Nature Neuroscience April 2004

Published on February 2017 | Categories: Documents | Downloads: 44 | Comments: 0 | Views: 1240
of 102
Download PDF   Embed   Report

Comments

Content

VOLUME 7 NUMBER 4 APRIL 2004

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

E D I TO R I A L
315
In competitive games, the outcome of one player’s choices often depends on the strategy chosen by each opponent. Lee and colleagues now show that activity in the prefrontal cortex may provide signals to update estimates of expected reward in monkeys playing a simple game against a computer opponent. Such signals could underlie the generation of random behavior for strategic purposes. The authors also used a reinforcement-learning algorithm to predict the monkeys’ choices. (pp 319 and 404)

Testing a radical theory

BOOK REVIEW
317 The Birth of the Mind: How a Tiny Number of Genes Creates the Complexities of Human Thought by Gary Marcus Reviewed by Charles Jennings

NEWS AND VIEWS
319 321 322 323 325 Unpredictable primates and prefrontal cortex Michael L Platt ̈ see also p 404 Synaptic vesicles really do kiss and run R Mark Wightman & Christy L Haynes ̈ see also p 341 Stiffening the spines Annette Markus ̈ see also p 357 Choices, choices, choices Jerold Chun Imaging gender differences in sexual arousal Turhan Canli & John D E Gabrieli ̈ see also p 411

REVIEW
327 Neural activity and the dynamics of central nervous system development Jackie Yuanyuan Hua & Stephen J Smith

CCK-mediated satiety and brainstem melanocortin (p 335)

B R I E F C O M M U N I C AT I O N S
333 335 Reverse propagation of sound in the gerbil cochlea T Ren Cholecystokinin-mediated suppression of feeding involves the brainstem melanocortin system W Fan, K L J Ellacott, I G Halatchev, K Takahashi, P Yu & R D Cone

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. Editorial Office: 345 Park Avenue South, New York, NY 10010-1707. Tel: (212) 726 9200, Fax: (212) 696 9635. Annual subscription rates: USA/Canada: US$199 (personal), US$99 (student), US$129 (postdoc). Canada add 7% GST #104911595RT001; Euro-zone: €289 (personal), €163 (student), €196 (postdoc); Rest of world (excluding China, Japan, Korea): £175 (personal), £99 (student), £119 (postdoc); 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. 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; Periodicals postage rate paid at New York, NY and additional mailing offices. CPC PUB AGREEMENT #40032744. POSTMASTER: Send address changes to Nature Neuroscience Subscription Department, P.O. Box 5054, Brentwood, TN 37024-5054. Printed by Publishers Press, Inc., Lebanon Junction, KY, USA. Copyright © 2004 Nature Publishing Group. Printed in USA.

i

VOLUME 7 NUMBER 4 APRIL 2004

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Tenascin-R mediates radial migration in the olfactory bulb (p 347)

337 339

A psychophysical test of the vibration theory of olfaction A Keller & L B Vosshall Parietal somatosensory association cortex mediates affective blindsight S Anders, N Birbaumer, B Sadowski, M Erb, I Mader, W Grodd & M Lotze

ARTICLES
341 347 Dopamine neurons release transmitter via a flickering fusion pore R G W Staal, E V Mosharov & D Sulzer ̈ see also p 321 Tenascin-R mediates activity-dependent recruitment of neuroblasts in the adult mouse forebrain A Saghatelyan, A de Chevigny, M Schachner & P Lledo Stability of dendritic spines and synaptic contacts is controlled by αN-catenin K Abe, O Chisaka, F van Roy & M Takeichi ̈ see also p 322 The X-linked mental retardation protein oligophrenin-1 is required for dendritic spine morphogenesis E Govek, S E Newey, C J Akerman, J R Cross, L Van der Veken & L Van Aelst Local structural balance and functional interaction of excitatory and inhibitory synapses in hippocampal dendrites G Liu Synaptic dynamics mediate sensitivity to motion independent of stimulus details H Luksch, R Khanbabaie & R Wessel Differential control over cocaine-seeking behavior by nucleus accumbens core and shell R Ito, T W Robbins & B J Everitt Glutamatergic activation of anterior cingulate cortex produces an aversive teaching signal J P Johansen & H L Fields Prefrontal cortex and decision making in a mixed-strategy game D J Barraclough, M L Conroy & D Lee ̈ see also p 319 Men and women differ in amygdala response to visual sexual stimuli S Hamann, R A Herman, C L Nolan & K Wallen ̈ see also p 325

357 364
αN-catenin regulates dendritic and synaptic stability (p 357)

373

380 389

Oligophrenin-1 is required for dendritic spine morphogenesis (p 364)

398

404 411

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.

Gender differences in amygdala responses to erotic images (pp 325 and 411)

NATURE NEUROSCIENCE

iii

E D I TO R I A L

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Testing a radical theory
he paper by Keller and Vosshall on page 337 of this issue is unusual; it describes a refutation of a theory that, while provocative, has almost no credence in scientific circles. The only reason for the authors to do the study, or for Nature Neuroscience to publish it, is the extraordinary—and inappropriate—degree of publicity that the theory has received from uncritical journalists. The theory, from Luca Turin (formerly of University College London), concerns the mechanism of olfactory transduction. Olfaction is not well understood compared to the other senses, but most experts believe that odorant molecules bind to specific receptors through conventional molecular interactions, causing a conformational change in the receptor that leads to activation of intracellular signals. Admittedly there are no clear demonstrations (apart from one study1 in C. elegans) that a specific receptor binding to an odorant mediates the perceptual response to that odorant, and there are some anomalies, such as molecules that smell the same despite their lack of chemical similarity. However, this could be explained through subsequent neural processing (if for example receptors with different specificity were to activate common targets within the brain). Turin proposed a very different theory, namely that olfactory receptors act like a spectroscope to detect intramolecular vibrations within the odorant molecule. According to this idea, the perceptual quality of an odorant is determined not by its shape but by its vibrational spectrum. This would be of great importance if true, but radical ideas require strong evidence, which Turin did not provide. Nor did he provide a detailed explanation of how these molecular vibrations could lead to neural activation. The magician James Randi, debunker of paranormal claims, once said that if you claim to have a goat in your back yard, people will probably believe you, but if you say you have a unicorn, you must expect closer scrutiny. The editors at Nature used to classify manuscripts on a ‘zoological scale’ that ranged from goats to unicorns, and Turin’s paper was toward the far end of that scale. Despite the forcefulness of his assertions, most scientists in the field were unconvinced by his proposal. Thus his paper was rejected by Nature, and it was eventually published (without review, according to Turin’s own account) by Chemical Senses in 1996. Turin’s theory would probably have vanished into obscurity but for two coincidences. First, one of his former students had become a producer for the BBC, and she decided to make a TV documentary about him. Second, he had a chance encounter with writer Chandler Burr, who was so taken with the theory that he wrote a popular book about it. The Emperor of Scent, which appeared in 2002, is effectively a mouthpiece for Turin’s views, and it is intensely hostile to the scientific establishment. It has attracted wide attention, and with the exception of a scathing review in this journal from Avery Gilbert2, the reviews have been almost uniformly favorable. The book is seductively written, and it was recently reissued in paperback, complete with a readers’ guide to promote book club discussions.

