent modules. Such modularity in plasticity signaling validates the basic assumption in mathematical models that the effect of complex spike trains can be decomposed into multiple pairing events22–26. During the induction of STDP, these modules may act as detectors of specific features in the spatiotemporal dynamics of calcium elevation resulting from pre- and postsynaptic spiking. The exact molecular mechanisms that underlie such detection remain to be identified. The 70-ms time window for P to override previously activated D does, however, indicate that the two modules are likely to reside in the same compartment where they interact nonlinearly over a timescale of tens of milliseconds. Timing-dependent supralinear D→P integration may reflect the existence of cellular processes that allow D to sensitize subsequent P; this sensitization decays over time. Alternatively, it may also result from a slowly developing D that is interrupted by the activation of P. In either scenario, the dominating effect of P diminishes as the D→P interval increases. It should also be noted that during the induction period, the triplet or quadruplet paradigm was repeated at 1-s intervals for 1 min. Therefore with longer T, the D→P integration will be more or less equivalent to the P→D integration, in which P and D cancel. In other words, the result of cancellation at long intervals is consistent with the periodic stimulation paradigm. It is also notable that the cancellation in P→D integration is reminiscent of the depotentiation observed in various preparations49, as well as the reversal of LTP and LTD under physiological conditions in vivo50. These phenomena at different timescales may reflect a certain intrinsic instability of the molecular signals that underlie the induction of LTP. Implications for network function How may the cellular properties of the interaction between P and D influence the computational functions of neuronal networks? In principle, the second-order rule could result in selective potentiation of synapses in response to stimuli with specific temporal structures. Even in the presence of noncorrelated background activity, however, the overall dominance of P in the P and D interaction could drive synapses to saturation. This undesirable consequence is likely to be prevented by the initial-strength dependence of potentiation (Fig. 6a). Because multiplicative potentiation but not depression decreases with increasing ‘present’ strength13, weak synapses are more easily potentiated and strong synapses are more easily depressed. With such ‘soft ceilings’ for first-order STDP, noncorrelated background firing leads to a unimodal equilibrium distribution of synaptic weights23,24,26; the involvement of second-order interaction between P and D may shift the equilibrium to a higher mean value but is unlikely to cause saturation. Further, at low frequency, nonstructured spike patterns could rarely trigger supralinear D→P integration, because the D→P interval is likely to fall outside the 70-ms window. Meanwhile, frequent P and D cancellation may limit diffusion-like drift of synaptic weights that is caused by the random coincidence of pre- and postsynaptic firing, thereby improving the stability and reliability of the network. METHODS
Cell culture. Low-density cultures of dissociated embryonic rat hippocampal neurons were prepared according to a previously described protocol with minor modifications13, as approved by the University of Pittsburgh Institutional Animal Care and Use Committee. Hippocampi were removed from embryonic day (E) 17–18 rats and were treated with trypsin for 20 min at 37 °C, followed by washing and gentle trituration. The dissociated cells were plated on poly-L-lysine-coated glass coverslips in 35-mm Petri dishes with 30,000–60,000 cells per dish. The culture medium was Dulbecco’s minimum essential medium (DMEM; BioWhittaker) supplemented with 10% heat-inactivated bovine calf serum (Hyclone), 10% Ham’s F12 with glutamine (BioWhittaker), 50 U/ml penicillin-streptomycin (Sigma) and 1× B-27 (Invitrogen/Gibco). One-third of the culture medium was replaced with the same medium supplemented with 20 mM KCl 24 h after plating. Cytosine arabinoside (Sigma) was added to the culture dish (final concentration, 5 µM) around 7–10 days in vitro (DIV) to prevent overgrowth of glial cells. The optimal period for using these cultures is 8–15 DIV, during which glutamatergic connections of 50–500 pA are commonly found. Electrophysiological recordings. Simultaneous whole-cell perforated-patch recordings were carried out with patch-clamp amplifiers (PC505A; Warner Instruments) at room temperature. The pipette solution contained the following: 136.5 mM K-gluconate, 17.5 mM KCl, 9 mM NaCl, 1 mM MgCl2, 10 mM HEPES, 0.2 mM EGTA and 200 µg/ml amphotericin B (pH 7.3). The external bath solution was a HEPES-buffered saline (HBS): 150 mM NaCl, 3 mM KCl, 3 mM CaCl2, 2 mM MgCl2, 10 mM HEPES and 5 mM glucose (pH 7.3). Stock solutions of KN62 (Calbiochem), CsA (Calbiochem) and nimodipine (Sigma/RBI) were first prepared in DMSO and then diluted (1:1,000) in HBS when being used. Throughout the recording, the culture was perfused with fresh bath solution at a constant rate of about 1 ml/min. Signals (filtered at 5 kHz) were acquired at a sampling rate of 10 kHz using a 16-bit digitizing board (E6035; National Instruments) interfaced with a custom program based on LabView (National Instruments), Igor Pro (WaveMetrics) or MatLab (MathWorks). Series resistances (10−30 MΩ) and input impedance (300−500 MΩ) were monitored by a test hyperpolarizing pulse (5 mV, 100 ms). Data were accepted for analysis only in the cases where series resistance and input impedance did not vary beyond 10% throughout the experiment. Trials showing significant run-up or run-down during the control period (>5% in 10 min) were also excluded from further analysis. To minimize the complication of connectivity with other neurons that were not monitored by our recording, we examined only pairs of neurons that were found on isolated patches of glial cells. Neighboring neurons that may have been connected to the pair were removed with a suction pipette. Only monosynaptic connections between two glutamatergic neurons were included in the current

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

192

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
study. Polysynaptic connections were identified based on the latency of EPSC or inhibitory postsynaptic current onset (>5 ms) and were excluded from the study because their activation timing could not be precisely controlled. The STDP ratio was calculated from the averaged EPSC amplitude during two time periods: within 10 min before and between 15 and 30 min after the stimulation paradigm.
ACKNOWLEDGMENTS We thank P. Lau for providing data for some STDP experiments; E. Aizenman, K. Kandler, P. Lau and J. Rubin for comments on the manuscript and members of the Bi lab for discussions. This work was supported by grants from Burroughs Wellcome Fund (Career Award in the Biomedical Sciences) and the National Institute of Mental Health (R01 MH066962) to G.Q.B. Note: Supplementary information is available on the Nature Neuroscience website. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 13 October; accepted 20 December 2004 Published online at http://www.nature.com/natureneuroscience/
1. Bliss, T.V. & Lømo, T. Long-lasting potentiation of synaptic transmission in the dentate area of the anaesthetized rabbit following stimulation of the perforant path. J. Physiol. (Lond.) 232, 331–356 (1973). 2. Hebb, D.O. The Organization of Behavior (Wiley, New York, 1949). 3. Constantine-Paton, M., Cline, H.T. & Debski, E. Patterned activity, synaptic convergence, and the NMDA receptor in developing visual pathways. Annu. Rev. Neurosci. 13, 129–154 (1990). 4. Bi, G.-Q. & Poo, M.-M. Synaptic modifications by correlated activity: Hebb’s postulate revisited. Annu. Rev. Neurosci. 24, 139–166 (2001). 5. Bliss, T.V. & Collingridge, G.L. A synaptic model of memory: long-term potentiation in the hippocampus. Nature 361, 31–39 (1993). 6. Linden, D.J. & Connor, J.A. Long-term synaptic depression. Annu. Rev. Neurosci. 18, 319–357 (1995). 7. Malenka, R.C. & Nicoll, R.A. Long-term potentiation—a decade of progress. Science 285, 1870–1874 (1999). 8. Abbott, L.F. & Nelson, S.B. Synaptic plasticity: taming the beast. Nat. Neurosci. 3 (Suppl.), 1178–1183 (2000). 9. Sjostrom, P.J. & Nelson, S.B. Spike timing, calcium signals and synaptic plasticity. Curr. Opin. Neurobiol. 12, 305–314 (2002). 10. Markram, H., Lubke, J., Frotscher, M. & Sakmann, B. Regulation of synaptic efficacy by coincidence of postsynaptic APs and EPSPs. Science 275, 213–215 (1997). 11. Magee, J.C. & Johnston, D. A synaptically controlled, associative signal for Hebbian plasticity in hippocampal neurons. Science 275, 209–213 (1997). 12. Mehta, M.R., Barnes, C.A. & McNaughton, B.L. Experience-dependent, asymmetric expansion of hippocampal place fields. Proc. Natl. Acad. Sci. USA 94, 8918–8921 (1997). 13. Bi, G.-Q. & Poo, M.-M. Synaptic modifications in cultured hippocampal neurons: Dependence on spike timing, synaptic strength, and postsynaptic cell type. J. Neurosci. 18, 10464–10472 (1998). 14. Debanne, D., Gahwiler, B.H. & Thompson, S.M. Long-term synaptic plasticity between pairs of individual CA3 pyramidal cells in rat hippocampal slice cultures. J. Physiol. (Lond.) 507, 237–247 (1998). 15. Zhang, L.I., Tao, H.W., Holt, C.E., Harris, W.A. & Poo, M.-M. A critical window for cooperation and competition among developing retinotectal synapses. Nature 395, 37–44 (1998). 16. Nishiyama, M., Hong, K., Mikoshiba, K., Poo, M.-M. & Kato, K. Calcium release from internal stores regulates polarity and input specificity of synaptic modification. Nature 408, 584–588 (2000). 17. Feldman, D.E. Timing-based LTP and LTD at vertical inputs to layer II/III pyramidal cells in rat barrel cortex. Neuron 27, 45–56 (2000). 18. Sjostrom, P.J., Turrigiano, G.G. & Nelson, S.B. Rate, timing, and cooperativity jointly determine cortical synaptic plasticity. Neuron 32, 1149–1164 (2001). 19. Froemke, R.C. & Dan, Y. Spike-timing-dependent synaptic modification induced by natural spike trains. Nature 416, 433–438 (2002). 20. Stevens, C.F. Strengths and weaknesses in memory. Nature 381, 471–472 (1996). 21. Berninger, B. & Bi, G.-Q. Synaptic modification in neural circuits: A timely action. Bioessays 24, 212–222 (2002). 22. Song, S., Miller, K.D. & Abbott, L.F. Competitive Hebbian learning through spike-timing-dependent synaptic plasticity. Nat. Neurosci. 3, 919–926 (2000). 23. van Rossum, M.C., Bi, G.-Q. & Turrigiano, G.G. Stable hebbian learning from spike timing-dependent plasticity. J. Neurosci. 20, 8812–8821 (2000). 24. Rubin, J., Lee, D.D. & Sompolinsky, H. Equilibrium properties of temporally asymmetric Hebbian plasticity. Phys. Rev. Lett. 86, 364–367 (2001). 25. Senn, W. & Buchs, N.J. Spike-based synaptic plasticity and the emergence of direction selective simple cells: mathematical analysis. J. Comput. Neurosci. 14, 119–138 (2003). 26. Gutig, R., Aharonov, R., Rotter, S. & Sompolinsky, H. Learning input correlations through nonlinear temporally asymmetric Hebbian plasticity. J. Neurosci. 23, 3697– 3714 (2003). 27. Hopfield, J.J. & Brody, C.D. Learning rules and network repair in spike-timing-based computation networks. Proc. Natl. Acad. Sci. USA 101, 337–342 (2004). 28. Lisman, J. A mechanism for the Hebb and the anti-Hebb processes underlying learning and memory. Proc. Natl. Acad. Sci. USA 86, 9574–9578 (1989). 29. Malenka, R.C. et al. An essential role for postsynaptic calmodulin and protein kinase activity in long-term potentiation. Nature 340, 554–557 (1989). 30. Malinow, R., Schulman, H. & Tsien, R.W. Inhibition of postsynaptic PKC or CaMKII blocks induction but not expression of LTP. Science 245, 862–866 (1989). 31. Artola, A. & Singer, W. Long-term depression of excitatory synaptic transmission and its relationship to long-term potentiation. Trends Neurosci. 16, 480–487 (1993). 32. Mulkey, R.M., Endo, S., Shenolikar, S. & Malenka, R.C. Involvement of a calcineurin/ inhibitor-1 phosphatase cascade in hippocampal long-term depression. Nature 369, 486–488 (1994). 33. Lisman, J., Schulman, H. & Cline, H. The molecular basis of CaMKII function in synaptic and behavioural memory. Nat. Rev. Neurosci. 3, 175–190 (2002). 34. Ito, I., Hidaka, H. & Sugiyama, H. Effects of KN-62, a specific inhibitor of calcium/ calmodulin-dependent protein kinase II, on long-term potentiation in the rat hippocampus. Neurosci. Lett. 121, 119–121 (1991). 35. Yao, H. & Dan, Y. Stimulus timing-dependent plasticity in cortical processing of orientation. Neuron 32, 315–323 (2001). 36. Schuett, S., Bonhoeffer, T. & Hubener, M. Pairing-induced changes of orientation maps in cat visual cortex. Neuron 32, 325–337 (2001). 37. Izhikevich, E.M. & Desai, N.S. Relating STDP to BCM. Neural Comput. 15, 1511–1523 (2003). 38. Bi, G.Q. Spatiotemporal specificity of synaptic plasticity: cellular rules and mechanisms. Biol. Cybern. 87, 319–332 (2002). 39. Shouval, H.Z., Bear, M.F. & Cooper, L.N. A unified model of NMDA receptor-dependent bidirectional synaptic plasticity. Proc. Natl. Acad. Sci. USA 99, 10831–10836 (2002). 40. Yang, S.N., Tang, Y.G. & Zucker, R.S. Selective induction of LTP and LTD by postsynaptic [Ca2+], elevation. J. Neurophysiol. 81, 781–787 (1999). 41. Liu, L. et al. Role of NMDA receptor subtypes in governing the direction of hippocampal synaptic plasticity. Science 304, 1021–1024 (2004). 42. Scannevin, R.H. & Huganir, R.L. Postsynaptic organization and regulation of excitatory synapses. Nat. Rev. Neurosci. 1, 133–141 (2000). 43. Sheng, M. & Sala, C. PDZ domains and the organization of supramolecular complexes. Annu. Rev. Neurosci. 24, 1–29 (2001). 44. Petersen, J.D. et al. Distribution of postsynaptic density (PSD)-95 and Ca2+/calmodulin-dependent protein kinase II at the PSD. J. Neurosci. 23, 11270–11278 (2003). 45. Graef, I.A. et al. L-type calcium channels and GSK-3 regulate the activity of NF-ATc4 in hippocampal neurons. Nature 401, 703–708 (1999). 46. Dolmetsch, R.E., Pajvani, U., Fife, K., Spotts, J.M. & Greenberg, M.E. Signaling to the nucleus by an L-type calcium channel-calmodulin complex through the MAP kinase pathway. Science 294, 333–339 (2001). 47. West, A.E. et al. Calcium regulation of neuronal gene expression. Proc. Natl. Acad. Sci. USA 98, 11024–11031 (2001). 48. Rubin, J.E., Gerkin, R.C., Bi, G.-Q. & Chow, C.C. Calcium time course as a signal for spike-timing dependent plasticity. J. Neurophysiol. (in the press). 49. Zhou, Q. & Poo, M.M. Reversal and consolidation of activity-induced synaptic modifications. Trends Neurosci. 27, 378–383 (2004). 50. Zhou, Q., Tao, H.W. & Poo, M.M. Reversal and stabilization of synaptic modifications in a developing visual system. Science 300, 1953–1957 (2003).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

193

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Invariant computations in local cortical networks with balanced excitation and inhibition
Jorge Mariño1,3, James Schummers1, David C Lyon1,3, Lars Schwabe2, Oliver Beck2, Peter Wiesing2, Klaus Obermayer2 & Mriganka Sur1
Cortical computations critically involve local neuronal circuits. The computations are often invariant across a cortical area yet are carried out by networks that can vary widely within an area according to its functional architecture. Here we demonstrate a mechanism by which orientation selectivity is computed invariantly in cat primary visual cortex across an orientation preference map that provides a wide diversity of local circuits. Visually evoked excitatory and inhibitory synaptic conductances are balanced exquisitely in cortical neurons and thus keep the spike response sharply tuned at all map locations. This functional balance derives from spatially isotropic local connectivity of both excitatory and inhibitory cells. Modeling results demonstrate that such covariation is a signature of recurrent rather than purely feed-forward processing and that the observed isotropic local circuit is sufficient to generate invariant spike tuning.

Processing networks in sensory cortex carry out transformations on their inputs so as to create outputs that are relevant for perception and action. These transformations are characteristic of an area, rely on discrete local circuits and are computed both dynamically and invariantly despite variations in functional architecture within the area1. For example, the computation of feature-selective responses, such as orientation selectivity in primary visual cortex (V1), involves the integration of excitatory and inhibitory inputs arising from a variety of sources to produce responses that are sharply tuned for the orientation of visual stimuli and that are also influenced adaptively by the history of stimulation2–6. An understanding of these computations requires a description of the behavior of neurons within the context of their cortical circuit2,7,8. The composition of the local circuit varies systematically across the orientation preference map in V1 (refs. 9,10); the local network near pinwheel centers contains a broad orientation distribution, whereas the network far from pinwheels, in orientation domains, contains a homogeneous representation. The impact of a neuron’s local neighborhood on its responses has recently been described in V1 of cats: neurons at pinwheel centers have more broadly tuned subthreshold responses compared with neurons in orientation domains11,12, yet the spike responses are sharply tuned for orientation whatever the orientation map location13,14. Recent studies in vitro and in vivo have emphasized the interplay between excitation and inhibition as an essential mechanism for stabilizing and shaping neural activity15–17. Here we have examined the mechanism by which invariant orientation tuning is created in V1 despite the diversity of local environments, by measuring electrophysiologically the excitatory and inhibitory synaptic conductances in neurons at different map
1Department

positions and describing anatomically the inputs to these neurons. Computational models of cellular and network behavior support the conclusion that the cortical network operates in a recurrent rather than a purely feed-forward mode, and that simple rules of spatial integration of excitation and inhibition can explain sharp orientation tuning at all locations in the orientation map. RESULTS Synaptic conductances at different map locations We combined optical imaging of intrinsic signals and whole-cell recordings in vivo (see Methods and Supplementary Figs. 1 and 2 online) to measure the orientation tuning of synaptic conductances of cells located at different sites in the orientation preference map of cat V1. Illustrated here are data from two cells, one located in an orientation domain (Fig. 1a–d) and another at a pinwheel center (Fig. 1e–h). The spike responses of both cells were sharply tuned for orientation (Fig. 1b,f). The visually evoked membrane potential response (Vm), obtained with different levels of intracellular current injection (Fig. 1d,h), was used to calculate changes in total conductance (g)18,19. For 7/7 orientation domain cells and 7/11 pinwheel cells, the preferred orientation of g was aligned (within ± 22.5°) with the preferred orientation of the spiking response. The remaining four pinwheel neurons showed a displacement of 45–67.5° in the peak g relative to spike tuning. This difference was due to broadly tuned conductances rather than to a rotation of narrow g tuning curves. The degree of selectivity was quantified using the orientation selectivity index (OSI), which is a global measure of tuning across the entire tuning curve11. Orientation domain cells displayed a clear difference in the magnitude of g between

2Department

of Brain and Cognitive Sciences and Picower Center for Learning and Memory, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA. of Computer Science and Electrical Engineering, Berlin University of Technology, FR2-1, Franklinstrasse 28/29, 10587, Berlin, Germany. 3Present addresses: Department of Medicine, Neuroscience and Motor Control Group (Neurocom), Univ. A Coruña, Fac. CC. da Saúde, Campus de Oza, 15006, A Coruña, Spain (J.M.), The Salk Institute, SNL-C, 10010 North Torrey Pines Road, La Jolla, California 92037, USA (D.C.L.). Correspondence should be addressed to M.S. ([email protected]). Published online 23 January 2005; doi:10.1038/nn1391

194

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES

a

b
Response
1.0
Spk

c
1.0

e

f
Response
1.0 0.5 0
Spk

g
1.0

OSI

OSI

0.5 0

0.5 0

0.5 0

Vm Spk g gi ge

d
© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience
Vm

h
Vm

250

190

g (%)
100 20

g (%)
100 30

∆ gi (ns) ∆ ge
0

∆ gi (ns) ∆ ge
0

Figure 1 Orientation tuning of synaptic conductances for a cell in an orientation domain (a–d) and at a pinwheel center (e–h). (a,e) Orientation preference maps from V1 (a small region is shown, from a large region that was imaged). Filled circles indicate the location of recording electrodes. Color-coded semicircle in a indicates the angle of orientation preference of each pixel. Scale bars: 500 µm. (b,f) Orientation tuning curves of the spike responses. (c,g) OSIs calculated for the membrane potential (Vm), spike response (Spk), total conductance (g) and inhibitory and excitatory conductances (gi and ge). Vm and the underlying conductances (g, gi and ge) are relatively narrowly tuned for the orientation domain cell but are broadly tuned for the pinwheel center cell. Spike tuning is invariantly sharp for both cells. (d,h) Traces of the average visually evoked changes in Vm, g, gi and ge in response to each of eight stimulus orientations spanning 180°. (Top rows) Vm traces. Black lines show the mean activity for three to five trials under resting conditions and for two levels of current injection (–0.2 and 0.1 nA). Red lines show the predicted Vm traces from linear regression fits used to obtain g (see Methods). The predicted and actual traces are in good agreement. Horizontal scale under g, 1 s (visual stimulation time); vertical scale, 10 mV in d and 5 mV in h. (Middle rows) g traces showing changes relative to rest. (Bottom rows) ∆gi and ∆ge traces showing absolute changes in gi and ge.

preferred and orthogonal stimulation (leading to high OSI values; Fig. 1c), whereas for pinwheel neurons there was a large increase in g for all stimulus orientations (leading to low OSI values; Fig. 1g). To reveal the synaptic mechanisms involved in the transformation of the diverse g tuning curves at different map locations into uniformly sharp spike tuning, we calculated the visually evoked changes in inhibitory and excitatory conductances (gi and ge)4,19. The mean absolute change in inhibitory conductance was always larger than for excitation, independent of orientation or map location. Similar to the differences observed for g, the OSIs for gi and ge were lower for pinwheel cells (Fig. 1c,g). Figure 2 shows, for four additional cells, the relationship between the OSI of visually evoked increases in gi and ge and map location. Orientation domain cells (Fig. 2a,b) responded to their preferred orientation with a large increase in gi and ge when compared with the orthogonal orientation, but this difference was smaller for pinwheel cells (Fig. 2c,d), giving rise to lower OSIs. For all cells, the OSI values for gi and ge covaried, indicating that regardless of location, inhibition always seemed to balance excitation. Laminar position and receptive field type20,21 (simple or complex) may also influence the integration of inputs by V1 cells. We conducted a threeway ANOVA to compare OSIs for g, gi and ge between map locations (pinwheel/domain), recording depth and cell type, and found a significant effect of orientation map location (all P values < 0.007) but no effect of the other two variables (all P values > 0.3). Subsequently, for the population analysis (Fig. 3), we did not differentiate between cell types or cortical depths.

Compared with orientation domains, the average tuning curve for ge in pinwheels was broader and showed a larger offset at orthogonal orientations (Fig. 3b). This could potentially be explained by an isotropic pattern of local connections, in which neurons located at or near pinwheel centers would receive inputs from neurons with different orientation preferences, flattening the Vm and conductance tuning curves (Fig. 3a), whereas orientation domain cells would be primarily driven by cells sharing the same orientation. Figure 3b shows that increases in gi run in parallel with ge; this inhibition, probably a mechanism to counteract the excitatory input, is especially prominent in pinwheel neurons at nonpreferred orientations, allowing these cells to display a sharp spike tuning despite broad excitatory inputs. The population analysis indicated that OSIs for both gi and ge (Fig. 3c) are significantly different between neurons located in orientation domains and pinwheel centers (P < 0.002 for gi, P < 0.001 for ge), indicating that V1 neurons receive different inputs depending on their location in the orientation map. The source of this heterogeneity is probably the structure of the map itself. Because we directed our recordings to orientation domains and pinwheel centers, these differences probably show the two extremes of a more gradual variation along the orientation map. To quantify the specificity of the surrounding orientation representation, we calculated the OSI of the orientation distribution of pixel counts in the map surrounding each neuron; cells located at pinwheels have low local input OSIs, whereas the OSIs are progressively higher as neurons move toward a domain center11,14. We found a significant correlation (r = 0.79, P < 0.0001) between the OSI of the conductance tuning curve and

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

Spk g gi ge
195

Vm

ARTICLES

a
1.0

c
1.0 0.5 0

0.5 0

15

20

∆ gi (ns) ∆ ge

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

0 Preferred orientation Orthogonal orientation

0

Figure 2 Additional examples of conductance tuning of cells in orientation domains and pinwheel centers. (a,b) Two examples of changes in gi and ge following visual stimulation (1 s, bars below traces) in orientation domain cells, and the associated OSIs for Vm, spike activity and changes in conductance. All conventions are as in Figure 1. (c,d) Two examples of conductance changes and OSIs in pinwheel cells. All conventions are as in Figure 1.

OSI

Vm Spk g

gi ge

b
1.0 0.5 0

d

∆ gi (ns) ∆ ge

6

14

0

0

the OSI of the local input region (Fig. 3d), suggesting a relationship between the tuning of synaptic inputs and the orientation representation in the local circuit across a cortical distance of only a few hundred microns. A similar relationship exists for gi and ge (see Fig. 5g,h), and indeed between gi and ge (r = 0.99, P < 0.0001). These relationships support the proposal that observed variations in the tuning of g, gi and ge are related to functional heterogeneities in the local inputs. The data suggest the existence of a single mechanism that is able to balance the different patterns of excitation and inhibition at different locations, keeping the spike response equally selective at any site. Thus, we proposed that the different tuning curves of conductance at different locations of the orientation map may be achieved by a common principle: a spatially isotropic pattern of local excitatory and inhibitory connections. The heterogeneity in tuning of synaptic inputs would then be a consequence of applying a uniform rule of anatomical pooling and synaptic integration to a heterogeneous functional map. To explore this possibility, we first examined the anatomical distribution of local excitatory and inhibitory inputs to neurons at different locations in the orientation map and subsequently used single-neuron and network models to analyze the effect of synaptic pooling within the anatomical input zone on g, Vm and spike responses. Anatomical inputs to different map locations We used extremely small injections (uptake zone <100 µm in diameter) of retrograde tracers to study the structure of local inputs to different sites in the orientation map, combined with labeling for GABAergic cells (Fig. 4a). Distributions of inhibitory neurons have been described less completely in previous reports22,23. Our technique of double-labeling cells from very small tracer injection and staining for the GABA antibody allowed us to demonstrate for the first time a more complete view of the distribution of local excitatory and inhibitory cells in V1. More importantly, our technique provided a fine-grained comparison of the projection patterns to orientation domains and pinwheel centers. Despite drastic differences in local orientation distributions at pinwheels and domains, we found no differences in the spatial distribution of either local inhibitory or excitatory cells labeled at these sites (Fig. 4b–d). An example of the pattern of retrogradely labeled cells from

a domain and a pinwheel injection is depicted (Fig. 4b). Independent of location, the pattern of labeled cells around the injection site was 1.0 always roughly circular, ignoring the distribution of orientation preferences24. This distri0.5 bution of cells, as shown for the two individual cases (Fig. 4c) and for the population (Fig. 4d), 0 indicated a local isotropic radius of influence of ∼250 µm. We computed orientation tuning curves of the inputs to each injection site by assigning an orientation preference to each labeled cell according to the optically imaged orientation preference map recorded in the same animal. Within this radius, the OSIs for cells labeled by pinwheel injections (n = 3) were much lower than those for orientation domains (n = 4) (Fig. 4e). Figure 4e also shows that the OSI differences between pinwheels and domains were similar for inhibitory and excitatory cells. Thus, the anatomical data provide a potential substrate for the electrophysiological measurements, suggesting that broader excitatory and inhibitory synaptic conductances at pinwheel centers arise naturally from spatially isotropic local projections. Model tuning at different map locations The pattern of anatomical connections can be combined with the strength of synaptic drive to explain the physiological responses of neurons. We used complementary single-neuron and network models to derive the excitatory and inhibitory conductances under which spike tuning of V1 neurons would be invariant with map location, which we then compared with the measured conductances. Thus, we first set up a Hodgkin-Huxley type single-neuron model25 (see Methods and Supplementary Notes online for a detailed description). By convolving the experimentally obtained spatial excitatory input profiles (Fig. 4d) with experimentally obtained orientation maps, we calculated the tuning of the excitatory conductance ge at locations ranging from pinwheels to domains. Then, given these ge curves, we determined the gi tuning curves that yield sharp spike tuning for each location (Fig. 5a,b, and Supplementary Fig. 3a online). These tuning curves are similar to the experimentally measured ge and gi tuning curves (Fig. 3b). Furthermore, the offset and slope for the OSIs as a function of local input OSIs fall within the 95% confidence interval of the experimentally measured values (see Supplementary Fig. 3a online). As with the experimental data, the difference between pinwheel locations and orientation domains is reflected in the subthreshold signal, but as a result of the appropriate inhibitory balance at orthogonal orientations, which keeps the membrane potential below threshold, it is not reflected in the spike responses (Fig. 5c,d and Supplementary Fig. 3 online). These model results indicate that, given the constraints of anatomical location within an orientation map, the inhibitory tuning that we measured resembles the tuning that is necessary, in theory, to balance excitation and yield sharp spike tuning at all locations.

196

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES

a
1.0
Amplitude

Domain Pinwheel

1.0 Vm g 0.5

0.5

0

0

90 Orientation

180

0

0

90

180

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

b
1.0
Amplitude

1.0 gi ge 0.5

0.5

Figure 3 Average tuning differences between cells at pinwheels and orientation domains. (a) Normalized tuning curves for mean changes in Vm and g in orientation domain and pinwheel neurons (± s.e.m.). A Gaussian function is fitted to the data for illustration purposes only; all quantitative comparisons are based on OSIs derived from actual data points for each cell. (b) Normalized tuning curves for changes in gi and ge in orientation domain and pinwheel cells. Both conductances were normalized together for each individual cell, and then population means were computed. Conventions are as in a. (c) Bar plot comparing mean gi and ge OSIs for the population of orientation domain and pinwheel cells. Asterisks: Student’s t-test, P < 0.002. (d) Scatter plot of the OSI values of g and the OSIs of the local input region (radius: 250 µm) for each recorded neuron. The orientation distribution of pixels in this local region was calculated from the actual orientation preference map. Triangles: pinwheel cells; circles: domain cells. Lower line is the least-squares linear fit to the data (r = 0.79; P < 0.0001). For comparison, the OSI values (squares) and linear fit (upper line; r = –0.06; P = 0.82) of the spike response (Spk) are also shown.

0

0

90 Orientation
Domain Pinwheel

180

0

0

90

180

c
0.6
OSI

d

*

*
0.5
g OSI

0.5

0

gi

ge

0 0 0 1 Local input OSI

Spk OSI

1.0

1.0

(the signature of a recurrent versus a feed-forward mode) no longer fell within the 95% confidence interval from the measurements. These simulations demonstrate that the tuning curves of the excitatory and inhibitory conductances deduced with the single-cell model indeed occur in a recurrent network, if the local excitatory and inhibitory synaptic inputs are balanced and both contribute significantly to the driving input. DISCUSSION Our results show that visual stimulation evokes a different pattern of synaptic inputs at orientation domains compared with pinwheel centers. We demonstrate that these response patterns result from diverse synaptic inputs impinging on different locations in the orientation map, acting through a locally isotropic and recurrent anatomical architecture. That is, the spatial distribution of excitatory and inhibitory neurons provides the necessary anatomical inputs, and their synaptic drive provides sufficient functional balance to preserve sharp spike tuning, particularly at pinwheel centers. The role of inhibition in orientation selectivity The generation of orientation selectivity in visual cortex includes mechanisms that shape two related aspects of a neuron’s response: its preferred orientation and its orientation selectivity or tuning strength. Recent evidence suggests that the preferred orientation of a V1 neuron arises from the feed-forward bias of its afferent inputs29–31. Orientation selectivity seems to be narrower than afferent spread and probably requires intracortical mechanisms for its generation12. These mechanisms potentially include the spike threshold of the neuron, recurrent excitation between cortical neurons and intracortical inhibition. The spike threshold is a nonlinearity that sharpens the selectivity of spike outputs relative to the selectivity of excitatory synaptic inputs32,33. It has been suggested that the spike threshold is dynamically regulated to enhance orientation selectivity34, but there is no evidence that the spike threshold varies with map location. The role of intracortical excitation and inhibition in generating orientation selectivity remains unresolved2,3,35,36. If intracortical mechanisms were to have a role, it is probable that their effect would be observed most clearly at pinwheels, where the local cortical network would provide broadly tuned excitation, and inhibition would be required to counter this spread. Our experimental results are consistent with recent findings that broad subthreshold excitation is present at pinwheels11,12, and we now demonstrate that inhibition balances excitation so that both are required for sharp spike tuning. These results help reconcile several findings that seem to contradict each other, particularly in relation to the preferred orientation and tuning

Finally, we examined whether the necessary covariation between excitation and inhibition also emerges in a cortical network similar to V1. In a previous modeling study of a V1 network that explicitly included pinwheels and orientation domains26,27, it was concluded that local isotropic connectivity leads to orientation tuning of spike responses that strongly depend on location in the orientation map, and in particular to sharper tuning at pinwheel centers (a prediction not supported by experimental findings11,13,14, including the present study) (Figs. 1,2). Thus, we set up a large-scale network of Hodgkin-Huxley type model neurons to determine whether and under what conditions it can produce the covarying excitation and inhibition leading to sharp orientation tuning invariant with map location. In the absence of evidence for location-specific feed-forward tuning, we assumed that the afferent drive is tuned similarly for cells across the map. Unlike previous models7,8,28, the model network had identical local excitatory and inhibitory connection length scales, determined from the results of our tracer injection experiments (Fig. 4d; see Supplementary Notes online) and was parameterized first to operate in a regime in which the recurrent excitation contributes considerably to visual responses. The model’s predicted tuning for gi and ge is shown in Figure 5e for pinwheels and orientation domains. Figure 5f–h shows that the tuning of g, gi and ge clearly depended on the local input OSI and hence on map location, matching the covarying excitatory and inhibitory conductance OSIs measured experimentally. The slope and intercept values for this relationship fell within the 95% confidence intervals from the measured values. As in the single-cell model, this led to location-independent spike tuning (see Supplementary Fig. 4 online). We then parameterized our model to operate in a range of conditions, including a regime in which recurrent excitation was weak, inhibition dominated and the neurons were mainly driven by feed-forward inputs. In this case, excitation and inhibition did not covary, and the spike tuning became dependent on map location (see Supplementary Fig. 5 online). Now the slope of ge
NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

197

ARTICLES
of inhibition and its contribution to orientation selectivity. Measurements of the excitatory and inhibitory synaptic conductances underlying orientation tuning in cat V1 have found large conductance changes18 but either strictly iso-oriented inhibition19 or diversity in the preferred orientation of inhibition4. We show that the diversity can be explained at least in part by map location: orientation domains contain iso-oriented excitation and inhibition, whereas pinwheels can show variable relationships among excitation, inhibition and spike tuning, owing mainly to broadly tuned conductances rather than to narrowly tuned crossoriented inhibition. Other reasons for divergent preferred orientations may include differences in the stimulus (full-field gratings versus bars) or uncontrolled differences in laminar position21. Measurements of the time course of responses in monkey V1 have found either stable tuning37,38 or dynamic changes in tuning over time, including strong suppression of nonpreferred orientations late in the response39. Our findings predict that response dynamics would be more variable at pinwheels than at orientation domains (and would be seen in a relatively small proportion of randomly sampled cells, because regions of rapid orientation change such as pinwheels and their neighborhood occupy a small fraction of the cortical surface14). Indeed, reverse correlation analyses at pinwheels and orientation domains in cat V1 support this prediction (J. Schummers, J. Mariño, M. Sur, Soc. Neurosci. Abstr. 818.6, 2003). Pharmacological manipulation of inhibitory inputs to neurons have also shown either no effect of intracellular inhibitory blockade on orientation selectivity40 or a broadening of tuning following extracellular iontophoresis of blockers at cross-oriented sites ∼500 µm away36. A plausible explanation for the former result is that intracellular blockade of inhibition at orientation domains may have little effect on tuning. For the latter result, the effect would be much greater at pinwheels, where nearby sites also probably have very different preferred orientations. Overall, it is probable that the tuning of neurons near pinwheel centers is more sensitive to changes in the balance of inhibition and excitation. This proposal is supported by the finding that visual pattern adaptation induces short-term shifts in the preferred orientation and tuning strength of neurons much more readily at pinwheels than at orientation domains14. Invariant tuning with balanced excitation and inhibition Consistent with a previous report19, we demonstrate a close relationship between the spread of excitation and inhibition, which are similarly tuned regardless of map location. The source of these synaptic inputs is probably the local neighborhood of a neuron24, although we cannot rule out at least excitatory inputs from iso-oriented sites that are located more distantly41,42. Importantly, although previous models of orientation selectivity have invoked broader inhibition compared with excitation7,8,28 to generate sharp tuning, our single-cell and network models show that balanced excitation and inhibition are sufficient to produce sharp tuning at all locations. Furthermore, in contrast to recent models that are based on an idealized structure of orientation maps26,27, our network model has the distinct advantage of incorporating parameters based on experimentally measured spatial profiles of excitatory and inhibitory convergence and projecting these spatial profiles onto experimentally obtained orientation maps. Future work will need to address the possible influence of the
Double labeled

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

a

Injection site

Tracer cells

GABA cells

b
Figure 4 Anatomical analysis of local excitatory and inhibitory projections to pinwheel centers and orientation domains. (a) Example of an injection site (scale bar: 100 µm), and three images of the same region showing retrogradely labeled, GABA-positive and double-labeled cells, respectively. For illustration purposes, the two middle images were high-pass filtered and merged to show the double-labeled neurons at right. (b) Examples of the distribution of cells within the orientation map following an injection of tracer in an orientation domain (left) and a pinwheel center (right). Black: excitatory cells; white: inhibitory cells. The two outer concentric circles are at 250-µm intervals from the injection site (inner circle). Scale bar: 250 µm. (c) Histograms and tuning curves computed from the orientation domain (left) and pinwheel (right) cells depicted in b. Histograms show the number of inhibitory and excitatory cells located at different distances from the injection. The orientation tuning curves indicate the number of inhibitory and excitatory cells found in the different orientation regions, within a distance of 250 µm. (d) Population curves of three pinwheel and four domain injections showing the proportion of inhibitory and excitatory cells found at different distances from the injection. (e) Bar plot of the mean OSIs for the population data. Error bars in d and e show s.e.m.

No. of cells

c 70
35 0 50

Excitatory Inhibitory 140 70 0 250 450 0 Distance (µm) Domain Pinwheel 90 180 Orientation

50 25 0

140 70 50 250 450 Distance (µm) 0 0 90 180 Orientation Domain Pinwheel

d
Mean % of cells

20

e
0.8 Mea n OS I

10 Excitatory 0 50 250 Distance (µm) Inhibitory 450

0.4

0

Inhibitory

Excitatory

198

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
Single-cell model
Figure 5 Model predictions for synaptic conductances, membrane potentials and spike responses underlying orientation selectivity across the orientation map. (a,b) Inhibitory and excitatory conductance tuning for a pinwheel cell and an orientation domain cell derived from the single-cell model. The afferent input in orientation space is described by a Gaussian function, σ = 25°, added to an offset of 10% of the maximum value. (c) Vm tuning at a pinwheel center and in an orientation domain with the input conductances from a and b. (d) Spike response tuning at a pinwheel center and in an orientation domain (the two are nearly identical but have different Vm tuning as shown in c, created by different gi and ge tuning as shown in a and b). (e) OSIs of gi and ge for pinwheels and orientation domains derived from the network model. (f) OSIs of the total conductance as a function of the local input OSI for the network model (gray), and data points from the experiments (black). (g,h) OSIs of gi and ge as a function of the local input OSI for the network model (gray), and data points from the experiments (black).

a
Amplitude

1.0 0.5 0
1.0

gi Domain Pinwheel

b

1.0 0.5 0

ge

0

90 180 Orientation

0 Spk

90

180

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

c
Amplitude

Vm

d

1.0 0.5 0

0.5 0

0

90 180 Orientation
Network model

0

90

180

e

0.6 OSI

f
Domain Pinwheel gi ge

0.9 g OSI

0

0

0 0.9 Local input OSI

g

0.9 gi OSI

h

0.9 ge OSI

0 0 0.9 Local input OSI

0

0 0.9 Local input OSI

temporal structure of voltage fluctuations34,43 and the relative timing of excitation and inhibition16. Our conductance measurements, together with our network model, constrain the regime in which visual cortex networks probably operate to generate orientation tuning. We show that a purely feed-forward regime is incompatible with the data: the regime predicts that the tuning of ge would show no relationship to map location, whereas the data show a strong relationship between the tuning of ge and the local orientation distribution. (Because the tuning of gi derives mainly from the local inhibitory network, it is always location dependent, being broad at pinwheels and narrow at orientation domains. A feed-forward model would thus generate sharper spike tuning at pinwheels than at orientation domains, which is not seen experimentally.) The model operates in a balanced regime in which recurrent excitation contributes significantly to neuronal responses, but not in an extreme recurrent regime, or ‘marginal phase’28, where the afferent input ‘selects’ predefined response patterns and the tuning width is strongly determined by the pattern of orientation-selective intracortical connections. In the model’s regime, predetermined response patterns do not exist; rather, the tuning depends on characteristics of the afferent input. Hence, broadly tuned afferent input does not lead to sharp output tuning as expected for a network operating in the marginal phase, but moderately tuned afferent input is sharpened by the cortical network (Supplementary Notes). Furthermore, in this regime, the tuning of ge correlates well with map location, matches the measured data closely, is balanced by a covarying gi, and generates location-independent spike tuning.