T

The villains of Burr’s book include many of the leading figures in the olfactory community, who are portrayed as ignorant and incompetent reactionaries, along with the journal editors who rely on their advice. Burr’s overall verdict is that Turin’s failure to convince the scientific establishment of his views reflects “scientific corruption…in the most mundane and systemic and virulent and sadly human sense of jealousy and calcified minds and vested interests.” Many olfactory researchers were dismayed by the book and by the apparent willingness of the media to accept Burr’s verdict. Keller and Vosshall were sufficiently upset that they decided to put Turin’s theory to an experimental test. As described in their paper, they tested three claims of the vibration theory, all of which feature prominently in Burr’s book. The experiments were conducted double-blind, and in all three cases the results were negative. Turin himself had no role in designing the study, and one could debate (as Turin probably will) whether this study constitutes a definitive refutation of his theory. A conservative statement would be that Turin’s claims are not reproducible based on the information provided in his own publications. At the least, the burden of proof for confirmation of his theory is now unambiguously transferred to Turin, where it should have been all along. In some sense it does not matter whether the public believes in the vibrational theory of olfaction; the truth will eventually come out. But of course this is not just about olfaction. It is about the public credibility of the scientific process and the biases that affect science reporting in the popular press. It is disturbing that Emperor of Scent received so much favorable publicity from reviewers who were ill qualified to judge its scientific content. The New York Times and The Washington Post, for instance, assigned it to their movie critic and fashion critic, respectively. The media loves controversy, and ever since David and Goliath, the story of a lone hero taking on the establishment has had enduring appeal. Of course, radical ideas from outside the mainstream do occasionally turn out to be right. Of course scientists are sometimes excessively attached to conventional ideas. But in science at least, the mainstream view is usually based on the accumulation of evidence over many years. Journalists are trained to report both sides of any argument, but this can be misleading when both sides are not equally credible. A mature body of scientific theory is like a large building, and the impulse to demolish it is often little more than a form of intellectual vandalism, an expression of frustration by those who did not succeed as architects. Some buildings outlive their usefulness, of course, but the threshold for knocking them down should be high. We hope that the paper from Keller and Vosshall will serve as a reminder of why that’s so.
1. Sengupta, P., Chou, J.H. & Bargmann, C.I. Cell 84, 899–909 (1996). 2. Gilbert, A.N. Nat. Neurosci. 6, 335 (2003).

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

315

NEWS AND VIEWS

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Unpredictable primates and prefrontal cortex
Michael L Platt
In many competitive games, players need to behave unpredictably so that their opponents cannot anticipate the next move. New recordings from monkeys playing a computer game support the idea that neurons in the prefrontal cortex may control this behavior and imply there is a cost for generating random behavior, which monkeys avoid unless the opponent is sophisticated. When the United States and China played to a 0–0 draw at the end of regulation time in the 1999 Women’s World Cup soccer finals, the tie-breaking shoot-out became one of the most memorable moments in sports. To be successful, kickers must be unpredictable or risk exploitation by the goalie1. As it only takes about a half second for the ball to reach the goal, goalies must anticipate whether the ball will be kicked left or right, and leap to block the kick before they can visually determine its path. With the 1999 shoot-out tied at 4 apiece, right-footed kicker Brandi Chastain unpredictably leftfooted a laser shot to the left of goalie Gao Hong for the US win. In this issue, Baraclough and colleagues2 demonstrate that unpredictable behavioral responses in such dynamic competitive contexts may be driven by signals carried by neurons in the prefrontal cortex. These cells apparently carry sufficient information about prior choices and their outcomes to guide the production of strategically unpredictable behavior. Despite great progress in understanding the neural mechanisms responsible for perception and movement, the neural basis for simple decision making has been poorly understood until recently. In the past few years, however, neural correlates of stimulus strength have been reported for a number of sensory brain areas in subjects making simple perceptual judgments3,4. Moreover, neural correlates of movement value have been found in several cortical and subcortical brain areas in subjects choosing between unequally rewarded alternatives5,6. Such signals are prerequisites for making informed and economically advantageous choices. In all these studies, the relationships between stimulus, response and outcome were fixed. That is, choosing the single movement with the highest expected value was always the optimal course of action. In social contexts, however, decision outcomes are not deterministic but vary depending on the choices made by other individuals, making prediction difficult. Game theory has been developed in the social sciences to predict and explain behavior under these circumstances7. Game theoretic models posit that players evaluate the costs and benefits of each alternative to themselves and their opponents and then adopt a behavioral strategy. Typically, these behavioral strategies comprise a probabilistic distribution of responses that settles at an equilibrium for all players. Equilibria of this sort are often known as Nash equilibria, after the Nobel prize-winning mathematician, and the resultant behavioral strategies dominate all others. For example, in the children’s game ‘rock-paperscissors’, the best response for all players is to unpredictably play each alternative one-third of the time. Any more predictable response can be easily exploited by other players. Baraclough and colleagues2 have significantly advanced the study of the neural basis of decision making by applying such a gametheoretic approach. The authors recorded the activity of neurons in the dorsolateral prefrontal cortex of monkeys playing a simple game against a computer opponent. Neurons in this area are sensitive to stimulus location, movement preparation, working memory8 and reward expectation9. Moreover, neurophysiological10 and neuroimaging11 studies have implicated prefrontal cortex in encoding information used to make decisions. In the new study2, two monkeys played an oculomotor version of ‘matching pennies,’ a

The author is in the Department of Neurobiology, Duke University Medical Center, Durham, North Carolina 27710, USA. e-mail: [email protected]

Figure 1 Experimental design. (a) Payoff matrix for a strategic eye movement game played by monkeys against a computer opponent. Each monkey stared at a central circle (not shown) on a computer monitor, two peripheral green circles were illuminated, and then the central circle was turned off. The monkey then looked to either the left or right circle. A red ring around the target revealed the computer’s selection after the monkey looked. If the computer and the monkey selected the same target, a squirt of fruit juice was delivered. (b) Action potentials were recorded extracellularly from neurons in the monkey prefrontal cortex (PFC), near the principal sulcus and anterior to the frontal eye fields.

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

319

NEWS AND VIEWS
standard game in which the optimal strategy is to behave randomly and unpredictably from trial to trial. The monkey looked at a computer screen, saw two targets, and then moved its eyes to either target (Fig. 1). If the monkey chose the target selected by the computer, he received a squirt of fruit juice. If he chose the other target, he received nothing. Monkeys playing this game developed different decision strategies depending on the algorithm implemented by the computer opponent. When the computer played randomly without regard to the monkeys’ choices, they developed spatial biases favoring one target over another. As the computer rewarded targets at random, this strategy was perfectly reasonable. When the computer tracked only the monkeys’ choices, however, they adopted a win–stay, lose–shift strategy. Thus, if a monkey chose left and was not rewarded, he chose the other target on the next trial. Most importantly, when the computer tracked both the monkeys’ choices and rewards, the monkeys developed a strategy of choosing randomly on each trial—the optimal strategy in the classic ‘matching pennies’ game. The authors then determined whether a reinforcement learning model could account for the monkeys’ choices. This model assumes that choices are made based on differences in the value functions associated with each alternative, which are determined by the prior history of rewards received for choosing each target. The model parameters differed in predictable ways depending on the monkeys’ strategies. Most importantly, when monkeys played against the most sophisticated computer program, the value differences were very small, indicating that choices were weakly, but systematically, influenced by the outcome of prior choices. Thus, the monkeys behaved in a way that made it difficult for the computer opponent to predict their choices reliably. These results suggest that the monkeys may have converged on the optimal strategy using a reinforcement learning algorithm. The authors found that many neurons in prefrontal cortex were systematically modulated by prior reward outcomes as well as by prior choices. Most importantly, many prefrontal neurons were modulated by a conjunction of these two factors. For example, one prefrontal neuron in this study fired preferentially when the monkey had selected the right-hand target on the previous trial and was not rewarded. Other prefrontal neurons were sensitive to different conjunctions of choice and reward outcome. The authors compared the activity of prefrontal neurons on the strategic decision task with activity evoked in a control task. In this task, one target turned red, cuing the monkey that shifting gaze to the second target would be rewarded. Comparing neuronal activity in the two tasks dissociated responses related purely to movement or reward from those related to making decisions. Such analyses revealed prefrontal signals specific to strategic decision making, which were absent or even reversed in the control task. This study is an elegant and novel application of game theory to understanding strategic decision making in monkeys. The data imply that there is a cost to generating random behavior, which monkeys avoid unless confronted with a sophisticated opponent. Indeed, normal human subjects appear to be quite poor at generating random sequences, but with extensive practice become more adept12, much like the simian juice experts in this study. The results also suggest that prefrontal neurons carry sufficient information to guide the behavioral choices made by monkeys in this task. The authors posit that neural circuits responsible for maintaining persistent activity in prefrontal cortex during standard working memory tasks may also serve to integrate reward and choice history in this simple strategic game. Theoretically, these circuits could provide signals needed to update value functions using a reinforcement learning model, and thereby guide the generation of optimally random choices. This study raises several intriguing questions. First, how are the prefrontal signals observed in this strategic game related to decision signals observed in other cortical and subcortical areas? After all, neurons in parietal cortex, anterior cingulate cortex, posterior cingulate cortex, superior colliculus and basal ganglia are sensitive to the value of a particular movement, and this value is often predicated on the prior history of choices13,14. It would be interesting to know whether neurons in these other areas continue to signal the value of a particular movement in these strategic contests. Indeed, the results of one such study suggest that parietal neurons encode the value of a particular eye movement, independent of movement probability, in monkeys playing a strategic game at various Nash equilibria (Dorris, M.C. & Glimcher, P.W., Soc. Neurosci. Abstr., 27, 58.10, 2001). A second question regards the necessity of dorsolateral prefrontal cortex for strategic decision making. Would reversible deactivation of this area, for example, render monkeys unable to generate random behavior in the oculomotor matching-pennies task? Moreover, would such deactivation result in the degeneration of value signals at other nodes in the decision network, such as parietal cortex? This would suggest that the value signals observed in these areas are computed from the prefrontal signals recorded by Baraclough and colleagues2. Finally, game theoretic approaches were originally developed to predict and explain the choices made by individuals interacting with others. The unpredictable random behavior of the monkeys in the Baraclough study is the optimal strategy when playing against an intelligent opponent. But did the monkeys actually treat the task as a competitive struggle with a savvy adversary? We might expect different outcomes if monkeys played this game face to face with another monkey. Under those conditions, social rules and conventions—such as dominance rank or a sense of fairness15— might exert a powerful influence on decision making. The influence of cultural norms on strategic decision making in humans is well documented12. The new paper2 will likely be a fundamental contribution to the literature. The application of game theory to the neurophysiology of decision making is new and noteworthy, and accomplished with elegance and finesse. Baraclough and colleagues have shown in their landmark study that formalisms developed in the social sciences to predict and explain strategic behavior offer a powerful tool for understanding the neural basis of decision making.
1. Palacios-Huerta, I. Brown University working paper, cited in Camerer, C.F. Behavioral Game Theory (Princeton Univ. Press, Princeton, New Jersey, 2003). 2. Baraclough, D.J., Conroy, M.L. & Lee, D. Nat. Neurosci. 7, 404–410 (2004). 3. Schall, J.D. & Hanes, D.P. Nature 366, 467–469 (1993). 4. Shadlen, M.N. & Newsome, W.T. Proc. Natl. Acad. Sci. USA 93, 628–633 (1996). 5. Platt, M.L. Curr. Opin. Neurobiol. 12, 141–148 (2002). 6. Glimcher, P.W. Decisions, Uncertainty, and the Brain (MIT Press, Cambridge, Massachusetts, 2003). 7. Kreps, D. Game Theory and Economic Modeling (Oxford Univ. Press, Oxford, UK, 1990). 8. Funahashi, S., Bruce, C.J. & Goldman-Rakic, P.S. J. Neurophysiol. 61, 1–19 (1989). 9. Leon, M.I. & Shadlen, M.N. Neuron 24, 415–425 (1999). 10. Kim, J.N. & Shadlen, M.N. Nat. Neurosci. 2, 176–185 (1999). 11. Sanfey, A.G., Rilling, J.K., Aronson, J.A., Nystrom, L.E. & Cohen, J.D. Science 300, 1673–1675 (2003). 12. Camerer, C.F. Behavioral Game Theory (Princeton Univ. Press, Princeton, New Jersey, 2003). 13. Platt, M.L. & Glimcher, P.W. Nature 400, 233–238 (1999). 14. Coe, B., Tomihara, K., Matsuzawa, M. & Hikosaka, O. J. Neurosci. 22, 5081–5090 (2002). 15. Brosnan, S.F. & De Waal, F.B. Nature 425, 297–299 (2003).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