Although the match between the tuning of conductance and of the local orientation network is comparable in our data and model, there is more scatter in the data. This may result from small errors in localization of electrode penetrations (Supplementary Fig. 2 online), pooling of data from different cortical layers, or other experimental variables. Another possible source of noise in the data is the influence of dendritic processing. We have measured synaptic conductances at the soma, but inputs to the dendrites may well be different from what is detectable at the soma44. In addition to location invariance, the orientation selectivity of V1 responses is also invariant with stimulus contrast, in which intracortical inhibition has a crucial role7,45. Recent simulations46 and experimental47 work have proposed the existence in cortical layer 4 of two functionally different types of inhibition, generated by simple cells with sharp orientation tuning and by untuned complex cells, respectively. The balance of excitation and inhibition described here is related to the functional architecture of the orientation preference map, without regard to laminar location or cell type. It is still unknown if the proposed difference between inhibitory neurons holds for other cortical layers and, if so, what might be the relative contribution of each type to their target cell responses. Our results are compatible with the presence of both types of cells: simple inhibitory cells would provide the observed tuned responses, whereas an orientation-independent offset could be regulated by inhibition from complex cells. In sum, the measurements of synaptic conductances and anatomical inputs, together with the models of single cells and local networks, provide a comprehensive description of the integration of inputs that underlies the computation of orientation tuning in V1. We have found that a simple rule of spatial integration ensures a balance of excitation and inhibition that produces sharp orientation tuning at all positions in the orientation map. A homeostatic balance between excitation and inhibition has been proposed as a mechanism for the regulation of synaptic strength in developing networks15,48 and for the consolidation of functional connections in cortex during a critical period of visual development49,50. Our results demonstrate, for the first time, the fundamental role of such a balance for a key emergent computation in the adult visual cortex. A similar mechanism based on the balance provided by local inputs may account for the tuning of other functional properties in visual cortex, and may be a general mechanism for generating and preserving response selectivity in sensory cortex16,17. METHODS
Animals. Experiments were done on 23 adult cats that were anesthetized and paralyzed. Stabilization for physiological recordings was achieved through stereotaxic fixation of the head, suspension at lumbar level, drainage of cerebrospinal fluid

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

199

ARTICLES
and bilateral pneumothorax. Subsequent to performance of a craniotomy and durotomy over V1, a steel chamber was mounted on the skull and filled with agar (2% in saline). Eyes were protected with contact lenses and focused on a computer monitor using appropriate lenses. EEG, EKG, expired CO2 and rectal temperature were continuously monitored and used to assess the state of the animal. All protocols were approved by MIT’s Animal Care and Use Committee and conformed to US National Institutes of Health (NIH) guidelines. Optical imaging. Techniques for intrinsic-signal optical imaging were similar to those we have described elsewhere9,11. Responses to full-field drifting square-wave gratings of eight orientations were used to compute orientation preference (angle) maps. Special care was taken in choosing highly accurate and reproducible maps. First, reference images of the surface vasculature were obtained after every 20 trials (∼100 min). Only maps in which there was zero movement were deemed acceptable. Second, angle maps were computed from independent blocks of 15–20 trials and compared. Third, angle maps were computed from the two independent sets of four orientations contained within the stimulus set. If shifts in pinwheel center location or other systematic discrepancies in the layout of the map were observed, recordings were discontinued until the preparation was stabilized. Imaging was only terminated after the independent maps showed extremely minor differences. Furthermore, the specific sites to be targeted for patch recordings or tracer injection were chosen to be the most stable sites, assessed as just described. In most cases, we were able to find pinwheel centers that shifted by no more than 1–2 pixels (13–26 µm). An analysis of the reliability of pinwheel center localization is described in Supplementary Notes online (see also Supplementary Fig. 2 online). Electrophysiology. Procedures for in vivo whole-cell recording were similar to those used previously by us11,40 and others4,19,47. Glass microelectrodes (resistance 6–12 MΩ) filled with a patch solution4,11 were carefully lowered into the cortex and directed to either orientation domains or pinwheel centers, using an image of the surface vasculature aligned to the orientation map as reference. Recordings were made in bridge mode; the extracellularly measured electrode resistance (Re) was neutralized at the beginning of each penetration. Visual stimuli were randomly generated drifting sine-wave gratings of eight orientations, moving in opposite directions, plus a blank stimulus, each of which was presented five times for 1 s. This protocol was repeated while injecting three to four different steady currents (Iinj) ranging from –0.2 to 0.1 nA. Cell parameters were monitored every stimulus cycle by means of I–V curves. The entire protocol was completed in 40 cells, from which 18 were chosen for further analysis according to their stable biophysical properties. The average series resistance was 70.5 ± 39.7 MΩ (mean ± s.d.), input resistance 27.8 ± 20.1 MΩ, time constant 17.4 ± 8 ms and resting potential –50.4 ± 15.1 mV. Compensation for series resistance, input conductance measurements and reliability of the measurements were made using quantitative methods similar to those described by others16,19. Briefly, Re was calculated offline by fitting a double exponential to the cell’s response to current pulses (I–V curves), and its contribution to the membrane potential (Vm) was subtracted from the traces. Total conductance g(t) at time t was estimated by regression as the inverse of the slope of a line fitted to the relation between Iinj and Vm. To calculate the inhibitory gi(t) and excitatory ge(t) conductances, we assumed g(t) = ge(t) + gi(t), and Vrest(t) = [ge(t)Ee + gi(t)Ei]/[ge(t) + gi(t)] where Vrest(t) is the membrane potential in the absence of current injection, and Ee and Ei are the equilibrium potentials for ge(t) and gi(t), respectively. Then, gi(t) and ge(t) can be derived as follows: gi(t) = [g(t)(Vrest(t) – Ee)]/Ei – Ee ge(t) = [g(t)(Vrest(t) – Ei]/Ee – Ei We used Ee = 0 mV and Ei = –80 mV (for the latter, we analyzed all cells using different values between –70 and –90 mV; these had no substantial effects on the main results). The g(t) values obtained from linear regression fits were used to predict the Vm values at each time point; these were compared to the actual values as a goodness-of-fit measure (Fig. 1). For each cell, we calculated the orientation tuning curve and OSI for spiking activity, Vm, g, gi and ge. The OSI is the magnitude of the vector average of the responses to all stimulus orientations, computed as OSI = [√((ΣR(θi) cos(2θi))2 + (ΣR(θi) sin(2θi))2)]/[ΣRi] where R is average response during grating presentation and θ is orientation from 0 to 157.5, indexed by i = 1 to 8. It is a continuous measure with values ranging from 0 (unselective) to 1 (perfectly selective). Anatomy. In V1 of nine cats, glass pipettes (tip diameter 10–20 µm) were used to place three distinct injections (25 nl; pressure injected with a Pico Spritzer II; General Valve) of fluorescent conjugates of cholera toxin subunit B (CTB) (2%; Alexa-Fluor 488, 594 and 647; Molecular Probes) in pinwheels and domains at a depth of ∼600 µm. After 36–48 h, the cats were given an overdose of sodium pentobarbital, perfused, and their brains processed. V1 was removed and sectioned (40 µm) tangential to the surface. The tissue was processed to reveal GABA+ neurons by using a GABA antibody (1:500, rabbit; Sigma) and Alexa-Fluor 350 goat anti-rabbit IgG (1:200). Images of the pattern of labeled neurons near the depth of the injection and laterally were acquired with a Zeiss Axiocam system. Inhibitory neurons were identified by positive staining for the GABA antibody and double labeling from tracer injection. The pattern of labeled cells was aligned to the orientation map using landmarks from three injection sites, visible penetrations from the patch pipettes and blood vessel patterns along the cortical surface and through a depth of 600 µm. To ensure the most detailed anatomical analysis of local connectivity from our sample of 27 injections, we limited analysis to three pinwheel and four domain injections, selected because their spread of tracer uptake was confined to a diameter <100 µm. For these injections, the numbers of excitatory and inhibitory neurons were counted every 50 µm from the boundary of the injection site to a radius of 450 µm. Computer simulations. The single-neuron model used (one compartment, Hodgkin-Huxley type neuron, with Na+, K+ and M currents and balanced background noise inputs25) is described in detail in Supplementary Notes online. Briefly, presynaptic activity (independent Poisson spike trains) was separated into background, feed-forward and recurrent components, which describe the continuing activity not dependent on the stimulus, the afferent stimulus–driven input and the inputs due to the activation of the local network neighborhood. Probabilities of intracortical synaptic connections with presynaptic excitatory neurons were estimated from optically imaged orientation maps. The total excitatory input conductance, given as a function of stimulus orientation, was then computed for local neighborhoods of varying OSI under the assumption that the spike tuning of all excitatory neurons is the same and independent of location in the orientation map. Then the tuning curve of the total inhibitory conductance necessary to obtain the observed sharp tuning curve of the firing rates was calculated. The large-scale network model is also described in detail in Supplementary Notes online. Its main features were as follows: first, the model was composed of Hodgkin-Huxley type point neurons similar to that used for the single-cell model, received synaptic background activity, and had synaptic currents modeled as originating from GABA, AMPA and NMDA receptors. Second, experimentally obtained optically imaged orientation maps were used for assigning orientation preferences to cortical locations. In addition, artificial orientation maps were used for comparison with other models26. The network was composed of up to 128 × 128 neurons and modeled a patch of cortex 2.25 × 2.25 mm2 in size. Third, the afferent inputs to cortical cells were broadly tuned (σ = 27.5°) and were described by Poisson spike trains with a time-independent firing rate. Fourth, spatially isotropic synaptic connections in cortical space, with experimentally determined radial profiles (r = 250 µm) for excitation and inhibition, were used. Fifth, we explored a range of parameters, importantly varying the relative strength of the afferent and the recurrent inputs.
Note: Supplementary information is available on the Nature Neuroscience website.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

200

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
ACKNOWLEDGMENTS This work was supported by Ministerio de Educación y Ciencia, Spain (J.M.), Howard Hughes Medical Institute (J.S.), Deutsche Forschungsgemeinschaft Sonderforschungsbereiche 618, Germany (L.S., O.B., K.O.), Wellcome Trust (P.W., K.O.) and National Institutes of Health (D.C.L., M.S.). COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 12 November 2004; accepted 21 December 2004 Published online at http://www.nature.com/natureneuroscience/
1693–1699 (2001). 25. Destexhe, A., Rudolph, M., Fellous, J.M. & Sejnowski, T.J. Fluctuating synaptic conductances recreate in vivo–like activity in neocortical neurons. Neuroscience 107, 13–24 (2001). 26. McLaughlin, D., Shapley, R., Shelley, M. & Wielaard, D.J. A neuronal network model of macaque primary visual cortex (V1): orientation selectivity and dynamics in the input layer 4C . Proc. Natl. Acad. Sci. USA 97, 87–92 (2000). 27. Wielaard, D.J., Shelley, M., McLaughlin, D. & Shapley, R. How simple cells are made in a nonlinear network model of the visual cortex. J. Neurosci. 21, 5203–5211 (2001). 28. Ben-Yishai, R., Bar-Or, R.L. & Sompolinsky, H. Theory of orientation tuning in visual cortex. Proc. Natl. Acad. Sci. USA 92, 3844–3848 (1995). 29. Mooser, F., Bosking, W.H. & Fitzpatrick, D. A morphological basis for orientation tuning in primary visual cortex. Nat. Neurosci. 7, 872–879 (2004). 30. Reid, R.C. & Alonso, J.M. Specificity of monosynaptic connections from thalamus to visual cortex. Nature 378, 281–284 (1995). 31. Lampl, I., Anderson, J.S., Gillespie, D.C. & Ferster, D. Prediction of orientation selectivity from receptive field architecture in simple cells of cat visual cortex. Neuron 30, 263–274 (2001). 32. Carandini, M. & Ferster, D. Membrane potential and firing rate in cat primary visual cortex. J. Neurosci. 20, 470–484 (2000). 33. Volgushev, M., Pernberg, J. & Eysel, U.T. Comparison of the selectivity of postsynaptic potentials and spike responses in cat visual cortex. Eur. J. Neurosci. 12, 257–263 (2000). 34. Azouz, R. & Gray, C.M. Adaptive coincidence detection and dynamic gain control in visual cortical neurons in vivo. Neuron 37, 513–523 (2003). 35. Shapley, R., Hawken, M. & Ringach, D.L. Dynamics of orientation selectivity in the primary visual cortex and the importance of cortical inhibition. Neuron 38, 689–699 (2003). 36. Crook, J.M., Kisvarday, Z.F. & Eysel, U.T. GABA-induced inactivation of functionally characterized sites in cat striate cortex: effects on orientation tuning and direction selectivity. Vis. Neurosci. 14, 141–158 (1997). 37. Mazer, J.A., Vinje, W.E., McDermott, J., Schiller, P.H. & Gallant, J.L. Spatial frequency and orientation tuning dynamics in area V1. Proc. Natl. Acad. Sci. USA 99, 1645–1650 (2002). 38. Dragoi, V., Sharma, J., Miller, E.K. & Sur, M. Dynamics of neuronal sensitivity in visual cortex and local feature discrimination. Nat. Neurosci. 5, 883–891 (2002). 39. Ringach, D.L., Bredfeldt, C.E., Shapley, R.M. & Hawken, M.J. Suppression of neural responses to nonoptimal stimuli correlates with tuning selectivity in macaque V1. J. Neurophysiol. 87, 1018–1027 (2002). 40. Nelson, S., Toth, L., Sheth, B. & Sur, M. Orientation selectivity of cortical neurons during intracellular blockade of inhibition. Science 265, 774–777 (1994). 41. Bosking, W.H., Zhang, Y., Schofield, B. & Fitzpatrick, D. Orientation selectivity and the arrangement of horizontal connections in tree shrew striate cortex. J. Neurosci. 17, 2112–2127 (1997). 42. Angelucci, A. et al. Circuits for local and global signal integration in primary visual cortex. J. Neurosci. 22, 8633–8646 (2002). 43. Volgushev, M., Pernberg, J. & Eysel, U.T. γ-Frequency fluctuations of the membrane potential and response selectivity in visual cortical neurons. Eur. J. Neurosci. 17, 1768–1776 (2003). 44. Williams, S.R. Spatial compartmentalization and functional impact of conductance in pyramidal neurons. Nat. Neurosci. 7, 961–967 (2004). 45. Troyer, T.W., Krukowski, A.E., Priebe, N.J. & Miller, K.D. Contrast-invariant orientation tuning in cat visual cortex: thalamocortical input tuning and correlation-based intracortical connectivity. J. Neurosci. 18, 5908–5927 (1998). 46. Lauritzen, T.Z. & Miller, K.D. Different roles for simple-cell and complex-cell inhibition in V1. J. Neurosci. 23, 10201–10213 (2003). 47. Hirsch, J.A. et al. Functionally distinct inhibitory neurons at the first stage of visual cortical processing. Nat. Neurosci. 6, 1300–1308 (2003). 48. Liu, G. Local structural balance and functional interaction of excitatory and inhibitory synapses in hippocampal dendrites. Nat. Neurosci. 7, 373–379 (2004). 49. Desai, N.S., Cudmore, R.H., Nelson, S.B. & Turrigiano, G.G. Critical periods for experience-dependent synaptic scaling in visual cortex. Nat. Neurosci. 5, 783–789 (2002). 50. Fagiolini, M. & Hensch, T.K. Excitatory-inhibitory balance controls critical period plasticity. in Excitatory-Inhibitory Balance: Synapses, Circuits, Systems (eds. Hensch, T.K. & Fagiolini, M.) 269–282 (Kluver Academic/Plenum, New York, 2003)

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

1. Mountcastle, V.B. Perceptual Neuroscience (Harvard University Press, Cambridge, MA, 1998). 2. Ferster, D. & Miller, K.D. Neural mechanisms of orientation selectivity in the visual cortex. Annu. Rev. Neurosci. 23, 441–471 (2000). 3. Martin, K.A. Microcircuits in visual cortex. Curr. Opin. Neurobiol. 12, 418–425 (2002). 4. Monier, C., Chavane, F., Baudot, P., Graham, L.J. & Fregnac, Y. Orientation and direction selectivity of synaptic inputs in visual cortical neurons: a diversity of combinations produces spike tuning. Neuron 37, 663–680 (2003). 5. Yao, H. & Dan, Y. Stimulus timing-dependent plasticity in cortical processing of orientation. Neuron 32, 315–323 (2001). 6. Dragoi, V. & Sur, M. Plasticity of orientation processing in adult visual cortex. in The Visual Neurosciences (eds. Chalupa, L.M. & Werner, J.S.) 1654–1664 (MIT Press, Cambridge, Massachusetts, 2003). 7. Somers, D.C., Nelson, S.B. & Sur, M. An emergent model of orientation selectivity in cat visual cortical simple cells. J. Neurosci. 15, 5448–5465 (1995). 8. Suarez, H., Koch, C. & Douglas, R. Modeling direction selectivity of simple cells in striate visual cortex within the framework of the canonical microcircuit. J. Neurosci. 15, 6700–6719 (1995). 9. Bonhoeffer, T. & Grinvald, A. Iso-orientation domains in cat visual cortex are arranged in pinwheel-like patterns. Nature 353, 429–431 (1991). 10. Das, A. & Gilbert, C.D. Topography of contextual modulations mediated by short-range interactions in primary visual cortex. Nature 399, 655–661 (1999). 11. Schummers, J., Mariño, J. & Sur, M. Synaptic integration by V1 neurons depends on location within the orientation map. Neuron 36, 969–978 (2002). 12. Sharon, D. & Grinvald, A. Dynamics and constancy in cortical spatiotemporal patterns of orientation processing. Science 295, 512–515 (2002). 13. Maldonado, P.E., Godecke, I., Gray, C.M. & Bonhoeffer, T. Orientation selectivity in pinwheel centers in cat striate cortex. Science 276, 1551–1555 (1997). 14. Dragoi, V., Rivadulla, C. & Sur, M. Foci of orientation plasticity in visual cortex. Nature 411, 80–86 (2001). 15. Turrigiano, G.G. & Nelson, S.B. Homeostatic plasticity in the developing nervous system. Nat. Rev. Neurosci. 5, 97–107 (2004). 16. Wehr, M. & Zador, A.M. Balanced inhibition underlies tuning and sharpens spike timing in auditory cortex. Nature 426, 442–446 (2003). 17. Zhang, L.I., Tan, A.Y., Schreiner, C.E. & Merzenich, M.M. Topography and synaptic shaping of direction selectivity in primary auditory cortex. Nature 424, 201–205 (2003). 18. Borg-Graham, L.J., Monier, C. & Fregnac, Y. Visual input evokes transient and strong shunting inhibition in visual cortical neurons. Nature 393, 369–373 (1998). 19. Anderson, J.S., Carandini, M. & Ferster, D. Orientation tuning of input conductance, excitation, and inhibition in cat primary visual cortex. J. Neurophysiol. 84, 909–926 (2000). 20. Hirsch, J.A. et al. Synaptic physiology of the flow of information in the cat’s visual cortex in vivo. J. Physiol. 540, 335–350 (2002). 21. Martinez, L.M., Alonso, J.M., Reid, R.C. & Hirsch, J.A. Laminar processing of stimulus orientation in cat visual cortex. J. Physiol. 540, 321–333 (2002). 22. Kisvarday, Z.F., Toth, E., Rausch, M. & Eysel, U.T. Orientation-specific relationship between populations of excitatory and inhibitory lateral connections in the visual cortex of the cat. Cereb. Cortex 7, 605–618 (1997). 23. Roerig, B. & Chen, B. Relationships of local inhibitory and excitatory circuits to orientation preference maps in ferret visual cortex. Cereb. Cortex 12, 187–198 (2002). 24. Yousef, T., Toth, E., Rausch, M., Eysel, U.T. & Kisvarday, Z.F. Topography of orientation centre connections in the primary visual cortex of the cat. Neuroreport 12,

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

201

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Bistability of cerebellar Purkinje cells modulated by sensory stimulation
Yonatan Loewenstein1–3,6, Séverine Mahon4,6, Paul Chadderton4, Kazuo Kitamura4, Haim Sompolinsky2,5, Yosef Yarom1,2 & Michael Häusser4
A persistent change in neuronal activity after brief stimuli is a common feature of many neuronal microcircuits. This persistent activity can be sustained by ongoing reverberant network activity or by the intrinsic biophysical properties of individual cells. Here we demonstrate that rat and guinea pig cerebellar Purkinje cells in vivo show bistability of membrane potential and spike output on the time scale of seconds. The transition between membrane potential states can be bidirectionally triggered by the same brief current pulses. We also show that sensory activation of the climbing fiber input can switch Purkinje cells between the two states. The intrinsic nature of Purkinje cell bistability and its control by sensory input can be explained by a simple biophysical model. Purkinje cell bistability may have a key role in the short-term processing and storage of sensory information in the cerebellar cortex. Persistent firing of neurons after a transient stimulus is a common feature of sensory processing in neural circuits that has been shown to be associated with short-term memory tasks1. Hebb proposed that such persistent firing states are generated by cell assemblies with mutually reinforcing excitatory feedback connections. Indeed, recent theoretical and experimental work2–4 has shown that local recurrent networks are capable of generating such sustained firing patterns. An alternative possibility is that persistent firing is sustained by the intrinsic properties of the neuron5. Several neuronal types show bistable behavior: brief excitation produces a prolonged increase in firing rate through activation of persistent inward currents that maintain depolarization5–8. Some neurons show multiple persistent firing states that reflect the history of their synaptic inputs9. Theoretically, single cell bistable or multistable states can function as short-term memory buffers or enhance the short-term memory capabilities of excitatory recurrent networks4,5,10–12. In contrast to the cerebral cortex, the cerebellum lacks a prominent recurrent excitatory synaptic network. Purkinje cells provide the sole output of the cerebellar cortex. They receive excitatory input from two distinct sources, the parallel fibers and the climbing fibers, and exert inhibitory control on their targets in the deep cerebellar nuclei. Each Purkinje cell is innervated by over 105 parallel fibers, which synapse on its elaborate leaf-like dendritic tree. Each Purkinje cell also receives input from a single climbing fiber axon, originating from a neuron in the inferior olive. The climbing fiber wraps around the proximal dendrites of the Purkinje cell, making hundreds of synaptic contacts. Purkinje cells can fire spontaneous sodium action potentials (known as ‘simple spikes’) in the absence of synaptic excitation13. In response to climbing fiber activation, they generate a stereotypic discharge pattern called a complex spike, which can be distinguished from ongoing simple spikes both intracellularly and extracellularly14. Previous intracellular in vitro studies have shown that Purkinje cells can show both spontaneous and current-evoked bistable behavior that is correlated with intermittent periods of simple spike discharge6,15. However, bistability in Purkinje cells in vivo and its potential implications for cerebellar information processing have not yet been investigated. Here we show that Purkinje cells in vivo show prominent ongoing bistability of membrane potential and spike output. Notably, the state of the neuron can be controlled by sensoryevoked climbing fiber input that can induce state transitions in both directions, triggering prolonged increases or decreases in the simple spike firing rate. RESULTS Purkinje cell membrane potential is bistable in vivo To explore the dynamic interplay between intrinsic properties and physiological synaptic inputs in Purkinje cells, we obtained in vivo whole-cell recordings from rat and guinea pig Purkinje cells located in the cerebellar vermis and hemispheres. Purkinje cells, identified from the spontaneous occurrence of complex spikes at their characteristic frequency (∼1 Hz)14, showed transitions in membrane potential between a hyperpolarized state (‘down state’) and a depolarized state (‘up state’). The hyperpolarized state was quiescent (that is, devoid of any simple spike activity), whereas the depolarized state was usually associated with simple spike discharge (Fig. 1a,b, left). Transitions between the down state and the up state were observed at either perisomatic (Fig. 1a) or dendritic (Fig. 1b) locations, and the existence of these two distinct states was apparent in the bimodal distribution of the membrane

1Department of Neurobiology, 2The Interdisciplinary Center for Neural Computation, and 5Racah Institute of Physics, The Hebrew University of Jerusalem, Jerusalem 91904, Israel. 3Department of Brain and Cognitive Sciences, The Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA. 4Wolfson Institute for Biomedical Research and Department of Physiology, University College London, Gower Street, London WC1E 6BT, UK. 6These authors contributed equally to this work. Correspondence should be addressed to Y.L. ([email protected]) or S.M. ([email protected]).

Published online 23 January 2005; doi:10.1038/nn1393

202

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
Figure 1 Spontaneous membrane potential bistability in Purkinje cells in vivo. (a) Left: whole-cell perisomatic recording of a rat Purkinje cell showing membrane potential fluctuations between two distinct states: a hyperpolarized quiescent state (‘down state’) and a depolarized spiking state (‘up state’). Right: the corresponding histogram of the membrane potential calculated from 84 s of recording (bin size 1 mV). (b) Presumed distal dendritic whole-cell recording of a rat Purkinje cell showing similar up and down state transitions (left) as reflected in the bimodal distribution of the membrane potential (right, 100 s of recording, bin size 1 mV). (c) Comparison of the average values of the two membrane potential states (down state, D; up state, U) obtained under different anesthetics (ketamine-xylazine (Ket.-xyl.), n = 24 Purkinje cells; barbiturate, n = 5 Purkinje cells). (d) Relationship between the duration of down and up states. Each point corresponds to a single Purkinje cell. Error bars represent s.e.m.

Membrane potential (mV)

a
Fraction of time (%)
30 20 10 0 –70 –60 –50 –40 –30

c

–40 –50 –60 –70 –80

U D D

U

–42 mV –57 mV 1s 20 mV

Membrane potential (mV)

Ket.-xyl. Barbiturate

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Down state duration (s)

b
Fraction of time (%)
30 20 10 0 –70 –60 –50 –40 –30

d

5 4 3 2 1 0 0 1 2 3 4 5

–45 mV –60 mV 2s 20 mV

Membrane potential (mV)

Up state duration (s)

potential (Fig. 1a,b, right). A dip test16, which gives the likelihood of drawing a sample from a unimodal distribution, was used to assess the statistical significance of this bimodality. In all recorded Purkinje cells (n = 31), the distribution of the membrane potential deviated significantly from a unimodal distribution (P < 0.0002, n = 30 cells; P < 0.05, n = 1 cell). Purkinje cells showed the characteristic electrophysiological properties reported for this cell type. As previously described15, Purkinje cells responded to long hyperpolarizing current pulses with a slow depolarizing sag towards the prestimulus membrane potential, indicative of the presence of an Ih current (see Supplementary Fig. 1). The ‘sag ratio’, defined as the steady-state versus peak deflection during a hyperpolarizing current pulse in the down state, was 0.50 ± 0.05 (mean ± s.e.m.; n = 11 rat Purkinje cells). Across the population, in ketamine-xylazine–anesthetized rats, the average values of membrane potential in down and up states were –61.6 ± 0.7 mV and –44.5 ± 0.8 mV, respectively (n = 24 cells), generating a voltage difference between the two states of 17.1 ± 1.1 mV (n = 24 cells). Spontaneous fluctuations in membrane potential under barbiturate anesthesia were similar to those observed under ketamine-xylazine anesthesia, with the average membrane potential in the down state (–61.7 ± 1.4 mV, n = 5 cells) and the up state (–44.9 ± 1.6 mV, n = 5 cells) not significantly different (P > 0.8; Fig. 1c). This suggests that bimodality in Purkinje cells does not depend on the type of anesthesia used. Similar bimodality was also observed in whole-cell recordings from Purkinje cells in ketamine-xylazine–anesthetized guinea pigs (n = 2). The durations of up and down states were quantified in rat Purkinje cells using long periods of continuous recording. The time spent by a

Purkinje cell in a given state varied considerably between and within cells (Fig. 1d). The mean durations of down and up states (averaged over the mean values of each cell) were 1.50 ± 0.25 s (range: 0.41–3.61 s) and 1.45 ± 0.28 s (range: 0.25–4.84 s), respectively (n = 17 cells; Fig. 1d), and the corresponding coefficient of variance (CV) values were 0.74 ± 0.06 for down states and 0.58 ± 0.05 for up states. We also examined the time between successive transitions to the down state. The mean duration between two successive up-to-down transitions was 2.98 ± 0.48 s with a corresponding CV of 0.54 ± 0.03 (n = 17 cells). These results suggest that the recurrence of up and down states in Purkinje cells is rather irregular and is thus unlikely to result from a simple underlying oscillatory process (see Supplementary Note). To determine whether membrane potential bimodality is an intrinsic property of Purkinje cells or whether it reflects bimodal synaptic input, we compared the intracellular activity of Purkinje cells with that of their input neurons: granule cells and molecular layer interneurons. Granule cells were identified by their depth within cerebellar cortex, their high input resistance (795 ± 88.4 MΩ, n = 6 cells) and their highfrequency non-accommodating spike discharge in response to strong depolarizing current pulses (Fig. 2a). Recordings from granule cells

a

Granule cell

d

Interneuron

20 mV 50 pA 100 ms

20 mV 100 pA 100 ms

Figure 2 Membrane potential bistability is a specific feature of Purkinje cells within the cerebellar cortex. (a–c) Electrophysiological properties of rat cerebellar granule cells in vivo. (a) Voltage responses of a granule cell to hyperpolarizing (–10 pA) and depolarizing (+30 pA) current injection (Rin = 872 MΩ). (b,c) Spontaneous activity of a granule cell recorded at rest (b, –59 mV) and the corresponding membrane potential distribution (c) calculated from 20 s of recording (bin size 1 mV). (d–f) Electrophysiological properties of rat molecular layer interneurons in vivo. (d) Voltage responses of a molecular layer interneuron to hyperpolarizing (–70 pA) and depolarizing (+65 pA) current injection (Rin = 103 MΩ). (e,f) Spontaneous activity of a molecular layer interneuron recorded at rest (e, –63 mV) and the corresponding membrane potential distribution (f) calculated from 20 s of recording (bin size 1 mV). Panels a and b are from different cells, as are d and e.

b

e

20 mV

20 mV

c
Fraction of time (%)
30 20 10 0 –70 –60 –50

500 ms

f
Fraction of time (%)
20 10 0 –70

500 ms

–40

–60

–50

–40

Membrane potential (mV)

Membrane potential (mV)

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

203

ARTICLES a
Figure 3 Membrane potential bistability is reflected in spike output pattern in vivo. (a) Left: extracellular recording from a guinea pig Purkinje cell in vivo. Firing pattern consisted of high-frequency bursts of simple spikes alternating with quiescent periods. Right, bimodal distribution of the instantaneous frequency of simple spikes calculated from 10 min of continuous recording (bin size 5 Hz). (b) Cell-attached patch-clamp recording showing the bimodal firing pattern of a rat Purkinje cell in vivo. (c) Intracellular activity of the same Purkinje cell obtained immediately after the formation of the whole-cell configuration, showing that the bimodal action potential discharge observed in the cell-attached recording reflects membrane potential bimodality.

Extracellular
0.85

Probability

0.01

0

0

100

200

Rate (Hz)

b
© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience
Cell-attached

1s

20 pA

c
Whole-cell

10 mV 400 ms

Hyperpolarized
10 mV

Fraction of time (%)

were characterized by frequent spontaneous excitatory postsynaptic potentials that could on occasion trigger spike discharge (Fig. 2b). The membrane potential distribution of granule cells was unimodal (P > 0.7; n = 6 cells) with a mean membrane potential of –64.3 ± 1.9 mV (Fig. 2c). These properties are consistent with those previously described in vivo for this cell type17. In agreement with previous studies18,19, molecular layer interneurons in vivo showed a lower input resistance (224.0 ± 34.3 MΩ, n = 10 cells; Fig. 2d) and spontaneous spiking at rest (Fig. 2e), which was reflected in the unimodal (P > 0.5, n = 8 cells; P > 0.1, n = 2 cells) distribution of their membrane potential around a mean value of –56.1 ± 1.4 mV (Fig. 2f). Granule cells and interneurons did not show bimodality even when recordings from bimodal Purkinje cells were obtained in the same preparation (n = 5). The existence of two distinct states of membrane potential in Purkinje cells should be reflected in extracellular measurements of spontaneous spiking. Indeed, transitions between quiescent and spiking states were observed in our extracellular recordings in vivo (Fig. 3a). The firing pattern of guinea pig Purkinje cells recorded extracellularly consisted of high-frequency trains of simple spikes alternating with quiescent periods (Fig. 3a, left). This firing pattern resulted in a bimodal

distribution of the instantaneous frequency of simple spikes (Fig. 3a, right). The prolonged periods devoid of simple spikes accounted for the peak at 0 Hz, whereas the peak around 100 Hz corresponded to the firing rate during the bursts of high-frequency simple spike firing. Across the population, the average durations of active and quiescent periods were 0.9 ± 0.2 s and 3.4 ± 0.8 s, respectively (n = 28 cells). Similar spiking patterns were also observed in cell-attached recordings from rat Purkinje cells (n = 9 cells, Fig. 3b). To assess the effect of the whole-cell configuration on the membrane bimodality, we performed cell-attached and whole-cell recording in the same Purkinje cells (n = 6 cells; Fig. 3b,c). The intracellularly recorded membrane potential, obtained immediately after the establishment of the whole-cell configuration, showed that the bimodal simple spike discharge observed in the cell-attached recording reflected underlying membrane potential bimodality (Fig. 3b,c). The mean instantaneous frequency of simple spikes in the cell-attached configuration (102.1 ± 3.8 Hz) was comparable to that measured in the whole-cell configuration during the up state (110.7 ± 3.3 Hz; P > 0.05). Across the population, no significant difference in the mean firing rate between the two configurations was observed (P > 0.6,

Rest

Fraction of time (%)

a

12 8 4 0 –90 –70 –50 –30

–65mV

Membrane potential (mV)

b
–80 mV

30 20 10 0 –90 –70 –50 –30

500 ms

Membrane potential (mV)

Figure 4 Intrinsic origin of membrane potential bistability in vivo. (a) Whole-cell perisomatic recording of bistable behavior of a rat Purkinje cell under barbiturate anesthesia (left). The corresponding histogram of membrane potential distribution calculated from 33 s of recording (right, bin size 1 mV) shows two distinct peaks reflecting the two states of membrane potential. (b) Bimodality in the same Purkinje cell was abolished by injection of negative DC current (–800 pA, left). Note the unimodal distribution of membrane potential (right, 30 s of recording, bin size 1 mV). The amplitude of spontaneous complex spikes has been truncated. (c) Whole-cell recordings (upper traces) of a rat Purkinje cell in vivo showing that the neuron can be switched from the down state to the up state and from the up state to the down state following the injection of brief depolarizing (left) and hyperpolarizing (right) current pulses, respectively (lower traces; 40 ms, 0.16 nA/–0.4 nA). (d) Whole-cell recording (upper trace) from a guinea pig Purkinje cell in vivo showing that the same hyperpolarizing current pulse (lower trace; 100 ms, 2 nA) can induce upto-down and down-to-up transitions depending on the initial state of the membrane potential.

c
20 mV

200 pA 500 ms

d
20 mV

1 nA 5s

204

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES a
In vitro
Up-down

10 mV 2s –57 mV CF stim. 20 mV 50 ms Down-up

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 5 Climbing fiber input can trigger transitions between states. (a) Left: whole-cell recording from a guinea pig Purkinje cell in vitro. Climbing fiber stimulation (CF stim., arrowheads) triggered transitions between up states and down states in both directions. Right, expanded traces (corresponding to the open arrowheads in the left panel) showing the complex spike-evoked transitions. (b) Temporal relationship between complex spikes and simple spike bursts in extracellular recordings from guinea pig Purkinje cells in vivo. Left, representative examples of simple spike bursts followed by (top) or preceded by (bottom) a complex spike (asterisk). Right, cross-correlation of complex spikes (CS) with the end (top) and with the beginning (bottom) of bursts of simple spikes (each bin is 5 ms).

b

In vivo / extracellular CS - burst end
No. of events CS

10

*
CS - burst onset
No. of events

0 –0.1

0

0.1

Time (s) CS
50

*

0 –0.1

0

0.1

100 ms

Time (s)

n = 6 cells), indicating that the whole-cell configuration did not affect the membrane potential dynamics. To understand the origin of Purkinje cell membrane bimodality, we examined its voltage dependence. Spontaneous up and down state transitions (Fig. 4a) could be abolished by injection of negative DC current (–380 ± 90 pA, n = 9 cells), which hyperpolarized Purkinje cells to –80.6 ± 2.5 mV (Fig. 4b). When Purkinje cells were hyperpolarized, only complex spikes were recorded (Fig. 4b, left) and the membrane potential histogram (Fig. 4b, right) showed a bell-shaped distribution around the mean holding potential. Similarly, constant depolarizing current resulted in a single up state (data not shown). Whereas steady current injection could prevent membrane potential bimodality, brief current injections were able to induce transitions between states. Short depolarizing current pulses (0.16–1 nA, 20–100 ms duration) delivered during the down state switched the neuron to the up state with a probability of 70.5 ± 8.3% (Fig. 4c, left; n = 6 cells), whereas brief hyperpolarizing current pulses (0.2–2 nA, 20–100 ms duration) delivered during the up state could induce a transition to the down state with an efficiency of 81.5 ± 10.5% (Fig. 4c, right, and Fig. 4d; n = 6 cells). These reliable and almost deterministic flip-flop-like transitions are a characteristic feature of bistable systems, where transient perturbations are sufficient to induce sustained changes. Notably, hyperpolarizing current pulses delivered during the down state could also induce a transition to the up state (Fig. 4d). In four of five cells, hyperpolarizing current pulses (0.2–2 nA, 20–100 ms duration; n = 5 cells) reliably induced a down-toup transition with a probability of 61.1 ± 14.0% (in the remaining neuron, the hyperpolarizing current pulse did not induce up states). These findings demonstrate that the bimodality of Purkinje cell membrane potential is a consequence of its bistable dynamics. Synaptic control of intrinsic bistability That the same current pulse can induce membrane potential transitions in both directions raised the possibility that in Purkinje cells,

the same synaptic input may induce transitions between up and down states. To test this hypothesis, we first examined whether climbing fiber activation, achieved by direct electrical stimulation, could induce transitions between quiescent and spiking states in a guinea pig slice preparation. During periods of quiescence, climbing fiber activation resulted in a transition to a depolarized potential associated with simple spike discharge. Similarly, the same climbing fiber activation during an active state was capable of inducing a transition towards the lower stable state (Fig. 5a), thereby terminating the spontaneous discharge. A careful examination of the records obtained in vivo confirmed this result. In many cases, the bursts of simple spikes recorded extracellularly were immediately followed (Fig. 5b, top; see Methods) or immediately preceded (Fig. 5b, bottom) by a complex spike (asterisk). We performed cross-correlations between complex spikes and the beginning or the end of simple spike bursts; we found a significant correlation (P < 0.05) between complex spikes and the end or the beginning of simple spike bursts in 75% (21 of 28) and 64% (18 of 28) of the cells, respectively (Fig. 5b, right). The correlation between complex spikes and state transitions was also observed in the in vivo whole-cell recordings. Across the population (n = 18 cells), 73 ± 4% of the transitions were preceded by a complex spike, occurring less than 100 ms before the transition (Fig. 6a, top). A significantly lower number of transitions apparently occurred spontaneously in absence of any climbing fiber input (Fig. 6a, bottom; P < 0.05, n = 18 cells). Transitions between states were characterized by a slow voltage change starting immediately after the complex spike (Fig. 6a, top), similar to the transitions triggered by direct climbing fiber stimulation obtained in the in vitro experiments (Fig. 5a, right). The temporal correlation between complex spikes and state transitions for one sample cell is shown (Fig. 6b). Up-to-down and down-to-up transitions typically occurred ∼40 and ∼70 ms, respectively, after a complex spike. These strong correlations allowed us to assume that state transitions occurring less than 100 ms after a complex spike were triggered by the climbing fiber activation. Across the population (n = 18 cells), 62 ± 7% of the transitions to the down state and 84 ± 5% of the transitions to the up state were triggered by a climbing fiber input (Fig. 6c). We calculated the efficiency with which a climbing fiber input induced a transition as the ratio of the number of complex spikes that induced transitions to the total number of complex spikes. The mean frequency of complex spike firing was 0.8 ± 0.1 Hz (range: 0.2–1.7 Hz, n = 18 cells) and the efficiency of a complex spike in inducing a transition was 76.8 ± 4.9% (n = 18 cells). We also assessed the climbing fiber input efficiency in a given state. Complex spikes occurring during the up state were followed by a transition to the down state in 66.0 ± 6.4% of the cases, whereas climbing fiber input in the down state resulted in transition to the up state in 88.2 ± 3.4% of the cases (n = 18 cells; Fig. 6d). These results indicate that climbing fiber input is capable of modulating, on-line, simple spike firing pattern in vivo.

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

205

ARTICLES a
Up-down Down-up

CS

–45 mV –58 mV

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Spont.
–33 mV

20 mV 50 ms

–53 mV

b

Percentage of transitions

Up-down

c
CS
100 80 60 40 20 0

Figure 6 Characterization of the complex spike–induced transitions in vivo. (a) Whole-cell recording from a rat Purkinje cell in vivo, showing representative transitions between up states and down states. Most transitions (92% in this cell) are associated with complex spikes (CS, top) and only a minority occur apparently in the absence of a complex spike (Spont., bottom). (b) Temporal relationship between complex spikes and state transitions in a 45-min recording from a guinea pig Purkinje cell in vivo. Cross-correlation of complex spikes (CS) with the transition from up-todown (top) and from down-to-up (bottom; each bin is 5 ms). (c) Population data from 18 rat Purkinje cell whole-cell recordings showing the percentage of up-down and down-up transitions associated with complex spikes (CS), or occurring spontaneously (Spont.). (d) Pooled data (n = 18 cells) showing the efficiency of a complex spike occurring during the up state in inducing transitions to the down state (Up-Down) and the efficiency of a complex spike occurring during the down state in inducing transitions to the up state (Down-Up).
Down-up

Up-down

10 5 0 –0.1

0

0.1

Time (s)

CS Spont. CS Spont.