320

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

NEWS AND VIEWS

Synaptic vesicles really do kiss and run
R Mark Wightman & Christy L Haynes
A new study demonstrates that small synaptic vesicles exocytose dopamine through a flickering fusion pore almost exclusively, a process known as ‘kiss-and-run’ exocytosis. This process is driven by the need for efficient use of few synaptic vesicles. Modern consumers realize that recycling is a good practice but that reusing products is even better. Scientists have long debated the relative contributions of vesicle recycling and reuse after neurotransmitter release as well. In the classic mechanism, dubbed ‘allor-none’ exocytosis (Fig. 1a), a synaptic vesicle fuses with the presynaptic membrane and releases its contents into the synapse; the vesicle membrane is then recycled. Alternately, a synaptic vesicle can form a transient fusion pore in the presynaptic membrane and release only part of its contents; in such ‘kiss-and-run’ exocytosis, the vesicle is then reused. In this issue, Staal and coworkers1 clearly establish that small synaptic vesicles in dopaminergic neurons almost exclusively use the kiss-and-run mechanism of exocytosis. Exocytosis is almost universally accepted as the primary means of chemical communication between neurons. Electron microscopy of freeze-fractured tissue has captured ‘omega’ structures at the surface of stimulated neurons2. Vesicular contents are released in concentration ratios that reflect their stored amounts3. Single-cell capacitance changes during exocytosis, indicating an increase in the cellular membrane area4. Finally, discrete packets of released chemicals can be detected with amperometry5, in which the number of easily oxidized molecules released from a vesicle is measured with carbon-fiber microelectrodes. Much of the current view of exocytosis from small synaptic vesicles has been extrapolated from results obtained with large, dense-core vesicles. Cells with large vesicles primarily use all-or-none exocytosis, and only occasionally kiss-and-run. Allor-none exocytosis of large vesicles is supported by whole-cell capacitance measurements4 as well as by amperometry, which shows that the amount of transmitter released from a variety of cell types corresponds well with known intravesicular amounts6. Video microscopy of chromaffin cells loaded with a fluorescent dye that accumulates in vesicles also illustrates allor-none exocytosis: the fluorescent vesicles can be seen to approach the plasma membrane and then completely lose their fluorescence7 (Fig. 1a). The kiss-and-run mechanism was clearly shown in mast cells by capacitance measurements with a patch-clamp electrode and simultaneous amperometry8. The correlated data showed that increases in membrane area are not always accompanied by full extrusion of the vesicle contents. In a study using fast-scan cyclic voltammetry, an intermediate ‘kiss-and-hold’ state was induced in both mast cells and chromaffin cells by increasing the osmolarity of the extracellular solution9. This removed the normal osmotic gradient between the vesicle interior and the extracellular space, preventing efflux of the intravesicular contents. The applicability of principles derived from large synaptic vesicles to small vesicles in neurons has been unclear. Theoretical considerations predicted that the modest surface tension changes that accompany fusion of a small synaptic vesicle would increase the likelihood of the kiss-and-run mode10. However, the restricted population of small vesicles within neurons—along with their extremely small size (diameters of 50 nm or less) and the limited number of molecules they contain—have made it very difficult to experimentally characterize the mode of small-vesicle exocytosis. Traditional capacitance measurements have been unsatisfactory because the fusion of a small surface-area vesicle has been undetectable. Only recently did Klyachko and Jackson manage to drastically reduce the noise level, enabling them to detect the fusion of secretory vesicles similar in size and shape to small synaptic vesicles using capacitance measurements11. In this proof-of-concept study, estimates of the fusion pore diameter were smaller than the neurotransmitter molecules that pass through the pore. Nevertheless, this technique now shows promise for the study of small vesicles. Fluorescence imaging of single neuronal vesicles is also difficult because the size of synaptic vesicles is smaller than the diffrac© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience tion limit of standard fluorophore emission wavelengths. Furthermore, fluorescence changes are difficult to interpret because they do not necessarily correlate with the neurotransmitter efflux12. However, Gandhi and Stevens13 used a photobleaching process to show small but discernable differences in fluorescence traces of individual dye-loaded vesicles that suggested kiss-and-run accompanying other forms of exocytosis. Direct efflux experiments are also difficult because neuronal vesicles typically contain only 3,000 to 30,000 neurotransmitter molecules, far fewer than the million or so contained in the large vesicles described above. Amperometry, however, can be used to detect even such a small number of molecules, and it allows realtime measurement of exocytotic events. Staal and colleagues have now clarified the situation by using amperometry with carbonfiber microelectrodes to directly monitor exocytosis of endogenous dopamine from cultured ventral midbrain neurons1. Measuring K+-stimulated dopamine release from individual neurons, the authors identi-

a

Time

b c
Time

Time

The authors are in the Department of Chemistry, University of North Carolina, Chapel Hill, North Carolina 27599-3290, USA. e-mail: [email protected]

Figure 1 Exocytotic mechanisms for small synaptic vesicles. Schematic drawings of the mechanisms (above) and amperometric current traces (below). (a) Full fusion of a small synaptic vesicle after initially forming a small fusion pore. After secretion, the vesicle is temporarily incorporated into the plasma membrane. (b) A simple kiss-and-run event. (c) A complex kissand-run event with three subunits, each decreasing in amplitude.

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

321

NEWS AND VIEWS
fied two classes of release events. In simple events, all the dopamine released from a given synaptic vesicle-membrane fusion site was measured in a single amperometric peak. In complex events, the dopamine released from a given synaptic vesicle-membrane fusion site was measured as a series of discrete peaks. The authors’ interpretation is that a simple event consists of a small synaptic vesicle generating a fusion pore in the presynaptic membrane, partially discharging its contents into the synaptic cleft, and then disconnecting from the membrane (Fig. 1b). A complex event occurs when the fusion pore flickers rapidly between an open and closed form, allowing repeated partial release of vesicle contents (Fig. 1c). Comparison of the amperometric traces from simple and complex events supports this interpretation: the number of dopamine molecules oxidized in a simple event is roughly equivalent to the number of dopamine molecules oxidized in the first subunit of a complex event. Thus, both simple and complex events seem to reflect kiss-andrun exocytosis. The authors found that small synaptic vesicles in midbrain dopaminergic neurons undergo kiss-and-run exocytosis almost exclusively. Kiss-and-run exocytosis is advantageous because it leads to increased longevity of a synaptic vesicle, thereby decreasing the importance of the relatively slow process of vesicle recycling through the endosomal compartment. The authors suggest that such efficient vesicle use is necessary because of the relatively small number of synaptic vesicles present in these midbrain neurons. Kiss-andrun exocytosis also avoids inefficient use of dopamine at synapses that lack well-defined active zones, as is typical of dopaminergic neurons. The complex form of kiss-and-run may represent a particularly economical form of exocytosis, which may be advantageous if transmitter-loaded vesicles are in short supply. To test this hypothesis, the researchers exposed the cultured neurons to pharmacological agents affecting the secondary messengers that regulate synaptic vesicle cycling. On addition of a phorbol ester, an agent that increases the number of releasable synaptic vesicles, amperometric traces revealed a relative decrease in the number of complex events from 20% to 6%. After inhibition of protein kinase C, reducing the number of releasable vesicles, the total number of exocytotic events per stimulus was decreased by 82%, but amperometric traces showed a relative increase from 20% to 40% in the number of complex events. Thus complex events appear to be favored when fewer releasable vesicles are available. If the nature of exocytotic mechanisms is determined by the number of vesicles and the nature of the synapse, comparison of the new data1 with similar data collected from cells with large dense-core vesicles8 should reveal significant variation. Kiss-and-run occurs in both cases, but there are some notable differences. First, the amperometric trace subunit duration is approximately 200 times shorter in small synaptic vesicles. Second, the fusion pore flickering occurs with a ten-fold increase in frequency in small synaptic vesicles compared to the large dense-core vesicles. Third, the small synaptic vesicles release 25–30% of their dopamine cargo with each flicker of the fusion pore, whereas the large dense-core vesicles release <1% of their dopamine. Clearly, although the same exocytotic mechanism is at work, the fusion pore flickering characteristics are greatly influenced by the size of the vesicle and the function of the cell. Some questions remain. The new research1 suggests that kiss-and-run exocytosis is driven by the need for efficient use of a relatively small number of synaptic vesicles. This hypothesis can be further tested by measuring the relative number of full fusion and kissand-run events at presynaptic terminals in neurons with a larger number of vesicles, and in neurons that use other neurotransmitters. Amperometry can only detect easily oxidized neurotransmitters such as dopamine, so new strategies will need to be developed for other transmitters such as glutamate. Extracellular calcium is central in regulating exocytosis and release probabilities, and it will be fascinating to explore its influence on the characteristics of kiss-and-run and the flickering pore. Ideally, this would entail amperometric measurements and simultaneous calcium imaging. Because the small synaptic vesicles release such a large percentage of their total neurotransmitter concentration with each flicker of the fusion pore, it would also be interesting to manipulate the intravesicular contents to see whether the kiss-and-hold state can be induced in neurons. Large vesicles forced into a kiss-and-hold state through increased extracellular osmotic pressure undergo massive release when returned to isotonic conditions. Will similar manipulations force the small synaptic vesicles from kiss-and-run to full fusion exocytosis? Future work will tell us whether non-dopaminergic neurons also use a nearly exclusive kiss-and-run mechanism of exocytosis and will explore the implications of kiss-and-run vesicle re-use for the synaptic vesicle recycling mechanism.
1. Staal, R.G.W., Mosharov, E.V. & Sulzer, D. Nat. Neurosci. 7, 341–346 (2004). 2. Heuser, J.E. Q. J. Exp. Physiol. 74, 1051–1069 (1989). 3. Viveros, O.H. in Handbook of Physiology Vol. 6 (eds. Blaschko, A. & Smith, A.D.) 389–426 (American Physiological Society, Washington, D.C., 1975). 4. Neher, E. & Marty, A. Proc. Natl. Acad. Sci. USA 79, 6712–6716 (1982). 5. Wightman, R.M. et al. Proc. Natl. Acad. Sci. USA 88, 10754–10758 (1991). 6. Finnegan, J.M. et al. J. Neurochem. 66, 1914–1923 (1996). 7. Steyer, J.A. & Almers, W. Biophys. J. 76, 2262–2271 (1999). 8. Alvarez de Toledo, G., Fernandez-Chacon, R. & Fernandez, J.M. Nature 363, 554–558 (1993). 9. Troyer, K.P. & Wightman, R.M. J. Biol. Chem. 277, 29101–29107 (2002). 10. Amatore, C., Bouret, Y., Travis, E.R. & Wightman, R.M. Angew. Chem. Int. Ed. Engl. 39, 1952–1955 (2000). 11. Klyachko, V.A. & Jackson, M.B. Nature 418, 89–92 (2002). 12. Aravanis, A.M., Pyle, J.L., Harata, N.C. & Tsien, R.W. Neuropharmacology 45, 797–813 (2003). 13. Gandhi, S.P. & Stevens, C.F. Nature 423, 607–613 (2003).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Stiffening the spines
The ability of dendritic spines to change shape in response to synaptic activity is crucial for synaptic plasticity. This motility is regulated by αN-catenin, report Abe et al. on page 357. Overexpression of αΝ-catenin (green; red is PSD95) stabilized spines in cultured neurons, reducing turnover and thereby increasing their number. Lack of αN-catenin increased spine motility, even at established synaptic contacts. Spine αN-catenin was regulated by synaptic activity: blocking activity with tetrodotoxin reduced αN-catenin staining (and increased spine motility), whereas blocking inhibitory neurotransmission increased αN-catenin.The catenins link cadherin cell adhesion molecules to the cytoskeleton, so αN-catenin is well placed to regulate spine dynamics.