Down-up

d
CS efficiency (%)

100 80 60 40 20 0

CS
No. of events
5

0 –0.1

0

0.1

Time (s)

Up-down

Down-up

and simple spike firing (n = 6; Fig. 7d). Evoked up-to-down and downto-up transitions were associated with prolonged decreases (Fig. 7d, top) and increases (Fig. 7d, middle) of simple spike discharge, respectively. After the sensory-evoked up-to-down transitions, the mean simple spike firing rate was significantly reduced (P < 0.0001), from 57 ± 3 Hz during the baseline to 22 ± 7 Hz during the first 500 ms after the stimulus (Fig. 7d, top). Conversely, associated with the sensory-evoked down-to-up transitions, the mean simple spike firing rate was significantly increased (P < 0.0001), from 5 ± 1 Hz to 53 ± 7 Hz (Fig. 7d, middle). When evoked complex spikes did not trigger state transitions, we did not observe any significant change in simple spike firing compared to baseline level (P > 0.5; Fig. 7d, bottom). In the same cells, when sensory stimuli did not evoke a complex spike in the recorded Purkinje cell (Fig. 7e), only a slight increase (+ 7%, P < 0.05) in the mean simple spike firing rate was observed (Fig. 7f), indicating that the sensory-evoked changes in simple spike discharge resulted mainly from the state transitions triggered by the activation of climbing fiber input. A model for bistability and state transitions To explore the biophysical determinants of Purkinje cell membrane potential dynamics, we constructed a simplified model that emulates the fundamental features of the observed behavior: bistability, bidirectional transitions induced by outward current pulses (Fig. 8a, left) and bidirectional transitions induced by climbing fiber–like input (Fig. 8a, right). This single-compartment model consists of three ionic currents: an instantaneous, non-inactivating inward current (modeled here as a sodium current), a slow h-like current and a voltage-independent outward current (see Supplementary Note). The dynamic variables of this model are the membrane potential (V) and the inactivation (h) of the h-like current. Their qualitative behavior is depicted in the V-h phase plane shown in Figure 8b. Each point in the phase plane corresponds to a possible state of the dynamic variables of the cell ˙ = 0 and model (V, h). The red and blue solid lines correspond to the h ˙ the V = 0 nullclines, respectively, and the two outermost points of intersection (circles) correspond to the two stable states (stable fixed points) of the system. This bistability is primarily due to the nonlinearity of the non-inactivating inward current, which is responsible ˙ = 0 nullcline. In the hyperpolarized state (left for the N-shape of the V open circle), the non-inactivating inward channels are closed and the membrane potential is determined by the combined effect of the h-like current and the voltage-independent outward current. In contrast, these channels are open in the depolarized state and will contribute substantially to the membrane potential value in the up state. The basins of attraction of the depolarized and hyperpolarized states are denoted by the blue and green areas, respectively. The separatrix, the

Purkinje cell bistability modulated by sensory stimulation The strong correlation between complex spikes and state transitions suggests that sensory stimuli that elicit complex spikes could modulate the state of the cell. To examine the influence of sensory stimulation on state transitions, we used air puff stimulation of the vibrissae or other perioral areas to activate climbing fiber inputs. In this set of experiments, recordings were performed in the Crus I and IIa of the cerebellar cortex, regions that have been shown to respond strongly to tactile stimulation of the perioral areas in anesthetized rats20. As illustrated by the raster plot in Figure 7a, air puff stimuli were effective in evoking complex spikes (filled circles). Across cells, complex spikes were evoked in 38 ± 4% of the trials (n = 7 cells), and comparable to previous findings20, the latency of evoked complex spikes from the onset of the stimulus was 70 ± 7 ms (range: 47.9–93.5 ms, n = 7 cells). The majority of the evoked complex spikes triggered up-to-down or down-to-up transitions, depending on the initial state. In the cell shown in Figure 7b, 93% of the evoked complex spikes were found to be effective in triggering state transitions, with 67% inducing a transition to the down state (Fig. 7b, top) and 26% a transition to the up state (Fig. 7b, middle). In only 7% of the cases did evoked complex spikes not induce any transition (Fig. 7b, bottom). Across the population, 66.4 ± 11% of the evoked complex spikes were effective in triggering transitions (n = 7 cells). The efficiency of a complex spike evoked during the up state in inducing up-to-down transitions was 57.8 ± 15.5%, compared to 81.6 ± 7.3% for the opposite transition (n = 7 cells; Fig. 7c). In recordings proximal enough to the soma of the Purkinje cell to allow a reliable detection of simple spikes6, we assessed the temporal relationship between peristimulus histograms of evoked complex spikes

No. of events

206

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
Figure 7 Sensory-evoked complex spikes in Purkinje cells can trigger membrane potential bistability in vivo. An air puff (40 ms) to the ipsilateral vibrissae was used to evoke responses in single Purkinje cells. (a) Raster plot showing the temporal relationship between sensory stimuli, complex spikes (filled circles) and simple spike firing (lines) in a rat Purkinje cell whole-cell recording in vivo (30 consecutive trials of 121). (b) Whole-cell recordings showing representative examples of sensory-evoked complex spikes (filled circles) triggering up-to-down, down-to-up, and no transitions, respectively. Note that sensoryevoked states could be interrupted by spontaneous transitions associated with spontaneous complex spikes (arrows) or occurring in the absence of complex spikes (crossed arrow). Open circles indicate spontaneously occurring complex spikes. (c) Pooled data (n = 7 cells) illustrating the efficiency of an evoked complex spike occurring during the up state in inducing transitions to the down state (Up-down) and the efficiency of an evoked complex spike occurring during the down state in inducing transitions to the up state (Down-up). (d) Temporal relationship between peristimulus histograms of evoked complex spikes (CS, bar chart) and simple spike firing (SS, dotted lines) on the same graph, separated into up-todown transitions (top, n = 5 cells), down-to-up transitions (middle, n = 6 cells) and no transitions (bottom, n = 6 cells) (bin size 50 ms). Note that up-to-down and down-to-up transitions associated with evoked complex spikes are accompanied by robust and prolonged changes in simple spike firing. Conversely, the mean simple spike firing rate was not significantly changed compared to the baseline period (100 ms baseline, n = 2 cells; 500 ms baseline for the remaining cells) when evoked complex spikes did not evoke state transitions. (e) Example of a Purkinje cell up state in the absence of any evoked complex spikes. Open circles indicate spontaneously occurring complex spikes. (f) Averaged peristimulus histogram (n = 6 cells) of simple spike firing (dotted line) computed for sweeps where sensory stimuli did not evoke complex spikes. Trials where stimuli were delivered either during an up or a down state of the Purkinje cell were pooled together. Bar charts represent spontaneously occurring complex spikes. Results presented in a, b and e are from the same cell. Calibration bars in b apply also to e.

a
30

Stim.

c
Evoked CS efficiency (%)
100 80 60 40 20 0

Trial no.

20

10 1 –0.1 0 0.1 0.2 0.3 0.4 0.5 0.6

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Up-down

Down-up

Time (s)

b

Up to down transitions

d
CS probability
0.6 0.4 0.2 0 –0.4 40 0 0 0.4 0.8 80

Stim.

Time (s)
0.6 0.4 0.2 0 –0.4 40 0 0 0.4 0.8

80

Stim.

Time (s) Stim. CS probability
0.6 0.4 0.2 0 –0.4 40 0 0 0.4 0.8

100 ms 20 mV

80

Stim.

Time (s)

e

f
CS probability
0.6 0.4 0.2 0 –0.4

Stim.
80 40 0 0 0.4 0.8

Stim.

Time (s)

border between these areas, passes through the middle, unstable, fixed point (open square). Any voltage perturbation at either of the two fixed points that is large enough to cross the separatrix will eventually lead to the convergence of the voltage to the other fixed point. For example, a brief depolarization from the down state to a potential that is more depolarized than –54.5 mV will cross the separatrix and drive the system to the up state (trajectory not shown). Notably, climbing fiber input (modeled here as a brief, large increase in sodium conductance) delivered during the up state can have a counterintuitive effect. As a result of the depolarization, the value of the inactivation term h during the input is decreased sufficiently to allow the system to cross the separatrix into the basin of attraction of the down state (right dashed line). Thus, when the input is terminated, the dynamics converge to the down state (dash-dotted line). Conversely, a sufficiently large hyperpolarizing pulse from the down state (left dashed

line) induces a substantial deinactivation of the h-like current, leading the system into the basin of attraction of the depolarized stable point. Thus, after the outward current pulse the dynamics will converge to the up state (left dash-dotted line). Varying the properties of the outward current could affect the ability of brief hyperpolarizing current pulses or depolarizations to induce transitions between states. To study this effect, we separated the voltageindependent current into a leak and a voltage-independent potassium component. A sufficient decrease in the potassium conductance will ˙ = 0 nullcline (blue line), such that simulated climbdownshift the V ing fiber input will induce an upward transition but not the opposite transition (Fig. 8c). Conversely, an increase in this conductance ˙ = 0 nullcline such that the simulated climbing fiber will upshift the V input will only generate a transition to the down state (Fig. 8d). To explain the spontaneous transitions between up states and down

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

SS frequency (Hz)

SS frequency (Hz)

No transitions

SS frequency (Hz)

Down to up transitions
CS probability

Stim.

SS frequency (Hz)

Stim.

CS SS

207

ARTICLES a b
0.6 20 mV 1s

0.4

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

0.2

0 –100

–80

–60

–40

–20

0

20

40

V (mV)

c

20 mV 0.5 s

d

20 mV 0.5 s

Figure 8 A model for bistability and state transitions. (a,b) Simulation of a model neuron incorporating a non-inactivating inward current, a slow h-like current, and a voltage-independent outward current. Transitions between the two stable states in both directions can be induced by brief outward current pulses (0.1 s, 7,200 nA cm–2) (a, left) or by simulated climbing fiber input (4 ms, 1,200 µS cm–2 sodium conductance) (a, right). (b) Phase plane for the two dynamic variables in the model, the membrane potential V ˙ = 0 and the and an inactivation term h. The solid lines (red, blue) are the h ˙ = 0 nullclines respectively; the circles denote the two stable states (fixed V points) and the square denotes the unstable fixed point, which is located on the separatrix, the border between the basins of attraction of the up state (blue) and the down state (green). Arrows mark the trajectory of the dynamic variables (V, h) during (dashed lines until the diagonal arrowheads) and after (dash-dotted lines) an outward current injection (left) or a simulated climbing fiber-like input (right). (c) Decreasing the potassium conductance ˙ = 0 nullcline down (bottom) such (see Results for details) shifts the V that the trajectory during a complex spike-like depolarization (dashed line) does not cross the separatrix. Thus, a climbing fiber input will only induce a down-to-up and not an up-to-down transition (top). (d) Inversely, increasing the potassium conductance shifts the nullcline up such that the trajectory during a complex spike-like depolarization (dashed line) crosses the separatrix twice (bottom). In this case, a climbing fiber input will only induce an up-to-down and not a down-to-up transition (top).

h

0.4

0.4

0 –80

–40

0

40

0 –80

–40

0

40

V (mV)

V (mV)

states observed in vivo (see Fig. 6a, bottom) and in vitro6,15,21,22, we modified the model by replacing the voltage-independent potassium current with a slowly activating potassium current. In the absence of climbing fiber input (similar to in vitro conditions), this modification generated spontaneous transitions between the states, resembling those observed in slice preparations6,15,21,22 (Supplementary Fig. 2). In the presence of climbing fiber–like input, this model reproduced both spontaneous and climbing fiber–evoked transitions, as well as the occasional failures of climbing fiber input to evoked a transition (see Supplementary Note and Supplementary Fig. 2). DISCUSSION We have demonstrated that Purkinje cells show bistability of membrane potential and spike output in vivo, which is a consequence of their intrinsic membrane properties. This bistability can be bidirectionally triggered by sensory-driven synaptic input, suggesting that it may represent an important intrinsic cellular mechanism for processing of sensory information in the cerebellar cortex. Bistability of Purkinje cell output in vivo Bistability of Purkinje cells in vitro is well-documented and has been observed to occur spontaneously6,21,23 or after modulation of intrinsic conductances15. Our intracellular recordings from Purkinje cells of rats and guinea pigs anesthetized with ketamine-xylazine or pentobarbital establish that this bistability is retained in vivo in different animal species and under different anesthetic conditions. We further demonstrate that this bistability of membrane potential is manifested in two modes of firing activity: tonic firing of simple spikes and quiescent periods. The presence of bistability in Purkinje cell spike output is supported by previous reports showing intermittent firing patterns where periods of high-frequency, simple spike firing are interspersed with

periods of quiescence in anesthetized24 and decerebrate animals25–27. Similar irregular alternations between quiescence and high-frequency simple spike activity have been observed in awake recordings from various animals, including frogs28, cats18,29,30, squirrel monkeys31 and rhesus monkeys32,33. In contrast, other cerebellar studies have not reported similar phenomena and typically find continuous firing of Purkinje cells rather than bimodal discharge, suggesting that in these cases the dynamics of Purkinje cells are not bistable. The reasons for this discrepancy are still unclear. As in other structures of the central nervous system showing intrinsic bistability, including spinal cord8, thalamus34 and olfactory bulb7, this may reflect heterogeneity in the expression of bistability within a particular cell type. Furthermore, the existence of bistability in a given cell is likely to depend on neuromodulation15,23 and on the balance of excitation and inhibition in the local network, which may depend both on the cerebellar region and the behavioral state. Our finding that complex spikes triggered by climbing fiber input can switch Purkinje cells between up states and down states may help to resolve the discrepancy in the literature regarding the effect of complex spikes on simple spiking. Complex spikes have been shown to be associated with both increases26,35,36 and decreases18,24,37 in simple spike activity. Our data suggest that the effect of a complex spike on simple spike discharge depends on the state of the Purkinje cell just before the complex spike. Thus, Purkinje cells predominantly in the down state will tend to respond with an increase in simple spike activity associated with the complex spike, whereas Purkinje cells predominantly in the up state will tend to respond with a decrease. This is in agreement with the prolonged dendritic plateaus associated with increased spiking reported to be triggered by climbing fiber input in intracellular and extracellular recordings from anesthetized cats38. Mechanisms of bistability Bistability is an intrinsic membrane property of Purkinje cells in vivo, as transitions between states can be triggered by brief hyperpolarizing and depolarizing current pulses and abolished by sufficiently strong negative or positive constant current injection. Further evidence that up states in Purkinje cells are not sustained by synaptic activity is provided by our recordings from interneurons of the molecular layer (Fig. 2e,f) and granule cells (Fig. 2b,c), the source of

h

208

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
parallel fiber input to both interneurons and Purkinje cells. Neither interneurons nor granule cells showed bistability, and thus it seems unlikely that up or down states are maintained by bimodal activity in the parallel fiber or interneuron input. Bistability in Purkinje cells is thus very different from the up and down states described in many cortical and striatal neurons in vivo. In these neurons, the up state is thought to result mainly from a continuous barrage of synaptic excitation reflecting widespread activity across the network3,39. In contrast, in Purkinje cells, the persistence of the two states is maintained by intrinsic voltage-dependent mechanisms. The transition between states in Purkinje cells can be triggered by the activation of the same single (albeit large) synaptic input. Notably, a recent in vitro study3 has shown that the same electrical stimulus applied to the white matter of the cerebral cortex can turn on and off a network-induced up state in cortical neurons. Although the ionic and synaptic currents involved in sustaining up states and down states in these cells are different from those in Purkinje cells, our dynamical model may serve as a conceptual framework for understanding this phenomenon. Previous simplified models for neuronal bistability in general40 and Purkinje cell bistability in particular41,42, as well as the present model, indicate that neither a detailed interplay between a large number of conductances nor a complex dendritic geometry are required to qualitatively generate this behavior. In our dynamical model, bistability relies on a non-inactivating inward current. However, many conductances are likely to influence the duration of states and the probability of transitions between states. Indeed, previous in vitro studies have shown that Purkinje cell bistability could result from interactions between non-inactivating sodium and calcium conductances with potassium conductances6,43. The role of the h-current (Ih) in generating the bistability is more controversial. Some in vitro studies have reported bistability in Purkinje cells in the presence of Ih6,21, whereas others have shown that its downregulation, for instance by serotonin, can unmask or enhance bistability15. Our experiments clearly show that Ih is present in bistable Purkinje cells in vivo (e.g. see Supplementary Fig. 1; see also the depolarizing sag in Figs. 1a,b, 3c and 4a,d). This current may also have an important role in setting the voltage of the states and in controlling the transitions between the states. In particular, the ability of outward current pulses to induce bidirectional transitions requires a ‘rebound’ response, which is likely to result from the slow deactivation of the hcurrent (see Supplementary Note). Purkinje cell bistability and olivo-cerebellar dynamics The long timescale associated with Purkinje cell bistability allows a large number of different dynamical states to be sustained in the cerebellar cortex for extended periods. Each of these states is associated with a specific configuration of up and down states in different Purkinje cells. These network states can be used to maintain specific functional networks within the olivary nucleus: the sustained firing state of a Purkinje cell inhibits neurons in the deep cerebellar nuclei, which in turn remove the inhibition from the electrically coupled dendrites of the inferior olive neurons, producing synchronized rhythmic activity in a subset of olivary neurons. Notably, recordings of deep cerebellar nucleus neurons show prolonged (up to 0.5 s) inhibitory responses to brief sensory stimulation (N.C. Rowland and D. Jaeger, Soc. Neurosci. Abstr. 75.10, 2003), which may correspond to synchronized up states in presynaptic Purkinje cells activated by sensory stimulation. Computational implications of bistability The presence of bistability may have important implications for integration of mossy fiber input to the cerebellar cortex mediated by the parallel fiber pathway. First, the responsiveness of the Purkinje cell to sensory-evoked parallel fiber input will depend on whether previous climbing fiber input has switched the Purkinje cell into the up state or down state. This suggestion is consistent with previous work showing that the gain of Purkinje cell responses to sensory-driven mossy fiber input depends on the recent history of climbing fiber activity on the time scale of hundreds of milliseconds44. Second, given that bistability is also observed in the distal dendrites, the activation of voltage-gated dendritic conductances6 will also be influenced. In particular, activation of dendritic calcium spikes by parallel fiber input is strongly voltage and time dependent45. Given that calcium spikes have been implicated in the induction of long-term depression (LTD) at parallel fiber synapses46, the fact that sensory-evoked climbing fiber inputs can induce prolonged depolarizations provides a possible cellular substrate for climbing fiber– dependent forms of parallel fiber plasticity in which a broad time window for induction (several hundred milliseconds) has been observed38. Purkinje cell bistability may therefore contribute to determining the rules for LTD induction in the intact cerebellar network. Chemical synapses have traditionally been classified as either excitatory or inhibitory according to their effect on the firing rate of the postsynaptic neuron. In this study we have demonstrated that the same synapse can play both roles. The activation of climbing fiber synapses can either increase or decrease the firing rate of the Purkinje cell, depending on its initial state. This property, in which the same input induces both transitions in a bistable element, is known in electronics as a ‘toggle switch’, which is widely used in electrical devices. We suggest that toggling of the Purkinje cells may serve as a higher-order reflex that generates an immediate and reflexive response of the system to the occurrence of an error by shifting the Purkinje cell activity away from its current erroneous operating state. METHODS
The care and experimental manipulation of the animals was carried out in accordance with the regulations of the Hebrew University of Jerusalem and the U.K. Home Office. In vivo recordings. Sprague-Dawley rats (P18–P27) or Dunkin-Hartley guinea pigs (180–300 g) were anesthetized by intraperitoneal injection of ketamine (50 mg kg–1)-xylazine (3 mg kg–1) or pentobarbital (60 mg kg–1). The level of anesthesia was routinely monitored by observing whisker movements or the response to a noxious stimulus to the hind limbs, and additional doses of anesthetic were added as needed. The animals were placed in a stereotaxic apparatus, and the occipital bone and the dura mater were removed, exposing a small region of the cerebellar vermis or hemispheres. The exposed area was covered with physiological saline or agar (3% in physiological or saline solutions). Sensory responses were evoked by an air puff (30–70 ms, 40 psi) timed by a custom pressure device and delivered to the ipsilateral perioral surface. Cell-attached and whole-cell patch-clamp recordings were made with a Multiclamp 700A or an Axoclamp 2B amplifier (Axon Instruments) using previously described techniques47. The pipette solution contained 130 mM potassium methanesulfonate, 7 mM KCl, 10 mM HEPES, 0.05 mM EGTA, 2 mM MgATP, 2 mM Na2ATP and 0.5 mM Na2GTP, pH 7.2, in recordings from rats, or 140 mM potassium gluconate, 4 mM NaCl, 10 mM HEPES, 4 mM MgATP, 1 mM EGTA and 0.1 mM CaCl2, pH 7.4, for guinea pig recordings. Extracellular recordings of guinea pig Purkinje cell single-unit activity were obtained using glass pipette electrodes, pulled to a DC resistance of 10–20 MΩ and filled with 2 M NaCl. A differential AC amplifier (DP-301; Warner Instruments) was used for monitoring ongoing activity. Recordings were filtered at 3–10 kHz and sampled at 20–50 kHz using an Instrutech (ITC-18) or National Instruments (PCI-MIO16XE-10) analog to digital board. Slice recordings. The in vitro experiments were performed on 300 µm thick sagittal slices of cerebellar vermis from guinea pigs (180–200g). Preparation of slices and recording techniques are described elsewhere48. Recordings were

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

209

ARTICLES
performed at 27–30 °C or at room temperature (22–25 °C) in physiological solution containing 124 mM NaCl, 5 mM KCl, 1.3 mM MgSO4, 1.2 mM KH2PO4, 26 mM NaHCO3, 10 mM glucose, and 2.4 mM CaCl2. Analysis. Correlation between complex spikes and state transitions in intracellular recordings was performed in Purkinje cells where the identification of complex spikes was unambiguous. The response of the membrane potential to climbing fiber input during the up and down states could differ considerably (see Figs. 5a and 6a). During the down state, complex spikes were characterized by a large-amplitude spike followed by a burst of smaller spikes and a long depolarization6. During the up state, complex spikes were characterized by a stereotypic waveform, composed of several fast distinct peaks. The rate of these complex spikes was independent of the state of the cell. Simple and complex spikes recorded extracellularly were sorted offline according to their amplitude and shape, using an adaptive template routine in MATLAB (version 6.0, Mathworks). To control the quality of sorting, all simple spikes and complex spikes were superimposed and examined visually. Cells were discarded from the analysis if the estimated number of errors in sorting exceeded 2%. Histograms of membrane potential distribution were constructed for each Purkinje cell, and Gaussian fits were applied to the individual peaks. The modal value of each Gaussian fit was taken as the mean membrane potential for the up state and the down state. To measure the duration of each individual sate, transitions between the up and down states were detected using the following thresholding procedure: two thresholds were set at one-fourth and three-fourths of the distance between the peaks of the membrane potential distribution. When the membrane potential rose above the lower threshold, a down-to-up transition was registered; when the membrane potential fell below the upper threshold, an up-to-down transition was registered. To avoid defining noise or individual complex spikes as a short state, we removed pairs of lower and upper thresholds generating a time difference of less than 100 ms. Up and down state durations were quantified in Purkinje cells where the duration of the recordings was long enough to allow measurement of at least 15 up and down states. The deviation from unimodality of the membrane potential of Purkinje cells, as well as that of granule cells and molecular interneurons, was tested using a ‘dip test’16, using software R (version 1.9.1; http://www.R-project.org). For each cell, the test was performed using 500 randomly chosen values of membrane potential. In extracellular and cell-attached recordings, we defined a burst of simple spikes as an event that is flanked by periods of at least 100 ms devoid of simple spikes (qualitatively similar results were obtained using other criteria). Correlations between complex spikes and the beginning of a burst were calculated using a window of 50 ms, starting 25 ms after the complex spike. Correlations between complex spikes and the end of a burst were calculated using a window of 40 ms preceding the complex spike. The use of different windows did not produce a qualitative difference in the results. Input resistance (Rin) was calculated from steady-state voltage deflections during (400–500 ms) step hyperpolarizing current injections (granule cells, 5–10 pA; molecular layer interneurons, 20–70 pA). Additional data analysis was carried out using Igor Pro (Wavemetrics). Data are mean ± s.e.m. unless otherwise indicated. Statistical comparisons were made using Student’s paired or unpaired t-test.
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS We thank H. Meiri, E. Chorev and P. Mann-Metzer for excellent technical assistance, J.T. Davie for help with programming, and J.I. Simpson and T. Margrie for encouragement and helpful discussions. This work was supported by grants from the European Commission (M.H. and Y.Y.), Wellcome Trust (M.H and S.M), Gatsby Foundation (M.H), JSPS (K.K.), US-Israel BSF (Y.Y.), the Israel Science Foundation (Y.Y.), the Israel Science Foundation Center of Excellence 8006-00 (H.S.) and the Yeshaya Horowitz Association (Y.L.). COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 20 October 2004; accepted 3 January 2005 Published online at http://www.nature.com/natureneuroscience/
1. Fuster, J.M. Unit activity in prefrontal cortex during delayed-response performance: neuronal correlates of transient memory. J. Neurophysiol. 36, 61–78 (1973). 2. McCormick, D.A. et al. Persistent cortical activity: mechanisms of generation and effects on neuronal excitability. Cereb. Cortex 13, 1219–1231 (2003). 3. Shu, Y., Hasenstaub, A. & McCormick, D.A. Turning on and off recurrent balanced cortical activity. Nature 423, 288–293 (2003). 4. Camperi, M. & Wang, X.J. A model of visuospatial working memory in prefrontal cortex: recurrent network and cellular bistability. J. Comput. Neurosci. 5, 383–405 (1998). 5. Marder, E., Abbott, L.F., Turrigiano, G.G., Liu, Z. & Golowasch, J. Memory from the dynamics of intrinsic membrane currents. Proc. Natl. Acad. Sci. USA 93, 13481– 13486 (1996). 6. Llinas, R. & Sugimori, M. Electrophysiological properties of in vitro Purkinje cell somata in mammalian cerebellar slices. J. Physiol. (Lond.) 305, 171–195 (1980). 7. Heyward, P., Ennis, M., Keller, A. & Shipley, M.T. Membrane bistability in olfactory bulb mitral cells. J. Neurosci. 21, 5311–5320 (2001). 8. Lee, R.H. & Heckman, C.J. Bistability in spinal motoneurons in vivo: systematic variations in persistent inward currents. J. Neurophysiol. 80, 583–593 (1998). 9. Egorov, A.V., Hamam, B.N., Fransen, E., Hasselmo, M.E. & Alonso, A.A. Graded persistent activity in entorhinal cortex neurons. Nature 420, 173–178 (2002). 10. Koulakov, A.A., Raghavachari, S., Kepecs, A. & Lisman, J.E. Model for a robust neural integrator. Nat. Neurosci. 5, 775–782 (2002). 11. Goldman, M.S., Levine, J.H., Major, G., Tank, D.W. & Seung, H.S. Robust persistent neural activity in a model integrator with multiple hysteretic dendrites per neuron. Cereb. Cortex 13, 1185–1195 (2003). 12. Loewenstein, Y. & Sompolinsky, H. Temporal integration by calcium dynamics in a model neuron. Nat. Neurosci. 6, 961–967 (2003). 13. Häusser, M. & Clark, B.A. Tonic synaptic inhibition modulates neuronal output pattern and spatiotemporal synaptic integration. Neuron 19, 665–678 (1997). 14. Ito, M. The Cerebellum and Neural Control (Raven, New York, 1984). 15. Williams, S.R., Christensen, S.R. Stuart, G J. & Häusser, M. Membrane potential bistability is controlled by the hyperpolarization- activated current I(H) in rat cerebellar Purkinje neurons in vitro. J. Physiol. (Lond.) 539, 469–483 (2002). 16. Hartigan, J.A. & Hartigan, P.M. The dip test of unimodality. Ann. Stat. 13, 70–84 (1985). 17. Chadderton, P., Margrie, T.W. & Häusser, M. Integration of quanta in cerebellar granule cells during sensory processing. Nature 428, 856–860 (2004). 18. Armstrong, D.M. & Rawson, J.A. Activity patterns of cerebellar cortical neurones and climbing fibre afferents in the awake cat. J. Physiol. (Lond.) 289, 425–448 (1979). 19. Jorntell, H. & Ekerot, C.F. Receptive field plasticity profoundly alters the cutaneous parallel fiber synaptic input to cerebellar interneurons in vivo. J. Neurosci. 23, 9620–9631 (2003). 20. Brown, I.E. & Bower, J.M. Congruence of mossy fiber and climbing fiber tactile projections in the lateral hemispheres of the rat cerebellum. J. Comp. Neurol. 429, 59–70 (2001). 21. Rapp, M., Segev, I. & Yarom, Y. Physiology, morphology and detailed passive models of guinea-pig cerebellar Purkinje cells. J. Physiol. (Lond.) 474, 101–118 (1994). 22. Chang, W., Strahlendorf, J.C. & Strahlendorf, H.K. Ionic contributions to the oscillatory firing activity of rat Purkinje cells in vitro. Brain Res. 614, 335–341 (1993). 23. Hounsgaard, J. & Midtgaard, J. Intrinsic determinants of firing pattern in Purkinje cells of the turtle cerebellum in vitro. J. Physiol. (Lond.) 402, 731–749 (1988). 24. Bell, C.C. & Grimm, R.J. Discharge properties of Purkinje cells recorded on single and double microelectrodes. J. Neurophysiol. 32, 1044–1055 (1969). 25. Brookhart, J.M., Moruzzi, G. & Snider, R.S. Spike discharges of single units in the cerebellar cortex. J. Neurophysiol. 13, 465–486 (1950). 26. McDevitt, C.J., Ebner, T.J. & Bloedel, J.R. The changes in Purkinje cell simple spike activity following spontaneous climbing fiber inputs. Brain Res. 237, 484–491 (1982). 27. Granit, R. & Phillips, C.G. Excitatory and inhibitory processes acting upon individual Purkinje cells of the cerebellum in cats. J. Physiol. (Lond.) 133, 520–547 (1956). 28. Nacimiento, R.C. Spontaneous and evoked discharges of cerebellar Purkinje cells in the frog. in Neurobiology of Cerebellar Evolution and Development (ed. Llinas, R.) 373–395 (American Medical Assn., Chicago, 1969). 29. McCarley, R.W. & Hobson, J.A. Simple spike firing patterns of cat cerebellar Purkinje cells in sleep and waking. Electroencephalogr. Clin. Neurophysiol. 33, 471–483 (1972). 30. Edgley, S.A. & Lidierth, M. Step-related discharges of Purkinje cells in the paravermal cortex of the cerebellar anterior lobe in the cat. J. Physiol. (Lond.) 401, 399–415 (1988). 31. Hirata, Y. & Highstein, S.M. Analysis of the discharge pattern of floccular Purkinje cells in relation to vertical head and eye movement in the squirrel monkey. Prog. Brain Res. 124, 221–232 (2000). 32. Bauswein, E., Kolb, F.P. & Rubia, F.J. Cerebellar feedback signals of a passive hand movement in the awake monkey. Pflugers Arch. 402, 292–299 (1984). 33. Kobayashi, Y. et al. Temporal firing patterns of Purkinje cells in the cerebellar ventral paraflocculus during ocular following responses in monkeys II. Complex spikes. J. Neurophysiol. 80, 832–848 (1998). 34. Williams, S.R., Toth, T.I., Turner, J.P., Hughes, S.W. & Crunelli, V. The ‘window’ component of the low threshold Ca2+ current produces input signal amplification and bistability in cat and rat thalamocortical neurones. J. Physiol. (Lond.) 505, 689–705 (1997). 35. Rawson, J.A. & Tilokskulchai, K. Repetitive firing of cerebellar Purkinje cells in response to impulse in climbing fibre afferents. Neurosci. Lett. 25, 131–135 (1981). 36. Sato, Y., Miura, A., Fushiki, H. & Kawasaki, T. Short-term modulation of cerebellar Purkinje cell activity after spontaneous climbing fiber input. J. Neurophysiol. 68, 2051–2062 (1992).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

210

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
37. Rawson, J.A. & Tilokskulchai, K. Suppression of simple spike discharges of cerebellar Purkinje cells by impulses in climbing fibre afferents. Neurosci. Lett. 25, 125–130 (1981). 38. Ekerot, C.F. & Kano, M. Long-term depression of parallel fibre synapses following stimulation of climbing fibres. Brain Res. 342, 357–360 (1985). 39. Mahon, S., Deniau, J.M. & Charpier, S. Relationship between EEG potentials and intracellular activity of striatal and cortico-striatal neurons: an in vivo study under different anesthetics. Cereb. Cortex 11, 360–373 (2001). 40. Rinzel, J. & Ermentrout, B. Analysis of Neural Excitability and Oscillations in Methods of Neuronal Modeling (eds. Koch, C. & Segev, I.) 251–291 (MIT Press, Cambridge, 1998). 41. Yuen, G.L., Hockberger, P.E. & Houk, J.C. Bistability in cerebellar Purkinje cell dendrites modelled with high- threshold calcium and delayed-rectifier potassium channels. Biol. Cybern. 73, 375–388 (1995). 42. Genet, S. & Delord, B. A biophysical model of nonlinear dynamics underlying plateau potentials and calcium spikes in Purkinje cell dendrites. J. Neurophysiol. 88, 2430–2444 (2002). 43. Llinas, R. & Sugimori, M. Electrophysiological properties of in vitro Purkinje cell dendrites in mammalian cerebellar slices. J. Physiol. (Lond.) 305, 197–213 (1980). 44. Ebner, T.J. & Bloedel, J.R. Role of climbing fiber afferent input in determining responsiveness of Purkinje cells to mossy fiber inputs. J. Neurophysiol. 45, 962–971 (1981). 45. Midtgaard, J., Lasser-Ross, N. & Ross, W. Spatial distribution of Ca2+ influx in turtle Purkinje cell dendrites in vitro: role of a transient outward current. J. Neurophysiol. 70, 2455–2469 (1993). 46. Wang, S.S., Denk, W. & Häusser, M. Coincidence detection in single dendritic spines mediated by calcium release. Nat. Neurosci. 3, 1266–1273 (2000). 47. Margrie, T.W., Brecht, M. & Sakmann, B. In vivo, low-resistance, whole-cell recordings from neurons in the anaesthetized and awake mammalian brain. Pflugers Arch. 444, 491–498 (2002). 48. Mann-Metzer, P. & Yarom, Y. Electrotonic coupling interacts with intrinsic properties to generate synchronized activity in cerebellar networks of inhibitory interneurons. J. Neurosci. 19, 3298–3306 (1999).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

211

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Central amygdala ERK signaling pathway is critical to incubation of cocaine craving
Lin Lu, Bruce T Hope, Jack Dempsey, Shirley Y Liu, Jennifer M Bossert & Yavin Shaham
Using a rat model of craving and relapse, we have previously found time-dependent increases in cue-induced cocaine seeking over the first months of withdrawal from cocaine, suggesting that drug craving incubates over time. Here, we explored the role of the amygdala extracellular signal–regulated kinase (ERK) signaling pathway in this incubation. Cocaine seeking induced by exposure to cocaine cues was substantially higher after 30 withdrawal days than after 1 withdrawal day. Exposure to these cues increased ERK phosphorylation in the central, but not the basolateral, amygdala after 30 d, but not 1 d, of withdrawal. After 30 d of withdrawal from cocaine, inhibition of central, but not basolateral, amygdala ERK phosphorylation decreased cocaine seeking. After 1 d of withdrawal, stimulation of central amygdala ERK phosphorylation increased cocaine seeking. Results suggest that the incubation of cocaine craving is mediated by time-dependent increases in the responsiveness of the central amygdala ERK pathway to cocaine cues.

Relapse to cocaine use after prolonged abstinence is a major clinical problem. This relapse can be induced by exposure to cues associated with cocaine use even after many months of abstinence1. To account for the persistent propensity for relapse, it has been suggested2 that craving induced by cocaine cues increases over the first several weeks of withdrawal and remains high over extended drug-free periods. Using a rat model of drug craving and relapse, we have identified an analogous phenomenon: time-dependent increases in cocaine seeking induced by exposure to cocaine cues over the first months of withdrawal3–5. These data suggest that craving, a motivational state elicited by exposure to drug cues that often precedes and accompanies drug seeking, incubates over time. The incubation of cocaine craving may be mediated by neuroadaptations within mesolimbic dopamine reward circuits that are induced by cocaine exposure and subsequent withdrawal6–9. In support of this idea, we found that the time-dependent increase in cocaine seeking after withdrawal is associated with an increase in the peptide levels of the plasticity-related growth factor BDNF (brain-derived nerve growth factor) in the ventral tegmental area (VTA, the cell body region of mesolimbic dopamine) and in its terminal areas, the nucleus accumbens and the amygdala10. We also found long-lasting increases in glutamate receptor expression (up to 90 d after withdrawal from cocaine self-administration) in the VTA and nucleus accumbens11. However, the precise role of these cocaine-induced neuroadaptations in the incubation of cocaine craving is not known, and as discussed elsewhere12,13, these neuroadaptations seem likely to potentiate an ongoing incubation process rather than directly mediating it12,13. Thus, the specific molecular alterations underlying the incubation of cocaine craving remain unknown. In the present study, we explored whether the time-dependent increase in cocaine seeking induced by cocaine cues involves activation of the

ERK signaling pathway in the amygdala. The ERK pathway is activated in mesolimbic dopamine areas by acute or repeated exposure to cocaine14–16. This pathway is known to be involved in synaptic plasticity and learning and memory17,18. In the amygdala, activation of ERK plays a key role in Pavlovian fear conditioning19 and extinction of learned fear20. Different amygdala subnuclei mediate different learning and motivational processes underlying behaviors controlled by appetitive or aversive stimuli21,22. In particular, the basolateral (BLA) and the central (CeA) nuclei of the amygdala have different roles in the acquisition and expression of cue-controlled drug and non-drug reward seeking23,24. However, the role of the ERK pathway in these nuclei in cocaine seeking induced by drug cues, as well as in the time-dependent increases in cocaine seeking after withdrawal, is unknown. Here we provide evidence that the central, but not basolateral, amygdala ERK pathway is critically involved in the time-dependent increase in cocaine seeking induced by exposure to cocaine cues after withdrawal, suggesting that this pathway mediates the incubation of cocaine craving. RESULTS In the present experiments, rats were trained for six hours per day for 10 d to self-administer intravenous cocaine (0.75 mg kg–1 per infusion) or saline (a control condition); each infusion was paired with a 5-s tone-light cue. During training, presses on one lever (the active lever) led to cocaine (or saline) delivery, whereas pressing on a second (inactive) lever had no consequences. We assessed cocaine seeking induced by exposure to cocaine-associated cues in 30-min extinction tests performed 1 or 30 d after withdrawal from drug self-administration (Fig. 1a). In these tests, rats were exposed to contextual cues previously associated with cocaine availability (such as houselight or lever

Behavioral Neuroscience Branch, Intramural Research Program/National Institute on Drug Abuse, National Institutes of Health, Department of Health and Human Services, 5500 Nathan Shock Drive, Baltimore, Maryland 21224, USA. Correspondence should be addressed to Y.S. ([email protected]). Published online 16 January 2005; doi:10.1038/nn1383

212

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES a
Cocaine or saline self-administration training (10 d) Withdrawal period (1 or 30 d)

Extinction test Day 1

Extinction test Day 30 Number of responses (30 min)

b
Number of infusions (6h)

80 60 40 20 0

Training

c

80 60 40 20 0

Extinction test

Saline Cocaine *

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Saline Cocaine

Figure 1 Time-dependent increases in cocaine seeking after withdrawal. (a) Timeline of the experimental procedure. (b) Training phase: number of saline or cocaine infusions (mean ± s.e.m.) over the ten 6-h daily selfadministration sessions. Cocaine (0.75 mg kg–1 per infusion) supported robust self-administration behavior, whereas with saline, the number of lever presses was very low. (c) Test for cocaine seeking: number of responses (mean ± s.e.m.) on the previously active lever during the tests for cocaine or saline seeking performed under extinction conditions after 1 d or 30 d of withdrawal. During the test sessions, cocaine or saline was not available and lever presses resulted in the delivery of the tone-light cue previously paired with cocaine or saline infusions. All rats were chronically housed in the self-administration chambers during training; the rats in the 30-d withdrawal groups were housed in their home cage in the animal facility after the self-administration training. Non-reinforced lever responding in cocainetrained rats tested after 30 d of withdrawal was higher than in the other experimental groups (*, P < 0.01, n = 8–9 per group).

1 2 3 4 5 6 7 8 9 10 Training day

1

30 Withdrawal day

extension), and presses on the previously active lever resulted in contingent presentations of the discrete tone-light cue; this discrete cue served as a conditioned reinforcer during testing. Conditioned reinforcers are previously neutral stimuli that acquire reinforcing effects through their prior association with an unconditioned reinforcer such as food or drug. Extinction procedures have been used to assess mechanisms underlying the motivational effects of cues previously paired with drug or non-drug reinforcers23,25–27. Cocaine cues CeA activate ERK after prolonged withdrawal In Experiment 1, we used eight groups of rats in a 2 (reward type: saline, cocaine) × 2 (withdrawal day: 1, 30) × 2 (extinction test: no test, test) factorial design to test the effect of exposure to cocaine cues on cocaine seeking and ERK activity in the CeA and BLA after withdrawal. Rats demonstrated reliable self-administration of cocaine but not saline (Fig. 1b). The statistical analysis revealed an interaction of training day × reward type (F9,594 = 13.7, P < 0.01) for the number of infusions. As in previous studies13, the number of lever responses after 30 d of withdrawal from cocaine was higher than after 1 d (Fig. 1c). The statistical analysis revealed an interaction of withdrawal day × reward type for presses on the active (F9,594 = 18.2, P < 0.01) but not the inactive (P > 0.05) lever. For all groups, the number of presses on the inactive lever (a lever that was not previously associated with cocaine or saline infusions) during the extinction tests was very low (fewer than three).
Figure 2 Exposure to cocaine cues after 30 d of withdrawal increases ERK phosphorylation in the central amygdala. Data from the experimental groups are presented as percent of phosphorylated ERK (mean ± s.e.m.) of naive-control rats that were not included in the behavioral experiment. (a,b) Phosphorylated and total ERK in the central amygdala. (c,d) Phosphorylated and total ERK in the basolateral amygdala. Rats in the extinction test condition were trained to self-administer cocaine or saline and were exposed to the cocaine or saline cues in a 30-min extinction test after 1 or 30 d of withdrawal. Rats in the ‘no test’ condition were trained to self-administer cocaine or saline and were not exposed the cocaine or saline cues after 1 or 30 d of withdrawal. In the central amygdala, exposure to cocaine cues in the extinction test induced a selective increase in phosphorylated ERK after 30 d of withdrawal. In the basolateral amygdala, cocaine self-administration led to increases in phosphorylated ERK after 1 d of withdrawal; this effect was not associated with the exposure to cocaine cues in the extinction tests. *, different from the other experimental groups, P < 0.05 (n = 8–9 per group). #, different from the P < 0.05. ERK1 and ERK2, extracellular signal–regulated kinases 1 and 2; p, phosphorylated; t, total; SA, self-administration.