Annette Markus
322 VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

NEWS AND VIEWS

Choices, choices, choices
Jerold Chun
Individual olfactory sensory neurons express only one of more than a thousand different odorant receptors, suggesting that DNA rearrangement may be involved. Based on a clever new technical approach, two groups now conclude that this is not the case. © 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Mammalian olfactory sensory neurons have a difficult decision to make. From over a thousand possible choices, each sensory neuron must pick only one type of olfactory receptor (OR) gene to express1. How this is accomplished is still unclear. In the immune system, the diversity of immunoglobulins and T-cell receptors arises through a process called ‘VDJ recombination’ (Fig. 1a)2. The mechanism (more generally called somatic DNA rearrangement) involves cutting segments of DNA from non-germline tissue and rejoining the segments to form composite genes, producing permanent changes in DNA and gene expression that are not passed on to future generations. The size of the OR gene family3 and its genomic organization, with over 1,000 OR genes dispersed in linear clusters and on different chromosomes4–6, raised the possibility of DNA rearrangement as a mechanism for receptor gene choice in the olfactory system, too. If such a mechanism were found in mammalian neurons, it might help to explain the brain’s complexity and diversity of connections and cell types. Now, in the first rigorous test of this hypothesis, two independent reports in Nature7,8 have cloned mice from single olfactory receptor neurons: neither finds evidence for DNA rearrangements. Of course, models for explaining how ORs are expressed need not invoke DNA rearrangement (Fig. 1b,c)1,4, and at least two key differences exist between OR genes and immune cell receptors. First, the entire OR is encoded by one contiguous stretch of DNA (a single exon), negating a need for combining gene segments. Second, there are no obvious rearrangementrecognition markers flanking OR genes (recombination signal sequences, heptamer/nonamer cis-elements), which are required for VDJ rearrangement. Thus, the mechanism for DNA rearrangement of OR genes would have to be

distinct from that observed in the immune system. One possible alternative mechanism might involve insertion of a transposable element with promoter/enhancer activity, which might direct expression of one OR from a basal promoter (Fig. 1d). To examine whether DNA rearrangement occurs in OR genes, one must first devise a method for detecting the rearrangement. Historically, clonal cell lines have been used, as they can be grown indefinitely to generate large amounts of DNA with the same rearrangement that, once amplified, can be detected by standard techniques. Such an approach was used to first identify

immunoglobulin DNA rearrangements2. However, no clonal cells lines expressing OR genes exist. Theoretically, single-cell PCR approaches might also work, but in the absence of defined OR genes and specific DNA sequences to target, PCR is not technically feasible. To get around this problem, Eggan et al.7 and Li et al.8 used the new—if involved—strategy of expanding a single OR neuronal genome by mouse cloning9. Indeed, cloning of lymphocytes that have already undergone immunological DNA rearrangements produce mice that maintain the original rearrangements in all tissues, yet also yield viable mice, validating this strategy for

Germ line arrangement

a

5′ 5′

V V V

D D D J

J
Site-specific DNA cleavage

3′ 3′

J
Non-homologous end joining

5′

3′

b c d

TFCB
5′

PA

ORA TFCB

PB

ORB

3′

5′

PA

ORA

LCR

PB

ORB

3′

5′ 5′

EA PA ORA TFCB PB ORB

3′ 3′

Kamal Masuta

5′

PA

ORA

EA

PB

ORB

3′

The author is in the Department of Molecular Biology at the Helen L. Dorris Institute for Neurological and Psychiatric Disorders, The Scripps Research Institute, ICND118, 10550 North Torrey Pines Road, La Jolla, California 92037, USA. e-mail: [email protected]

Figure 1 Models for regulating gene expression in the immune and olfactory systems. (a) Classical ‘VDJ’ recombination that occurs in the immune system to generate immunoglobulins. Component gene segments that do not themselves encode mature immunoglobulins are brought together to form a composite coding region that serves as the antigen recognition portion of an antibody2. In the olfactory system, receptor expression controlled by short promoter (P) elements (b) or by distant loci (locus control region, LCR) (c) could provide sufficient information to allow appropriate expression of OR genes in the presence of appropriate transcription factors (TCF)4. (d) Olfactory receptor expression controlled by DNA rearrangement3, in which a distant segment of DNA with promoter/enhancer activities is placed, through rearrangements, in proximity to a basal promoter to provide specific expression4; other variations are also possible.

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

323

NEWS AND VIEWS a
CNS neuron

b
Ofactory sensory neuron

Nuclear transfer

Nuclear transfer

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Enucleated oocyte

Enucleated oocyte

Blastocyst Blastocyst

ES cells Enucleated oocyte Tetraploid blastocyst No mice Blastocyst
Kamal Masuta

ES cell nucleus

Figure 2 Different mouse-cloning strategies. In all approaches, a nucleus from a single neuron is isolated and transferred to a recipient enucleated egg, which further develops in culture into a blastocyst that, following intrauterine implantation, could become a mouse. (a) Nuclei from CNS neurons were unable to generate viable mice9,11. (b) In the new studies7,8, permanently labeled nuclei (shown in green) from olfactory sensory neurons were used in a two-step cloning approach10 in which totipotential ES cells derived from the cloned blastocyst were injected into specially modified tetraploid blastocysts. The approach generated viable cloned mice derived only from the transplanted ES cells (versus contributing to the placenta, which is of distinct embryological origin). To determine totipotentiality, Eggan et al. also used an ES cell nucleus that itself had been derived from an olfactory neuron. The resulting cloned mouse is a clearer indication of the totipotential state of the neuronally derived ES cell nucleus.

Viable mouse Viable mouse

revealing DNA rearrangements10. To assess the neuronal genome of a single, identified OR gene, the two research groups extended the cloning approach with elegant variations that blended a range of other molecular genetic techniques including embryonic stem (ES) cells (semi-immortal cells that can reconstitute a mouse), targeted knockout/knock-ins and lineage tracing. One way to think about cloned mice is as a genome magnifier, as a single neuronal nucleus produces an entire mouse in which all non-immune cells are classically expected to be genomically identical. Whereas cloning of mice was reported several years ago using non-neural cumulus cells9, attempts to produce mice from CNS neurons (Fig. 2a) had proven unsuccessful9,11. Unlike earlier cloning reports9, Eggan et al. and Li et al. used a two-step cloning process (Fig. 2b), whereby donor nuclei are first transferred into enucleated oocytes, which are allowed to form blastocysts from which ES cells are derived. These ES cells are then transferred to a modified recipient embryo (a tetraploid rather than diploid blastocyst) before transfer to a host mouse for in utero development; the end result is that cloned mice are actually

produced from the ES cells rather than directly from a neuronal nucleus itself. Both groups first used permanently tagged olfactory sensory neuron nuclei to produce ES cells and then cloned mice, allowing identification of the tag in subsequent steps. The results showed that at least some olfactory neuronal nuclei were competent to produce cloned, apparently normal mice. If restricted expression of a single OR gene in the differentiated olfactory sensory neuron was permanent, then the researchers might have expected to see mice expressing only one OR type, along with olfactory sensory neurons showing a single, stereotyped neuroanatomical projection pattern. However, further analyses of these mice indicated normal OR gene expression, with expression of multiple receptors and normal neuroanatomy, including spatial distribution of olfactory sensory neurons. However, this first approach could not identify which single OR subtype was being used by the neuron from which a mouse was cloned. Thus, the two groups went further by permanently tagging individual neurons of defined OR identity followed by cloning. As in the first experiment, OR neurons in the mice again showed a normal

range of expressed ORs, despite having originated from a neuron expressing an identified OR subtype. Finally, the researchers used ES cell DNA derived from a tagged OR neuron to search by classical means for possible rearrangements surrounding that receptor—no rearrangements were identified. Therefore, DNA rearrangements are not necessary for OR expression. That the two research groups used different ORs also strengthens this conclusion. The results of Eggan et al.7 and Li et al.8 refocus attention on epigenetic mechanisms of OR expression. In this active field, novel interactions relevant to OR expression are being identified, such as intracellular negative feedback by ORs themselves, which may account for the one-receptor/one-neuron rule12. Eggan et al. look beyond olfaction per se, by providing further data on the totipotentiality of a neuron. They cloned mice using direct transfer of a neuron-derived ES cell nucleus into an oocyte (Fig. 2b). The normal-appearing mice that resulted from these apparently totipotent nuclei suggested that even a postmitotic neuronal nucleus could be reprogrammed to produce an entire organism. As with most scientific studies, some caveats may be worth considering. Although Eggan et al. validly concluded that their cloned mice did not contain DNA rearrangements that interfered with development of a viable mouse, it is notable that biologically important DNA rearrangements of the immune system are maintained in, and are fully compatible with, normal cloned mice10. Therefore, ‘clonablity’ cannot be equated with an absence of DNA rearrangement. It also remains formally possible that a subset