In the CeA, exposure to cocaine cues after 30 d of withdrawal induced selective increases in phosphorylated, but not total, ERK (Fig. 2a,b). The statistical analysis revealed significant or near-significant interactions of extinction test × withdrawal day (F1,59 = 6.4, P < 0.05), extinction test × reward type (F1,59 = 3.3, P = 0.074) and withdrawal day × reward type (F1,59 = 3.9, P = 0.05). A subsequent one-way ANOVA revealed an effect of experimental group (F7,55 = 3.9, P < 0.01) and a post-hoc analysis showed that the phosphorylated ERK levels of the rats from the cocaine-trained group exposed to the cocaine cues after 30 d of withdrawal were higher than those of the other groups (Fig. 2a). In the BLA, prior exposure to cocaine increased phosphorylated, but not total, ERK after 1 d, but not 30 d, of withdrawal (Fig. 2c,d). This effect, however, was not associated with the exposure to cocaine cues in the extinction tests. The statistical analysis revealed a significant effect of withdrawal day × reward type (F1,59 = 5.1, P < 0.05). Subsequent oneway ANOVA revealed a significant effect of experimental group (F7,60 = 2.3, P < 0.01) and post-hoc analysis found that phosphorylated ERK

a
Percentage of naive rats

Central amygdala: phosphorylated ERK

c

Basolateral amygdala: phosphorylated ERK

Day

Day

Day

Day

Day

Day

Day

Day

b
Percentage of naive rats

Central amygdala: total ERK

d

Basolateral amygdala: total ERK

Day

Day

Day

Day

Day

Day

Day

Day

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

213

ARTICLES a
Figure 3 Exposure to cocaine cues after 30 d of withdrawal increases ERK phosphorylation in the central amygdala: a replication using rats chronically housed in the cocaine self-administration environment during the withdrawal period. (a) Test for cocaine seeking: number of responses (mean ± s.e.m.) on the previously active lever and on the inactive lever during the tests for cocaine seeking performed under extinction conditions after 1 or 30 d of withdrawal. Non-reinforced lever responding in rats tested after 30 d of withdrawal was higher than in rats tested after 1 d of withdrawal. All rats were chronically housed in the self-administration chambers during the training phase (10 d) and the subsequent withdrawal periods. (b–e) Phosphorylated and total ERK in the central and basolateral amygdala. Data from the experimental groups are presented as a percentage of phosphorylated or total ERK of naive control rats (mean ± s.e.m.) that were not included in the behavioral experiment. In the central amygdala, exposure to cocaine cues in the extinction test induced a selective increase in phosphorylated ERK after 30 d of withdrawal. In the basolateral amygdala, cocaine self-administration led to increases in phosphorylated ERK after 1 d of withdrawal; this effect was not associated with the exposure to cocaine cues in the extinction tests. *, different from the other groups, P < 0.01. #, different from the day 30 groups, P < 0.05 (n = 7–9 per group). ERK, extracellular signal–regulated kinases; p, phosphorylated; t, total.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

b
Percentage of naive rats

Central amygdala: phosphorylated ERK

d

Basolateral amygdala: phosphorylated ERK

c
Percentage of naive rats

Central amygdala: total ERK

e

Basolateral amygdala: total ERK

ERK (P > 0.05) (Fig. 3b,c). In the BLA, prior exposure to cocaine selfadministration increased phosphorylated ERK (F1,32 = 62.1, P < 0.01 for effect of withdrawal day), but not total ERK (P > 0.05), after 1 d, but not 30 d, of withdrawal (Fig. 3d,e). Together, these findings suggest that the environmental context to which the rats were exposed during the withdrawal period does not mediate the time-dependent increases in extinction responding or CeA ERK phosphorylation. CeA ERK inhibition attenuates cocaine seeking In Experiment 3, we used four groups of rats to determine the functional role of activation of the ERK pathway in the amygdala in enhanced cocaine seeking after 30 d of withdrawal. For this purpose, we infused U0126, which inhibits ERK phosphorylation28, or its vehicle (20% DMSO) into the CeA or BLA 30 min before the extinction tests. Active lever responding was decreased by infusions of U0126 into the CeA but not the BLA (Fig. 4a,b, left). Statistical analysis, which included the factors of U0126 dose (0, 100 ng per side) and amygdala site (CeA, BLA), showed an interaction between these two factors for active (F1,33 = 5.6, P < 0.05), but not inactive (P > 0.05), lever responding. The analysis of the data from the western blot assays revealed that intra-CeA infusions of U0126 decreased phosphorylated ERK in the CeA (F1,16 = 10.6, P < 0.01) but not the BLA (P > 0.5), whereas intra-BLA infusions of U0126 decreased phosphorylated ERK in the BLA (F1,14 = 9.1, P < 0.01) but not the CeA (P > 0.5) (Fig. 4a,b, right). U0126 infusions into the CeA or BLA had no effect on total ERK (data not shown). Representative placements of the tips of the injectors within the CeA and BLA are shown in Figure 4c. The pictures were taken from naive rats that were injected with 20% DMSO. Naive rats were used because the CeA and BLA of the experimental rats were dissected for the western blot assays. The increase in phosphorylated ERK in the CeA after exposure to cocaine cues during the day 30 extinction test may be mediated by increases in glutamate transmission; glutamate is known to activate the ERK pathway through its action on the NMDA (N-methyl-Daspartate) receptor17. This possibility was assessed in Experiment 4 in two groups of rats by infusing AP-5, an antagonist of the NMDA receptor, or saline vehicle into the CeA before the extinction tests that were conducted after 30 d of withdrawal. CeA infusions of AP-5 decreased active (F1,16 = 18.1, P < 0.01), but not inactive (P > 0.05),
VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

levels were higher for the cocaine-trained rats from the 1-d withdrawal groups than for the other groups (Fig. 2c). The rats from the 30-d cocaine withdrawal group of Experiment 1 were housed in the animal facility after self-administration training and were brought back to the test chambers on the day of the extinction test. Thus, selective increases in CeA ERK phosphorylation after 30 d of withdrawal may be due to a context switch in the late withdrawal group. In order to examine this possibility, in Experiment 2, four groups of rats were trained to self-administer cocaine for 10 d and were tested, or not tested, in 30-min extinction sessions after 1 or 30 d of withdrawal; during the withdrawal periods, rats were housed in the test chambers (the drug environment). The number of cocaine infusions per 6 h (mean ± s.e.m.; data collapsed from the four groups) on the last three training days was 62.9 ± 5.8, 61.1 ± 4.1 and 61.2 ± 4.0, respectively. The number of lever responses in the extinction tests was higher after 30 d of withdrawal from cocaine than after 1 d of withdrawal (F1,16 = 18.9, P < 0.01 for an effect of withdrawal period) (Fig. 3a). In the CeA, exposure to cocaine cues after 30 d of withdrawal induced a selective increase in phosphorylated ERK (F1,29 = 33.6, P < 0.01 for interaction of withdrawal day × extinction test) but not total

214

ARTICLES
Figure 4 Inhibition of ERK phosphorylation in the central amygdala decreases cocaine seeking after 30 d of withdrawal. (a,b) Left column: number of responses (mean ± s.e.m.) on the previously active lever (a measure of cocaine seeking) and on the inactive lever after infusions of U0126 (an inhibitor of ERK phosphorylation) or its vehicle (20% DMSO) into (a) the central amygdala or (b) the basolateral amygdala 30 min before the extinction tests that were conducted after 30 d of withdrawal from cocaine. Bilateral infusions of U0126 (100 ng per side) into the central but not the basolateral amygdala decreased cocaine seeking. Right column: percentage of phosphorylated ERK in the central and basolateral amygdala after infusions of U0126 into (a) the central amygdala or (b) the basolateral amygdala 30 min before the extinction tests. Data are presented as a percentage of the values obtained for the rats infused with the vehicle. *, different from the vehicle condition, P < 0.05 (n = 7–9 per group). (c) Representative cannula placements in the CeA and BLA of naive rats injected with 20% DMSO.

a
Extinction responding 80 Responses (30 min) 60 40

Central amygdala U0126 infusions Phosphorylated ERK Percentage of vehicle values 150 Vehicle U0126 100 * 50

Inactive lever Active lever *

20 0 Vehicle U0126

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

0

Central amygdala

Basolateral amygdala

b
Extinction responding 80 Responses (30 min) 60 40 20 0 Vehicle

Basolateral amygdala U0126 infusions Phosphorylated ERK Percentage of vehicle values 150

100 * 50

0

U0126

Central amygdala

Basolateral amygdala

c
Central amygdala
M2 RS A S1Tr M1

Basolateral amygdala

S1 DZ cg RS Gb

IG alv df cc CA1 CA2 dhc DH C CA3 LV ec fi D3V st PV iml LH b MD C MD L IM D MD M eml ic IA M Rh rf CM PC VL VPM VPL Rt CPu LGP Sub VM IPAC mt Re B VRe AStr CeL LaDL I CeC CeM AI P GI CL S2 MH b DG LD DM LDVL S1BF

DI

CeA
cst al opt

Xi ns DA A13 Stg f mf b 3V Pe AH P

ZI

MG P SI LH

PaPo

VM HD M TC VM HC sox Ar cM Ar cD MEI Ar cL MEE SO R VM HV L BAOT

MeAD

I MeAV BMA

BLA
IM G BLA BLP VEn BLV Pir CxA

DEn

ACo PLCo

lever responding (Fig. 5a). The analysis of the data from the western blot assays revealed that intra-CeA infusions of AP-5 decreased phosphorylated ERK in the CeA (F1,16 = 12.3, P < 0.01) but not the BLA (P > 0.5) (Fig. 5b). AP-5 infusions had no effect on total ERK (data not shown). CeA ERK activation increases cocaine seeking In Experiment 5, to further determine the role of the ERK pathway in the CeA in the time-dependent increases in cocaine seeking after withdrawal, NMDA (which activates ERK) and U0126 were infused into the CeA on day 1 of withdrawal. We tested whether the induction of ERK phosphorylation in the CeA would restore cocaine seeking on day 1 of withdrawal. Because there are no selective agonists of the ERK pathway, we used NMDA to induce ERK phosphorylation29 and U0126 to reverse the effect of NMDA. Initially, we examined the

effects of intra-CeA infusions of NMDA on cocaine seeking and ERK phosphorylation in three groups of rats. Subsequently, we determined whether U0126 would attenuate the effects of NMDA in four groups of rats. CeA infusions of NMDA increased responding on the active lever (Fig. 6a, left column). The statistical analysis revealed a significant effect of NMDA dose for active (F2,21 = 5.6, P < 0.05), but not inactive (P > 0.2), lever responding. Western blot assays showed that intra-CeA infusions of NMDA increased phosphorylated ERK in the CeA (F2,20 = 4.8, P < 0.05) but not the BLA (P > 0.5) (Fig. 6a, right column). NMDA infusions had no effect on total ERK (data not shown). Pretreatment of the CeA with U0126 attenuated NMDA-induced increases in responding on the active lever (Fig. 6b, left column). Data were analyzed using the factors of NMDA dose (0, 250 ng per side) and U0126 dose (0, 100 ng per side). This analysis revealed an interaction between these two factors for active (F1,29 = 7.5, P < 0.05), but not inactive (P > 0.3), lever responding. Pretreatment with U0126 also attenuated NMDA-induced increases in phosphorylated ERK in the CeA (Fig. 6b, right column). The statistical analysis revealed effects of NMDA dose (F1,27 = 28.7, P < 0.01) and U0126 dose (F1,27 = 18.0, P < 0.01). Total ERK was not altered by infusing NMDA, U0126 or both (data not shown). CeA ERK inhibition does not alter cocaine or food intake We determined the specificity of the behavioral effects of intra-CeA infusions of U0126 on cocaine seeking induced by exposure to the drug cues. In Experiment 6, we tested whether these infusions would alter cocaine selfadministration or high rates of lever responding reinforced by high-fat palatable food. The number of cocaine infusions or food pellets (mean ± s.e.m.) earned in the last three daily 6-h sessions before testing was 58.3 ± 2.2, 62.7 ± 3.5 and 63.8 ± 4.8, and 160.8 ± 4.4, 163.5 ± 6.8 and 158.3 ± 7.6, respectively. U0126 had no effect on lever responding reinforced by cocaine or palatable food (P > 0.3 for the effects of U0126 dose or U0126 dose by hour) (Fig. 7).

Figure 5 Blockade of NMDA receptors in the central amygdala decreases cocaine seeking and ERK phosphorylation after 30 d of withdrawal. (a) Number of responses (mean ± s.e.m.) on the previously active lever and on the inactive lever after infusions of the NMDA antagonist AP-5 (3 µg per side) or its saline vehicle into the central amygdala 30 min before the extinction tests that were conducted after 30 d of withdrawal from cocaine. Bilateral infusions of AP-5 (3 µg per side) into the central amygdala decreased cocaine seeking. (b) Percentage of phosphorylated ERK in the central and basolateral amygdala after infusions of AP-5 into the central amygdala. Data are presented as a percentage of the values obtained for the rats infused with the vehicle. *, different from the vehicle condition, P < 0.05 (n = 8–10 per group).

a
Extinction responding 80 Responses (30 min) 60 40 * 20 0 Inactive lever Active lever Percentage of vehicle values

b
Phosphorylated ERK 200 150 100 * 50 0 Central amygdala Basolateral amygdala Vehicle AP-5

0

3.0

AP-5 dose (µg/side)

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

215

ARTICLES
Central amygdala infusions of NMDA Figure 6 Induction of ERK phosphorylation in the central amygdala by NMDA increases Extinction responding Phosphorylated ERK cocaine seeking after 1 d of withdrawal. (a) Vehicle 200 80 Left column: number of responses (mean ± NMDA (25 ng) s.e.m.) on the previously active lever and on * Inactive lever 150 NMDA (250 ng) 60 the inactive lever after infusions of NMDA or Active lever its vehicle into the central amygdala 5 min 100 * 40 before the extinction tests that were conducted after 1 d of withdrawal from cocaine. Bilateral 50 20 infusions of NMDA into the central amygdala increased cocaine seeking. Right column: 0 percentage (mean ± s.e.m.) of phosphorylated 0 Basolateral Central 0 25 250 ERK in the central and basolateral amygdala amygdala amygdala NMDA dose (ng/side) after infusions of NMDA into the central amygdala. Data are presented as a percentage of the values obtained for the rats infused Central amygdala infusions of NMDA and U0126 with the vehicle. *, different from the vehicle Extinction responding Phosphorylated ERK condition, P < 0.05 (n = 7–9 per group). (b) Left column: number of responses (mean ± 200 80 s.e.m.) on the previously active lever and on the inactive lever after the combined infusions of 150 60 NMDA (250 ng per side), U0126 (100 ng per * * side) or their vehicles into the central amygdala 100 40 *# 5 or 30 min, respectively, before the extinction tests conducted after 1 d of withdrawal from * 50 20 cocaine. U0126 infusions into the central * * amygdala reversed NMDA-induced potentiation 0 0 of cocaine seeking and NMDA-induced ERK Veh Veh U0126 U0126 Veh Veh U0126 U0126 phosphorylation. Right column: percentage Veh NMDA Veh NMDA Veh NMDA Veh NMDA (mean ± s.e.m.) of phosphorylated ERK in the Experimental condition Experimental condition central amygdala after infusions of U0126, NMDA and their vehicles before the extinction tests. Data are presented as a percentage of the values obtained for the rats infused with the vehicles. *, different from the vehicle-NMDA group, P < 0.05; #, different from the vehicle-vehicle group, P < 0.05 (n = 8–9 per group). Veh, vehicle (sterile saline for NMDA, and 20% DMSO for U0126).

a

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

b

DISCUSSION As in previous studies13, we have found time-dependent increases in cocaine seeking induced by exposure to drug cues, suggesting that cocaine craving incubates over time. These time-dependent increases in cocaine seeking were accompanied by enhanced ERK phosphorylation in the CeA. Furthermore, attenuation of ERK phosphorylation by U0126 or AP-5 infusions into the CeA decreased cocaine seeking after 30 d of withdrawal, and NMDA infusions into the CeA increased ERK phosphorylation and augmented cocaine seeking after 1 d of withdrawal. These findings suggest that timedependent increases in the responsiveness of central amygdala ERK to cocaine cues mediate the incubation of cocaine craving. Cocaine self-administration experience also resulted in increased ERK phosphorylation in the BLA. However, this effect was observed only after 1 d of withdrawal and was not dependent on exposure to cocaine cues during testing. Furthermore, BLA infusions of U0126 had no effect on cocaine seeking after 30 d of withdrawal. Thus, it is unlikely that

the short-lasting cocaine-induced changes in ERK activity in the BLA are related to the incubation of cocaine craving. Different roles of CeA and BLA in cue-induced drug seeking Previous studies have shown that permanent or reversible lesions of either the BLA or CeA attenuate cue-controlled cocaine seeking as measured by the cue-induced reinstatement or second order schedule procedures9,24; in these procedures, cocaine cues serve as conditioned reinforcers. Based on these reports, the selective activation of ERK in the CeA, but not in the BLA, by cocaine cues after 30 d of withdrawal, as well as the selective decrease in cocaine seeking at the same time point by infusions of U0126 in the CeA, but not in the BLA, were both unexpected findings. A conceptual framework that might account for the present and previous findings is that the BLA is involved in cocaine seeking induced by cocaine cues but is not critical for the time-dependent potentiation of this cocaine seeking after withdrawal. Rather, this potentiation is mediated by time-dependent increases in the responsiveness of the CeA ERK pathway to cocaine cues. This framework for the present and previous findings is consistent with studies showing a dissociation between the neuroanatomical

a
40 Active lever responses 30 20 10 0

Cocaine self-administration Vehicle U0126

b
160 Active lever responses 120 80 40 0

Food self-administration

1

2

3

4 Hour

5

6

1

2

3 4 Hour

5

6

Figure 7 Inhibition of ERK activity in the central amygdala has no effect on lever responding reinforced by cocaine or palatable food. (a) Number of active lever presses (mean ± s.e.m.) reinforced by cocaine (0.75 mg kg–1 per infusion) after bilateral infusions of U0126 (100 ng per side) or its vehicle (20% DMSO). (b) Number of active lever presses (mean ± s.e.m.) reinforced by palatable high-fat 45 mg pellets after bilateral infusions of U0126 or its vehicle. U0126 infusions had no significant effect on lever presses reinforced by cocaine or palatable food.

216

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

Percentage of vehicle values

Responses (30 min)

Percentage of vehicle values

Responses (30 min)

ARTICLES
sites underlying the conditioned reinforcing effects of cues paired with unconditioned reinforcers such as food and water, and the neuroanatomical sites underlying psychostimulant-induced potentiation of the behavioral responses to conditioned reinforcers23,30. Particularly relevant here are the findings that lesions of the BLA disrupt the ability of neutral cues paired with food or water to serve as conditioned reinforcers. However, when these lesions are performed after the conditioning phase, they had a very modest effect on psychostimulant-induced potentiation of the responses to conditioned reinforcers31. In contrast, CeA lesions have no effect on the ability of neutral cues to serve as conditioned reinforcers, but they significantly attenuate psychostimulant-induced potentiation of their conditioned responses23,32. On the basis of the different roles for the BLA and CeA in the behavioral effects of conditioned reinforcers and the present results, we speculate that the incubation of cocaine craving is due to time-dependent alterations in the motivational impact of cocaine cues. Alternatively, the incubation of cocaine craving may be due to cocaine-induced alterations in amygdala-mediated associative learning processes. In this regard, there is evidence that psychostimulant exposure induces changes in cellular mechanisms underlying learning and memory processes33, and several studies demonstrate a double dissociation in the involvement of the CeA and BLA in the consolidation of the memory of different appetitive conditioned responses23,34. ERK, synaptic plasticity and incubation of cocaine craving The ERK pathway is involved in learning and memory–associated synaptic plasticity processes such as long-term potentiation and depression in several brain areas, including the amygdala18. The ERK pathway in the amygdala also mediates learning and memory processes underlying the consolidation of conditioned fear memories19. Both acute and repeated exposure to cocaine activate the ERK pathway14,15 and induce long-term potentiation and depression35,36 in mesolimbic dopamine areas. The ERK pathway is activated by several different mechanisms, one of which involves activation of the TrkB receptor by BDNF17. In this regard, we have previously found that intra-VTA infusions of U0126 reverse the potentiation of cocaine seeking induced by BDNF infusions into this brain area12 when BDNF and U0126 are infused at the end of the 10-d cocaine self-administration training phase and extinction responding is measured after 3 and 10 d of withdrawal. However, intra-VTA infusions of U0126 alone do not prevent the subsequent emergence of the time-dependent increases in cocaine seeking after withdrawal, and acute activation of the ERK pathway in the VTA by BDNF has no effect on extinction responding during early withdrawal (day 3). Based on these and other findings, we have suggested that cocaine-induced alterations of mesolimbic BDNF facilitate an ongoing incubation process rather than mediating this process directly13. The ERK pathway is also activated by NMDA receptor–mediated calcium influx, which occurs when glutamate binds to this receptor17. In the present study, we have found that after 30 d of withdrawal, blockade of NMDA receptors in the CeA attenuated cocaine seeking and ERK phosphorylation. Furthermore, after 1 d of withdrawal, activation of NMDA receptors in the CeA increased cocaine seeking and ERK phosphorylation; these effects were reversed by U0126. These data indicate that glutamate is a likely upstream neurotransmitter that contributes to the activation of the CeA ERK pathway by exposure to cocaine cues. Previous studies also have shown that cocaine-induced ERK phosphorylation is blocked by NMDA receptor antagonists15. Our findings on the role of glutamate in relapserelated behavior extend findings from previous studies7,37 on the role of glutamate in other mesocorticolimbic areas (prefrontal cortex and nucleus accumbens) in relapse to cocaine seeking induced by acute re-exposure to the drug. Methodological considerations Several methodological issues should be considered in the interpretation of the present results. The decrease in lever responding after CeA infusions of U0126 or AP-5 may be due to motor disruption, but this is unlikely because these infusions have no effect on high rates of lever responding reinforced by palatable food. The anatomical specificity38 of the findings could be questioned because of possible diffusion of U0126 or AP-5 from the CeA to adjacent areas. However, this interpretation is also unlikely because CeA infusions of U0126 or AP-5 had no effect on ERK phosphorylation in the nearby BLA. Moreover, BLA infusions of U0126 had no effect on cocaine seeking after 30 d of withdrawal. The effect of U0126 on extinction responding may be due to statedependent learning processes39. In other words, U0126 infusions into the CeA may have attenuated extinction responding after 30 d of withdrawal because these infusions induce an interoceptive state different from the state experienced by the rats when exposed to the cocaine cues during self-administration. This unfamiliar interoceptive state may have interfered with the retrieval of the memory of the drug-associated cues during the extinction test. This state-dependent account is unlikely, however, because infusions of U0126 into the BLA, a site mediating the formation of learned associations between cocaine and cocaine cues40, had no effect on extinction responding. Also, results from a previous study indicate that state dependency cannot account for the effect of inhibition of BLA ERK activity on extinction of fear memories20. Furthermore, findings from several studies indicate that the predominant effect of ERK inhibitors is the disruption of memory consolidation rather than interference with the retrieval of existing memories18,41,42. Concluding remarks It has been suggested that drug relapse and craving are due to druginduced maladaptive alterations in synaptic processes underlying learning and memory9,43 and motivation8,44. The present data are in agreement with these ideas and provide new evidence to implicate the activation of the ERK pathway (a cellular pathway known to be involved in learning and memory45) in the incubation of cocaine craving. The present results are also potentially relevant for human cocaine addiction because craving-inducing cocaine cues activate amygdala neurons46. Furthermore, the present findings suggest a new function of activation of the ERK pathway in the amygdala in associative learning: enhancement of the behavioral impact of learned reward cues, a function different from the established role of this signaling pathway in the consolidation of memory18. Finally, a question for future research is whether the molecular mechanism of the incubation of cocaine craving identified here generalizes to other related phenomena such as incubation of palatable food craving13 or incubation of fear47. METHODS
Subjects. The subjects were 335 male Long-Evans rats (Charles River). They were maintained on a reverse 12-h light-dark cycle with food and water freely available in the home cage. We followed the ‘Principles of Laboratory Animal Care’ (NIH publication no. 86-23, 1996), and procedures were approved by the local Animal Care and Use Committee. Intracranial and intravenous surgery. Rats (350–400 g) were anesthetized with xylazine plus ketamine (10 and 100 mg kg–1, respectively) or sodium pentobarbital plus chloral hydrate (60 and 225 mg kg–1, respectively), and guide cannulae (Plastics One) were implanted bilaterally 1 mm above the

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

217

ARTICLES
BLA or CeA. The coordinates48 for the BLA and CeA were AP –2.5 mm, L ± 5.3 mm (2° angle), DV 7.8 mm; and AP –2.5 mm, L ± 4.5 mm (2° angle), DV –7.5 mm, respectively. After cannula implantation, catheters were inserted into the jugular vein as previously described4. Buprenorphine (0.1 mg kg–1) was given after surgery. Intracranial injections. Injections of U0126 (Calbiochem), NMDA (RBI) and AP-5 (Tocris) were made with Hamilton syringes (Hamilton) that were connected to 30-gauge injectors (Plastics One). NMDA and AP-5 were dissolved in sterile saline and U0126 was dissolved in 20% DMSO. The pH of the drug solutions was normalized with NaOH. A volume of 0.5 µl was infused into each side over 1 min and the injector was left in place for 1 min after the injections. The rats were tested within 5 min after NMDA infusions or 30 min after U0126 or AP-5 infusions. Drug doses are based on previous reports12,19,41,49. Tissue sample preparation. At the end of the 30 min extinction sessions, the rats were decapitated, and the brains were extracted and frozen in –50 °C isopentane. The brains of the rats from the ‘no test’ groups (see Figs. 2 and 3) of Experiments 1 and 2 were taken at the same time as those of the matched groups that were exposed to the cocaine cues in the extinction tests. After extraction, the brains were stored at –80 °C. Using a freezing cryostat (–20 °C), bilateral tissue punches (16 gauge) of the CeA and BLA (approximately –2.5 mm from bregma48) were taken from 1-mm-thick coronal sections. Tissue punches were sonicated for 10–15 s in 1% SDS. The protein concentrations of all samples were determined using the bicinchoninic acid assay (Pierce). Samples were further diluted in 1% SDS to equalize the protein concentrations. In Experiments 1 and 2, the protein concentrations of samples were determined in a single assay together with brain samples from 8–10 naive rats (350–400 g). Western blot assays. The samples were treated as previously described11. Loading buffer (16% glycerol, 20% β-mercaptoethanol and 0.05% bromophenol blue) was added to each sample (3:1, sample/loading buffer) before boiling for 3 min. Samples were cooled and subjected to SDS-polyacrylamide gel electrophoresis (10% acrylamide/0.27% N,N´-methylenebisacrylamide resolving gel) for 4 h at 150 V. For each electrophoresis, increasing amounts of protein pooled from all samples were electrophoresed to produce a standard curve. Proteins were transferred electrophoretically to Immobilon-P transfer membranes (Millipore) at 0.3 A for 2 h. Membranes were washed four times, for 15 min each, in blocking buffer: 2% polyvinyl pyrrolidone in PBST (phosphate-buffered saline plus 0.05% Tween-20) for the anti–phospho-ERK antibody, or 2% dry milk in PBST for the anti-ERK antibody. Membranes were then incubated overnight at 4 °C with anti–phospho-ERK antibody (1:1,000; New England Biolabs) or anti-ERK antibody (1:1,000; Upstate Biotechnology) in their blocking buffers plus 0.05% sodium azide. After four 15-min washes in the blocking buffer, blots were incubated for 2 h at 22 °C with horseradish peroxidase–conjugated secondary antibody (goat anti–rabbit IgG; PI-1000; Vector Labs) diluted 1:2,000 in blocking buffer. The blots were washed six times, for 10 min each, in PBST and developed for 60 s using the enhanced chemiluminescence (ECL) procedure (Amersham Pharmacia Biotech). Luminescence from the blots was detected using Amersham ECL Hyperfilm followed by digital scanning in transparency mode. Band intensities were quantified using Quantity One software (Version 4.0.3, Bio-Rad). Band intensities from each test sample were compared to the band intensities from the standard curves. The amount of the protein of interest in each sample was interpolated from the standard curve. Intravenous cocaine self-administration training (Experiments 1–5). The chambers, controlled by a Med Associates system, had two levers, but only one (an active, retractable lever) activated the infusion pump. Rats were chronically housed in the chambers and were trained to self-administer cocaine-HCl (0.75 mg kg–1 per infusion) during six 1-h daily sessions that were separated by 5 min over 10 d; the sessions started at the onset of the dark cycle. Presses on the active lever resulted in cocaine infusions that were accompanied by a 5-s tone-light cue. A fixed-ratio onereinforcement schedule was used, with a 40-s timeout period after each infusion. Each session began with the insertion of the active lever and the illumination of a houselight that remained on for the entire session. At the end of each session, the houselight was turned off and the active lever retracted. To facilitate the acquisition of cocaine self-administration, food was removed from the chambers during the 6-h sessions of the first 5 d. The number of cocaine infusions was limited to 20 per hour. At the end of the training phase, the groups in the different conditions of each experiment were matched for their cocaine intake during training. Seventy-seven rats were excluded because of loss of catheter patency, poor health or failure to acquire cocaine self-administration. At the end of the training phase of Experiment 1 and Experiments 3–5, the rats tested on day 1 of withdrawal were kept in the self-administration chambers until testing on the following day, whereas the rats tested on day 30 were brought back to the animal facility and were handled three times per week. In Experiment 2, all rats remained in the selfadministration chambers during the training and withdrawal phases. Intravenous saline self-administration (Experiment 1). The experimental conditions were identical to those described above for cocaine selfadministration, with the exception that lever presses led to infusions of saline (0.1 ml per infusion). Tests for cocaine seeking. These tests consisted of a 30 min extinction session. The testing conditions were the same as in training, except that presses on the previously active lever were not reinforced with cocaine. The tests started at the onset of the dark cycle. Each session began with the insertion of the active lever and the illumination of the houselight, which remained on for the entire session. Lever responding during the test sessions resulted in contingent presentations of the tone-light cue that was previously paired with cocaine infusions. Cocaine and food self-administration (Experiment 6). The training conditions for the cocaine-trained rats were the same as those described above. The food-trained rats learned to lever press for high-fat palatable food (45 mg, 25% fat precision pellets) (Bio-Serv) during six 1-h daily sessions that were separated by 5 min. The experimental conditions were similar to those of the cocainetrained rats, except that presses on the active lever resulted in the delivery of food pellets, and the maximum number of earned rewards was 30 per hour. The food-trained rats were subjected to food restriction (10–15 g of supplemental regular food pellets per day). After 7–8 d of training, the rats were infused with U0126 (100 ng per side) or its vehicle (20% DMSO) into the CeA 30 min before two test sessions that were separated by a regular training day. The order of the vehicle and U0126 infusions was counterbalanced. At the end of the experiment, cannula placements were verified histologically. Statistical analyses. The data from the extinction sessions were analyzed separately for total (non-reinforced) active and inactive lever responses. Because the experimental manipulations had similar effects on ERK1 and ERK2 phosphorylation, these values were added and the analyses were performed on the combined values of phosphorylated ERK50. Lever presses, phosphorylated ERK and total ERK (ERK1 plus ERK2) were analyzed with ANOVAs. Posthoc analyses of significant effects were performed by a Bonferroni/Dunn test (two-tailed).
ACKNOWLEDGMENTS We thank S. Gray, D. Nagarkar, D. Chuang and C. Scheidweiler for technical assistance, and J. Stewart, D.E. Epstein, B.J. Everitt and R.A. Wise for helpful comments and critical discussions of the present data. Research was supported by the National Institute on Drug Abuse Intramural Research Program. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 17 November; accepted 13 December 2004 Published online at http://www.nature.com/natureneuroscience/
1. O’Brien, C.P. A range of research-based pharmacotherapies for addiction. Science 278, 66–70 (1997). 2. Gawin, F.H. & Kleber, H.D. Abstinence symptomatology and psychiatric diagnosis in cocaine abusers. Clinical observations. Arch. Gen. Psychiatry 43, 107–113 (1986). 3. Grimm, J.W., Hope, B.T., Wise, R.A. & Shaham, Y. Incubation of cocaine craving after withdrawal. Nature 412, 141–142 (2001). 4. Lu, L., Grimm, J.W., Dempsey, J. & Shaham, Y. Cocaine seeking over extended withdrawal periods in rats: different time courses of responding induced by cocaine cues versus cocaine priming over the first 6 months. Psychopharmacology (Berl.) 176, 101–108 (2004). 5. Neisewander, J.L. et al. Fos protein expression and cocaine-seeking behavior in rats after exposure to a cocaine self-administration environment. J. Neurosci. 20, 798–805 (2000).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

218

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
6. Nestler, E.J. Molecular basis of long-term plasticity underlying addiction. Nat. Rev. Neurosci. 2, 119–128 (2001). 7. Kalivas, P.W. Glutamate systems in cocaine addiction. Curr. Opin. Pharmacol. 4, 23–29 (2004). 8. Stewart, J. Pathways to relapse: factors controlling the reinitiation of drug seeking after abstinence. Nebr. Symp. Motiv. 50, 197–234 (2004). 9. Everitt, B.J. & Wolf, M.E. Psychomotor stimulant addiction: a neural systems perspective. J. Neurosci. 22, 3312–3320 (2002). 10. Grimm, J.W. et al. Time-dependent increases in brain-derived neurotrophic factor protein levels within the mesolimbic dopamine system after withdrawal from cocaine: implications for incubation of cocaine craving. J. Neurosci. 23, 742–747 (2003). 11. Lu, L., Grimm, J.W., Shaham, Y. & Hope, B.T. Molecular neuroadaptations in the accumbens and ventral tegmental area during the first 90 days of forced abstinence from cocaine self-administration in rats. J. Neurochem. 85, 1604–1613 (2003). 12. Lu, L., Dempsey, J., Liu, S.Y., Bossert, J.M. & Shaham, Y. A single infusion of brainderived neurotrophic factor into the ventral tegmental area induces long-lasting potentiation of cocaine seeking after withdrawal. J. Neurosci. 24, 1604–1611 (2004). 13. Lu, L., Grimm, J.W., Hope, B.T. & Shaham, Y. Incubation of cocaine craving after withdrawal: a review of preclinical data. Neuropharmacology 47 (Suppl. 1), 214–226 (2004). 14. Berhow, M.T., Hiroi, N. & Nestler, E.J. Regulation of ERK (extracellular signal regulated kinase), part of the neurotrophin signal transduction cascade, in the rat mesolimbic dopamine system by chronic exposure to morphine or cocaine. J. Neurosci. 16, 4707–4715 (1996). 15. Valjent, E. et al. Involvement of the extracellular signal-regulated kinase cascade for cocaine-rewarding properties. J. Neurosci. 20, 8701–8709 (2000). 16. Licata, S.C. & Pierce, R.C. The roles of calcium/calmodulin-dependent and Ras/mitogen-activated protein kinases in the development of psychostimulant-induced behavioral sensitization. J. Neurochem. 85, 14–22 (2003). 17. Thomas, G.M. & Huganir, R.L. MAPK cascade signalling and synaptic plasticity. Nat. Rev. Neurosci. 5, 173–183 (2004). 18. Adams, J.P. & Sweatt, J.D. Molecular psychology: roles for the ERK MAP kinase cascade in memory. Annu. Rev. Pharmacol. Toxicol. 42, 135–163 (2002). 19. Schafe, G.E. et al. Activation of ERK/MAP kinase in the amygdala is required for memory consolidation of pavlovian fear conditioning. J. Neurosci. 20, 8177–8187 (2000). 20. Lu, K.T., Walker, D.L. & Davis, M. Mitogen-activated protein kinase cascade in the basolateral nucleus of amygdala is involved in extinction of fear-potentiated startle. J. Neurosci. 21, RC162 (2001). 21. LeDoux, J.E. Emotion circuits in the brain. Annu. Rev. Neurosci. 23, 155–184 (2000). 22. Gallagher, M. & Chiba, A.A. The amygdala and emotion. Curr. Opin. Neurobiol. 6, 221–227 (1996). 23. Everitt, B.J. et al. Associative processes in addiction and reward. The role of amygdalaventral striatal subsystems. Ann. NY Acad. Sci. 877, 412–438 (1999). 24. See, R.E. Neural substrates of conditioned-cued relapse to drug-seeking behavior. Pharmacol. Biochem. Behav. 71, 517–529 (2002). 25. Shaham, Y., Shalev, U., Lu, L., De Wit, H. & Stewart, J. The reinstatement model of drug relapse: history, methodology and major findings. Psychopharmacology (Berl.) 168, 3–20 (2003). 26. Shalev, U., Grimm, J.W. & Shaham, Y. Neurobiology of relapse to heroin and cocaine seeking: a review. Pharmacol. Rev. 54, 1–42 (2002). 27. Balleine, B.W. & Dickinson, A. Goal-directed instrumental action: contingency and incentive learning and their cortical substrates. Neuropharmacology 37, 407–419 (1998). 28. Davies, S.P., Reddy, H., Caivano, M. & Cohen, P. Specificity and mechanism of action of some commonly used protein kinase inhibitors. Biochem. J. 351, 95–105 (2000). 29. Fiore, R.S., Murphy, T.H., Sanghera, J.S., Pelech, S.L. & Baraban, J.M. Activation of p42 mitogen-activated protein kinase by glutamate receptor stimulation in rat primary cortical cultures. J. Neurochem. 61, 1626–1633 (1993). 30. Jentsch, J.D. & Taylor, J.R. Impulsivity resulting from frontostriatal dysfunction in drug abuse: implications for the control of behavior by reward-related stimuli. Psychopharamacology 146, 373–390 (1999). 31. Burns, L.H., Robbins, T.W. & Everitt, B.J. Differential effects of excitotoxic lesions of the basolateral amygdala, ventral subiculum and medial prefrontal cortex on responding with conditioned reinforcement and locomotor activity potentiated by intra-accumbens infusions of D-amphetamine. Behav. Brain Res. 55, 167–183 (1993). 32. Robledo, P., Robbins, T.W. & Everitt, B.J. Effects of excitotoxic lesions of the central amygdaloid nucleus on the potentiation of reward-related stimuli by intra-accumbens amphetamine. Behav. Neurosci. 110, 981–990 (1996). 33. Wolf, M.E., Sun, X., Mangiavacchi, S. & Chao, S.Z. Psychomotor stimulants and neuronal plasticity. Neuropharmacology 47 (Suppl. 1), 61–79 (2004). 34. Holland, P.C. & Gallagher, M. Double dissociation of the effects of lesions of basolateral and central amygdala on conditioned stimulus-potentiated feeding and Pavlovianinstrumental transfer. Eur. J. Neurosci. 17, 1680–1694 (2003). 35. Ungless, M.A., Whistler, J.L., Malenka, R.C. & Bonci, A. Single cocaine exposure in vivo induces long-term potentiation in dopamine neurons. Nature 411, 583–587 (2001). 36. Thomas, M.J., Beurrier, C., Bonci, A. & Malenka, R.C. Long-term depression in the nucleus accumbens: a neural correlate of behavioral sensitization to cocaine. Nat. Neurosci. 4, 1217–1223 (2001). 37. Baker, D.A. et al. Neuroadaptations in cystine-glutamate exchange underlie cocaine relapse. Nat. Neurosci. 6, 743–749 (2003). 38. Wise, R.A. & Hoffman, D.C. Localization of drug reward mechanisms by intracranial injections. Synapse 10, 247–263 (1992). 39. Overton, D.A. Experimental methods for the study of state-dependent learning. Fed. Proc. 33, 1800–1813 (1974). 40. Whitelaw, R.B., Markou, A., Robbins, T.W. & Everitt, B.J. Excitotoxic lesions of the basolateral amygdala impair the acquisition of cocaine-seeking behaviour under a second-order schedule of reinforcement. Psychopharmacology (Berl.) 127, 213–224 (1996). 41. Davis, M., Walker, D.L. & Myers, K.M. Role of the amygdala in fear extinction measured with potentiated startle. Ann. NY Acad. Sci. 985, 218–232 (2003). 42. Berman, D.E. & Dudai, Y. Memory extinction, learning anew, and learning the new: dissociations in the molecular machinery of learning in cortex. Science 291, 2417–2419 (2001). 43. Nestler, E.J. Common molecular and cellular substrates of addiction and memory. Neurobiol. Learn. Mem. 78, 637–647 (2002). 44. Robinson, T.E. & Berridge, K.C. Addiction. Annu. Rev. Psychol. 54, 25–53 (2003). 45. Sweatt, J.D. The neuronal MAP kinase cascade: a biochemical signal integration system subserving synaptic plasticity and memory. J. Neurochem. 76, 1–10 (2001). 46. Grant, S. et al. Activation of memory circuits during cue-elicited cocaine craving. Proc. Natl. Acad. Sci. USA 93, 12040–12045 (1996). 47. Eysenck, H.J. A theory of the incubation of anxiety-fear responses. Behav. Res. Ther. 6, 309–321 (1968). 48. Paxinos, G. & Watson, C. The Rat Brain in Stereotaxic Coordinates (Academic, San Diego, 1998). 49. Hayes, R.J., Vorel, S.R., Spector, J., Liu, X. & Gardner, E.L. Electrical and chemical stimulation of the basolateral complex of the amygdala reinstates cocaine-seeking behavior in the rat. Psychopharmacology (Berl.) 168, 75–83 (2003). 50. Kelleher, R.J. III, Govindarajan, A., Jung, H.Y., Kang, H. & Tonegawa, S. Translational control by MAPK signaling in long-term synaptic plasticity and memory. Cell 116, 467–479 (2004).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

219

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Dynamics of motion signaling by neurons in macaque area MT
Matthew A Smith1,2, Najib J Majaj1 & J Anthony Movshon1
Most neurons in macaque area MT are selective for the direction of stimulus motion. By comparing direction selectivity for gratings and plaids, we classified MT neurons as pattern direction selective (PDS) or component direction selective (CDS). We compared the time course of responses in CDS and PDS neurons in opiate-anesthetized macaques, using a rapid pseudorandom sequence of gratings and plaids that moved in different directions. On average, responses began 6 ms earlier in CDS neurons than in PDS neurons. More importantly, the pattern-selective responses of PDS neurons did not reach their fully selective state until 50–75 ms after the responses of CDS neurons had stabilized. The population motion response of MT is therefore initially dominated by component motion signals, and does not completely represent pattern motion until substantially later. The circuits that compute pattern motion take more time to finish their work than those signaling component motion.