324

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

NEWS AND VIEWS
of ORs might still use a rearrangement mechanism: although there is no evidence for this, these two studies only analyzed two of more than 1,000 expressed ORs. An unresolved issue, which may be technical and/or biological in nature, is that no clones have yet been reported using direct transfer of a neuronal nucleus into an oocyte (Fig. 2), despite expert attempts to do so with nuclei from other neuronal populations9,11. Even with the use of an ES cell intermediate, the overall success rate of cloning with neuronal nuclei seems to be ∼1%. Neuronal nuclear ‘reprogramming’ (might it also include some forms of DNA repair?) seems to require the ES cell intermediate step, although precisely what this step might do to the clonability of neuronal nuclei is currently unclear. The state of the remaining 99% of neuronal nuclei that cannot be cloned remains unknown. It is conceivable that DNA rearrangements exist in some of these neurons, although the nature of such rearrangements remains purely speculative and, as noted above, might not be expected to hamper cloning. By contrast, this 99% most certainly contains nuclei with global changes in chromosome number (aneuploidy) that exist among developing and postmitotic neurons11,13–15. Although the function and total extent of this aneuploidy have yet to be clarified, it could in part account for the low percentage of successful clones. It could also account for the developmental failures observed by Eggan et al. and Li et al., as well as place limits on the percentage of totipotential neurons identified by Eggan et al. That said, none of these considerations detracts from these first glimpses into a single OR neuronal genome, and these impressive technical and scientific achievements will no doubt yield further insights into both olfaction and other neural systems in the near future.
1. 2. 3. 4. Reed, R.R. Cell 116, 329–336 (2004). Jung, D. & Alt, F.W. Cell 116, 299–311 (2004). Buck, L. & Axel, R. Cell 65, 175–187 (1991). Kratz, E., Dugas, J.C. & Ngai, J. Trends Genet. 18, 29–34 (2002). 5. Lane, R.P. et al. Proc. Natl. Acad. Sci. USA 98, 7390–7395 (2001). 6. Zhang, X. & Firestein, S. Nat. Neurosci. 5, 124–133 (2002). 7. Eggan, K. et al. Nature 428, 44–49 (2004). 8. Li, J., Ishii, T., Feinstein, P. & Mombaerts, P. Nature 428, 393–399 (2004). 9. Wakayama, T., Perry, A.C., Zuccotti, M., Johnson, K.R. & Yanagimachi, R. Nature 394, 369–374 (1998). 10. Hochedlinger, K. & Jaenisch, R. Nature 415, 1035–1038 (2002). 11. Osada, T., Kusakabe, H., Akutsu, H., Yagi, T. & Yanagimachi, R. Cytogenet. Genome Res. 97, 7–12 (2002). 12. Serizawa, S. et al. Science 302, 2088–2094 (2003). 13. Rehen, S.K. et al. Proc. Natl. Acad. Sci. USA 98, 13361–13366 (2001). 14. Kaushal, D. et al. J. Neurosci. 23, 5599–5606 (2003). 15. Yang, A.H. et al. J. Neurosci. 23, 10454–10462 (2003).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Imaging gender differences in sexual arousal
Turhan Canli & John D E Gabrieli
Men tend to be more interested than women in visual sexually arousing stimuli. Now we learn that when they view identical stimuli, even when women report greater arousal, the amydala and hypothalamus are much more strongly activated in men. “A man falls in love through his eyes, a woman through her ears,” wrote Woodrow Wyatt in 1918. In this issue, Hamann and colleagues1 use functional magnetic resonance imaging to test whether males and females indeed differ in their brain responses to sexually arousing images. The authors find greater activation in males than females in the amygdala, a brain region involved in emotional arousal, and in the hypothalamus, a brain region central to reproductive functions. What distinguishes this study from a previous effort2 is that the investigators went to great lengths to select stimuli and subjects that would ensure similar degrees of selfreported arousal in both sexes. Thus, the observed brain differences are less likely to reflect sex differences in arousal; instead they reflect sex differences in the processing of sexually arousing stimuli. Hamann and colleagues scanned 28 healthy, heterosexual volunteers, an equal number of males and females. Participants passively viewed neutral images of couples interacting in nonsexual ways (such as weddings, dancing or therapeutic massage), nude photographs of opposite-sex individuals in modeling poses (opposite-sex stimuli) and photographs of couples engaged in explicit sexual acts (couples stimuli), as well as a fixation cross condition to establish brain activation at baseline. Participants subsequently rated their sexual attraction and physical arousal in response to each image on a threepoint scale. Analysis of the imaging data contrasted brain activation to the couples stimuli versus activation to neutral or fixation stimuli, thus revealing regions of significant activation for each sex separately, as well as significant differences between, and commonalities across, the sexes (Fig. 1). Both sexes reported comparable sexual attraction and physical arousal in response to the images; both groups found the couples stimuli to be the most attractive and arousing. The most sensitive direct comparison between males and females looked at the contrast in brain activation between the couples and neutral stimuli. Both classes of stimuli depicted couples interacting, differing only in the sexual aspect of the interaction. In this contrast, males showed significantly greater activation than females in the amygdala. This differential activation in the amygdala stands in striking contrast to many brain regions that were commonly activated for both males and females—regions associated with visual processing, attention, motor and somatosensory function, emotion and reward. Several additional observations are noteworthy. First, brain activation data remained unchanged when the one female subject who reported low sexual arousal was excluded from the analysis. Removal of this subject caused the average arousal of the females to significantly exceed that of the males, yet it was the males who exhibited greater amygdala activation. This is perhaps the strongest indicator that amygdala activation is not related to sexual arousal per se. Second, the average differences between the sexes were striking. Not only did men show greater activation than women in response to sexually explicit couple images in

Turhan Canli is at the Department of Psychology, SUNY Stony Brook, Stony Brook, New York 11794-2500, USA. e-mail: [email protected] John D.E. Gabrieli is at the Department of Psychology, Stanford University, Jordan Hall, Stanford, California 94305, USA. e-mail: [email protected]

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

325

NEWS AND VIEWS
Figure 1 Gender differences in sexual arousal. When viewing sexually arousing visual stimuli, men show greater activation in the amygdala (blue), a brain region involved in emotional arousal, and in the hypothalamus (green), a region involved in reproductive function. Men showed greater activation in these regions even when women reported equal or greater sexual arousal.

the left amygdala, right amygdala and hypothalamus, but also women did not show any greater activation in these regions for the sexually explicit stimuli than for the neutral scenes. It is unclear, therefore, which neural system mediates the sexual arousal reported by the women in this study. Third, males, but not females, showed significant activation in another region associated with sexual behavior, the hypothalamus, when viewing neutral stimuli depicting couples (albeit at a lower level of statistical significance). The authors speculate that this may represent the male’s propensity to view even neutral interactions with females as vaguely sexual, a point that is unlikely to be missed by late-night comedians. Fourth, males and females differed greatly in their amygdala responses to couples and opposite-sex stimuli. Males showed greater activation for those stimuli that generated the greatest arousal: there was no significant activation to the nudes depicted in the oppositesex set, but highly significant activation to the sexually explicit couples, relative to the neutral pictures. Females showed the opposite pattern: they had significantly greater activation to the less arousing opposite-sex stimuli, but no significant activation to the copulating couples. The authors speculated that greater amygdala activation in males may represent their propensity for varied, explicit sexual activity, but the paper offers no explanation for women’s amygdala responses. The distinction between males’ and females’ amygdala reactivity appears to map onto that of ‘hard’ versus ‘soft’ pornography and is likely to invite commentary from many different schools of thought on human sexuality. The benefit of recruiting males and females matched in their ability to experience and express sexual arousal comes at the cost of differential recruitment across the sexes. For example, participants were pre-screened

to respond similarly to sexually explicit material. Intuition (and general life experience) suggests that this process generated a greater yield for males than females. Indeed, none of the males reported lack of arousal to visual erotica, whereas 16% of the female prospects were excluded because of insufficient selfreported arousal. This suggests that the data reported here may not necessarily generalize to all women. Another noticeable sex difference that emerged during the screening of prospective participants was related to self-reported same-sex desire or experience. Only 12% of prospective males, but 36% of prospective females, were excluded from the study for this reason. The basis of this difference remains unclear. The only other study to directly compare brain responses to sexual images between males and females failed to detect any sex difference2. In that study, males reported greater sexual arousal than females, and no significant activation differences were noted when controlling for arousal. Two other studies looking only at males reported conflicting data3,4. It is possible that the extent to which males show amygdala activation to sexually explicit stimuli varies as a function of other factors, such as personality. Indeed, amygdala activation to positive stimuli such as pleasant scenes or happy faces varies as a function of the personality trait of extraversion5,6. Whether this trait may also predict individual differences in amygdala activation to sexual stimuli is unknown. Asymmetries in left versus right amygdala function are of interest, but poorly understood at present. Hamann and colleagues report greater activation in the left than right amygdala of males for the explicit images. The only other study to report male amygdala activation to sexual stimuli observed it in the right hemisphere3. Consistent with the results of Hamann et al.1, left amygdala activation has been reported to be a function of emotional arousal to non-sexual emotional stimuli7,8, although one of these experiments involved highly negative stimuli8. Studies of the encoding of emotional scenes into longterm memory have consistently reported a stronger relation between successful encod-