Many studies of visual cortex concern the mean activity of neurons measured over periods of seconds, but recently the dynamics of neural response in visual cortex have drawn increased attention. Visual cortical neurons can change their response characteristics over quite brief periods, affecting such fundamental properties as orientation selectivity1–3 and various forms of contextual modulation4–7. Substantial time-dependent changes in tuning have also been reported for responses to complex stimuli in V1 and in inferotemporal cortex8–10 and to complex motion stimuli in macaque area MT/V511,12. The interpretation of these dynamic variations in response pattern is somewhat controversial: some take the view that temporal variations themselves encode important aspects of visual stimuli13, whereas others remain agnostic as to the meaning of temporal variations and use them instead to probe the neuronal circuits that give rise to selectivity2,14. The circuitry and response properties of MT neurons are particularly appealing for a study of this kind because these neurons have rather precise temporal response properties15 and perform a few well-defined visual computations. Area MT contains a high proportion of directionally selective neurons16–19 and plays an important role in visual motion perception20,21. Motion processing in primate visual cortex occurs in at least two stages. The first stage, most likely located in primary visual cortex (V1), encodes orientation, spatial frequency and motion energy in a local region of space22. To decode more complex motion signals, a second stage takes inputs from the first stage and combines them to compute the true direction and speed of a moving stimulus. Plaid stimuli, made by adding two sinusoidal gratings with different orientations, have proved useful in probing the circuitry of these stages of motion processing. When presented with a plaid stimulus, V1 neurons signal only the direction of motion of the component gratings, and not

the true direction of the pattern17,23 (Fig. 1a,b). However, although some cells in area MT behave similarly to those in V1, others respond to the true direction of motion of the plaid stimulus17,24 (Fig. 1a,b). The former are termed CDS and the latter PDS. The V1 neurons that project to MT are CDS23, consistent with the idea that circuits within MT compute pattern motion and thus represent the neural substrate for the second stage of motion processing17. There is evidence from both psychophysics and physiology that the neural representation of two-dimensional motion evolves over tens to hundreds of milliseconds12,25–28. Here we report a difference in the time course of response of CDS and PDS neurons in area MT whose dynamics match recent behavioral and psychophysical data in humans27,28, suggesting that perceptual dynamics may be closely linked to the action of MT circuits. The additional time required for pattern direction selectivity to become manifest is considerable, and suggests that pattern motion is computed by circuits that are more complex than a simple feed-forward network. RESULTS We recorded from 143 neurons in area MT of 11 macaque monkeys. For each neuron, we first measured responses to drifting sinusoidal gratings presented for several seconds with a blank period between stimuli. We determined the optimal direction, spatial and temporal frequency, and area for grating stimuli, and then tested the dynamics of each neuron’s response over time using a novel ‘streaming’ stimulus. Figure 1c shows this stimulus schematically. For each cell, we presented a continuous sequence of drifting gratings and plaids. We interleaved gratings of 50% contrast drifting in 12 evenly spaced directions, 12 plaid stimuli created by adding together two gratings separated by 120°, and four mean gray

1Center

for Neural Science, 4 Washington Place, New York University, New York, New York 10003, USA. 2Present address: Center for the Neural Basis of Cognition, Carnegie Mellon University, 4400 Fifth Avenue, 115 Mellon Institute, Pittsburgh, Pennsylvania 15232, USA. Correspondence should be addressed to J.A.M. ([email protected]). Published online 16 January 2005; doi:10.1038/nn1382

220

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES a b
Figure 1 MT stimulus and response. (a) Most cells in MT are tuned for the direction of a drifting grating. A sample response is shown here in the polar plot. (b) When a plaid stimulus is presented, we might predict two possible responses. The pattern prediction (dotted line) is that the neuron integrates the motion signals and responds to the plaid as it does to the individual grating. The component prediction (solid line) is that the neuron responds to individual grating components of the plaid as if they were presented alone. (c) We used a stimulus which rapidly changed between gratings and plaids drifting in different directions. The stimulus remained on the screen, drifting, for 320 ms, after which a new stimulus was chosen randomly. This figure shows a sample sequence with a spike raster below it. Each cell has a response latency (the time it takes for a change in the stimulus to be reflected in a response change), which is indicated by ∆t. We collected multiple repeats and then parsed out the response for each stimulus to arrive at a detailed response histogram. (d) For three sample cells, these lines represent the normalized variance in the tuning curves as a function of ∆t. The peak variance values were at ∆t = 56 ms, 70 ms and 84 ms. (e) This panel shows a frequency distribution of precision values obtained by a bootstrap method. The mean of this distribution was 2.1 ms, and 92% of the data fell below 5 ms.

c
© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience
...

Time

320 ms ...

Spikes

∆t

Analysis epochs

d
Normalized variance 1.0

e
Proportion of cells 0.3 0.2 0.1 0.6 40 80 120 ∆t (ms) 0 4 8 Precision (ms) 12 0

0.8

i–l, respectively) were PDS cells, as can be seen by noting the similarity of their tuning for gratings and plaids (gray curves in Fig. 2e,f,i,j). The latencies determined as detailed above for these three cells were 73 ms, 63 ms and 50 ms, indicated by the arrows beneath the response histograms. These latencies can be seen to correspond closely with the times of response onset estimated by eye from the histograms. The response details highlighted by the red and blue stripes are discussed below. Classification of pattern and component cells The difference between PDS and CDS cells is captured by comparing tuning for gratings with that for plaids. We took each cell’s tuning curve for gratings and generated two predictions—one for a pattern response (Fig. 1b, dotted curve) and one for a component response (Fig. 1b, solid curve). We computed pattern and component correlations (Rp, Rc) of the actual response (measured over the optimal 320-ms interval determined as described above) with these predictions, using the standard technique17,23. These correlation measures were also converted into Z-scores using Fisher’s r-to-Z transformation (see Methods). The r-to-Z transformation is a variance-stabilizing transformation that makes it possible to compute quantities such as the difference of correlation values. When using the raw correlations, the meaning of a difference between two numbers depends on their values. For example, the difference between r-values of 0.91 and 0.92 is in no meaningful sense the same as that between 0.51 and 0.52. With Z-transformation, the differences between values are in units of their standard deviation. Figure 3 shows the distributions of both the partial correlations (Rp, Rc, Fig. 3a) and their Z-transforms (Zp, Zc, Fig. 3b). The class boundaries17 separate the cells into CDS cells (blue), PDS cells (red), and intermediate or ‘unclassed’ cells. Approximately 25% of cells were classified as PDS (36/143) and 41% as CDS (58/143), whereas 34% were unclassed (49/143). These proportions are similar to those previously observed for this measurement in awake29 (G. Stoner and T. Albright, personal communication; D. Bradley, personal communication) and anesthetized17,24,30 monkeys using the same stimuli. The data in the two panels are the same, but note that the curved class boundaries in Figure 3a are transformed into straight lines in Figure 3b. Response latency for pattern and component cells Distributions of response latency are shown in Figure 4a–c for CDS, PDS and unclassed cells. PDS cells have a significantly longer latency than CDS cells by an average of about 6 ms (ANOVA, P = 0.025). The latency for PDS cells was also longer than for unclassed cells, but this

blank stimuli. Every 320 ms we changed the stimulus, holding spatial and temporal frequency and size constant at the optimal values. The stimuli were presented in random order together in a block. This block was typically repeated five times (each time in a new random order), followed by several seconds of blank screen. This entire procedure was usually repeated 5–40 times to give 25–200 repetitions of each stimulus. Determining response latency When using a stimulus that runs continuously without pause, parsing the spike train is not as simple as with discrete, separately presented stimuli. Below the schematic stimulus in Figure 1c is a spike trace showing a cell’s response. Because of response latency, the onset of firing induced by a stimulus lagged behind the onset of that stimulus. To relate the spikes to the stimuli that evoked them, we shifted the spike times by ∆t. To optimize ∆t for each cell, we computed the mean responses to all stimuli and took the variance about these means as a measure of how much of the cell’s response variation could be driven by our stimuli (see Methods). Figure 1d shows variance as a function of ∆t for three sample cells. For these and all others, the curves had a single clear maximum. We took this maximum—the value of ∆t that produced the largest variance and therefore the most tuning-curve modulation—as the latency. We used a bootstrap method to compute the precision of each latency measure (see Methods) (distribution of these values is shown in Fig. 1e). Precision was 2.1 ms on average, and better than 5 ms for 92% of cells. Figure 2 shows the responses of three MT cells to gratings and plaids presented with the streaming method. For each cell, average response histograms for gratings and plaids moving in different directions are shown on the right, and polar plots of response amplitude versus direction are shown on the left. The first cell (Fig. 2a–d) was a CDS cell, as can be seen by comparing the single-peaked tuning curve for gratings (Fig. 2a, gray line) with the bi-lobed tuning curve for plaids (Fig. 2b, gray line), whose peaks are separated by the 120° that separated the component gratings in the stimulus. The second and third cells (Fig. 2e–h and

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

221

ARTICLES
Grating Plaid Grating Plaid
Figure 2 Three types of responses to plaids and gratings. (a,b) These two polar plots show the orientation tuning of a sample cell to gratings (left) and plaids (right). This cell had a response latency of 72.5 ms and was classified as CDS using its response from the entire 320-ms stimulus interval. The red curve indicates the cell’s tuning in a 15-ms window near the beginning of the cell’s response, whereas the blue curve indicates the cell’s tuning in a later 15-ms window after the response has stabilized. The cell’s response over a 320-ms window starting at the optimal offset is shown with the gray curve. The small black circle at center indicates the cell’s baseline response to a gray screen. (c,d) These two panels show histograms of the same cell’s responses to gratings (left) and plaids (right). The vertical bars indicate the early (red bar) and late (blue bar) windows from which the polar plots in a and b were taken. (e–h) These polar plots and histograms are constructed from the responses of another cell with a response latency of 62.7 ms that was classified as PDS. Like the component cell in a–d, this neuron did not show any change in its directionality to plaid stimuli over time. (i–l) Here the same plots are shown for a PDS neuron with a response latency of 50.1 ms. In this case, the cell’s plaid response was markedly different over time. In the earlier window, the response matched the component prediction. In the later window, the pattern prediction was a better match for the cell’s response.

a

b

c
0˚ 90˚ 180˚

d

30 60 ips

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

30 60 ips 85–100 ms 200–215 ms Best 320 ms 270˚
0

200 ips 100 200 300
0

100

200

300

e

f

g
0˚ 90˚ 180˚ 30 75 ips 270˚
0

h

30

75 ips

75–90 ms 200–215 ms Best 320 ms

100 ips 100 200 300
0

100

200

300

i

j

k
0˚

l

but were not individually significant (P = 0.27, 0.08, and 0.23), presumably owing to the smaller number of cells in each group. 180˚ The peak firing rates of PDS cells (38.8 ± 30 60 ips 26.4 impulses per second (ips)) and CDS 50–65 ms 270˚ cells (46.2 ± 24.0 ips) across all stimuli 175–190 ms were not significantly different (ANOVA, Best 320 ms P = 0.17). In CDS cells, the peak firing rate to 0 100 200 300 0 100 200 300 grating and plaid stimuli was nearly the same Time (ms) (41.6 ips versus 41.4 ips). In PDS cells, the peak response to plaids (38.1 ± 26.8 ips) was much difference was not statistically significant (ANOVA, P = 0.09). Across higher than to gratings (23.9 ± 13.7 ips). This difference of 12.7 ips was all cells, the mean response latency was 63.5 ± 13.9 ms, similar to that statistically significant (t-test, P < 0.0001). The baseline rates of PDS cells (4.1 ± 2.8 ips) and CDS cells (4.8 ± 4.7 ips) were very similar (ANOVA, found in other studies31–34. Although PDS and CDS neurons show a latency difference as a P = 0.40), and there was no correlation between the spontaneous rate group, the division of cells into those groups is based on somewhat and response latency (r = –0.05, P = 0.53). To assess whether the response latencies for gratings and plaids were arbitrary significance criteria. To analyze the relationship between pattern response and latency in our entire population of neurons, we the same for each cell, we used the same automated method to determine computed Zp–Zc, which we term ‘patternness’, and its relationship with response latency for each neuron’s responses to gratings and plaids separesponse latency. Across the entire population of 143 neurons, there was rately (Fig. 4d–f). For all three classes of cells, grating and plaid latencies a significant positive correlation between ‘patternness’ and response were highly correlated (r = 0.90, 0.73 and 0.88, respectively; P < 0.0001) latency (Pearson’s r = 0.25, P = 0.003). This statistic indicates that across but not identical. We computed the latency difference (plaid latency the full spectrum of PDS and CDS behavior, cells that tended to show minus grating latency) for each class (Fig. 4g–i), which revealed a delay of about 4 ms for plaid response relative to grating response for PDS more PDS behavior also tended to have longer response latencies. The vigor and latency of cortical responses are often negatively cor- cells (t-test, P = 0.006) and unclassed cells (t-test, P = 0.0003). CDS related1,33,35. This may be because cells with very low firing rates have cells showed a delay of 1 ms that was not statistically significant (t-test, higher thresholds and therefore longer integration times. In the popu- P = 0.43). Across all cells, responses to gratings were on average 2.9 ± 8.5 lation as a whole we found a weak but significant negative correlation ms earlier than responses to plaids. Cells tend to respond with shorter between peak firing rate and response latency (Pearson’s r = –0.21, latencies to high-contrast stimuli36–38. It is notable that the response P = 0.013), consistent with data from other studies. Pattern, compo- latency to plaids (100% contrast) was slightly longer than to gratings nent and unclassed cells grouped separately showed correlations that (50% contrast) even though the plaids had higher contrast, for which were in the same direction (r = –0.19, –0.23, and –0.18, respectively) we might expect shorter latency responses. One simple interpretation is
30 60 ips 90˚ 100 ips

222

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES a
1.0 Pattern
Pattern correlation (Rp)

b
Z-transformed pattern correlation (Zp) 8 Pattern 6 4 2 0 Component n = 143

0.5

0.0

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

–0.5
Component n = 143 –1.0 –1.0 –0.5 0.0 0.5 1.0 Component correlation (Rc)

Figure 3 Pattern and component direction selectivity. (a) Scatter plot of Rc versus Rp used to divide MT cells into component (CDS, blue), pattern (PDS, red) and unclassed groups. The class boundaries are previously defined17,23 and represent differences between Rc versus Rp values that have a 0.1 probability of arising by chance. (b) Scatter plot of Zc versus Zp showing data from the same calls as in a after applying Fisher’s r-to-Z transformation to express the correlation values in units normalized by their standard deviation (Z-scores). The solid lines bound the same three plot regions as in a.

–2

–2 0 2 4 6 8 Z-transformed component correlation (Zc)

that latency for plaids is determined not by the pattern’s total contrast, but by the contrast of each component grating, which is 50%. Time course of response for individual neurons The difference we found in mean response latency between PDS and CDS cells prompted us to examine the time course of response for individual cells, with the particular view of determining the selectivity of early and late parts of the response. The blue and red bars under the average response histograms in Figures 2c and d, 2g and h, and 2k and l indicate for each cell an early and a late response interval, respectively, of 15 ms duration, and the similarly colored curves in the polar tuning diagrams in Figures 2a and b, 2e and f, and 2i and j show the magnitude of response in these intervals. For the first and second examples, tuning curves taken from both windows show similar responses, indicating that these cells showed a stable direction preference for gratings and plaids over time. The third cell, however, exhibited a different response pattern (Fig. 2i–l). Here, the polar plots of direction tuning show the same a pattern for gratings in early and late response 0.2 windows (Fig. 2i). However, the response to
0.1
Figure 4 Response latency for pattern, component and unclassed cells. (a–c) These three PSTHs show frequency distributions of response latency in the three classes of MT neurons. The mean of each distribution is indicated with the black arrow. (a) Pattern cells had the longest latency on average, 68.1 ± 16.5 ms (s.d.), about 6 ms longer than for both other classes. (b) Component cells showed the shortest latency, 61.8 ± 10.5 ms. (c) Unclassed cells had an average latency that was nearly the same as for component cells, 62.3 ± 14.8 ms. (d–f) We determined the response latency separately for grating and plaid responses. These three scatter plots show these data for pattern, component and unclassed cells. (g–i) We computed the latency difference (plaid response latency minus grating response latency) for each class of cells. On average (black arrows), all cells responded earlier to gratings than to plaids, although that difference was very small: 3.9 ± 7.9 ms for pattern cells, 1.0 ± 9.2 ms for component cells and 4.3 ± 7.8 ms for unclassed cells.

plaids shows a markedly different pattern in the early and late response windows (Fig. 2j). Near the beginning of the response (red line), the cell behaved as a typical CDS cell would, with two peaks separated by 120° on its direction tuning curve. After the response had stabilized (blue line), the directionality for plaids was similar to that for gratings, as is characteristic of PDS cells. This change is easily seen in the average response histograms for plaids (Fig. 2l), where it is evident that the cell responded to a wide range of directions in the early (blue) window but showed narrower tuning in the late window. From these three examples, it is clear that some cells in area MT show dynamic changes in their direction tuning over time, whereas others maintain stable preferences throughout their responses. To add to this qualitative view of the data, we quantitatively assessed the frequency of the two behaviors described above for pattern cells. We aligned all of the data to the response onset of each neuron and analyzed the data in 20-ms windows starting at this latency. Of the 36 neurons classified as PDS (see Methods), 9 (25%) showed significant CDS tuning (defined as a Z-score difference larger than 1.28) in the first 20-ms epoch (like the example in Fig 2l), whereas only 9 (25%) showed significant PDS

150 PDS (36) 100 50

d
0.4 0.2
0

g
Plaid leads Grating leads

0.0 0 Proportion of cells

20 40 60 80 100 120 Grating latency (ms)

0 150 100 50 0 150

50

100 150
Proportion of cells

0.0

–20

0

20

b
0.2 CDS (58) 0.1

e
0.4 0.2 0.0

h

0.0 0

20 40 60 80 100 120

0

50

100 150

–20

0

20

c
0.2 Unclassified (49) 0.1

f
0.4 0.2 0 50 100 150 Plaid latency (ms) 0.0

i

100 50

0.0 0

20 40 60 80 100 120 Latency (ms)

0

–20 0 20 Latency difference (ms)

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

223

ARTICLES
8 6 4 2 0 –2 8 6

a

30–50 ms

8 6 4 2 0 –2 8 6

b

30–70 ms

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Z-transformed pattern correlation (Zp)

Figure 5 Scatter plot of Z-scores over time in sliding or cumulative windows. Here we present scatter plots of Z-transformed pattern and component correlation. The blue dots represent component cells, the red dots represent pattern cells, and the black dots represent those cells which are not classified. This assignment was done based on the response of the neurons over a full 320-ms stimulus window (see Fig. 3b). The time window ranges are indicated above each plot. (a–d) These four scatter plots show responses starting at 30–50 ms (a), with the cumulative window expanding by 20 ms as you move to each successive plot. The final window includes responses from 30–110 ms (d).

c

30–90 ms

d

30–110 ms

time after stimulus onset; the labels indicate the end of each time window plotted. CDS cells (blue line) develop their characteristic response 4 4 tuning much earlier (within 60–65 ms) than do PDS cells (red line, 125–130 ms). In addition, 2 2 whereas CDS cells show an increase in Zc at very early times with little change in Zp, in PDS 0 0 cells both correlations increase for some time before Zp dominates. This population behavior presumably reflects a general tendency, evident –2 –2 in some individual cells (Fig. 2i–l), for CDS–2 0 2 4 6 8 –2 0 2 4 6 8 type responses or broadly tuned responses to Z-transformed component correlation (Zc) predominate early in the response of many PDS and unclassed cells. Another way to visualize this result is through tuning (like the example in Fig 2h). The remaining 18 (50%) initially the difference in the Z-scores for pattern and component correlation: the showed broad tuning that was not classifiable as CDS or PDS, and became deviation from the diagonal in the space of Figure 6a. We plotted Zc–Zp PDS at variable times after response onset. For comparison, we analyzed (which one might term ‘componentness’) for CDS cells, and Zp–Zc (or the 58 neurons which were classified as CDS (see Methods). In the first ‘patternness’) for PDS cells, against time (Fig. 6b), using the same cumula20 ms of response, 44 cells (76%) showed significant CDS tuning (like the tive bins as in Figure 6a. It is again obvious that the response time courses example in Fig. 2d), whereas only 1 (2%) showed significant PDS tuning. of pattern and component cells are quite different. The horizontal black It is clear that although the large majority of CDS cells were component line in this figure (and the parallel black lines in Fig. 6a) represents the selective from the very beginning of their response, most PDS cells were significance value we used in previous analysis (Fig. 3). This value is a not pattern selective until their responses were well under way. measure of significance for an individual cell, and the plotted lines represent the Z-correlation of an average neuron in each class. Taking this value Population dynamics of pattern and component direction as a threshold, there is a difference of approximately 60–70 ms between selectivity the time at which the average CDS cell and the average PDS cell become To quantify these dynamics in our population, we generated scatter plots significantly selective. of Z-transformed pattern and component correlation for different time windows. We first classified the cells based on their responses over a full DISCUSSION 320-ms time window (Fig. 3) and then plotted their responses in small The signaling of true pattern direction by neurons in area MT is one time windows. The results of this type of analysis for four windows are of the hallmarks of direction selectivity in this area17. Our results show shown in Figure 5a–d; cells are color-coded to indicate their selectivity in that this signal evolves during the first 100–150 ms after the presena full 320-ms window. We used a cumulative response window starting tation of a complex stimulus such as a plaid. CDS cells, thought to 30–50 ms after the stimulus transition (Fig. 5a) and extended the represent an earlier stage of motion processing than PDS cells, give window by 20 ms in each successive plot. Many CDS cells (blue circles) responses whose selectivity is stable and consistent from the time showed early responses (that is, by the second time window (Fig. 5b)) they are first activated. But PDS cells often respond with different and that had significant component selectivity. PDS cells (red circles) tended broader selectivity when first activated, sometimes even resembling to develop their selectivity later (Fig. 5c–d), and many remained unclas- CDS cells. After some tens of milliseconds, their responses evolve to be sifiable or unselective even at the end of this long interval. more PDS-like. Only one CDS cell behaved analogously, first respondTo capture the aggregate behavior of the three groups of cells, we ing to the motion of the pattern and later to the motion of the compotook their mean Zp and Zc values for each of a series of cumulative time nents. Population analyses confirmed that PDS cells tend to develop windows, starting with 30–35 ms and extending to 30–155 ms. A plot their characteristic response more slowly than CDS cells. Although a of Zp against Zc for PDS, CDS and unclassed cells is shown in Figure 6a. 6-ms latency difference between the classes contributes to this effect, The data are connected to form a trajectory in the space of pattern and the tuning dynamics show a much more substantial difference than is component selectivity that describes the growth of their selectivity with explained by response latency alone. In other words, the computation

224

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES a
Z-transformed pattern correlation (Zp ) 5 4 3 2 1 0 Pattern Component Unclassified 140 120 120 100 100 90 90 80 80 70 60 50 60 50 50 60 70 80 70
Figure 6 Line plots showing the evolution of Z-correlation values over time. (a) In this panel we show the evolution of pattern and component responses. The red line represents PDS cells, the blue line represents CDS cells, and the black line represents unclassed cells. We plotted each data point as the average value of Zc and Zp at that time for each class of cells. The data are analyzed cumulatively, so each point represents the Z-correlation for the average response from time zero up that time. The numbers along each line indicate the time of the closest data point. For example, the blue number ‘60’ indicates that the adjacent data point represents the average value of Zc and Zp for component cells up to and including the time window beginning at 60 ms (and ending at 65 ms). (b) The evolution of the Z-transformed pattern and component correlation over time. Here we plot the deviation from the diagonal in the space of Figure 6a, which is the difference in the Z-scores for pattern and component correlation (Zc–Zp for CDS cells, plotted with a blue line, and Zp–Zc for PDS cells, plotted with a red line). The data points are spaced by 5 ms, and each point represents the correlation measured from a tuning curve calculated from all the data up to that time (error bars show ± 1 s.e.m.). The first point is the correlation from 30–35 ms, and the last point from 30–155 ms. The horizontal black line indicates the significance value used throughout this chapter for Z-correlation significance for one cell. CDS cells crossed this significance line approximately 60–70 ms earlier than PDS cells.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

120 100 90

Difference in Z-transformed correlation

b

–1 –1 0 1 2 3 4 5 Z-transformed component correlation (Zc ) 3 CDS cells, Zc–Zp

2

1

0

PDS cells, Zp–Zc

–1

40

60

80 100 120 Time (ms)

140

of pattern direction proceeds more slowly after response onset than the computation of component direction. These results suggest that cortical processing of two-dimensional motion signals by MT neurons is a dynamic process that is continuously shaped in the first hundred milliseconds of the neurons’ responses. Plaid patterns with component gratings of unequal speed have a twodimensional ‘pattern’ velocity that differs in direction from the mean of the component velocities17. Human observers viewing these ‘type 2’ plaids perceive mostly component motion in brief presentations and pattern motion only after a delay27. The delayed selectivity of PDS neurons offers a direct explanation for this perceptual effect: the shift from perceiving component velocity to perceiving pattern velocity corresponds to a shift in population activity from an early phase dominated by the early selective response of CDS cells to a later one in which the delayed selective response of PDS cells becomes important. Although we did not use ‘type 2’ stimuli in our physiological experiments, one need only assume that the difference in response dynamics in our experiments is characteristic of CDS and PDS responses to other kinds of stimuli. Our data also parallel the behavior of short-latency ocular following responses to grating and plaid stimuli in human subjects28. With conventional plaids (made from two orthogonal moving gratings) or single gratings, motion onset elicits a very fast response closely aligned from the outset to the true motion direction. With ‘unikinetic plaids’ (‘type 2’ plaids made from one stationary and one moving grating separated by 45° in orientation), the ocular following response initially follows the grating motion and only later follows the plaid motion. This result demonstrates behaviorally that the encoding of two-dimensional motion in plaid patterns takes additional processing time compared with

gratings, and agrees qualitatively with our physiological data showing that the processing of plaid patterns evolves over time. Two recent studies have shown changes in direction preference during the responses of cells in motion-sensitive areas of cortex. One study11 examined responses in the posteromedial lateral suprasylvian area (PMLS) of the cat evoked by fields of randomly placed iso-oriented line segments in which the angle between orientation and direction of movement was varied. The early responses tended to be more selective to orthogonal (line) motion than the later responses, which reflected the true direction. These results are broadly consistent with ours in that they show an initial predominance of component-dominated response. In area MT of the alert macaque, a stimulus of terminated bars evoked similar behavior12. When the bars moved obliquely to their orientation, cells initially responded to the motion of the bar segments, and only later signaled the true direction of motion. Our data show that CDS cells have shorter latencies and reach their characteristic response faster than PDS cells. CDS signals would therefore dominate the early average population response, and PDS cells would contribute later. However, the terminated line stimuli in those experiments, unlike plaid stimuli, contain motion signals that can be disambiguated by CDS neurons alone, as shown in recent imaging studies of primary visual cortex39. Moreover, the results of these experiments in cat PMLS and macaque MT are very similar, even though PMLS contains few, if any, PDS cells11,12,40. The similarity between our results and those obtained with terminated line stimuli may therefore be more illusory than real, and whether they all reflect the processes we have uncovered remains open to discussion. It appears that the difference between the dynamics of CDS and PDS cells reflects a fundamental difference in the neural circuitry of these two cell types. What kind of circuits might account for the very substantial timing differences we observe? One model that has been proposed to account for the behavior of PDS neurons in MT is an extension of the normalization model in V1 (ref. 41). In this framework, pattern selectivity arises from a recurrent circuit that implements a divisive gain control. In principle, such a recurrent network might take some time to stabilize, and although it is stabilizing, one would expect to see the CDS-like behavior of input neurons expressed in responses, much as we observed in some PDS cells (as in Fig 2i–l). Such a gain control circuit is thought to act in V1, where it can be shown to act very quickly (within a few milliseconds42). If MT’s gain control is responsible for the slow (40–60 ms) evolution of PDS responses, it

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

225

ARTICLES
must be based on very different circuitry and involve some very slow form of recurrent inhibition not evident in V1. A second idea is that pattern motion signals might derive from feedback from higher cortical areas, which presumably represent more advanced processing. These signals might enter MT after a delay and influence responses. The earliest responses would then reflect feed-forward processing, whereas the slower dynamics of PDS responses could reflect the influence of the feedback signal. Arguing against this idea is the likelihood that feedback signals from higher areas are attenuated under anesthesia and the evidence from studies of lower visual areas which suggests that feedback signals modulate but do not directly activate cortical neurons43. Another proposal is that pattern and component motion might be computed in two separate pathways, as proposed on psychophysical grounds44. If the ‘pattern’ pathway were slower, dynamics in MT might reflect the additional time needed for signals from this pathway to propagate. However, there is no obvious candidate for a separate cortical ‘pattern’ pathway, because studies of pattern and component selectivity in areas V2 and V3 do not suggest an important separate contribution of those areas to PDS behavior in MT45,46. Finally, within MT, cells might vary in their dynamics based on the position they occupy within the local cortical circuitry (for instance, input layers versus output layers). We determined the laminar location for only a fraction of our cells. Examining the laminar distribution of PDS and CDS neurons and laminar variations in latency does not reveal strong trends in the predicted direction for this subset of our data, but further experiments may prove more illuminating. Regardless of the underlying neural mechanism, our results show a clear difference in response dynamics between two functionally distinct signals evident in neurons in a single cortical area; we are aware of no comparable difference in response dynamics in any previous work. The dynamics we observe match well with psychophysical data and are likely to be connected directly to the perception of complex motion stimuli in the natural environment. Because there are principled reasons to believe that pattern selectivity is computed directly from component selective inputs17,23, our findings define a uniquely favorable system with which to probe the timing and architecture of cortical computation using natural stimuli.
We recorded with quartz-platinum-tungsten microelectrodes (Thomas Recording) advanced with a mechanical microdrive through a small durotomy made within a craniotomy of approximately 10 mm diameter. The craniotomy was typically centered 4 mm posterior to the lunate sulcus and 15 mm lateral to the midline. The electrode was advanced 20° anterior and down in the parasagittal plane. MT cells were recorded at eccentricities ranging from 2° to 33°, but the great majority of cells were between 3° and 12°. Apart from the larger scale of the more peripheral receptive fields, we noticed no differences related to eccentricity. Signals from the microelectrode were amplified and bandpass filtered, and we isolated single units with a dual-window time-amplitude discriminator (Bak). The time of each action potential was recorded with a resolution of 0.25 ms by a CED-1401 Plus laboratory interface (Cambridge Electronic Design). We made small electrolytic lesions at the end of each electrode track by passing DC current (2 µA for 5 s, tip negative) through the recording electrode. Once the experiment was finished, the animals were killed with an overdose of Nembutal and perfused through the heart with 0.1 M PBS followed by 4% paraformaldehyde in 0.1 M PBS. Sections of the superior temporal sulcus were taken every 40 µm and stained for Nissl substance with cresyl violet or myelin using Gallyas' method48. We were able to confirm most recording locations directly, but in some cases we relied on proximity to histologically confirmed recording sites and a high proportion of directional cells with relatively small receptive fields to determine that the recordings were made in MT49. Visual stimulus generation. We displayed all visual stimuli at a resolution of 1,024 × 731 pixels and a video frame rate of 100 Hz on an Eizo T550 monitor. The video monitor was placed between 80 and 180 cm from the animal’s eye, where it subtended between 10° and 22° of visual angle. We used look-up tables to correct for nonlinearities in the relation between input voltage and phosphor luminance in the monitors. We generated grating stimuli for basic characterization with a Cambridge Research Systems VSG 2/2 board running on an Intel x86-based host computer. The dynamic grating and plaid stimulus was generated with a Silicon Graphics workstation. The mean luminance of the display was 33 cd m–2. We presented gratings and plaids to the dominant eye in a circular aperture surrounded by a gray field of the average luminance. The starting phase was set to the same value for all stimuli. The stimuli were equal in duration (typically 1–2 s) and were separated by presentation of a uniform mean gray background for about 1.5 s. For each isolated neuron, we determined the optimal direction, spatial and temporal frequency, position and size of a 100% contrast drifting sine wave grating. After the initial characterization of each neuron with gratings, we presented a dynamic random plaid and grating stimulus. Gratings were presented at 50% contrast, and plaids were constructed by adding two such gratings separated by 120° in orientation. We presented each of these two types of stimuli drifting in 12 different directions, in addition to four periods of blank (mean gray) screen. Trials lasted 45 s and consisted of all 28 of the stimuli presented in random order for 320 ms each, with five presentations of each stimulus. These trials were typically repeated 20 times, to achieve 100 presentations of each of the 28 stimuli. The actual number of presentations was chosen for each cell based on the variability of its response (the range was 15–300). Determining response latency. We used a novel method to determine response latency in MT neurons with our ‘streaming’ stimulus (Fig. 1c). For each neuron, we computed the mean firing rate for each of the 28 stimuli (12 gratings, 12 plaids, 4 blank screens) in a sliding window of duration 320 ms (the stimulus duration) beginning ∆t ms after stimulus onset. When ∆t was set at any value other than the response latency, the cell’s response was not well aligned with the window used to compute the mean firing rate, and spikes due to any stimulus would spill into an adjacent analysis window, diluting differences in response and bringing the firing rate for a given interval closer to the average firing rate to all stimuli. As ∆t approached the true latency, the mean firing rate increased for the preferred stimuli over the window and decreased for the non-preferred stimuli. When the value of ∆t was at the true response latency, the tuning curve would have the maximum modulation, which we took as the variance of the responses to each stimulus about the overall mean. Preliminary exploration showed that the appropriate values of ∆t typically fell between 40 and 100 ms; for each cell, we used a binary search method to find the optimal value in the range of 20–200 ms, within which the relationship between ∆t and tuning curve variance was invariably of inverted-U shape with a single maximum.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

METHODS
Electrophysiology. We recorded extracellularly from single units in area MT of seven Cynomolgus macaques (Macaca fascicularis), two bonnet macaques (M. radiata) and two pig-tailed macaques (M. nemestrina), ranging in weight from 4.0 to 7.9 kg. The data from all three species were indistinguishable in all respects. The techniques used in our laboratory for recording from the visual cortex of anesthetized, paralyzed monkeys have been reported in detail elsewhere47. Briefly, animals were premedicated with atropine sulfate (0.05 mg kg–1) and diazepam (Valium, 1.5 mg kg–1) 30 min before induction of anesthesia with ketamine HCl (10.0 mg kg–1). Anesthesia was maintained throughout the experiment by a continuous infusion of sufentanil citrate (typically 4 µg kg–1, adjusted for each animal). Under this anesthetic regime, MT responses to grating and texture stimuli are similar in magnitude, reliability and time course to those in awake animals (L.P. O’Keefe and J.A.M., unpublished observations).To minimize eye movements, the animal was paralyzed with a continuous intravenous infusion of vecuronium bromide (Norcuron, 0.1 mg kg–1 hr–1). Vital signs (EEG, ECG, end-tidal PCO2, temperature and lung pressure) were monitored continuously. The pupils were dilated with topical atropine and the corneas protected with gas-permeable hard contact lenses. We used supplementary lenses to bring the retinal image into focus by direct ophthalmoscopy. We later adjusted the refraction further to optimize the response of recorded units. Experiments typically lasted 4–5 d. All procedures complied with guidelines approved by the New York University Animal Welfare Committee.

226

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
We tested the precision of this latency estimate with a bootstrap method. For each cell, we randomly chose one-half of the presentations of each stimulus and computed the latency only using that subset of the data. We repeated this 100 times for each cell, randomly choosing a different subset of trials each time, and computed the standard deviation of the latencies. The resulting values were based on a bootstrap using only half of the trials for each cell, so we divided this number by √2 to estimate the standard error of the latency; the distribution of these values for all cells is shown in Figure 1e. To test the generality of this method, we compared it with another procedure for computing latency. For 37 neurons, we collected grating and plaid tuning curves using a more conventional stimulus period of 1–2 s with a blank period of approximately 2 s between stimuli. We calculated latency for this data as the time required to reach 50% of peak response in a mass histogram of responses to all stimuli. Latency values computed in this way correlated very well with those from the streaming method (Pearson’s r = 0.53, P < 0.001). The streaming method gave slightly longer latencies on average (by 4.8 ms), although this difference was not statistically significant (P = 0.123). We conclude that our automated method is an efficient and unbiased way to determine response latency. It is important to note that this method captures ‘average’ latency, but not variations in latency across stimuli. For instance, it might not work well were contrast the experimental variable, because there is a systematic variation in the time of response onset with contrast36–38. As response offset typically has a shorter latency than response onset35, the method has the ‘leeway’ it needs to work when the latency variations across stimuli are neither large nor systematic. Analysis of MT pattern and component data. We computed the partial correlation for the pattern and component predictions using standard methods17. For each cell, we used the response latency (determined using the method described above) to determine the window over which we calculated these correlations. That is, if a cell’s latency was determined to be 50 ms, we calculated the correlations based on tuning curves made from responses 50–370 ms (the response latency plus the stimulus duration) after the transitions in our dynamic MT stimulus. The partial correlations for the pattern and component predictions are of the form (rp– rc rpc) (1– rc2)(1– rpc2) exceed the value of Zc (or zero, if Zc is negative) by this amount. Similarly, Zc had to exceed Zp by that same amount (1.28) for a cell to be judged as CDS. If a cell met neither of these conditions, it remained unclassed.
ACKNOWLEDGMENTS This work was supported by a research grant from the NIH (EY02017), and by an HHMI Investigatorship to J.A.M. M.A.S. was supported in part by a National Eye Institute Institutional Training Grant (T32-7136). We thank A. Kohn, N. Rust and S. Schultz for assistance with some of the data collection, R. Young for technical assistance, and M. Hou and N. Doron for help with histology. We are grateful to W. Bair and A. Kohn for helpful advice and discussion. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 9 July; accepted 10 November 2004 Published online at http://www.nature.com/natureneuroscience/

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Rp =

and (rc– rp rpc) (1– rp2)(1– rpc2)

Rc =

where rc is the correlation of the data with the component prediction, rp is the correlation of the data with the pattern prediction and rpc is the correlation of the two predictions. Because the sampling distribution of Pearson’s r is not normal, we used Fisher’s r-to-Z transformation for its variance-stabilizing effect. We took each value of Rp or Rc and converted it to a Z-score using the following equation (shown for Rp): (1 + Rp) (1 – Rc) 1 df where df is the degrees of freedom, equal to the number of values in the tuning curve minus 3 (there were 12 directions in our tuning curves). The numerator of the equation is the Fisher r-to-Z transformation. Each value of Zc or Zp was then tested for significance. We used a criterion of 1.28, equivalent to P = 0.90, for this purpose. For a cell to be judged as a PDS cell, the value of Zp had to

0.5 ln Zp =

1. Celebrini, S., Thorpe, S., Trotter, Y. & Imbert, M. Dynamics of orientation coding in area V1 of the awake primate. Vis. Neurosci. 10, 811–825 (1993). 2. Ringach, D.L., Hawken, M.J. & Shapley, R. The dynamics of orientation tuning in the macaque monkey striate cortex. Nature 387, 281–284 (1997). 3. Ringach, D.L., Hawken, M.J. & Shapley, R. Dynamics of orientation tuning in macaque v1: the role of global and tuned suppression. J. Neurophysiol. 90, 342–352 (2003). 4. Bair, W., Cavanaugh, J.R. & Movshon, J.A. Time course and time-distance relationships for surround suppression in macaque V1 neurons. J. Neurosci. 23, 7690–7701 (2003). 5. Knierim, J.J. & Van Essen, D.C. Neuronal responses to static texture patterns in area V1 of the alert macaque monkey. J. Neurophysiol. 67, 961–980 (1992). 6. Lee, T.S., Yang, C.F., Romero, R.D. & Mumford, D. Neural activity in early visual cortex reflects behavioral experience and higher-order perceptual saliency. Nat. Neurosci. 5, 589–597 (2002). 7. Zipser, K., Lamme, V.A. & Schiller, P.H. Contextual modulation in primary visual cortex. J. Neurosci. 16, 7376–7389 (1996). 8. Richmond, B.J., Optican, L.M., Podell, M. & Spitzer, H. Temporal encoding of twodimensional patterns by single units in primate inferior temporal cortex. I. response characteristics. J. Neurophysiol. 57, 132–146 (1987). 9. Richmond, B.J., Optican, L.M. & Spitzer, H. Temporal encoding of two-dimensional patterns by single units in primate primary visual cortex. I. Stimulus-response relations. J. Neurophysiol. 64, 351–369 (1990). 10. Sugase, Y., Yamane, S., Ueno, S. & Kawano, K. Global and fine information coded by single neurons in the temporal visual cortex. Nature 400, 869–873 (1999). 11. Li, B., Chen, Y., Li, B.W., Wang, L.H. & Diao, Y.C. Pattern and component motion selectivity in cortical area PMLS of the cat. Eur. J. Neurosci. 14, 690–700 (2001). 12. Pack, C.C. & Born, R.T. Two-dimensional substructure of MT receptive fields. Nature 409, 1040–1042 (2001). 13. McClurkin, J.W., Optican, L.M., Richmond, B.J. & Gawne, T.J. Concurrent processing and complexity of temporally encoded neuronal messages in visual perception. Science 253, 675–677 (1991). 14. McLaughlin, D., Shapley, R., Shelley, M. & Wielaard, D.J. A neuronal network model of macaque primary visual cortex (V1): orientation selectivity and dynamics in the input layer 4cα. Proc. Natl. Acad. Sci. USA 97, 8087–8092 (2000). 15. Bair, W., Koch, C., Newsome, W. & Britten, K. Power spectrum analysis of bursting cells in area MT in the behaving monkey. J. Neurosci. 14, 2870–2892 (1994). 16. Albright, T.D. Direction and orientation selectivity of neurons in visual area MT of the macaque. J. Neurophysiol. 52, 1106–1130 (1984). 17. Movshon, J.A., Adelson, E.H., Gizzi, M.S. & Newsome, W.T. The analysis of visual moving patterns. in Pattern Recognition Mechanisms (eds. Chagas, C. Gattass, R. & Gross, C.) 117–151 (Springer, New York, 1985). 18. Van Essen, D.C., Maunsell, J.H.R. & Bixby, J.L. The middle temporal visual area in the macaque: Myeloarchitecture, connections, functional properties and topographic organization. J. Comp. Neurol. 199, 293–326 (1981). 19. Zeki, S.M. Functional organization of a visual area in the posterior bank of the superior temporal sulcus of the rhesus monkey. J. Physiol. (Lond.) 236, 549–573 (1974). 20. Britten, K.H., Shadlen, M.N., Newsome, W.T. & Movshon, J.A. The analysis of visual motion: a comparison of neuronal and psychophysical performance. J. Neurosci. 12, 4745–4765 (1992). 21. Newsome, W.T. & Paré, E.B. A selective impairment of motion perception following lesions of the middle temporal area MT. J. Neurosci. 8, 2201–2211 (1988). 22. DeValois, R.L., Albrecht, D.G. & Thorell, L.G. Spatial frequency selectivity of cells in macaque visual cortex. Vision Res. 22, 545–559 (1982). 23. Movshon, J.A. & Newsome, W.T. Visual response properties of striate cortical neurons projecting to area MT in macaque monkeys. J. Neurosci. 16, 7733–7741 (1996). 24. Rodman, H.R. & Albright, T.D. Single-unit analysis of pattern-motion selective properties in the middle temporal visual area (MT). Exp. Brain Res. 75, 53–64 (1989). 25. Kooi, F.L., DeValois, K.K., Switkes, E. & Grosof, D.H. Higher-order factors influencing the perception of sliding and coherence of a plaid. Perception 21, 583–598 (1992). 26. Lorenceau, J., Shiffrar, M., Wells, N. & Castet, E. Difference motion sensitive units are involved in recovering the direction of moving lines. Vision Res. 33, 1207–1217 (1993).

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

227

ARTICLES
27. Yo, C. & Wilson, H.R. Perceived direction of moving two-dimensional patterns depends on duration, contrast and eccentricity. Vision Res. 32, 135–147 (1992). 28. Masson, G.S. & Castet, E. Parallel motion processing for the initiation of short-latency ocular following in humans. J. Neurosci. 22, 5149–5163 (2002). 29. Stoner, G.R. & Albright, T.D. Neural correlates of perceptual motion coherence. Nature 358, 412–414 (1992). 30. Priebe, N.J., Cassanello, C.R. & Lisberger, S.G. The neural representation of speed in macaque area MT/V5. J. Neurosci. 23, 5650–5661 (2003). 31. Maunsell, J.H.R. Physiological evidence for two visual subsystems. in Matters of Intelligence (ed. L.M. Vaina) 59–87 (Reidel, Dordrecht, The Netherlands, 1987). 32. Raiguel, S., Lagae, L., Gulyás, B. & Orban, G.A. Response latencies of visual cells in macaque areas V1, V2, and V5. Brain Res. 493, 155–159 (1989). 33. Raiguel, S.E., Xiao, D-K., Marcar, V.L. & Orban, G.A. Response latency of macaque area MT/V5 neurons and its relationship to stimulus parameters. J. Neurophysiol. 82, 1944–1956 (1999). 34. Schmolesky, M.T. et al. Signal timing across the macaque visual system. J. Neurophysiol. 79, 3272–3278 (1998). 35. Bair, W., Cavanaugh, J.R., Smith, M.A. & Movshon, J.A. The timing of response onset and offset in macaque visual neurons. J. Neurosci. 22, 3189–3205 (2002). 36. Albrecht, D.G. Visual cortex neurons in monkey and cat: effect of contrast on the spatial and temporal phase transfer functions. Vis. Neurosci. 12, 1191–1210 (1995). 37. Carandini, M., Heeger, D.J. & Movshon, J.A. Linearity and normalization in simple cells of the macaque primary visual cortex. J. Neurosci. 17, 8621–8644 (1997). 38. Gawne, T.J., Kjaer, T.W. & Richmond, B.J. Latency: another potential code for feature binding in striate cortex. J. Neurophysiol. 76, 1356–1360 (1996). 39. Basole, A., White, L.E. & Fitzpatrick, D. Mapping multiple features in the population response of visual cortex. Nature 423, 986–990 (2003). 40. Gizzi, M.S., Katz, E., Schumer, R.A. & Movshon, J.A. Selectivity for orientation and direction of motion of single neurons in cat striate and extrastriate visual cortex. J. Neurophysiol. 63, 1529–1543 (1990). 41. Simoncelli, E.P. & Heeger, D.J. A model of neuronal responses in visual area MT. Vision Res. 38, 743–761 (1998). 42. Albrecht, D.G., Geisler, W.S., Frazor, R.A. & Crane, A.M. Visual cortex neurons of monkeys and cats: temporal dynamics of the contrast response function. J. Neurophysiol. 88, 888–913 (2002). 43. Bullier, J., Hupé, J.M., James, A.C. & Girard, P. The role of feedback connections in shaping the responses of visual cortical neurons. Prog. Brain Res. 134, 193–204 (2001). 44. Wilson, H.R., Ferrera, V.P. & Yo, C. A psychophysically motivated model for two-dimensional motion perception. Vis. Neurosci. 9, 79–97 (1992). 45. Gegenfurtner, K.R., Kiper, D.C. & Levitt, J.B. Functional properties of neurons in macaque area V3. J. Neurophysiol. 77, 1906–1923 (1997). 46. Levitt, J.B., Kiper, D.C. & Movshon, J.A. Receptive fields and functional architecture of macaque V2. J. Neurophysiol. 71, 2517–2542 (1994). 47. Cavanaugh, J.R., Bair, W. & Movshon, J.A. Nature and interaction of signals from the receptive field center and surround in macaque V1 neurons. J. Neurophysiol. 88, 2530–2546 (2002). 48. Gallyas, F. Silver staining of myelin by means of physical development. Neurol. Res. 1, 203–209 (1979). 49. Desimone, R. & Ungerleider, L.G. Multiple visual areas in the caudal superior temporal sulcus of the macaque. J. Comp. Neurol. 248, 164–189 (1986).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

228

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Using visual direction in three-dimensional motion perception
Julie M Harris1,2 & Vit F Drga1,2
The eyes receive slightly different views of the world, and the differences between their images (binocular disparity) are used to see depth. Several authors have suggested how the brain could exploit this information for three-dimensional (3D) motion perception, but here we consider a simpler strategy. Visual direction is the angle between the direction of an object and the direction that an observer faces. Here we describe human behavioral experiments in which observers use visual direction, rather than binocular information, to estimate an object’s 3D motion even though this causes them to make systematic errors. This suggests that recent models of binocular 3D motion perception may not reflect the strategies that human observers actually use.

The human visual system is exquisitely sensitive to binocular disparity1,2. In principle, binocular information could also be used to obtain an estimate of 3D object motion3–5. A top-down schematic view of an observer and a small object moving along a 3D trajectory are shown in Figure 1a. To obtain the angle of the motion trajectory, θ, which equals zero when the object moves directly toward the observer, the visual system could estimate the x and z components of the motion or their retinal correlates, such as the ratio of left and right eye motions4,5, or the ratio of lateral motion to change in binocular disparity3,6. Research in this area has previously focused on establishing the source of binocular information used to detect and discriminate 3D motion direction. Here we suggest that such a sophisticated mechanism may not be used. Instead, observers may use a simpler strategy not tested in previous investigations. Visual direction can be defined as the angle between where an object of interest is and where the observer is facing (α in Fig. 1a). We considered whether observers make an estimate of the visual direction of an object at the start and end of its motion and use that estimate to determine how the object has moved. The advantage of using visual direction is its simplicity: all that is required is a measurement of direction. The use of visual direction is also a general strategy. For example, we know that it is used for other tasks including selfmotion direction during locomotion7–10. The disadvantage of using visual direction is that α does not specify a unique motion direction; additional visual information is needed, such as object distance and size. Thus, one visual direction can correspond to a variety of object trajectories (as illustrated in Fig. 1b). Under this scheme, systematic errors in perception would be expected. In the experiments presented here we show that observers do seem to use α rather than θ even though this produces the expected systematic errors in the indicated direction of motion. We suggest that the visual system may not explicitly represent the direction of 3D motion, as suggested by recent binocular models.

RESULTS We devised an experiment to distinguish between the use of visual direction, α, and trajectory angle, θ. Observers viewed a small point-like object moving along a 3D trajectory (indicated by large black arrow at the top of Fig. 1c) and moved a wooden pointer to reproduce the perceived 3D motion direction of the target. Fixation distance must be used to scale disparity to obtain metric depth2. In this experiment there were two sets of disparities to be scaled: those associated with the visual stimulus and those associated with the pointer. If observers used binocular information to estimate the angle θ and correctly scaled the disparities of both the visual stimulus and the pointer, they should have set the pointer angle, θ′, to equal the perceived angle (small black arrow, Fig. 1c). If the scaling of either were done incorrectly, errors in perceived angle would be expected. If α were used, it might also be necessary for the brain to scale it. It has been suggested that visual mechanisms do not have access to the raw angle but that rather the angle is scaled, assuming a particular viewing distance11; here again, if scaling were incorrect, systematic errors would be expected. However, if observers used visual direction, α, they should have set θ′ so that the end of the pointer lay along the perceived visual direction of the target at the endpoint of the motion (small gray arrow, Fig. 1c). With judicious choice of trajectory angle and the x and z components of motion (see Methods), it is possible to set up an experiment so that observers should respond differently if they use θ or α, as we have done here. θ and α can be defined in terms of an x and z component of motion or in terms of distance moved during motion (Fig. 1a). Here we define x as the lateral distance moved, z as the distance moved in depth and d as the distance to the start point of the motion (the fixation point). It follows that: tan θ = x/z tan α = x/(d – z) (1) (2)

1School of Psychology, University of St. Andrews, St. Mary’s College, South Street, St. Andrews, Fife, Scotland, KY16 9JP, UK. 2Psychology Brain and Behaviour, University of Newcastle upon Tyne, Newcastle NE2 4HH, UK. Correspondence should be addressed to J.M.H. ([email protected]).

Published online 23 January 2005; doi:10.1038/nn1389

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

229

ARTICLES
Figure 1 Schematics of stimulus geometry. (a) Observer and typical target trajectory (solid arrow) at an angle θ to straight ahead are illustrated. Visual direction of the moving target at the end of its motion is given by angle α. (b) Many different trajectories correspond to a single visual direction. (c) Location of the pointer (gray circle). If observers used θ to set the pointer, they would respond as shown by the black arrow. The gray arrow shows a typical response if observers used α.

Experiment 1: use of α or θ to perceive motion direction For the experiments described here we used a range of relatively small θs, centered on zero. Typically z was large compared with x, providing a good opportunity for stimulation of binocular mechanisms sensitive to the z component of the motion. We used two conditions in Experiment 1. In Experiment 1A, z was constant and θ varied by changing x. From equations (1) and (2) we see that x varies linearly with the tangent of either angle: for constant z, as θ increases, θ′ should increase whether θ or α is used. In Experiment 1B, x was constant and θ varied by changing z. A larger increase in z is required to increase θ by a constant amount for small θ than for large θ. Thus, if disparity-sensitive brain mechanisms were used to obtain the distance moved in depth, and hence θ, they would be most useful for small values of θ. However, from equation (2) it can be seen that if z << d, α changes very little with z. If α were used to perceive motion direction, θ′ should be almost constant with increasing θ. If θ were used, we might expect observers to set the same θ′ as in Experiment 1A, because the same range of θ values was used in both conditions. Table 1 shows x and z distances moved for each condition. For Experiment 1A, plotting θ′ as a function of θ (Fig. 2a) showed that different observers behaved very differently, but all observers set wider angles than would be expected using veridical values of either θ or α. However, the behavior of each observer was internally consistent: for all observers, θ′ was an increasing function of θ. For Experiment 1B, we presented the same range of θ′ values, yet the θ′ values were very different from those in Experiment 1A (Fig. 2b). For each observer, θ′ values for Experiment 1A were significantly different from those for Experiment 1B (based on χ2, 3 degrees of freedom (d.f.), P < 0.001 for each observer, Bonferroni-corrected P < 0.01). These data were not consistent with a strategy based on estimating θ from x and z.
Table 1 Parameters for Experiment 1
Angle θ (°) x at trajectory endpoint (m) z at trajectory endpoint (m)

The data show that there are some θs that are perceived as the same when they are in fact physically different. To determine whether observers would be able to discriminate between these trajectories, we conducted a control discrimination experiment, using a new population of observers who, for completeness, also performed the original experiment (see Supplementary Note 1). The experiment was set up so that the trajectory angle should have been easily discriminable for all angles tested if observers used θ for the discrimination. Differences in α were large for two trajectories and small for two others. Observers’ performance was good when the differences in α were large but poor when the differences in θ were large, consistent with the use of α and not θ (see Supplementary Note 2). The observers’ use of α in Experiment 1 is not veridical; θ′ was set to be much larger than expected if the correct value of α was used. Previous studies have found that observers probably do not have direct access to estimates of visual angle; instead, angular judgments are thought to be scaled by apparent distance or disparity11. Furthermore, we expect apparent distance to be underestimated in sparse visual displays12 and, hence, visual angles to be overestimated. Assuming that systematic errors in estimates of α or θ are consistent (in other words, that scaling is constant between conditions), it is possible to predict how observers should respond in Experiment 1B, given their performance in Experiment 1A. Data from Experiment 1A were averaged across observers and used to predict observer performance in Experiment 1B. The relationship between θ and θ′ was calculated from the 1A data. This was used in two ways: first, to predict observer performance in Experiment 1B, assuming observers estimate θ; and second, to calculate the relationship between α and estimated visual angle, α′. This second relationship can be used to predict performance in Experiment 1B, assuming that observers estimate α, not θ. When we compared data from Experiment 1B to predictions from Experiment 1A based on the use of either θ or α (Fig. 2c), the Experiment 1B data were significantly different from the θ line (χ2 = 26.23, d.f. = 3, P < 0.001) but not from the α line (χ2 = 0.29, d.f. = 3, P = 0.96; see Supplementary Note 3 for individual observer data and predictions).
Table 2 Conditions and predictions for Experiment 4
a Experiment 4 conditions Condition A B C D Information used Extent of visual direction Speed of visual direction Duration
aPredictions

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Distance (m) 0.012 0.024 0.012 0.006

Speed (m s–1) 0.012 0.024 0.024 0.012

Duration (s) 1 1 0.5 0.5

Experiment 1A 5 10 15 20 Experiment 1B 5 10 15 20 0.012 0.012 0.012 0.012 0.137 0.068 0.045 0.033 0.012 0.024 0.037 0.050 0.137 0.137 0.137 0.137

b Experiment 4 predictionsa Conditions for which performance is predicted to be the same A and C A and D, B and C A and B, C and D
for expected observer responses if using extent, speed or duration.

230

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 2 Results and predictions for Experiment 1. (a) θ′ as a function of θ for 10 naive observers in Experiment 1A. Gray lines show data from individual observers. Average performance is shown by the solid line. (b) Results presented in the same way for Experiment 1B. (c) Average data from Experiment 1B (filled squares; error bars, ± 1 s.e.m.). The black line is the prediction calculated from Experiment 1A, if observers used θ. The gray line is the prediction line calculated from the results of Experiment 1A, if observers used α. These lines are explicit predictions: we are not showing the results for a model and there are no free parameters varied to fit the data.

Experiment 2: perceived direction for larger motion extents We further tested the idea that observers used α to set perceived direction by performing Experiment 2. The extent of motion was increased from that in Experiment 1, but observers were presented with the same range of θ. The x component of motion was constant as θ was varied, but it was either twice (Experiment 2A) or three times (Experiment 2B) that used in Experiment 1B. If α were used, we would expect observers to respond by indicating a larger constant angle θ′ (see equation (2)). We plotted the average results compared with the same six observers' data from Experiment 1B (Fig. 3a). For any one physical angle, for larger extents of motion, the estimated trajectory angle was larger. As before, it is possible to predict how observers should behave in Experiment 2 based on their behavior in Experiment 1A. In Figure 3b and 3c we re-plotted the data from Figure 3a along with the predictions for the use of θ and α. The data were significantly different from the line showing the θ prediction (Experiment 2A: χ2 = 10.55, d.f. = 3, P = 0.014; Experiment 2B: χ2 = 18.57, d.f. = 3, P = 0.003) but not from the α line (Experiment 2A: χ2 = 0.91, d.f. = 3, P = 0.82; Experiment 2B: χ2 = 1.25, d.f. = 3, P = 0.74). Observers’ performance was consistent with use of the perceived visual direction of the endpoint of the motion trajectory. Experiment 3: depth discrimination of endpoint z locations The data are also consistent with observers behaving as if they attempted to estimate θ but used a constant estimate of the z component of the motion. In Experiment 3, we tested whether observers could

discriminate between z values corresponding to the set of endpoints of the motion in Experiment 1B. In a two-alternative forced choice experiment, observers discriminated between adjacent z locations in the set (see Methods). Observers can distinguish between the depths of adjacent z endpoints (Fig. 4). On average, performance was better than 72% correct across all depth pairs tested (Fig. 4a), but some individual observers performed more poorly than others (Fig. 4b). Performance was clearly not perfect, but it should be noted that the discrimination task performed in Experiment 3 was between adjacent endpoint depths. The difference in z between any two trajectories in Experiment 1 was therefore at least (and often two to three times) the threshold depth. If observers were attempting to use z but were finding it difficult to discriminate between the different z endpoints in Experiment 1B, we would expect the response variability to be greater than for Experiment 1A, and we would expect those observers with worse performance in Experiment 3 to be more variable in Experiment 1B. An assessment of standard errors for each observer for both conditions (Fig. 5) indicated that although some observers were much more variable in Experiment 1B (ALG, LSD), they were not the ones with the worst performance in Experiment 3 (CJW, JAF, YAL). The inability to perceive z well is therefore not a consistent explanation for poor observer performance in Experiment 1B. Despite depth differences in the endpoint of motion being discriminable, this information seems not to be used for the trajectory direction task. Experiment 4: is α used or its rate of change? We have so far discussed the data in terms of observers using the visual direction of the endpoint of the motion. To determine whether they

Figure 3 Comparison of Experiments 1 and 2. (a) Results for Experiment 1B (open squares) and Experiment 2 (2A: black squares, 2B: open circles), for six naive observers (error bars, ± 1 s.e.m.). The graph shows θ′ as a function of θ. (b) Repeat of data for doubled motion extent (Experiment 2A), with prediction lines for the competing response hypotheses. The black line is the prediction line calculated from Experiment 1A if observers use θ. The gray line is the prediction line calculated from the results of Experiment 1A if observers use α. (c) Repeat of data for tripled motion extent (Experiment 2B) along with prediction lines as described above.

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

231

ARTICLES
Figure 4 Experiment 3 results. (a) Proportion correct for each standard/test pair. Pairing is coded as ‘a’ (standard, 0.045 m; test, 0.033 m); ‘b’ (standard, 0.045 m; test, 0.068 m); ‘c’ (standard, 0.068 m; test, 0.045 m); ‘d’ (standard, 0.068 m; test, 0.137 m). (b) Proportion correct for each observer (indicated by initials) across all conditions. Error bars, s.e.m.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

were truly using the endpoint or whether they could respond to the rate of change of the angle, we varied either motion extent (and speed) for a fixed duration, or we held speed fixed and varied duration and distance in Experiment 4. Performance was very similar only for conditions where the extent of motion was constant (Fig. 6). Thus, observers seem to respond to the visual angle at the endpoint of motion; rate of change of angle seems to play no part in the observers’ response. DISCUSSION The emphasis in previous research on binocular 3D motion has been on how both binocular disparity and motion signals are used to form retinal image correlates of the direction of motion in depth, and how the brain could use them to estimate direction3–5,13. We have shown here that for small trajectory angles close to the nose, observers may use neither of these sources of information to specifically obtain the angle of 3D motion. Instead, observers use an estimate of visual direction to set perceived angle rather than explicit estimates of distances moved in depth. This work complements other recent work showing the importance of visual direction for perception and the control of action7–10,14,15. This conclusion is striking and unexpected. Because it causes large systematic errors, the visual direction strategy is not very useful if used alone. In a natural context, it could be used in combination with other sources of visual information. Alternatively, observers’ use of α in this context may reflect a general inability of the visual system to make explicit estimates of trajectory. In support of this unexpected conclu-

sion, many studies have described strategies that use simple prospective, rather than predictive, information to obtain collision achievement or avoidance16, possibly removing the need for explicit representation of 3D object motion. In the context of such direct interaction with the visual stimulus, our results must be treated with caution. We asked observers to respond after the event, not during it, and we do not yet know whether similar errors would occur during more direct interaction with the visual stimulus. However, other reports in the literature do report systematic errors similar to those found here when observers attempt to intercept real objects17, as well as when reporting trajectory angles in computer-generated displays6,18. There has been no work yet on how errors in trajectory perception affect collision avoidance. Other questions remain. For example, if observers use the endpoint of the motion, then the further an object travels, the wider the perceived trajectory should be. This was the case for the conditions used here (Experiment 2). Would observers really see a long displacement trajectory as almost fronto-parallel? Does a trajectory that is double the length of another appear to be one angle for half of its motion, and another for the remainder? How do observers accurately interact with objects in the real environment? Answering such questions will require a more extensive sweep of the available parameter space than is described in this manuscript. METHODS
Experiments 1 and 2. Active stereogoggles synchronized to the screen refresh rate (100 Hz) allowed left and right eyes to view stereoscopic images via alternate CRT monitor frames. Ray traced and anti-aliased images were used to simulate 3D motion toward the observer, starting at the screen (1 m from observer). The initial stimulus comprised a fixation point (luminance 6.6 cd m–2) in the center of the image (subtending 8.3 min arc), which remained in view during all trials. When a trial began, a single target point (8.3 min arc, 6.6 cd m–2, 0.32° below fixation target) moved toward the observer for 1 s along a 3D trajectory left or right of straight ahead. There were no other visible references. Experiments

Figure 5 Standard error of the mean for each observer for each trajectory angle. (a) Experiment 1A. (b) Experiment 1B. Key shows initials of subjects.

Figure 6 Experiment 4 results. Indicated angle as a function of actual angle for four conditions (see Table 2a), averaged across five observers. Error bars, s.e.m.

232

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
were conducted in a dark room. We did not monitor eye movements, so it is possible that observers moved their eyes to follow the target motion. However, it has previously been shown that performance is not significantly different for this task whether observers are asked to move their eyes or to keep them stationary18. Observers were told that any trajectory along any angle from +90° to –90° might appear. Values of the trajectory angle, θ, and the magnitude of the x and z distances moved through in Experiment 1 are given in Table 1. Both positive and negative x components were used so that trajectories could be left or right of straight ahead. For Experiment 2, x distances were constant at 0.024 m or 0.036 m; θ values were the same as for Experiment 1. The task was to indicate the motion direction perceived by moving a pointer to indicate the direction, as described previously6. Before each trial began, the room lights were extinguished and observers were asked to fixate the single reference point in the center of the display. The stimulus appeared and moved for 1 s. A light was then turned on, allowing observers to see the pointer while they moved it to the desired angle. The lights were extinguished and the pointer reset to zero before the next trial began. For each condition, observers performed two blocks of trials. In each block, they were asked to make two settings for each of nine trajectory angles (θ in Fig. 1), ±5°, ±10°, ±15°, ±20° and 0°. These are trajectory angles, measured with respect to the start point on the screen, not eccentricities. We have repeated the experiments presented here using real 3D motion of a target moving along a track, controlled by a stepper motor. Experimental results were similar to those presented here, giving us confidence that the use of active stereogoggles was not detrimental to observer performance (see Supplementary Note 4). Experiment 3. We used the same apparatus and general stimulus form as for Experiments 1 and 2 but now conducted a two-alternative forced choice experiment. Observers were asked to fixate the stationary reference point. In the first interval, the stationary target point appeared for 1 s, with an x location of 0.012 m from the fixation point (as for Experiment 1B). It was always in front of the fixation point, with its z location chosen from the set of trajectory endpoints in Experiment 1B (Table 1). After an inter-stimulus interval of 300 ms, the second interval (also 1 s) contained a stimulus comprising the fixation point and a point whose depth was different from the first but was one of the adjacent depths in the set. The target point could appear to the left or right of fixation. The task was to indicate the order of presentation (nearer-farther or farther-nearer). There were four different blocks of trials. In one, the standard stimulus was 0.045 m from the screen (tests at 0.033 m and 0.068 m) and 0.012 m to the right of fixation. In the second, the standard was as 0.068 m (tests at 0.045 m and 0.137 m) and 0.012 m to the right of fixation. The third and fourth blocks were the same depth arrangements but 0.012 m to the left of the fixation points. Experiment 4. We used the same apparatus and general stimulus form as Experiment 2. A target object moved in depth along a trajectory that could be toward or away from the observer. Five observers were used. Observers were shown trajectories at angles θ of ±20 or ±10°. The task was to indicate the perceived angle using the pointer. Either we varied motion extent (and speed) for a fixed duration, or we held speed fixed and varied both duration and distance. Table 2a shows values of the x component of the distance moved, the speed of that x component and the stimulus duration. Table 2b shows predictions for response behavior based on the use of the different sources of information. If observers respond to the visual direction of the endpoint, performance should be the same for trajectories A and C. If instead they respond to speed, performance should be the same for A and D and for B and C, but larger angles should be perceived for the latter two conditions because the speed is faster. The final alternative is that if observers respond to duration, performance should be the same for A and B and for C and D. Informed written consent was obtained from all observers before their participation in any of the experiments.
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS We would like to thank P. Dean for collecting some of the data presented here. The work was funded by an Engineering and Physical Sciences Research Council project grant and an EPSRC Advanced Fellowship to J.M.H. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 29 July; accepted 28 December 2004 Published online at http://www.nature.com/natureneuroscience/
1. Westheimer, G. The spatial sense of the eye. Invest. Ophthalmol. Vis. Sci 18, 893–912 (1979). 2. Howard, I.P. & Rogers, B.J. Depth Perception. Seeing in Depth Vol. 2 (I. Porteous, Toronto, 2002). 3. Cumming, B.G. & Parker, A.J. Binocular mechanisms for detecting motion-in-depth. Vision Res. 34, 483–495 (1994). 4. Regan, D. Binocular correlates of the direction of motion in depth. Vision Res. 33, 2359–2360 (1993). 5. Brooks, K.R. Interocular velocity difference contributes to stereomotion speed perception. J. Vis. 2, 218–231 (2002). 6. Harris, J.M. & Dean, P.J.A. Accuracy and precision of binocular 3D motion perception. J. Exp. Psych. Hum. Perc. Perf. 29, 869–881 (2003). 7. Rushton, S.K., Harris, J.M., Lloyd, M.R. & Wann, J.P. Guidance of locomotion on foot uses perceived target location rather than optic flow. Curr. Biol. 8, 1191–1194 (1998). 8. Wilkie, R.M. & Wann, J.P. Driving as night falls: the contribution of retinal flow and visual direction to the control of steering. Curr. Biol. 12, 2014–2017 (2002). 9. Harris, J.M. & Bonas, W. Optic flow and scene structure do not always contribute to the control of human walking. Vision Res. 42, 1619–1626 (2002). 10. Fajen, B.R. & Warren, W.H. Visual guidance of intercepting a moving target on foot. Perception 33, 689–715 (2004). 11. McKee, S.P. & Welch, L. The precision of size constancy. Vision Res. 32, 1447–1460 (1992). 12. Foley, J.M. Binocular distance perception. Psychol. Rev. 87, 411–434 (1980). 13. Beverley, K.I. & Regan, D. The relation between discrimination and sensitivity in the perception of motion in depth. J. Physiol. (Lond.) 249, 387–398 (1975). 14. Wann, J.P. & Swapp, D.K. Why you should look where you are going. Nat. Neurosci. 3, 647–648 (2000). 15. Wann, J. & Land, M. Steering with or without the flow: is the retrieval of heading necessary? Trends Cogn. Sci. 4, 319–324 (2000). 16. Chardenon, A., Montagne, G., Laurent, M. & Bootsma, R.J. The perceptual control of goal-directed locomotion: a common control architecture for interception and navigation? Exp. Brain Res. 158, 100–108 (2004). 17. Peper, L., Bootsma, R.J., Mestre, D.R. & Bakker, F.C. Catching balls: how to get the hand to the right place at the right time. J. Exp. Psych. Hum. Perc. Perf. 20, 591–612 (1994). 18. Welchman, A.E., Tuck, V.L. & Harris, J.M. Human observers are biased in judging the angular approach of a projectile. Vision Res. 44, 2027–2042 (2004).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

233

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

A representation of the hazard rate of elapsed time in macaque area LIP
Peter Janssen1,2 & Michael N Shadlen1
The capacity to anticipate the timing of environmental cues allows us to allocate sensory resources at the right time and prepare actions. Such anticipation requires knowledge of elapsed time and of the probability that an event will occur. Here we show that neurons in the parietal cortex represent the probability, as a function of time, that a salient event is likely to occur. Rhesus monkeys were trained to make eye movements to peripheral targets after a light dimmed. Within a block of trials, the ‘go’ times were drawn from either a bimodal or unimodal distribution of random numbers. Neurons in the lateral intraparietal area showed anticipatory activity that revealed an internal representation of both elapsed time and the probability that the ‘go’ signal was about to occur (termed the hazard rate). The results indicate that the parietal cortex contains circuitry for representing the time structure of environmental cues over a range of seconds.

Humans and animals rely on a sense of elapsed time to plan actions, anticipate salient events, infer causal regularities and learn associations1. Yet little is known about the neural mechanisms that underlie the encoding and use of elapsed time2,3. Traditionally, the focus has been on the cerebellum and basal ganglia, but several recent studies suggest that the association areas of the neocortex may play an important role4–6. Neurons in association areas perform computations that span gaps in time between sensation and action in order to mediate working memory7, motor planning8,9 and decision making10–12. These processes depend on a representation of time to infer temporal order, plan sequences of actions and control the tradeoff between speed and accuracy. In general, processes not controlled by immediate external events may still need to know when information is useful. Behavioral performance is enhanced if subjects can anticipate the point in time that a stimulus is likely to appear or an instruction is likely to occur4,13–15. To anticipate the timing of behaviorally relevant events, the brain must represent the passage of time and use this representation to estimate the probability that an event is likely to occur, given that it has not occurred already. This computation is termed a hazard rate14. Although several brain regions have been shown to encode the probability that an action will ensue12,16,17, little is known about how such probabilities are represented by the brain when they change as a function of time. Neurons in various brain areas have occasionally shown climbing activity in the interval preceding a test stimulus or ‘go’ instruction, an effect that has been interpreted as a neural correlate of anticipation18–23. A recent study showed that neurons in the lateral intraparietal area (LIP) undergo time-dependent changes in their responses when monkeys make decisions about the duration of a time

interval5. Because LIP neurons show sustained changes in firing rate during delayed eye movement tasks9,24–26, we hypothesized that this persistent activity might encode the hazard rate used to anticipate the time of a pending ‘go’ signal. RESULTS Two rhesus monkeys were trained to make eye movements to a peripheral target (Fig. 1a). The monkeys were required to hold steady fixation until the fixation spot had dimmed and were then rewarded for initiating an accurate eye movement as soon as possible (see Methods). The waiting time between target onset and the ‘go’ signal (that is, the ‘go’ time) was a random variable whose probability distribution was fixed throughout a block of trials. We used two time schedules in alternating blocks of trials. In the bimodal time schedule (Fig. 1b, top row left), the ‘go’ signal could come either early or late, but not in the interval between 0.75 and 1.75 s, whereas in the unimodal time schedule (Fig 1b, top row right), the ‘go’ times were distributed between 0.5 and 2 s. We reasoned that exposure to these schedules might allow the monkeys to anticipate the arrival of the ‘go’ signal. The time course of such anticipation is formalized by the hazard rate. Because the brain cannot estimate elapsed time precisely1, the mathematical functions are replaced by blurred versions, which we term subjective hazard rates or anticipation functions (see Methods). There are clear differences between the anticipation functions associated with the unimodal and bimodal schedules (Fig. 1b, bottom). When ‘go’ times are drawn from the unimodal distribution, the anticipation function mainly increases with waiting time. In contrast, when ‘go’ times are drawn from the bimodal distribution, the anticipation is triphasic: it rises, falls and

Hughes Medical Institute, National Primate Research Center and Department of Physiology and Biophysics, University of Washington, Box 357290, Seattle, Washington 98195. 2Present address: Laboratorium voor Neuro-en Psychofysiologie, KU Leuven Medical School, Herestraat 49, B-3000 Leuven, Belgium. Correspondence should be addressed to M.N.S. ([email protected]).
Published online 16 January 2005; doi:10.1038/nn1386

1Howard

234

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
Figure 1 Methods. (a) Delayed eye movement task. The monkey made eye movements to the red target as soon as the fixation point dimmed. The target appeared in different locations on each trial. Here we report trials in which the target appeared in the response field of the LIP neuron we recorded. A bracket demarcates the random waiting time between target onset and ‘go’ signal. (b) Probability distributions, hazard rates and subjective hazard rates. The top row illustrates the probability distributions for drawing ‘go’ times under the two schedules used in the experiments: bimodal (left) and unimodal (right). The middle row shows the hazard rates for each of these time schedules (equation (3)). The bottom row shows the subjective hazard rates or anticipation functions for each of these time schedules (Ab(t) and Au(t), φ = 0.26, equation (4)). The functions employ a common scale, which is normalized to the peak of the Au(t). (c) Representative magnetic resonance image from monkey J. Neurons in both monkeys were recorded from the posterior portion of LIPv27, outlined by the arrows; ips, intraparietal sulcus.

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

rises again. The scale of these anticipation functions was set from 0 to 1 to facilitate interpretation of the behavioral and physiological data, as described below. Anticipation of the ‘go’ cue affects saccade reaction time Both monkeys learned to anticipate the time of the ‘go’ signal, as evidenced by their reaction times. Figure 2 shows running means of reaction times measured under the unimodal (purple) and bimodal (red) schedules. Each point shows the mean of 51 observations. In both types of blocks, reaction time showed a clear dependency on the amount of time that elapsed before the ‘go’ signal. In ‘unimodal schedule’ blocks, reaction time decreased with longer waiting times (regression analysis,

P < 0.001). In ‘bimodal schedule’ blocks, reaction time decreased and rose again for ‘go’ times that happened to be drawn from the first mode and was fastest for ‘go’ times that were drawn from the second mode. This triphasic pattern was more striking for monkey J (Fig. 2a), but it was evident in both monkeys, as shown below. The reaction times from both monkeys clearly decreased between the early and later modes (for monkey J, mean reaction time was 268 ± 1.2 and 237 ± 0.8 ms in epochs within ±350 ms of the first and second modes, respectively; for monkey H, the corresponding means were 295 ± 0.7 and 259 ± 1). The monkeys' reaction times were inversely related to the anticipation functions associated with the schedules of random 'go' times. For either schedule, the data were well fit by a weighted sum of the two anticipation functions delayed by 56 ms (Methods, equation (5)). These fits furnish estimates of the number of milliseconds that reaction time is reduced per unit change in anticipation (Table 1). Under the bimodal schedule, the fit to the data from monkey J (Fig. 2a) is dominated by the bimodal anticipation function, whereas under the unimodal schedule, the fit is dominated by the unimodal anticipation function (Table 1). For monkey H, the unimodal anticipation function explains most of the decline in reaction time under both schedules. Nevertheless, the bimodal anticipation function has a significant role under the bimodal schedule of ‘go’ times (P < 0.02); (Table 1). The influence of the bimodal anticipation function can be understood more directly by examining the partial correlation between the reaction times from individual trials and the anticipation function appropriate for the schedule. Because both anticipation functions can affect the pattern of reaction times under either schedule, we performed a partial (conditional) correlation to factor out the influence of the potentially confounding anticipation function. Under the bimodal schedule, the partial correlation between reaction time and the bimodal anticipation function, rRTb ,Ab|Au, was –0.36 for monkey J (P < 0.001, Fisher z). Under the unimodal schedule, there was a significant inverse correlation between reaction time and the unimodal anticipation function (rRTu,Au|Ab = –0.24; P < 0.001). A similar inverse correlation was present

Table 1 Weighting coefficients for the unimodal and bimodal anticipation functions fit to reaction time data in Figure 2, using equation (5)
Weights of anticipation functions fit to reaction times (ms change in reaction time per unit anticipation) Monkey J Distribution of ‘go’ times Bimodal schedule Unimodal schedule wu –42.2 ± 1.4 –17.9 ± 2.2 wb –69.8 ± 4.2 –9.7 ± 6.9* R2 0.96 (0.29) 0.77 (0.06) wu –40.0 ± 2.0 –32.9 ± 2.6 Monkey H wb –9.7 ± 3.9 –22.1 ± 9.1 R2 0.95 (0.39) 0.91 (0.11)

The coefficients multiply the anticipation functions, Au(t) and Ab(t). One unit of anticipation is the range of Au(t) shown in Figure 1. Both anticipation functions affected reaction time inversely under both schedules (P < 0.02, except for asterisk). R2 describes the fraction of the running mean variance explained by the fit; parenthetical values describe the fraction of variance for individual trials.

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

235

ARTICLES
Figure 2 Reaction time (RT) is modulated by anticipation of the ‘go’ cue. Eye movement RTs are plotted as a function of the waiting time between target onset and the ‘go’ signal (the ‘go’ time), during the bimodal (red) and the unimodal (purple) time schedules. Points represent running means of RT from 51 consecutive ‘go’ times. Black curves are fits to the data using a weighted sum of the anticipation functions (equation (5); weights are in Table 1). The dashed red curve is the bimodal anticipation function sampled at the ‘go’ times that occur under the bimodal schedule. The dashed purple curve shows the unimodal anticipation function sampled at the ‘go’ times under the unimodal schedule. Note that RTs are only measured when there is a finite probability that a ‘go’ cue could occur. (a) Data from monkey J (n = 2,322 unimodal trials and n = 1,876 bimodal trials). (b) Data from monkey H (n = 1,480 unimodal trials and n = 1,456 bimodal trials).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

in monkey H (rRTb ,Ab|Au = –0.09; rRTu ,Au|Ab = –0.32; P < 0.001 for both values). These behavioral observations imply that the brain is capable of representing both the passage of time and the time-dependent probability that the ‘go’ signal is about to occur. Anticipation of the ‘go’ cue affects LIP neural activity We recorded 70 LIP neurons, screening them using an eye movement task in which monkeys made saccadic eye movements to the remembered location of a briefly flashed target. We studied only those neurons that showed strong, spatially selective activity during the memory period between the flashed eye movement target and the ‘go’ instruction (see Methods). These neurons were commonly

encountered in the ventral portion of LIP27 (Fig. 1c). We then tested these neurons using the delayed saccadic eye movement task with a bimodal or unimodal distribution of random ‘go’ times. In half the trials, the saccade target was in the response field of the LIP neuron, giving rise to a persistent elevation in the neural activity. Our focus is on the modulation of this persistent activity while the monkey awaited the ‘go’ instruction. Many neurons in area LIP altered their firing rates as a function of time, in accordance with the changing anticipation of the ‘go’ signal. Figure 3a shows responses of a representative neuron recorded in blocks of trials using either the bimodal or the unimodal schedule of ‘go’ times. Each curve represents averaged spike rates measured during

Figure 3 LIP responses represent anticipation as a function of time. (a) Single-neuron example. Top: anticipation functions for the unimodal (purple) and the bimodal (red) time schedule. Bottom: average neural activity recorded during waiting period for the ‘go’ signal under the bimodal (red) and the unimodal (purple) time schedule. Shading, s.e.m.; black curves, fits to equation (5). (b) Population responses from blocks using bimodally and unimodally distributed ‘go’ times. Averaged normalized responses (± s.e.m.) plotted as a function of waiting time for monkey J (n = 39) and monkey H (n = 31). Spike rates from each neuron were normalized to the mean activity 320–1,360 ms after target onset using all trials in both schedules. Black curves, fits to equation (5); red, bimodal schedule; purple, unimodal schedule. (c) Effect of schedule on the temporal pattern of the response from single neurons. Average response from a block using either bimodal or unimodal schedule of ‘go’ times was described as a weighted sum of anticipation functions, Au(t) and Ab(t). Upper scatter plot compares amount of response modulation attributed to Ab(t) in unimodal and bimodal testing blocks. The bimodal anticipation function explained more of the response modulation during the block of trials using bimodally distributed ‘go’ times (mean wb was +14.1 and +0.8 during the bimodal and unimodal schedules, respectively). Differences are summarized by the frequency histogram. Lower scatter plot compares amount of response modulation attributed to Au(t) in unimodal and bimodal testing blocks. The unimodal anticipation function explained more of the response modulation during the block of trials using unimodally distributed ‘go’ times (mean wu was +4.8 and –1.1 during unimodal and bimodal schedules, respectively). Differences are summarized by the frequency histogram. Error bars, s.d. of parameter estimates. Shaded histograms indicate significant cases, P < 0.01. Green symbols indicate example neuron in a.

236

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
Table 2 Weighting coefficients for the unimodal and bimodal anticipation functions fit to response averages in Figure 3b, using equation (5)

Weights of anticipation functions fit to LIP response (Percentage change in firing rate per unit anticipation) Monkey J Distribution of ‘go’ times Bimodal schedule wu –19 ± 1 +25 ± 2 wb +26 ± 2.3 wu +11 ± 1.4 Monkey H wb +39 ± 2.7 +22 ± 2.3

+51.4 ± 1.9 –7.1 ± 1.8

Unimodal schedule

The coefficients multiply the anticipation functions, Au(t) and Ab(t). Units for the population responses are percentage change in firing rate (relative to the average delay period activity) per unit change in the anticipation function. One unit of anticipation is the range of Au(t) shown in Figure 1b. Schedule affected the weights in the appropriate direction (compare weights in each column; P < 10–5 for all comparisons).

the waiting period from the onset of the eye movement target at the cell’s preferred location until the dimming of the fixation point. When the neuron was studied in a bimodal ‘go’ time schedule, the firing rate showed an early peak in activity at ∼0.4 s after target onset. Then, if the 'go' signal did not arrive, the activity declined by 41 spikes s–1 over the following 0.4 s. Then, from ∼0.8 s, the response increased gradually by 24 spikes s–1 as the monkey awaited a ‘go’ signal drawn from the later mode of the bimodal distribution. When the neuron was studied with a unimodal ‘go’ time schedule, the early rise in activity was conspicuously absent. After the onset response, the firing rate declined by 9 spikes s–1 until ∼0.4 s, and then increased steadily by 20 spikes s–1 from 0.4 to 1.4 s as the monkey waited for the ‘go’ signal to arrive. The time course of activity from this neuron was strongly influenced by the monkey’s knowledge of the random ‘go’ time schedule. Notably, the responses depicted by the purple and red curves in Figure 3a were recorded under identical physical conditions: while viewing the same target and fixation point and awaiting the ‘go’ signal. The marked difference in the time course of the neural response reflects only a difference in the monkey’s state of anticipation. The response functions in Figure 3a can be approximated by a weighted sum of the anticipation functions associated with the unimodal and bimodal schedules (black curves). The weights derived from these fits furnish a test of the hypothesis that the LIP response is dominated by the hazard function associated with the schedule that the monkey experienced. For the neuron in Figure 3a, the weights assigned to the bimodal and unimodal anticipation functions (wb and wu, respectively) were +36.6 ± 1.7 and –8.6 ± 1.1, respectively, during testing with the bimodal schedule. In contrast, during testing with the unimodal schedule, the fit was dominated by the unimodal anticipation function: wb = –10.7 ± 1.3, wu = +8.9 ± 1.3. The negative weights associated with the ‘wrong’ anticipation function suggest that LIP activity also reflects the knowledge that under each of the schedules certain ‘go’ times will not occur.

These weighting coefficients estimate the change in spike rate associated with the range of the anticipation functions shown in Figure 3a (top); they have the unit ‘spikes per second per unit of anticipation’, where one unit is the range of the subjective hazard associated with the unimodal distribution (see Methods). The weights and their standard errors demonstrate that the response modulations observed in either block are dominated by the appropriate bimodal or unimodal anticipation function. Figure 3b shows the averaged responses from all 70 neurons in the two monkeys. To make these graphs, we normalized the responses from each neuron to its average firing rate during the delay period using combined data from both schedules. When the ‘go’ times were drawn from a bimodal distribution, the response averages showed a triphasic pattern: increasing in anticipation of ‘go’ times drawn from the early mode, then decreasing when ‘go’ signals were unlikely to occur, then increasing again in anticipation of ‘go’ times drawn from the later mode (Fig. 3b, red). When the ‘go’ times were drawn from the unimodal distribution, the triphasic pattern was less apparent. The responses showed a steady rise (monkey J) or remained stable (monkey H) from ∼0.4 s onward. The graphs demonstrate that anticipation-related modulation can constitute a large fraction of the delay period activity. For example, during the waiting period for bimodally distributed 'go' signals, the responses from monkey J modulated between –40% and +40% of the delay period activity (Fig. 3b, upper red curve). These effects were smaller in monkey H, but the effect of schedule on the responses was highly significant for both monkeys. Fits to the response averages using weighted sums of the anticipation functions (equation (5); Fig. 3, black curves) indicate that the schedule affected the pattern of firing in both monkeys (Table 2). When the schedule changed from unimodal to bimodal, the fitted response functions showed an increase in the weight of the bimodal anticipation function on the response and a decrease in the weight of the unimodal schedule (P < 10–5 for all comparisons in both monkeys; see Table 2). Individual neurons showed a variety of response patterns, but most underwent modulations similar to the response averages and example in Figure 3a,b. The spike rate functions for 68 of 70 neurons (97%) were well described by a weighted sum of the two anticipation functions (P < 0.05; H0: wb = wu = 0 ; mean R2 = 0.67; interquartile range = 0.53–0.87) delayed by 55 ms on average (τ, equation (5)). The scatter plot in the upper panel of Figure 3c compares the contribution of the bimodal anticipation function to each neuron’s response under the two schedules. This bimodal anticipation function was weighted significantly more during the block using the bimodal schedule of ‘go’ times (mean difference = +12.2 ± 2.8, P < 0.001, paired t-test). The corresponding analysis of the unimodal anticipation functions is shown in the lower panel of Figure 3c. The unimodal anticipation functions tended to exert less weight overall but were stronger during blocks of trials using the unimodal schedule of ‘go’ times (mean difference = –5.9 ± 1.3, P < 0.01).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 4 Neural activity is inversely related to reaction time on a trial-bytrial basis. For each trial, we compared the reaction time to the neural activity in an epoch from 150 ms before to 50 ms after the ‘go’ signal. We removed the potentially confounding effect of time by subtracting the mean reaction time and spike rate from each value. The correlation coefficient was obtained using these residual values. The histogram shows r-values from 70 cells (shading indicates significance; P < 0.05 in 28/76 neurons; Fisher z) using all trials from unimodal and bimodal schedules (results are similar using either schedule alone). The weak but significant inverse correlation indicates that variability in neural activity in LIP affects the reaction time.

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

237

ARTICLES
Figure 5 Time-dependent anticipatory activity is associated with motor preparation. (a) Variant of the delayed eye movement task. The monkey made an eye movement to the green target as soon as the blue stimulus dimmed. The monkey must attend to both targets but plan an eye movement only to the green target. Either target could appear in the neuron’s response field. (b) Single-neuron responses. The average activity is shown for trials in which the saccade target appeared in the response field (green) and for trials in which the ‘go’ cue appeared in the response field (blue). (c) Population responses. Left, response averages from 25 neurons (monkey J) on trials in which the saccade target appeared in the response field (green) and trials in which the ‘go’ cue appeared in the response field (blue). Right, population responses of the same 25 neurons on trials in which both the saccade target and the ‘go’ signal appeared in the response field (black), along with the average activity of these 25 neurons on trials using the unimodal time schedule with the saccade target in the response field (purple).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

sequences, wb was +33.6, +4.2, and +18.6 spikes s–1 per unit anticipation, respectively. In unimodal-bimodal-unimodal sequences, wu was +8.3, +0.1, and +6.9 spikes s–1 per unit anticipation, respectively. Trial-to-trial correlation between neural activity and behavior Does the anticipatory activity in LIP help the monkey prepare its eye movement, or are the behavioral and neural manifestations of anticipation merely coincidental? Although it is impossible to exclude the latter possibility without manipulating the discharge of the neurons we recorded, some insight can come from examining the relationship between neural activity and reaction time on single trials. We asked whether the firing rate in the epoch around the time of the ‘go’ signal predicted the monkey’s reaction time on that trial. By subtracting the mean neural activity and mean reaction time from the data, we removed the effect of time and anticipation from both sets of measurements, leaving only the residual changes about the time-dependent means. We then calculated a correlation coefficient between these residual values for each neuron in our sample. The histogram of correlation coefficients (Fig. 4) shows that the majority of neurons showed a negative correlation between firing rate and reaction time on single trials (median correlation coefficient = –0.09, P < 0.001; for neurons with significant anticipation modulation, the median correlation coefficient was –0.10, P < 0.001). The size of this correlation was weak, but it is notable that the variable discharge of a single neuron on a single trial has any impact at all on the monkey’s reaction time. Thus, in addition to reflecting the monkeys’ state of anticipation, the time-dependent activity of LIP neurons is probably partially responsible for the behavioral manifestation of this anticipation in the saccadic eye movements. Anticipatory activity is associated with eye movement planning Previous studies suggest that area LIP plays a role in the allocation of spatial attention and the planning of eye movements9,26,28. The anticipatory activity we observed could reflect temporal variation in either of these functions. To distinguish these possibilities, we trained the monkeys on a variant of our delayed eye movement task (Fig. 5a). Instead of anticipating the dimming of the fixation point, the monkey was required to monitor a second peripheral target, which was distinguished by its color. The sole purpose of this second target was to provide the ‘go’ signal, again by dimming slightly, thereby instructing an eye movement to the other target. Thus the monkey had to attend to the ‘go’ cue and possibly to the saccade target, but the eye movement plan was directed only to the latter. By placing either the ‘go’ cue or the eye movement target in the neuron’s response field, we could determine whether anticipatory activity in LIP is associated with the direction of the intended eye movement, spatial attention or both.

Note that the weights for the appropriate anticipation function tended to be positive, indicating that greater anticipation leads to higher spike rates, on average, in LIP. Together, these findings indicate that persistent activity in LIP is modulated by the passage of time. All of the neurons in our sample respond selectively to targets in their response fields. Yet, during the delay period in which the monkeys knew where but not when to make an eye movement, LIP neurons modulated their discharge by tens of spikes per second (on the order of one-third the level of sustained activity) in a pattern resembling the theoretical anticipation functions associated with the schedules of random ‘go’ times. The changes in neural activity associated with the change in task timing evolved on a short time scale (on the order of minutes) and were highly flexible. For the neuron in Figure 3a, reliable changes in the modulation pattern emerged as early as 20 trials after the change in the time schedule. For example, wb decreased from +36.6 ± 1.7 to +2.6 ± 1.7 and wu increased from –8.6 ± 1.1 to +22.4 ± 2.0 over the first 20 trials (P < 0.001, data not shown). Notably, this rapid change was also evident in the monkeys’ behavior. As is apparent from Figure 2a, reaction time tended to be shorter on average under the unimodal schedule across all ‘go’ times. This change was apparent within 30 trials after the change from the bimodal to the unimodal time schedule in monkey J (P < 0.01, t-test). To further examine this flexibility, we conducted a third block of testing in 11 neurons (monkey J) in which we reverted to the first schedule of random ‘go’ times used in the first block (either unimodal or bimodal). The neural activity in this third block changed again and became similar to the activity in the first block of trials. In bimodal-unimodal-bimodal

238

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
The neuron in Figure 5b showed robust anticipatory activity associated with the bimodal time schedule when the eye movement target appeared in the response field (wb = +65.2 ± 3.6, wu = –13.2 ± 2.2). In contrast, when the 'go' cue was in the response field, this anticipatory modulation was much weaker (Fig. 5b; wb = +6.9 ± 1.8, wu = –13.8 ± 1.0). This pattern of results was similar across the population of 25 neurons tested in this manner (Fig. 5c, left). The average wb was +24.7 ± 2.5 spikes s–1 per unit anticipation when the target appeared in the response field compared to +5.1 ± 1.9 when the 'go' cue appeared in the response field. Although it is not apparent in the figure, the weaker representation of the bimodal anticipation function in the latter configuration was significant (P < 0.01, t-test). In contrast, when these neurons were tested under the unimodal schedule (target in response field), the responses did not show a positive influence of the bimodal anticipation function (Fig. 5c, right; mean wb = –4.1 ± 1.2, mean wu = +6.3 ± 0.9). To test whether we were suppressing modulation at the attended location of the 'go' cue because of competition with the saccade target in the opposite hemifield, we added a control condition in which both the saccade target and the 'go' cue were in the neuron's response field. The pattern of neural activity in this condition (Fig. 5c, right) was virtually indistinguishable from the activity when the saccade target was presented in the RF and the ‘go’ cue outside the response field (green in Fig. 5c, left). Hence the location of the ‘go’ cue did not seem to influence the pattern of anticipation responses in LIP. Based on these control experiments, we conclude that anticipatory modulation was strongest in LIP neurons that represent the locus of the intended eye movement. DISCUSSION We trained rhesus monkeys to anticipate the timing of a ‘go’ signal in a delayed saccade task. Unlike in previous studies29, the monkeys had to learn two probabilistic time schedules that were used in alternating blocks of trials. To maximize the chances that the animals would learn both schedules, we used a bimodal and unimodal distribution of ‘go’ times with minimal overlap. Examination of the monkeys’ reaction times demonstrated that they had learned features of these probabilistic schedules in order to anticipate the time of the ‘go’ instruction. We also found that single neurons in area LIP modulated their firing rate as a function of time in a way that reflected the monkeys’ state of anticipation. The spike rate was strongly influenced by the probability that the ‘go’ instruction would occur in the next moment, based on the monkey’s experience with the two schedules. The spike rate modulation in LIP is thus related to the hazard rates associated with the unimodal and bimodal probabilistic schedules of ‘go’ times (Figs 3 and 5). The pattern of modulation also reflected the well-known fact that the experience of elapsed time carries with it a degree of uncertainty that is proportional to the true duration (Weber’s law)3,5,30. This uncertainty implies that the distribution of ‘go’ times that the monkey experienced during training is a distorted version of the probability distributions that we programmed into the computer, giving rise to the subjective hazard rates reflected in LIP. Notably, the temporal pattern of the LIP spike rate often changes substantially within a single experimental session upon a change in the probabilistic schedule. It is important to bear in mind that markedly different patterns of response (in Fig. 3, for example) were observed from the same neuron while the monkey viewed the identical display and awaited the ‘go’ instruction. Evidently, the brain can detect the change of schedule after a few samples of ‘go’ times (<30 trials) and adjusts its circuitry to achieve the appropriate pattern of anticipation in LIP. Further studies are needed to clarify the mechanism of this flexibility. In the control experiment used to dissociate an intention to move the eyes from the spatial attention allocated to the detection of the ‘go’ instruction (Fig. 5), we found that the representation of the hazard rate was markedly diminished when the ‘go’ cue was in the response field and the saccade target was not. At first glance, this seems to contradict a recent study that demonstrated hazard-like modulation of attentionrelated signals in area V4 of the monkey29. However, a small amount of modulation was present when spatial attention was directed to the LIP response field (∼5 spikes s–1 per unit of anticipation). It is possible that this is sufficient to explain the modest degree of modulation seen in the V4 study of attention. Alternatively, the allocation of spatial attention may be difficult to dissociate from an intention to make an eye movement31. Indeed, the neural basis of attention-related modulation in area V4 may be mediated through high-level oculomotor structures32. It is therefore plausible that the intention-related signals seen in our experiments could underlie a shift in spatial attention that is not in competition with an eye movement plan33. The present study cannot determine whether the timing-related anticipatory activity arises in area LIP or is simply passed to LIP from other structures that have been implicated in interval timing, such as the prefrontal cortex18,20, the basal ganglia34 or the cerebellum35. However, two observations lead us to suspect that the anticipation signals measured in LIP may have more than a coincidental role in shaping the monkey’s behavior. First, the anticipation functions are reflected by the neurons and by the reaction time at the same latency with respect to the ‘go’ instruction. The anticipation of the ‘go’ instruction best matches the neural responses with a latency of 55 ms and is maximally (inversely) correlated with the monkey’s reaction time 56 ms before the ‘go’ signal (Fig. 2). Second, we observed a weak correlation between the variable responses on single trials and the monkey’s reaction time (Fig. 4). The weakness of the reaction time–spike rate correlation distinguishes LIP from motor structures, such as the frontal eye field and the superior colliculus, with which it is interconnected36,37. Indeed, much larger negative correlations between spike rate and reaction time have been reported in these areas38,39. The weak correlation between the variable spike rate from LIP neurons on single trials and the monkey’s subsequent reaction time on that trial suggests that either LIP neurons directly influence reaction time, or they mirror with great fidelity the structures that lie along the causal chain. This implies that the marked anticipationrelated modulation represented in LIP is likely to explain the behavioral manifestation of anticipation: reduced saccadic latency. Our findings extend a previous study that has demonstrated a possible role for area LIP in interval timing5. As in the present study, LIP neurons were found to encode the salience of potential eye movement targets in a dynamic fashion. In that study, the monkey’s perception of the duration of a test light was explained by comparing the activity of LIP neurons whose activity increased or decreased as a function of time. In light of the present findings, we suspect that the modulation of activity is governed by the anticipation of the termination of the test light, whose random durations resemble the unimodal schedule of ‘go’ times used in the present study. By adding and subtracting the subjective hazard rate to or from low and high background firing rates, respectively, LIP could produce a pair of crisscrossing functions (as in ref. 5) to represent elapsed time with respect to a memorized standard duration. Together, these studies suggest that LIP encodes elapsed time insofar as it affects the meaningfulness of visual objects that are potential gaze targets. Indeed, the observation that some LIP neurons track the motion of occluded objects40 might be explained by the prediction of a salient object emerging from behind an occluder in time. In addition to elapsed time, it has been shown that persistent neural activity in area LIP can

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

239

ARTICLES
be influenced by a variety of factors associated with reward expectation12, response bias41 and decision formation42. All of these functions require the neurons to operate on a time scale governed neither by immediate changes in the sensory environment nor by real-time constraints of moving body parts. Hence, the brain must keep an internal representation of time to determine by what time an action must occur or when information becomes relevant. Throughout the association cortex4,20,43,44, neurons with persistent activity may therefore represent elapsed time to infer the temporal structure of the environment. METHODS
Task. Two rhesus monkeys were trained on the delayed eye movement task45. The monkeys maintained their gaze within 1° of a fixation spot in the center of the display. After 400 ms of stable fixation, a red target was presented at a position between 3° and 15° eccentricity. After a variable delay, the fixation point dimmed by 30% of its luminance. This dimming served as the 'go' signal to make an eye movement to the target. A liquid reward was given for accurate saccades (within 4° of the target) initiated 150–500 ms after the 'go' signal. To encourage fast responses, reward size was governed by an exponential function of reaction time (minus 150 ms). The time between target onset and the 'go' signal was a random variable drawn from either a bimodal or a unimodal distribution (Fig. 1b, upper row). The bimodal distribution, B(t), was the sum of two non-overlapping Rayleigh distributions, delayed by d1 and d2: B(t) = 2 (R1 ϩ R2)
1

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

~ ~ substituting f(t) and its definite integral, F(t) into equation (3). We refer to these subjective hazard rates as anticipation functions, Au(t) and Ab(t), below. For ease of interpretation, they are scaled in all graphs and fits by a common factor (2.03, for φ = 0.26) so that the functions range from 0 to 1 (Fig. 1b, bottom row). In the cue experiment, a blue and a green stimulus were presented on opposite sides of the fixation point. The monkeys' task was to make a saccade to the green stimulus (the target) as soon as the blue stimulus (the cue) dimmed. We compared the activity during trials in which the ‘go’ cue was in the neuron's response field and the target was outside the response field, to trials in which the opposite configuration was present. To determine any potential color selectivity46 we also recorded trials in which the monkey made saccades to either a green or a blue target in interleaved blocks of trials. Recording procedure. Action potentials from single neurons were recorded extracellularly using standard procedures10 and stored with 1 ms precision. Eye position was monitored using a scleral search coil (sampled at 1 kHz). We screened neurons using a memory-guided saccade task45. Only neurons with high spatially selective delay period activity were studied: the average delay activity (computed 320–720 ms after target onset) was at least 60% of the visual response (80–320 ms after target onset). Most of these neurons also showed strong presaccadic activity (88% had larger responses in a 100 ms epoch ending at saccade initiation than in the 300 ms epoch preceding this; P < 0.05, t-test). This property is typical of LIP neurons in other studies10,24,37,47. All training, surgical and recording procedures complied with the guidelines of the National Institutes of Health and the research protocols approved by the University of Washington. Data analysis. For each trial, the spike rates were truncated at 50 ms after the dimming of the fixation point. The first 50 trials after the change in the time schedule were excluded from the analysis. We fitted the mean spike rates (80-ms bins) after the initial visual response (starting at 160 ms after target onset) using a weighted sum of subjective hazard rates:

where
Ri =

{

2αi(t – di)e – αi(t–di)2
0

for t > di otherwise

(1)

(α1 = 18, d1 = 0.1, α2 = 15, d2 = 1.75). The ‘go’ time was drawn with equal probability from R1 or R2. The unimodal distribution of ‘go’ times was a single Weibull function delayed by 0.5 s:

r (t ) = w e + wuAu(t – τ ) + wbAb( t – τ ) + ε

(5)

U(t) =

{

3α(t –

1 2 – α(t– 1 )3 ) e 2 2 0

for t >

1 2

otherwise

(2)

The probability distribution of the ‘go’ times was fixed throughout a block of trials. After 200–300 trials, the time schedule was changed without notification. Ideally, anticipation should be governed by the conditional probability that an event will occur given that it has not yet occurred, termed the hazard rate (Fig. 1b, middle row). This is the probability that the ‘go’ signal will occur at time t divided by the probability that it has not yet occurred: h(t) = ƒ(t) 1 – F(t)

(3)

where f(t) is either U(t) or B(t), and F(t) is the cumulative distribution, t ∫0 ƒ(s)ds. To obtain our predicted anticipation functions, we calculated ‘subjective’ hazard rates based on the assumption that elapsed time is known with uncertainty that scales with time. The probability density function f(t) = U(t) or B(t) was first blurred by a normal distribution whose standard deviation is proportional to elapsed time. 1 φt
∞

ƒ(t) =

~

∫ƒ(t)e –(τ–t)2/(2φ2t2)dτ 2π –∞

(4)

The coefficient of variation, φ, is a Weber fraction for time estimation (for most analyses, φ = 0.26). Equation (4) implements the idea that the monkey’s estimate of elapsed time carries uncertainty. Thus, an event at objective time t0 is sensed as if it occurred at t0 ± σ. The subjective hazard rate is then obtained by

where r is the neuronal response, we is a constant term, and wu and wb are the weights for the unimodal (Au) and bimodal (Ab) anticipation functions, respectively, delayed by time shift τ. ε represents noise, which is assumed to be Gaussian with uncertainty derived from the sample means. Equation (5) was also used to fit the reaction times on each trial with a weighted sum of subjective hazard rates. Because the functions range from 0 to 1, the weights can be interpreted as the magnitude of the spike rate modulation attributed to these theoretical waveforms (an approximation to a basis set) in units of spikes per second per unit anticipation (for the fits to the reaction times, the units are milliseconds per unit anticipation). We used a maximum-likelihood fitting procedure to obtain the fits, parameter estimates and their standard errors. Standard errors of parameters were estimated from the Hessian matrix of second partial derivatives of the log likelihood48 and were used to generate t-statistics cited throughout the paper. We fit the data from each neuron and each schedule independently to obtain the weights shown in the scatter plots (Fig. 3c). The Weber fraction was fixed (equation (4), φ = 0.26). We tried to estimate φ using the fits, but found that only strongly modulated neurons gave reliable estimates (mean φ from 48 reliable cases was 0.33 ± 0.03). A similar strategy was used to fit the population response data (Fig. 3b) except that fits to the two response averages were constrained to use a common time delay, τ, and the Weber fraction was free. The large Weber fraction obtained from these fits (0.41 and 0.5 for monkeys J and H, respectively) was probably induced by smearing of the response functions because of averaging. For the all other analyses, we assumed φ = 0.26, consistent with previous human and animal studies30,49 and with behavioral measurements in monkeys in our laboratory5. The partial correlation coefficients between the theoretical anticipation function and the reaction times (RT) observed under either the unimodal or bimodal schedule were calculated by partitioning the 3 × 3 covariance matrix based on the ordered triplets [RT(t), Au(t – τ), Ab(t – τ)], where t is the ‘go’ time and τ = 56 ms (from the fit shown in Fig. 2). The partitioning effectively factors out the contribution of the potentially confounding variable (for example, Au(t – τ) for data obtained with the bimodal schedule) on the correlation between the other two variables50.

240

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
For the spatial control experiment (Fig. 5), we also tested for color selectivity by interleaving blocks of trials in which the monkey made saccades to either a green or a blue target. Only 2 of 18 neurons responded significantly more to one color. The mean absolute response difference between the two colors was 3.6 spikes s–1 (s.d. = 4.0, which is >0.75 times the mean, consistent with the expected distribution of absolute values under hypothesis H0: no difference).
ACKNOWLEDGMENTS We thank M. Mihali and L. Jasinski for technical assistance, and T. Yang, T. Hanks, M. Leon and J. Palmer for helpful comments on the manuscript. Work was supported by Howard Hughes Medical Institute, the International Human Frontiers Science Program Organization, the Fonds voor Wetenschappelijk Onderzoek Vlaanderen, the National Center for Research Resources (RR00166) and the National Eye Institute (EY11378). COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 17 September; accepted 20 December 2004 Published online at http://www.nature.com/natureneuroscience/
1. Gallistel, C.R. & Gibbon, J. Time, rate and conditioning. Psychol. Rev. 107, 289–344 (2000). 2. Mauk, M.D. & Buonomano, D.V. The neural basis of temporal processing. Annu. Rev. Neurosci. 27, 307–340 (2004). 3. Meck, W.H. Internal clock and reward pathways share physiologically similar informationprocessing pathways. in Quantitative Analyses of Behavior: Biological Determinants of Reinforcement vol. 7 (eds. Commons, M.L., Church, R.M., Stellar, J.R. & Wagner, A.R.) 121–138 (Erlbaum, Hillsdale, New Jersey, USA, 1988). 4. Coull, J.T. & Nobre, A.C. Where and when to pay attention: the neural systems for directing attention to spatial locations and to time intervals as revealed by both PET and fMRI. J. Neurosci. 18, 7426–7435 (1998). 5. Leon, M.I. & Shadlen, M.N. Representation of time by neurons in the posterior parietal cortex of the macaque. Neuron 38, 317–327 (2003). 6. Onoe, H. et al. Cortical networks recruited for time perception: a monkey positron emission tomography (PET) study. Neuroimage 13, 37–45 (2001). 7. Funahashi, S., Chafee, M.V. & Goldman-Rakic, P.S. Prefrontal neuronal activity in rhesus monkeys performing a delayed anti-saccade task. Nature 365, 753–756 (1993). 8. Tanji, J. & Hoshi, E. Behavioral planning in the prefrontal cortex. Curr. Opin. Neurobiol. 11, 164–170 (2001). 9. Andersen, R.A. & Buneo, C.A. Intentional maps in posterior parietal cortex. Annu. Rev. Neurosci. 25, 189–220 (2002). 10. Shadlen, M.N. & Newsome, W.T. Neural basis of a perceptual decision in the parietal cortex (area LIP) of the rhesus monkey. J. Neurophysiol. 86, 1916–1936 (2001). 11. Schall, J.D. Neural correlates of decision processes: neural and mental chronometry. Curr. Opin. Neurobiol. 13, 182–186 (2003). 12. Platt, M.L. & Glimcher, P.W. Neural correlates of decision variables in parietal cortex. Nature 400, 233–238 (1999). 13. Karlin, L. Development of readiness to respond during short foreperiods. J. Exp. Psychol. 72, 505–509 (1966). 14. Luce, R.D. Response Times: Their Role in Inferring Elementary Mental Organization (Oxford Univ. Press, New York, 1986). 15. Schall, J.S. & Hanes, D.P. Saccade latency in context: regulation of gaze behavior by supplementary eye field. Behav. Brain Sci. 16, 588–589 (1993). 16. Basso, M.E. & Wurtz, R.H. Modulation of neuronal activity by target uncertainty. Nature 389, 66–69 (1997). 17. Dorris, M.C. & Munoz, D.P. Saccadic probability influences motor preparation signals and time to saccadic initiation. J. Neurosci. 18, 7015–7026 (1998). 18. Niki, H. & Watanabe, M. Prefrontal and cingulate unit activity during timing behavior in the monkey. Brain Res. 171, 213–224 (1979). 19. Chafee, M.V. & Goldman-Rakic, P.S. Matching patterns of activity in primate prefrontal area 8a and parietal area 7ip neurons during a spatial working memory task. J. Neurophysiol. 79, 2919–2940 (1998). 20. Brody, C.D., Hernandez, A., Zainos, A. & Romo, R. Timing and neural encoding of somatosensory parametric working memory in macaque prefrontal cortex. Cereb. Cortex 13, 1196–1207 (2003). 21. Reutimann, J., Yakovlev, V., Fusi, S. & Senn, W. Climbing neuronal activity as an event-based cortical representation of time. J. Neurosci. 24, 3295–3303 (2004). 22. Durstewitz, D. Neural representation of interval time. Neuroreport 15, 745–747 (2004). 23. Loveless, N.E. & Sanford, A.J. Slow potential correlates of preparatory set. Biol. Psychol. 1, 303–314 (1974). 24. Bracewell, R.M., Mazzoni, P., Barash, S. & Andersen, R.A. Motor intention activity in the macaque’s lateral intraparietal area. II. Changes of motor plan. J. Neurophysiol. 76, 1457–1464 (1996). 25. Gnadt, J.W. & Andersen, R.A. Memory related motor planning activity in posterior parietal cortex of macaque. Exp. Brain Res. 70, 216–220 (1988). 26. Colby, C.L. & Goldberg, M.E. Space and attention in parietal cortex. Annu. Rev. Neurosci. 22, 319–349 (1999). 27. Lewis, J.W. & Van Essen, D.C. Mapping of architectonic subdivisions in the macaque monkey, with emphasis on parieto-occipital cortex. J. Comp. Neurol. 428, 79–111 (2000). 28. Bisley, J.W. & Goldberg, M.E. Neuronal activity in the lateral parietal area and spatial attention. Science 299, 81–86 (2003). 29. Ghose, G.M. & Maunsell, J.H.R. Attentional modulation in visual cortex depends on task timing. Nature 419, 616–620 (2002). 30. Gibbon, J., Malapani, C., Dale, C.L. & Gallistel, C.R. Toward a neurobiology of temporal cognition: advances and challenges. Curr. Opin. Neurobiol. 7, 170–184 (1997). 31. Rizzolatti, G., Riggio, L., Dascola, I. & Umilta, C. Reorienting attention across the horizontal and vertical meridians: evidence in favor of a premotor theory of attention. Neuropsychologia 25, 31–40 (1987). 32. Moore, T. & Armstrong, K.M. Selective gating of visual signals by microstimulation of frontal cortex. Nature 421, 370–373 (2003). 33. Kowler, E., Anderson, E., Dosher, B. & Blaser, E. The role of attention in the programming of saccades. Vision Res. 35, 1897–1916 (1995). 34. Rao, S.M., Mayer, A.R. & Harrington, D.L. The evolution of brain activation during temporal processing. Nat. Neurosci. 4, 317–323 (2001). 35. Ivry, R.B. & Spencer, R.M.C. The neural representation of time. Curr. Opin. Neurobiol. 14, 225–232 (2004). 36. Ferraina, S., Pare, M. & Wurtz, R. Comparison of cortico-cortical and cortico-collicular signals for the generation of saccadic eye movements. J. Neurophysiol. 87, 845–858 (2002). 37. Pare, M. & Wurtz, R.H. Monkey posterior parietal cortex neurons antidromically activated from superior colliculus. J. Neurophysiol. 78, 3493–3497 (1997). 38. Dorris, M.C., Pare, M. & Munoz, D.P. Neuronal activity in monkey superior colliculus related to the initiation of saccadic eye movements. J. Neurosci. 17, 8566–8579 (1997). 39. Everling, S. & Munoz, D.P. Neuronal correlates for preparatory set associated with prosaccades and anti-saccades in the primate frontal eye field. J. Neurosci. 20, 387–400 (2000). 40. Eskandar, E.N. & Assad, J.A. Dissociation of visual, motor and predictive signals in parietal cortex during visual guidance. Nat. Neurosci. 2, 88–93 (1999). 41. Coe, B., Tomihara, K., Matzuzawa, M. & Hikosaka, O. Visual and anticipatory bias in three cortical eye fields of the monkey during an adaptive decision-making task. J. Neurosci. 22, 5081–5090 (2002). 42. Shadlen, M.N. & Newsome, W.T. Motion perception: seeing and deciding. Proc. Natl. Acad. Sci. USA 93, 628–633 (1996). 43. Roux, S., Coulmance, M. & Riehle, A. Context-related representation of timing processes in monkey motor cortex. Eur. J. Neurosci. 18, 1011–1016 (2003). 44. Coull, J.T., Vidal, F., Nazarian, B. & Macar, F. Functional anatomy of the attentional modulation of time estimation. Science 303, 1506–1508 (2004). 45. Hikosaka, O. & Wurtz, R. Visual and oculomotor functions of monkey substantia nigra pars reticulata. III. Memory-contingent visual and saccade responses. J. Neurophysiol. 49, 1268–1284 (1983). 46. Toth, L.J. & Assad, J.A. Dynamic coding of behaviourally relevant stimuli in parietal cortex. Nature 415, 165–168 (2002). 47. Colby, C. L., Duhamel, J.-R., & Goldberg, M. E. Visual, presaccadic, and cognitive activation of single neurons in monkey lateral intraparietal area. J. Neurophysiol. 76, 2841–2852 (1996). 48. Meeker, W.Q. & Escobar, L.A. Statistical Methods for Reliability Data (Wiley, New York, 1998). 49. Rakitin, B.C. et al. Scalar expectancy theory and peak-interval timing in humans. J. Exp. Psychol. Anim. Behav. Process. 24, 15–33 (1998). 50. Anderson, T.W. An Introduction to Multivariate Statistical Analysis edn. 2 (Wiley, New York, 1984).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

241

CO R R I G E N DA

Corrigendum: High frequency, synchronized bursting drives eye-specific segregation of retinogeniculate projections
Christine L Torborg, Kristi A Hansen & Marla B Feller Nat. Neurosci. 8, 72-78 (2005) A sentence in the discussion of this paper contained some inaccurate references. © 2005 Nature Publishing Group http://www.nature.com/natureneuroscience The second sentence (first column) on page 76 should read “First, local infusion of general nAChR antagonists directly into the dLGN during the first postnatal week does not prevent eye-specific segregation2”. The authors regret the error.

Corrigendum: A representation of the hazard rate of elapsed time in macaque area LIP
Peter Janssen & Michael N Shadlen Nat. Neurosci. 8, 234-241 (2005) In the Methods section, equation 4 contained an error. The corrected version should read as follows:










The authors regret the error.

396

VOLUME 8 | NUMBER 3 | MARCH 2005 NATURE NEUROSCIENCE

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Prefrontal white matter volume is disproportionately larger in humans than in other primates
P Thomas Schoenemann, Michael J Sheehan & L Daniel Glotzer
Determining how the human brain differs from nonhuman primate brains is central to understanding human behavioral evolution. There is currently dispute over whether the prefrontal cortex, which mediates evolutionarily interesting behaviors, has increased disproportionately. Using magnetic resonance imaging brain scans from 11 primate species, we measured gray, white and total volumes for both prefrontal and the entire cerebrum on each specimen (n = 46). In relative terms, prefrontal white matter shows the largest difference between human and nonhuman, whereas gray matter shows no significant difference. This suggests that connectional elaboration (as gauged by white matter volume) played a key role in human brain evolution.

Although the human brain is around three times larger than expected for a primate of our body size, it does not seem to be simply a scaledup version of a primate brain1. Because neural tissue is evolutionarily expensive (for metabolic2 and maturational3 reasons), changes in relative proportions in different parts of the brain are likely to be behaviorally adaptive. Thus, determining the various ways in which the human brain is different from nonhuman primate brains is of central importance to understanding human evolution. It is clear that at least some areas of the human brain are proportionately smaller than predicted based on primate scaling trends. For example, the human olfactory bulb is only ∼30% as large and Brodmann's area 17 (primary visual cortex) only ∼60% as large as predicted for a primate brain our size4,5. Given that the entire human brain is much larger than predicted overall, at least some areas must therefore be significantly larger than predicted. One area of particular interest for human evolution is the prefrontal cortex, which mediates such important behaviors as planning6, working memory7 and memory for serial order and temporal information8, aspects of language (Broca’s area and symbolic behavior1,9,10), attention11 and social information processing12. To the extent that the human prefrontal cortex is disproportionately large, it would suggest that some combination of these behavioral dimensions were particularly important to our evolutionary history. Comparative studies of the entire frontal cortex (of which the prefrontal cortex is a subcomponent) have not reported disproportionate increases13–16. However, because some data suggest that portions of the human frontal cortex are smaller than primate data predict, other portions must necessarily be larger. A cytoarchitectural study of seven primate genera (Homo, Pan, Pongo, Hylobates, Papio, Cercopithecus, Callithrix)17 has suggested that primary motor and premotor areas of the frontal cortex (Brodmann’s areas 4 and 6, respectively) occupy a much smaller proportion of the cortex in humans than in primates1.

However, a magnetic resonance imaging (MRI) study of the relative size of the precentral gyrus in humans compared to five hominoid species (Pan paniscus, Pan troglodytes, Pongo pygmaeus, Gorilla gorilla, Hylobates lar) reported no significant difference15. Because the precentral gyrus includes most of area 4 but only a portion of area 6 (ref. 18), these studies are not necessarily contradictory. Comparative studies focusing specifically on the prefrontal cortex itself have come to conflicting conclusions regarding its relative size in humans: some suggest substantial disproportionate increases17,19–22, whereas others suggest much more moderate increases, if any23,24. Methodological differences may explain these disagreements. Many have focused on cortical gray matter exclusively, using cytoarchitectural criteria to define areas17,19,21,25, sometimes also incorporating thalamic projection patterns24. Only one study has used MRI to estimate the volume of the prefrontal cortex itself, though the analysis was limited to female specimens of just two species: Homo sapiens and Papio cynocephalus, thereby precluding allometric analysis23. Other studies have used indices of the gyrification (degree of folding) of the cortex, measured on coronal sections, as a proxy for cortical surface area20,22. These studies have suggested disproportionate prefrontal increases. One component that has not been reported in previous comparative studies is the volume of white matter underlying prefrontal areas. This is potentially of great interest, because the executive role played by the prefrontal cortex depends critically on its connections to posterior processing regions. The prefrontal is known to have extensive reciprocal connections with the diencephalon, mesencephalon and limbic system, as well as numerous cortical areas that mediate higher sensory functions26. Overall, cortical white matter increases disproportionately with increasing brain size across mammals, though apparently not enough to maintain equal degrees of connectivity between existing regions27. Thus, quantifying white matter in prefrontal areas is an important goal.

Department of Anthropology, University of Pennsylvania, 3260 South St., Philadelphia, Pennsylvania 19104-6398, USA. Correspondence should be addressed to P.T.S. ([email protected]). Published online 23 January 2005; doi:10.1038/nn1394

242

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
Because it is difficult to delimit the prefrontal cortex unambiguously using gross sulcal landmarks, it has been argued that definitive comparative quantitative analysis will require extensive detailed cytoarchitectural studies that, because of their expense, are not likely to be carried out in the near future15. However, a reasonable proxy for the prefrontal cortex can be defined as all portions of the frontal cortex anterior to the genu of the corpus callosum, in a plane perpendicular to the line connecting the anterior and posterior commissures. A consideration of primate cytoarchitectural maps19 indicates that this method will actually underestimate human prefrontal size relative to that of other primates. Human cytoarchitectural maps (for example, Figures 49 and 50 of ref.19) suggest that a substantial amount of prefrontal cortex extends

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience
Figure 1 Three-dimensional renderings, from left lateral, anterior, and superior views (in correct left-right orientation as shown), indicating the cortical portions designated prefrontal in this study for a representative of each species. Images have been rescaled to approximately the same size for comparison. (a) Homo sapiens. (b) Pan paniscus. (c) Pan troglodytes. (d) Gorilla gorilla. (e) Pongo pygmaeus. (f) Hylobates lar. (g) Cercocebus torquatus atys. (h) Papio cynocephalus. (i) Macaca mulatta. (j) Cebus apella. (k) Saimiri sciureus.
NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

243

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 2 Percentage of cerebral volume that is prefrontal for individual specimens in 11 primate species. (a) Total (gray plus white) volume. (b) Gray volume. (c) White volume.

further posterior than the most anterior point of the corpus callosum, whereas nonhuman primate maps (for example, of gibbon (Hylobates lar) in Figures 44 and 45 and marmoset (Hapale jacchus) in Figures 33 and 34 of ref. 19) suggest that this method overestimates prefrontal cortex superiorly and underestimates prefrontal cortex inferiorly to approximately the same extent. Thus, defining prefrontal cortex in this way will result in a conservative estimate of any differences between humans and nonhuman primates that might be found. Furthermore, this method benefits from the fact that it can be objectively applied to MRI scans of different species, which avoids problems of postmortem tissue shrinkage found in cytoarchitectural studies of cadaver specimens15,16,22. MRI scans also have excellent gray-white differentiation, allowing for the measurement of white matter volumes, and also allow for the estimation of volume using standard stereological techniques28. Using these criteria, gray matter, white matter, and total volumes for both prefrontal and total cortex were measured on 46 high-resolution MRI scans of individuals from 11 primate species: 12 Homo sapiens (6 male (M), 6 female (F)), 4 Pan paniscus (3 M, 1 F), 6 Pan troglodytes (3 M, 3 F), 2 Gorilla gorilla (1 M, 1 F), 4 Pongo pygmaeus (3 M, 1 F), 2 Hylobates lar (1 M, 1 F), 4 Cercocebus torquatus atys (3 M, 1 F), 2 Papio cynocephalus (2 M), 3 Macaca mulatta (3 M), 3 Cebus apella (2 M, 1 F) and 4 Saimiri sciureus (3 M, 1 F). Figure 1 shows the regions designated prefrontal using these criteria on three-dimensional renderings of a sample specimen from each of the species used in the present study. The effect of species differences in voxel size, ratio of females to males and subtle differences in image quality were assessed and shown to be unlikely to affect the conclusions about humans and nonhuman primates presented here (see Methods). In this dataset the amount of human prefrontal cerebral volume anterior to the corpus callosum is disproportionately large on average, with the greatest human–nonhuman difference evident for the proportion of white matter that is prefrontal (‘prefrontal percentage white’). Though humans do not seem to differ substantially from other hominoids in the proportion of gray matter that is prefrontal (‘prefrontal percentage gray’), hominoids as a group differ significantly from nonhominoid primates (which were all monkey species in the present study). Although positive allometry is evident, average prefrontal white matter volume in humans nevertheless slightly but significantly exceeds the volume predicted from primate trends based on the size of non-prefrontal cerebral white matter volume. The absolute amounts of extra volume predicted by our analysis are substantial, being close to the size of an entire chimpanzee prefrontal cortex for total (gray plus white) and white matter volume measures.

RESULTS Prefrontal percentages We measured total, gray and white volumes of both prefrontal and total cerebral portions for all specimens (Table 1). We calculated the proportion of total cerebrum that was prefrontal (prefrontal percentage) for gray, white and total (gray plus white) volumes (data for each species are shown in Table 2 and data from individual specimens is shown in Fig. 2). Average percentages were larger in Homo sapiens than all other primate species for each prefrontal measurement, consistent with a subjective assessment of the individual images (from Fig. 1). An ANOVA using Dunnett’s method with Homo sapiens as the control group showed that human percentage prefrontal was significantly larger than for all except Gorilla gorilla (P = 0.25) for total cerebral volume. The effect seems to be almost entirely due to white matter: prefrontal percentage white in Homo sapiens was significantly different from all species except Gorilla gorilla (P = 0.14) and Cebus apella (P = 0.11), whereas prefrontal percentage gray differed significantly only from Macaca mulatta, Cercocebus torquatus atys, Cebus apella and Saimiri sciureus. The lack of statistically significant

Figure 3 Relationship between the proportion of total white matter volume anterior to the corpus callosum (prefrontal percentage white) compared with proportion of total gray matter anterior to the corpus callosum (prefrontal percentage gray). The solid line represents least-squares regression based on nonhuman species average values (prefrontal percentage white volume = 4.794 + 0.212 (prefrontal percentage gray volume)) and the dotted lines represent the 95% confidence intervals.

244

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
differences between Homo sapiens and Gorilla gorilla may partly be due to the fact that only two Gorilla brains were available. When nonhuman primate individuals were pooled into the taxonomic categories of Hominoidea (apes: Pan Paniscus, Pan troglodytes, Gorilla gorilla, Pongo pygmaeus, Hylobates lar), Cercopithecoidea (Old World Monkeys: Cercocebus turquatus atys, Papio cynocephalus, Macaca mulatta) and Platyrrhini (New World Monkeys: Cebus apella, Saimiri sciureus), Homo sapiens was significantly different from these groups in prefrontal percentage total cerebrum and prefrontal percentage white (ANOVA, Dunnett’s method with Homo sapiens as control, P < 0.0001 for all comparisons). Although Homo sapiens did not differ significantly from the other hominoid genera for prefrontal percentage gray, both of these groups differed significantly from cercopithecoids and platyrrhines on this measure (ANOVA with Tukey-Kramer HSD). These differences in prefrontal percentage translated into large absolute differences. Using the average pongid proportions (Table 2) to calculate predicted absolute values for the different components in humans, Homo sapiens had 25.7 ml (23.6%) more prefrontal total volume, 5.3 ml (7.3%) more gray matter and 17.0 ml (42.5%) more white matter than would be expected for a pongid brain of its size (ignoring allometric considerations— which are discussed below—for the moment). To put these values into perspective, we note that the entire estimated prefrontal cortex for Pan paniscus averaged 25.4 ml total volume, 15.4 ml gray matter and 10.0 ml white matter (Table 1). Although the gray volume difference was relatively small, using published neuronal density values for area 10 of the prefrontal cortex in humans21, the extra prefrontal gray matter in humans suggests an additional ∼180 million neurons beyond that predicted by cortical size difference alone. Another way to assess these differences is to compare how much larger human values are from Pan species for prefrontal compared to non-prefrontal portions. The total non-prefrontal cerebral volume in humans averaged 3.7 times larger than the average of Pan paniscus and Pan troglodytes, but the prefrontal portion averaged 4.9 times larger. The pattern for gray and white volumes was consistent with the general findings above: non-prefrontal gray volume was 4.2 times larger in humans, and the prefrontal portion was 4.8 times larger; yet non-prefrontal white volume was 3.3 times larger, whereas the prefrontal portion was 5.0 times larger. As delineated in this study, white and gray matter volume differences were not distributed equally in prefrontal compared with non-prefrontal regions.
Table 1 Total, gray and white volumes of both prefrontal and total cerebral portions for all specimens studied
Species/individual specimen identifiers (sex) Whole cerebral volume (mm3) Total Homo sapiens 7830 (M) 7897 (M) 7933 (M) 7972 (M) 8201 (M) 8574 (M) r16 (F) r27 (F) r30 (F) r62 (F) r64 (F) r85 (F) Averagea Pan paniscus Bo (M) Brian (M) Lorel (M) Jill (F) Averagea Pan troglodytes Merv (M) Laz (M) Jimmy Carter (M) Mary (F) Lulu (F) Kengee (F) Averagea Gorilla gorilla Kekla (M) Kinyani (F) Averagea Pongo pygmaeus Mentubar (M) Minyak (M) Molek (M) Hati (F) Averagea Hylobates lar Buddy (M) Cleo (F) Averagea Cercocebus turquatus atys FSO (M) FWJ (M) FYF (M) FFK (F) Averagea 77,990 87,131 85,664 81,394 82,495 41,589 37,113 36,106 37,701 37,985 36,401 50,018 49,558 43,694 44,510 6,267 5,604 7,841 6,627 6,599 4,503 2,887 3,623 3,937 3,804 1,764 2,716 4,218 2,689 2,794 66,450 65,426 65,938 31,796 33,040 32,418 34,654 32,386 33,520 5,908 6,696 6,302 3,537 4,452 3,994 2,371 2,244 2,307 339,266 365,674 424,841 289,556 333,075 162,421 164,367 199,384 133,196 154,293 176,845 201,308 225,457 156,360 178,781 39,067 39,930 46,176 25,802 33,763 24,338 23,791 28,429 16,866 21,193 14,728 16,139 17,747 8,936 12,570 377,732 401,630 389,681 165,004 176,006 170,505 212,728 225,624 219,176 38,470 44,839 41,655 21,297 25,542 23,419 17,173 19,297 18,235 334,724 237,988 263,789 271,766 248,056 310,735 277,843 135,354 111,072 130,447 112,168 102,013 152,098 123,858 199,370 126,917 133,343 159,599 146,042 158,638 153,985 44,385 20,878 28,507 23,006 24,556 40,613 30,324 23,368 10,540 19,051 11,087 14,966 24,389 17,234 21,017 10,338 9,456 11,919 9,590 16,223 13,091 240,699 261,576 384,535 250,071 272,837 108,102 125,128 163,587 118,666 125,469 132,597 136,448 220,948 131,405 147,368 22,014 22,027 35,759 24,149 25,374 11,806 14,339 20,098 15,349 15,382 10,207 7,688 15,660 8,800 9,993 1,206,129 1,014,949 1,157,341 1,085,317 1,185,223 1,018,913 999,296 915,355 1,187,709 840,174 1,038,834 1,009,091 1,054,861 535,956 479,788 539,275 501,414 551,192 436,578 537,146 518,469 698,689 476,431 581,307 588,281 537,044 670,174 535,161 618,066 583,903 634,031 582,335 462,150 396,885 489,020 363,744 457,527 420,810 517,817 156,443 151,317 145,393 141,237 151,090 121,015 116,587 102,242 149,914 104,516 127,895 144,950 134,383 82,346 78,913 80,649 74,359 77,917 53,490 72,744 68,217 98,880 68,422 81,350 92,849 77,511 74,097 72,404 64,744 66,878 73,174 67,525 43,843 34,025 51,033 36,094 46,546 52,101 56,872 Gray White Prefrontal cerebral volume (mm3) Total Gray White

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

(continued)

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

245

ARTICLES
ter volume. Using these regressions, human total prefrontal volume averaged 20.7 ml (18%) larger, gray matter averaged Species/individual Prefrontal cerebral volume (mm3) specimen Whole cerebral volume (mm3) 2.1 ml (3%) smaller, and white matter averidentifiers (sex) aged 16.5 ml (41%) larger than predicted. Total Gray White Total Gray White Note that these estimates are very close to Papio cynocephalus those obtained above simply using average Boon1 (M) 108,135 50,685 57,450 11,427 5,962 5,466 pongid prefrontal percentage value for white volume and are only slightly smaller for total Boon2 (M) 153,065 66,546 86,519 11,892 7,098 4,794 volume. The gray volume estimates were small Average 130,600 58,615 71,985 11,660 6,530 5,130 when obtained using pongid prefrontal perMacaca mulatta centage predictions, and become slightly neg153C (M) 63,202 28,771 34,430 5,094 2,555 2,540 ative using allometric predictions. Prefrontal Reg3 (M) 71,028 33,323 37,705 6,460 3,103 3,357 white matter volume in humans was signifiRue-1 (M) 75,395 33,430 41,965 5,183 2,781 2,402 cantly larger than predicted (based on nonAverage 69,875 31,841 38,034 5,579 2,813 2,766 human species average values, one-tailed Cebus apella P = 0.03). The other measures were not sigAndy (M) 63,315 27,982 35,333 6,474 3,464 3,009 nificantly larger, though it is important to Vincent (M) 66,571 37,495 29,075 6,474 4,068 2,406 reiterate that our method for delineating the Binkey (F) 52,219 25,977 26,242 5,196 2,820 2,377 prefrontal is highly conservative. Thus, these Averagea 58,581 29,358 29,223 5,835 3,293 2,542 values should be seen as minimum estimates Saimiri sciureus for the entire prefrontal cortex. There was a large degree of individual variS104 (M) 20,191 8,548 11,644 1,424 748 677 ation within species, including Homo sapiens S105 (M) 21,818 9,069 12,750 1,665 837 828 (Figs. 2 and 4). For total prefrontal volumes, Squirrel1 (M) 22,279 10,551 11,728 2,152 1,326 826 human individual residual values ranged from Squirrel2 (F) 21,654 10,181 11,472 1,894 1,217 677 2.9 ml to 45.2 ml (3% to 43%) larger; for gray Averagea 21,542 9,785 11,756 1,820 1,094 727 volumes, from 11.5 ml (18%) smaller to aAverage of male and female means. 10.5 ml (15%) larger; and for white volumes, from 2.7 ml to 31.9 ml (9% to 79%) larger. Overall, the data presented here suggest that gray and white matter Although six individual human values fell below the upper confidence proportions are not tightly constrained by each other over evolutionary intervals for prefrontal white volume, all individual human values nevtime. The association between prefrontal percentage white and prefron- ertheless fell above the regression based upon species means (only tal percentage gray was very weak: r2 = 0.16 (Fig. 3; adjusted r2 = 0.05, Cebus apella, n = 3, and Gorilla gorilla, n = 2, showed this pattern n = 10 nonhuman primate species averages, P > 0.25). Nevertheless, among nonhuman primates). All individual human values also fell the human prefrontal percentage white value still fell outside the 95% above the regressions for total prefrontal volume (also true only for confidence intervals for predicting prefrontal percentage white from Cebus apella and Gorilla gorilla among nonhuman primates), though prefrontal percentage gray, an additional indication that human pre- one individual fell very close to the line. frontal white is disproportionately large. DISCUSSION The difference found in the proportion of prefrontal white to gray Prefrontal scaling relationships If prefrontal values showed positive allometry, the relatively high matter is not necessarily in conflict with an earlier finding that total human proportion might nevertheless be expected because of the cerebral white matter is predicted by total neocortical gray matter across large overall size of the human brain (this says nothing, however, primates22. This same study also reported that gyrification (degree of about the possible behavioral relevance of any difference; see below). folding) in the prefrontal cortex is significantly larger in humans than Positive allometry has been shown for the frontal cortex as a whole16 predicted, even though overall gyrification of the cortex as a whole is as well as for Brodmann’s area 10 within the prefrontal21,29. Plots of not22. A tight relationship between overall gray and white volumes for the relationships between prefrontal measures against corresponding the whole cerebrum does not require a uniform relationship within non-prefrontal measures are shown in Figure 4 (all values log- all subregions. transformed). The lines represent least-squares regressions calculated This difference between proportions of prefrontal white versus gray on the nonhuman species averages (of male and female means), matter has important implications for neural development. Neural although individual values are plotted to allow qualitative assessment Darwinist accounts of connectional development suggest that early of the spread within species. All of the regressions had slopes greater neuronal proliferation in some areas biases connection patterns to and than 1, though these were significant only for total volume and gray from those areas, thereby substantially biasing functional processing1. To volume. (One-tailed probabilities: total volume, P = 0.02; gray volume, the extent that gray matter volume is a proxy for neuronal proliferation, P = 0.02; white volume, P = 0.12. Including Homo sapiens in the such a model predicts that gray matter increases should go hand-in-hand calculations made all slopes significantly greater than 1.) This sug- with white matter increases in elaborated regions. Yet the present research gests positive allometry for prefrontal cerebral volume anterior to the suggests that connectional patterns themselves can vary independently corpus callosum in primates. of such neuronal proliferation and may not simply be the result of them. The average human values fell above the regression line for At the very least, it suggests that connectional biases (as gauged by white total volume and white matter volume but not for gray mat- matter distributional patterns) may be more evident than differences
Table 1 Total, gray and white volumes of both prefrontal and total cerebral portions for all specimens studied (continued)

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

246

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
Table 2 Average percentage of cortical measure designated prefrontal

Total volume Species N Mean s.e.m.

Gray volume Mean s.e.m.

Homo sapiens

12

12.7

0.3

14.4

0.3

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Pan paniscus Pan troglodytes Gorilla gorilla Pongo pygmaeus Pongid average

4 6 2 4 4.0

9.1 10.9 10.7 10.6 10.3

0.3 0.8 0.5 0.6 0.5

11.9 14.1 13.7 14.1 13.4

0.4 1.1 0.8 0.5 0.7

Hylobates lar

2

9.6

0.7

12.3

1.2

Cercocebus torquatus atys Papio cynocephalus Macaca mulatta Cebus apella Saimiri sciureus Cercopithecoid and ceboid average

4 2 3 3 4 3.2

7.9 9.2 8.0 10.0 8.3 8.7

0.6 1.4 0.6 0.1 0.6 0.7

9.8 11.2 8.8 11.4 10.6 10.4

0.7 0.5 0.3 0.5 1.0 0.6

in neuronal proliferation in certain cortical areas. Recent advances in diffusion tensor MRI30 open the possibility of detailed quantitative analysis of white matter tract patterns across species. Determining the specific behavioral implications of the prefrontal elaboration during human evolution is of great interest. One notable unresolved question is exactly which neuroanatomical measure is the most behaviorally relevant, particularly for understanding human evolution. Should we ascribe more importance to the percentage of cerebral volume that is prefrontal, or the absolute excess prefrontal over average ape value (as shown above), or the residual (either absolute or relative) from primate expectations, or some other quantitative measure? A conclusion on this question cannot be made a priori, but only empirically, and may well vary for different behavioral domains. Having absolutely more cerebral volume in a region might well have

important behavioral implications, regardless of allometric scaling trends31. To answer this question (if it is indeed answerable), it is White volume necessary to determine, for a given behavioral Mean s.e.m. domain or task, which neuroanatomical measure best predicts species differences in these 10.9 0.4 behaviors. It has long been known that species 6.8 0.4 differences in behavioral abilities are gener8.3 0.7 ally reflected in differences in their cortical maps. Echolocating bats have greatly expanded 8.3 0.2 auditory cortical regions (accounting for 7.5 0.6 more than half the cortex of the ghost bat, 7.7 0.5 Macroderma gigas, for example), whereas primarily subterranean mole species have mark6.9 0.0 edly reduced visual cortical areas32. Within humans, variation in prefrontal cortex size has 6.2 0.8 been found to be positively associated with at 7.5 2.0 least one cognitive task known to be mediated 7.3 0.9 by the prefrontal: the Stroop test, which tests 8.6 0.2 the ability to extract and focus on relevant cues 6.3 0.3 in the face of distractors33. Gray matter differences in the frontal lobe have been shown to be 7.2 0.8 correlated with general cognitive ability, or g34 (although this study did not control for possible between-family confounds33). Thus, it is a reasonable starting assumption that the elaboration of the prefrontal cortex did in fact have behavioral implications during human evolution, specifically involving the increased importance of the kinds of behaviors mediated by the prefrontal. Given the general executive role of the prefrontal cortex, connections to posterior regions, to regions within the prefrontal or to both are clearly essential. As increased brain size seems to be strongly correlated with an increased number of distinct cortical areas35, one would expect connectivity to and from the prefrontal to be particularly enhanced in humans. Such an effect would explain positive allometry of prefrontal scaling with respect to the rest of the cerebrum. That human white matter exceeds the amount predicted allometrically, however, suggests that some additional explanation is necessary. One obvious possibility is the evolution of human language, which is more complex

Figure 4 Relationships between prefrontal and non-prefrontal cerebral volume for all 46 specimens from 11 primate species. The mean value for Homo sapiens is indicated by the black dot. The lines represent least-squares regressions based on nonhuman species average values: (a) total (gray + white) cerebral volume; (log prefrontal cerebral volume) = –1.413 + 1.085 × (log non-prefrontal cerebral volume); (b) cerebral gray volume; (log prefrontal cerebral gray volume) = –1.430 + 1.118 × (log non-prefrontal cerebral gray volume); (c) cerebral white volume; (log prefrontal cerebral white volume) = –1.367 + 1.055 × (log non-prefrontal cerebral white volume). All relationships: r 2 = 0.99, P < 0.0001.

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

247

ARTICLES

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience
Figure 5 Segmentation examples. Two coronal (above) and two transverse (below) slices are shown for each species (scaled approximately to the same size for comparison). The leftmost coronal and transverse images show the original grayscale values; the rightmost show the corresponding segmented versions (dark gray, CSF; medium gray, gray matter; white, white matter). (a) Homo sapiens. (b) Pan paniscus. (c) Pan troglodytes. (d) Gorilla gorilla. (e) Pongo pygmaeus. (f) Hylobates lar. (g) Cercocebus torquatus atys. (h) Papio cynocephalus. (i) Macaca mulatta. (j) Cebus apella. (k) Saimiri sciureus.

and functional than any communication system seen in primates. In this regard it is important to note that an anterior portion of Broca’s area is probably included in our delineation of the prefrontal (Fig. 1). However, a variety of anterior prefrontal regions in addition to Broca’s area have been implicated in critical aspects of language processing, particularly those involving semantic information1,9,10. Such processing is clearly central to language. There are a number of models of natural language that specifically emphasize the importance of semantics, and it has been argued that the evolution of semantic and conceptual complexity is likely to have been the engine driving the evolution of language generally, and grammar and syntax specifically36,37. In addition, the importance of interconnectivity between the prefrontal cortex and numerous other brain areas has been emphasized in discussions of language evolution. Specifically,

posterior cortical regions (particularly inferior parietal and temporal areas), basal ganglia, thalamus, midbrain, cerebellum and brainstem have all been implicated 1,38,39. Other behavioral domains might also have had important roles in the increase in size of prefrontal white matter during human evolution. Brain size across primates correlates strongly with typical size of the social group, and human social group sizes are on average larger than those found in other living hominoid species40. This suggests social group size increased during human evolution. If so, it would have placed a premium on social information processing in general41 (of which language is likely a special case40). Other behaviors mediated by the prefrontal cortex (such as planning, working memory and attention) would also have likely been useful for an increasingly complex social life.

248

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
Homo sapiens brain scans were obtained from healthy volunteers who had given informed written consent (approval was obtained from the institutional review boards of the University of California at San Francisco and the University of Pennsylvania). The female scans were from a previously reported sample33. These scans were T1-weighted, with TR = 32 ms, TE = 8 ms, with in-plane resolution of 0.94 mm2. Female scans had slice thickness of 1.5 mm; male scans had slice thickness of 0.99 mm. Image processing. Brain scans were processed as follows. (i) Primate scans were corrected for fluctuations in average intensity across slices (which were evident on visual inspection) using mean-based homomorphic filtering (Analyze image processing software (AnalyzeDirect), inhomogeneity correction). Because slice differences in average intensity where not evident on the human scans, they were not filtered at this point. However, another inhomogeneity correction method was applied to all scans (human and nonhuman) during gray-white segmentation (step (iv) below). Any remaining inhomogeneity effects are unlikely to bias our results (see below). (ii) Cerebral portions of the brain were semimanually extracted using standard flood-fill thresholding techniques (Analyze image processing software). Non-brain tissues were removed, followed by cerebellar and brain stem tissues, which were obtained as follows: in coronal view, non-cerebral tissues were removed starting at the most posterior point and proceeding anteriorly until no obvious break was evident between the midbrain and thalamus; then in transverse view, non-cerebral tissues were removed starting at the most inferior slice and proceeding superiorly until no obvious break was evident between midbrain and posterior limb of internal capsule (transition between cerebral peduncle and posterior limb of internal capsule). (iii) Extracted cerebrums were aligned to anterior commissure–posterior commissure orientation in transverse view. (iv) Gray and white matter regions were segmented using FAST (FMRIB’s Automated Segmentation Tool, http://www.fmrib.ox.ac.uk/fsl/). This method implements a hidden Markov random field model and an associated expectation-maximization algorithm to categorize voxels into gray, white and cerebrospinal fluid (CSF) tissue types45. FAST also implements a correction for magnetic field inhomogeneities46. FAST was initialized to categorize nonzero voxels into three tissue types (corresponding to CSF, gray and white matter). Two segmented volumes were created for each specimen: white matter only and white plus gray matter. Although designed for human brain MRI, visual inspection confirmed that FAST works equally well for the nonhuman species in our sample. No gross errors of classification could be found. Example segmentations on two slices (one coronal, one transverse) for a specimen of each species are shown (Fig. 5). Biases potentially caused by possible species differences in image quality (because of different scanning protocols for human and nonhuman scans), voxel size or both do not seem to differentially affect prefrontal as compared with non-prefrontal volume sufficiently to account for the differences found between human and nonhuman specimens (see detailed analysis below). (v) Stereological methods were used to estimate volume of brains in coronal orientation28. Stereology has long been used to accurately estimate brain volumes47. A three-dimensional grid of lines spaced at equal distances running in the all three dimensions is superimposed on the volume, and points where this grid falls on the object of interest (that is, white matter or gray plus white matter) are then counted. Volume (Vest) is proportional to the number of times the intersection points of the grid (that is, points where the grid lines in the x, y and z dimension intersect each other) fall on the object of interest. Specifically, Vest = T(a/p)(Ptotal), where T is the distance between grid intersection points in the z dimension, a/p is the area associated with each point in the x and y dimensions, and Ptotal is the sum of grid intersection points that fall on the object28. The specific algorithms and formulas used to calculate volume were those implemented in Analyze image processing software. We found that volume estimates were relatively insensitive to large changes in grid size, but we nevertheless used a grid size with interstices that approximated 3 mm as closely as possible (given a scan’s voxel dimensions), thereby maintaining grid sizes across species. To ensure an accurate estimate, ten randomly chosen grid alignments for the x dimension were used for each subject for each measurement, and results were averaged (within subject and measurement). Measurements of white matter only and total (white plus gray) volumes were taken for both the total cerebral and prefrontal

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 6 Effect of blurring on segmentation. Two coronal (left) and two transverse (right) slices are shown for each degree of segmentation. The leftmost coronal and transverse images show the original grayscale values; the rightmost show the corresponding segmented versions (dark gray: CSF; medium gray: gray matter; white: white matter). (a) Original image. (b) Blurred version.

An additional possible explanation involves the increasing importance of processing temporal information during human evolution. The prefrontal cortex is known to play an important role in mediating such information8. An understanding of causality is predicated on the ability to remember the temporal order of past events (if one cannot pay attention to the temporal order of events, one cannot reconstruct cause and effect). In addition, prefrontal regions have been shown to have a role in general (‘fluid’) intelligence42, which is thought to be a measure of general problem-solving ability. Both of these abilities (keen understanding of causal relationships and general-problem solving ability, which are likely to be related at some level) would have been useful for effectively navigating an increasingly complex social existence. However, they would also have been critical for the development of elaborate technology, the extensive use of which is another behavioral domain in which humans differ from other animals. It is perhaps of interest that the earliest evidence of stone tool manufacturing (which is likely to demarcate a shift in the degree of focus on technology) occurs at ∼2.5 million years ago43. This coincides with the beginning of hominid brain size expansion as determined from the fossil record44. The timing of expansion of the prefrontal cortex (let alone prefrontal white matter) is not currently known, however. It is important to keep in mind that none of these possible explanations are mutually exclusive. Barring further information, the most prudent position to take is that all of these behavioral domains are likely to have been involved in the evolution of the prefrontal cortex. Further research into the associations between behavioral differences and prefrontal cortex size, or subdivisions of the prefrontal, both within and between species, may help further elucidate possible interpretations of the findings reported here. What is clear is that the prefrontal cortex, and specifically its connections with other cortical areas, seems to have increased disproportionately during human evolution, in contrast to the entire frontal lobe itself. This strongly suggests that the prefrontal cortex played a key role in human behavioral evolution. METHODS
MRI dataset. The primate brain scans were obtained from Yerkes Regional Primate Research Center22. The scans were T1 weighted, TR = 19.0 ms, TE = 8.5 ms; slice thickness varied from 1.2 to 2 mm depending on scan; and in-plane spatial resolution varied from 0.47 to 0.70 mm2.

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

249

ARTICLES
portions. Gray matter volumes were calculated by subtraction (white plus gray minus white matter only). Prefrontal measurements were made on all coronal slices anterior to the corpus callosum (after aligning the brain to the anteriorposterior commissure line). Cerebral white and gray matter volumes defined in this way do not correspond exactly to the measures reported in ref. 22, which focused just on the cortex. The present study included all gray and white matter in the cerebrum (e.g., internal capsule, basal ganglia, thalamus). Analysis of potential sources of bias. Sex differences. Male brains are bigger than female brains in humans and other primates48,49. Because the present dataset contains an uneven distribution of males and females across species (that is, several species are represented by more males than females, and for two (Papio cynocephalus and Macaca mulatta) there were no females), the possibility that this might bias the results was specifically addressed. Two-way ANOVAs using data from species with both male and female specimens (that is, excluding Papio cynocephalus and Macaca mulatta) showed that sex was not a statistically significant factor in predicting prefrontal percentage. This was the case for cortical white matter (effect of sex, P = 0.41; sex-species interaction, P = 0.84), gray matter (effect of sex, P = 0.53; sex-species interaction, P = 0.74) and total (gray plus white) volume (effect of sex, P = 0.94; sex-species interaction, P = 0.81). Consequently, sex was not used as a factor in statistical comparisons of prefrontal percentage measures between species. For inter-species regressions to predict human residuals, the average of male and female means was used (except for Papio cynocephalus and Macaca mulatta, for which only there were male specimens, in which case the average of male specimens only was used). However, separate regressions were also carried out using both male-only and female-only species averages. These within-sex human residual values show the same patterns and relationships seen for the overall data, indicating that the conclusions presented here are not explained by different proportions of males to females in different species. Scan quality. Differences in scanning parameters between human and nonhuman species could conceivably affect the quantification of volume measurements if they resulted in differences in image quality between human and nonhuman scans. As mentioned above, we consider this highly unlikely based on a qualitative assessment of individual segmentations (Fig. 5). However, to assess the extent to which a subtle bias in scan quality might affect our conclusions, a subset of six of the human scans were purposely blurred before segmentation by FAST, and the results were compared to the values obtained with the original non-blurred versions. Blurring was accomplished by successively applying a binomial filter (Analyze) with a 3-pixel × 3-pixel mask ten times. This filter implements a nearest-neighbor average along each volume dimension, which approaches a convolution with a Gaussian as the number of iterations increases. Effectively, this spreads and smoothes the voxel values across the intensity range, thereby reducing the gray-white differentiation and successively degrading the resolution of the image. Qualitatively, this resulted in significantly poorer resolution than we observed in any of our nonhuman scans (compare Fig. 6 to examples in Fig. 5). After segmentation via FAST, tissue volumes were calculated stereologically exactly as in the original non-blurred versions (Fig. 6). The resulting average differences in volume between blurred and original non-blurred versions for these test images are included in Supplementary Table 1 online. Volume estimates vary between –5.8% (blurred versions yielding values 5.8% smaller than the original image) and +8.6% (blurred versions 8.6% larger). Blurring the image tends to make white volumes slightly larger, but makes gray volumes slightly smaller. The critical issue for the present study, however, concerns the extent to which the prefrontal portion is differentially affected by blurring relative to the total cerebral volume. Any biasing effects with respect to differing proportions of gray compared with white will not be critical to our conclusions unless they affect prefrontal regions substantially differently than non-prefrontal regions. For all measurements, blurring affected prefrontal and non-prefrontal portions in the same direction, though not exactly to the same extent (Supplementary Table 1 online). The average differences of the estimates of prefrontal percentage were quite small: +0.24 (s.e.m. = 0.19) absolute percentage points for total volume, +0.87 (s.e.m. = 0.16) for gray volume, and –0.14 (s.e.m. = 0.26) for white volume. Thus, white volume prefrontal percentage estimates tended to be slightly smaller in blurred images, whereas total volume and gray volume prefrontal percentage estimates tended to be slightly larger. This analysis suggests large amounts of blurring have relatively small effects on estimates of prefrontal percentage when compared to the average Homo sapiens–non-Homo sapiens differences that were found. Homo sapiens values exceed the pongid averages by 2.4% for prefrontal percentage total volume, and by 3.2% for white volume (Table 2). In addition, we note that total and gray volumes are larger in blurred images. Thus, if Homo sapiens scans were clearer than the average primate scan, our analyses suggest that this would artifactually decrease the measured Homo sapiens–non-Homo sapiens difference for these measures. For these reasons, we believe that any subtle differences in image quality that might exist in our sample (which are less obvious than the differences between blurred and non-blurred test images) are highly unlikely to materially affect the conclusion that Homo sapiens differs in prefrontal percentage, particularly with respect to white matter. The possible effects of image quality differences on average human residual prefrontal values (caused by possible changes in the scaling relationships of prefrontal to non-prefrontal measurements) can be estimated by artificially adjusting nonhuman primate data to make a worst-case assessment. The analyses above suggest that any image quality biases that might exist are likely to differentially affect prefrontal areas by significantly less that 10% more than non-prefrontal areas (Supplementary Table 1 online). Furthermore, image quality degradation seems to artificially decrease only prefrontal white volume estimates. If nonhuman scans were systematically of lower quality, then only these measures would be expected to result in inflated human residuals (through lower nonhuman primate regression lines). A worst-case assessment of such effects can therefore be accomplished by artificially inflating nonhuman prefrontal values by 10% and then recalculating human residuals. This would model a larger, more systematic image-quality bias pervading nonhuman scans than is evident on visual inspection (Fig. 5). Nevertheless, even in this case average human residual prefrontal white volume is still 29% larger than nonhuman primate trends predict (compared to 41% calculated on the actual data), which approaches statistical significance (P = 0.08, one-tailed). If nonhuman primate prefrontal total and gray volumes are similarly adjusted down by 10% (reflecting the fact that image degradation seems to artificially inflate these measures), average human residual prefrontal total volume is 31% larger than nonhuman primate trends predict (compared to 18% for actual data), and average prefrontal gray volume becomes 8% larger (compared to 3% smaller for actual data). These analyses suggest that even if pervasive, systematic biases exist as a result of differing quality of human as compared to nonhuman images, average human residual values would likely still exceed primate allometric predictions for prefrontal total and white volumes. Inhomogeneity. Although the scans were processed to correct variations in signal intensity in different regions caused by inhomogeneities in the magnetic field, no such method of correction is perfect. If species varied with respect to the presence of subtle inhomogeneities in prefrontal but not in non-prefrontal regions, or vice versa, it is possible that this could bias the estimation of prefrontal cerebral volume compared with non-prefrontal cerebral volume. If a greater degree of inhomogeneity persists in the prefrontal cortex even after correction, then the prefrontal cortex might tend to be brighter relative to the rest of the cerebrum, and therefore more of the prefrontal cortex might be incorrectly classified as white than in the rest of the cerebrum. This could potentially bias results if it occurred differentially in human and nonhuman scans. However, if this is truly a pervasive problem in the present dataset, it should result in a positive association between the relative brightness of the prefrontal cortex (indexing the degree that inhomogeneities are biased towards the prefrontal compared with non-prefrontal areas) and the relative increase in percentage white in the prefrontal compared with non-prefrontal areas. In the present dataset, the correlation between ratio of prefrontal average intensity to non-prefrontal average intensity ((average intensity of prefrontal)/(average intensity of non-prefrontal)) and the ratio of percentage of prefrontal that is white to percentage of nonprefrontal that is white ((percentage of white in prefrontal)/(percentage of white in non-prefrontal)) is essentially zero for nonhuman primate species (r2 = 0.0002, P = 0.98, n = 10). That is, species in this dataset with relatively brighter prefrontal cortices do not tend to have greater proportions of the prefrontal designated as white matter. This suggests that even if subtle inhomogeneity effects remain after the corrections applied in this study, they do not seem to affect the conclusions presented here. Voxel size. Voxel size varied considerably across subjects, from 0.26 ml (Saimiri sciureus: 0.47 mm × 0.47 mm × 1.20 mm) to 1.32 ml (human females: 0.94 mm

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

250

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE

ARTICLES
× 0.94 mm × 1.50 mm). Conceivably, these differences might affect the ability of the algorithm to segment gray and white volumes because of partial volume effects. Two kinds of voxel resolution might be relevant here: (i) absolute size of voxels regardless of brain size and (ii) size of voxels relative to overall brain size (that is, proportion of total brain size accounted for by each voxel). Given that cortical thickness varies as a function of brain size50, smaller brains are likely to have greater gray/white partial volume effects, holding absolute voxel size constant. Human brains in this sample have the coarsest absolute resolution but the finest relative resolution (see Supplementary Table 2 online for average absolute and relative voxel sizes for each species in this study). The crucial question for the present study is again whether prefrontal percentage estimates are significantly affected by differences in either relative or absolute voxel size. The possible effects were assessed in two different ways. First, we artificially inflated voxel sizes in a sample of our human scans and measured the extent to which this influenced prefrontal percentage estimates. Second, we assessed the extent to which voxel size (either relative or absolute) explained any of the variation between species in prefrontal percentage measures. The six female human scans were reformatted to contain 2 mm cubic voxels, which thereby contain six times more volume per voxel than the original versions. Because actual brain size for each of these subjects of course remains the same, this reformatting results in large increases in relative voxel size (proportion of total brain per voxel), which approximate the coarsest relative resolutions used in other species in this study (Supplementary Table 2 online). Analysis showed that although coarser resolutions do affect the relative segmentation of gray and white, prefrontal and non-prefrontal portions are affected very similarly (and just as with the blurring study, always in the same direction for prefrontal volume compared with total cerebral volume; see Supplementary Table 3 online). As a result, the change in prefrontal percentage amounts to only +0.12 (s.e.m. = 0.20) percentage points for total volume, –0.35 (s.e.m. = 0.11) percentage points for gray, and +0.69 (s.e.m. = 0.31) percentage points for white (Supplementary Table 3 online). Again we note that the Homo/non-Homo differences found are substantially larger than these values (see Table 2). Notably, with the exception of prefrontal percentage total, these are in the opposite direction from the effects documented for artificially blurring images. Thus, the effects of these two independent potential sources of error will likely often partially offset. Also, larger relative voxel sizes (as seen in nonhuman primates) seem to artificially inflate prefrontal percentage white values, which would have the effect of decreasing the measured prefrontal percentage white differences between humans and nonhuman primates in this study, thus increasing the likelihood that the difference found for white matter volume is real. Prefrontal percentage total is largely unaffected, and prefrontal percentage gray is marginally affected (but as prefrontal percentage gray was not found to be significantly different in humans, this does not affect the conclusions presented here). Another way to assess the possible effects of voxel size differences across species is to directly assess its association with the measures of interest. If species differences in voxel size meaningfully affect the estimation of either prefrontal percentage total, white, or gray volumes, then these measures should correlate with differences in either absolute or relative voxel size in nonhuman primate values (including Homo sapiens in the assessment conflates the specific species difference we are interested in with any biasing effect we are trying to estimate). For the present dataset, none of the correlations with absolute voxel size are significant in nonhuman primates (correlation between absolute voxel size and prefrontal percentage total: r = 0.05, P = 0.89; prefrontal percentage gray: r = –0.03, P = 0.93; prefrontal percentage white: r = 0.25, P = 0.49). The threefold range of variation in nonhuman primates’ absolute voxel size suggests that restriction of range is not a likely explanation for these low correlations. With respect to relative voxel size, correlations with the proportion of total brain size accounted for by each voxel are significant only for prefrontal percentage total (r = –0.76, P = 0.01) and prefrontal percentage gray (r = –0.79, P = 0.01) but not for prefrontal percentage white (r = –0.38, P = 0.27). Human prefrontal percentage total falls outside the 95% confidence intervals, though the human prefrontal percentage gray does not. This is additional confirmation that our conclusions are not materially affected by possible biasing effects of relative or absolute voxel size. Taken together, these analyses suggest that differences in human and nonhuman prefrontal percentage white and total volumes are not likely to be explained by image quality, voxel size or sex representation differences between species.
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS We thank J. Rilling and T. Insel for allowing us to analyze their collection of primate brain scans, and M. Grossman for six human male brains. S. Langin-Hooper and P. Silverman helped with data processing. We also thank the subjects who allowed themselves to be scanned. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 7 December 2004; accepted 3 January 2005 Published online at http://www.nature.com/natureneuroscience/

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

1. Deacon, T.W. The Symbolic Species: the Co-evolution of Language and the Brain (Norton, New York, 1997). 2. Hofman, M.A. Energy metabolism, brain size, and longevity in mammals. Q. Rev. Biol. 58, 495–512 (1983). 3. Harvey, P.H. & Clutton-Brock, T.H. Life history variation in primates. Evolution Int. J. Org. Evolution 39, 559–581 (1985). 4. Stephan, H., Frahm, H. & Baron, G. New and revised data on volumes of brain structures in insectivores and primates. Folia Primatol. (Basel) 35, 1–29 (1981). 5. Holloway, R.L. The failure of the gyrification index (GI) to account for volumetric reorganization in the evolution of the human brain. J. Hum. Evol. 22, 163–170 (1992). 6. Damasio, A.R. The frontal lobes. in Clinical Neuropsychology (eds. Heilman, K. & Valenstein, E.) 339–375 (Oxford University Press, Oxford, 1985). 7. Goldman-Rakic, P.S. The prefrontal landscape: implications of functional architecture for understanding human mentation and the central executive. Phil. Trans. R. Soc. Lond. B Biol. Sci. 351, 1445–1453 (1996). 8. Fuster, J.M. The prefrontal cortex, mediator of cross-temporal contingencies. Hum. Neurobiol. 4, 169–179 (1985). 9. Gabrieli, J.D., Poldrack, R.A. & Desmond, J.E. The role of left prefrontal cortex in language and memory. Proc. Natl. Acad. Sci. USA 95, 906–913 (1998). 10. Thompson-Schill, S.L. et al. Verb generation in patients with focal frontal lesions: a neuropsychological test of neuroimaging findings. Proc. Natl. Acad. Sci. USA 95, 15855–15860 (1998). 11. Vendrell, P. et al. The role of prefrontal regions in the Stroop task. Neuropsychologia 33, 341–352 (1995). 12. de Bruin, J.P.C. Social behavior and the prefrontal cortex. in Progress in Brain Research (eds. Uylings, H.B.M. et al.) 485–497 (Elsevier, New York, 1990). 13. Von Bonin, G. The frontal lobe of primates: cytoarchitectural studies. in The Frontal Lobes: Proceedings of the Association for Research in Nervous and Mental Disease, December 12 and 13, 1947 67–83 (Williams & Wilkins, Baltimore, 1948). 14. Holloway, R.L. The evolution of the primate brain: some aspects of quantitative relations. Brain Res. 7, 121–172 (1968). 15. Semendeferi, K., Lu, A., Schenker, N. & Damasio, H. Humans and great apes share a large frontal cortex. Nat. Neurosci. 5, 272–276 (2002). 16. Bush, E.C. & Allman, J.M. The scaling of frontal cortex in primates and carnivores. Proc. Natl. Acad. Sci. USA 101, 3962–3966 (2004). 17. Blinkov, S.M. & Glezer, I.I. The Human Brain in Figures and Tables (Plenum, New York, 1968). 18. Brodmann, K. Vergleichende Lokalisationsiehre der Grosshirnrinde in ihren Prinzipien Dargestellt auf Grund des Zellenbaues (Verlag, Leipzig, 1909). 19. Brodmann, K. Neue Ergebnisse über die vergleichende histologische localisation der grosshirnrinde mit besonderer berücksichtigung des stirnhirns. Anat. Anz. 41 (suppl.), 157–216 (1912). 20. Armstrong, E., Zilles, K., Curtis, M. & Schleicher, A. Cortical folding, the lunate sulcus and the evolution of the human brain. J. Hum. Evol. 20, 341–348 (1991). 21. Semendeferi, K., Armstrong, E., Schleicher, A., Zilles, K. & Van Hoesen, G.W. Prefrontal cortex in humans and apes: a comparative study of area 10. Am. J. Phys. Anthropol. 114, 224–241 (2001). 22. Rilling, J.K. & Insel, T.R. The primate neocortex in comparative perspective using magnetic resonance imaging. J. Hum. Evol. 37, 191–223 (1999). 23. McBride, T., Arnold, S.E. & Gur, R.C. A comparative volumetric analysis of the prefrontal cortex in human and baboon MRI. Brain Behav. Evol. 54, 159–166 (1999). 24. Uylings, H.B.M. & Van Eden, C.G. Qualitative and quantitative comparison of the prefrontal cortex in rat and in primates, including humans. in Progress in Brain Research, Vol. 85 (eds. Uylings, H.B M., Van Eden, C.G., De Bruin, J.P.C., Corner, M.A. & Feenstra, M.G.P.) 31–62 (Elsevier, New York 1990). 25. Semendeferi, K., Armstrong, E., Schleicher, A., Zilles, K. & Van Hoesen, G.W. Limbic frontal cortex in hominoids: a comparative study of area 13. Am. J. Phys. Anthropol. 106, 129–155 (1998). 26. Fuster, J.M. The Prefrontal Cortex: Anatomy, Physiology, and Neuropsychology of the Frontal Lobe edn. 2 (Raven, New York, 1989). 27. Ringo, J.L. Neuronal interconnection as a function of brain size. Brain Behav. Evol. 38, 1–6 (1991). 28. Roberts, N., Puddephat, M.J. & McNulty, V. The benefit of stereology for quantitative radiology. Br. J. Radiol. 73, 679–697 (2000). 29. Holloway, R.L. Brief communication: how much larger is the relative volume of area 10 of the prefrontal cortex in humans? Am. J. Phys. Anthropol. 118, 399–401 (2002). 30. Melhem, E.R. et al. Diffusion tensor MR imaging of the brain and white matter tractography. AJR Am. J. Roentgenol. 178, 3–16 (2002).

NATURE NEUROSCIENCE VOLUME 8 | NUMBER 2 | FEBRUARY 2005

251

ARTICLES
31. Rumbaugh, D.M., Savage-Rumbaugh, E.S. & Wasburn, D.A. Toward a new outlook on primate learning and behavior: complex learning and emergent processes in comparative perspective. Jpn. Psychol. Res. 38, 113–125 (1996). 32. Krubitzer, L. The organization of neocortex in mammals: are species differences really so different? Trends Neurosci. 18, 408–417 (1995). 33. Schoenemann, P.T., Budinger, T.F., Sarich, V.M. & Wang, W.S.-Y. Brain size does not predict general cognitive ability within families. Proc. Natl. Acad. Sci. USA 97, 4932–4937 (2000). 34. Thompson, P.M. et al. Genetic influences on brain structure. Nat. Neurosci. 4, 1253– 1258 (2001). 35. Changizi, M.A. Principles underlying mammalian neocortical scaling. Biol. Cybern. 84, 207–215 (2001). 36. Schoenemann, P.T. Syntax as an emergent characteristic of the evolution of semantic complexity. Minds Machines 9, 309–346 (1999). 37. Savage-Rumbaugh, E.S. & Rumbaugh, D.M. The emergence of language. in Tools, Language and Cognition in Human Evolution (eds. Gibson, K.R. & Ingold, T.) 86–108 (Cambridge Univ. Press, Cambridge, 1993). 38. Aboitiz, F. & Garcia, V.R. The evolutionary origin of the language areas in the human brain. A neuroanatomical perspective. Brain Res. Brain Res. Rev. 25, 381–396 (1997). 39. Lieberman, P. On the nature and evolution of the neural bases of human language. Yearb. Phys. Anthropol. 45, 36–62 (2002). 40. Dunbar, R. Grooming, Gossip and the Evolution of Language (Faber and Faber, London, 1996). 41. Humphrey, N. The social function of intellect. in Consciousness Regained 14–28 (Oxford Univ. Press, Oxford, 1984). 42. Gray, J.R. & Thompson, P.M. Neurobiology of intelligence: science and ethics. Nat. Rev. Neurosci. 5, 471–482 (2004). 43. Wynn, T. Archaeology and cognitive function. Behav. Brain Sci. 25, 389–438 (2002). 44. Holloway, R.L. Toward a synthetic theory of human brain evolution. in Origins of the Human Brain (eds. Changeux, J.-P. & Chavaillon, J.) 42–54 (Clarendon, Oxford, 1995). 45. Zhang, Y., Brady, M. & Smith, S. Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm. IEEE Trans. Med. Imaging 20, 45–57 (2001). 46. Guillemaud, R. & Brady, M. Estimating the bias field of MR images. IEEE Trans. Med. Imaging 16, 238–251 (1997). 47. Elias, H., Hennig, A. & Schwartz, D.E. Stereology: applications to biomedical research. Physiol. Rev. 51, 158–199 (1971). 48. Ankney, C.D. Sex differences in relative brain size: the mismeasure of woman, too? Intelligence 16, 329–336 (1992). 49. Falk, D., Froese, N., Sade, D.S. & Dudek, B.C. Sex differences in brain/body relationships of Rhesus monkeys and humans. J. Hum. Evol. 36, 233–238 (1999). 50. Jerison, H.J. Allometry, brain size, cortical surface, and convolutedness. in Primate Brain Evolution (eds. Armstrong, E. & Falk, D.) 77–84 (Plenum, New York, 1982).

© 2005 Nature Publishing Group http://www.nature.com/natureneuroscience

252

VOLUME 8 | NUMBER 2 | FEBRUARY 2005 NATURE NEUROSCIENCE