ing and left amygdala activation for females versus right amygdala activation for males8,9. Although the specific patterns of laterality are difficult to synthesize, amygdala activation often seems to depict some sort of sex difference in the context of emotionally provocative visual stimulation. It is natural to question whether such brain activation differences reflect genetic or social influences on the human brain and mind. That is a question, however, that brain imaging cannot answer. Men and women are, by definition, genetically different. Men and women are also powerfully socialized into gender roles, a socialization that begins shortly after birth. Both genetic and social influences shape brain function and its consequent behavior, so imaging differences could arise from either nature or nurture or both. Hamann et al.1 have reported a thoughtfully controlled study of one aspect of human sexuality. Human sexuality, however, would remain unfulfilled without a climax. Psychologists have long distinguished between ‘appetitive’ and ‘consummatory’ sexual behaviors, that is, those that lead up to, and those that conclude the sexual act. One imaging study went right to the point, imaging the male brain during ejaculation10. The authors of this study were not only intrepid in their choice of research topic and subject participation, but they were also undeterred by concerns about motion artifacts. Remarkably, ejaculation in males was associated with decreased amygdala activation. Thus, the appetitive phase of sexual arousal seems to coincide with increased amygdala activation that is then reversed during the consummatory phase. This activation change parallels the rise and rapid fall in sexual excitement from one phase to the other. It remains to be seen whether decreased amygdala activation associated with ejaculation is causally linked to males’ subsequent unwillingness to snuggle.
1. Hamann, S., Herman, R.A., Nolan, C.L. & Wallen, K. Nat. Neurosci. 7, 411–416 (2004). 2. Karama, S. et al. Hum. Brain Mapp. 16, 1–13 (2002). 3. Beauregard, M., Levesque, J. & Bourgouin, P. J. Neurosci. 21, RC165 (2001). 4. Redoute, J. et al. Hum. Brain Mapp. 11, 162–177 (2000). 5. Canli, T. et al. Behav. Neurosci. 115, 33–42 (2001). 6. Canli, T., Sivers, H., Whitfield, S.L., Gotlib, I.H. & Gabrieli, J.D. Science 296, 2191 (2002). 7. Hamann, S.B., Ely, T.D., Hoffman, J.M. & Kilts, C.D. Psychol. Sci. 13, 135–141 (2002). 8. Canli, T., Desmond, J.E., Zhao, Z. & Gabrieli, J.D.E. Proc. Natl. Acad. Sci. USA 99, 10789–10794 (2002). 9. Cahill, L. et al. Neurobiol. Learn. Mem. 75, 1–9 (2001). 10. Holstege, G. et al. J. Neurosci. 23, 9185–9193 (2003).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

326

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

REVIEW

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Neural activity and the dynamics of central nervous system development
Jackie Yuanyuan Hua1,2 & Stephen J Smith1
Recent imaging studies show that the formation of neural connections in the central nervous system is a highly dynamic process. The iterative formation and elimination of synapses and neuronal branches result in the formation of a much larger number of trial connections than is maintained in the mature brain. Neural activity modulates development through biasing this process of formation and elimination, promoting the formation and stabilization of appropriate synaptic connections on the basis of functional activity patterns. The construction of neural circuits is a dynamic process consisting of both formation and elimination events1,2. Stereotyped pruning guided by molecular cues contributes to axon targeting3. Axonal and dendritic arborization in target areas similarly consists of both formation and elimination of neuronal branches and synapses. How the formation and elimination are coordinated and what role neural activity has in this process have been subjects of intensive study. Synapse elimination is perhaps best characterized at the vertebrate neuromuscular junction (NMJ) and at the cerebellar climbing fiber synapses. Abundant functional and morphological evidence suggests that in these synapses, postsynaptic targets initially receive input from multiple axons. As a result of synapse elimination and axon retraction, only one of all axons maintains its input on a given postsynaptic cell in the adult animal1,4,5. In most parts of the central nervous system (CNS), the heterogeneity of neural connections makes it difficult to identify and study multiple axon inputs onto the same target cell. Nevertheless, functional evidence suggests that a common rule in much of CNS development is that postsynaptic cells initially receive widespread inputs, and input elimination sharpens neural response during maturation. For instance, Hubel and Wiesel noted that neurons in the binocular zone of the visual cortex initially respond weakly to visual stimuli applied to either eye, and the visual response becomes crisper and dominated by one eye as the visual system matures6. Immature LGN neurons similarly receive multiple retinal inputs, only 1–3 of which are maintained into adulthood7,8. Developmental refinements of receptive fields have also been observed in visual systems of non-mammalian species9 and in auditory systems10. Observations of developmental refinement of functional connections naturally lead to questions regarding their structural basis. Is the elimination of selected synaptic inputs a result of synapse disassembly or the change of synaptic strength at stable synaptic sites? If nonselective synapses and neuronal branches are produced and then eliminated during development, then how are the processes of formation and elimination choreographed; specifically, are there separate phases of formation and elimination? To one extreme, a ‘sequential’ formation-elimination model may be proposed, where a given neuron initially forms excessive and unconstrained branches and synapses, and then the inappropriate connections are subsequently removed during a phase dominated by elimination. On the other hand, CNS development may proceed through a ‘concurrent’ mode, where the formation and elimination of branches and synapses occur within the same time frame and are nearly balanced (Fig. 1). Anatomical studies provide strong evidence for elimination of structural synapses in CNS. Cajal originally noted that pyramidal neuron spine density is higher during early postnatal development than it is in adulthood11. Later, electron microscopy studies confirmed that synapse density in the primate cortex decreases during late childhood and adolescence12–14. Great efforts to reconstruct all the synaptic connections formed between one retinal ganglion cell and its postsynaptic partners in the adult cat LGN resulted in the estimate that a single retinal ganglion cell axon arbor forms synapses with only four postsynaptic cells15. This estimate indicates that structural synapse elimination must underlie the functional input elimination observed during development7,8. Morphological reconstructions also provided clues about how formation and elimination are coordinated during development. Individual retinogeniculate and geniculocortical arbors reconstructed from brain tissues of young animals have sparse and nonselective branches, while axon arbors reconstructed from older animals have more restricted projection areas and higher branch densities16–18. Thus, the axon arborization process must involve elimination of a limited number of immature connections coupled with the elaboration of remaining ones. Crowley and colleagues recently reported that geniculate fibers innervate proper visual cortical targets before local branch elaboration19, raising concerns that elimination

1Department

of Molecular and Cellular Physiology, Beckman Center B141, Stanford University, Stanford, California 94305, USA. 2Neurosciences Program, Stanford University, Stanford, California 94305, USA. Correspondence should be addressed to S.J.S. ([email protected]). Published online 26 March 2004; doi:10.1038/nn1218

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

327

REVIEW
related, activity-dependent motility of neuronal filopodia24–29, which are fine, motile branch protrusions enriched in the immature nervous system. Dendritic filopodia have average lifetimes on the order of minutes24–26. Widely thought to be potential synapse precursors30, motile filopodia may have a crucial role in regulating the rate of synapse turnover during development. Time-lapse imaging studies also suggest that synapses remodel rapidly in developing neural circuits (Fig. 2c,d). Some of these studies characterize the dynamics of dendritic spines as proxies for synapses and report extensive remodeling of dendritic spines in both developing and adult nervous systems31,32. However, not all synapses form on spines, and not every spine bears a synapse32,33. Thus, spines are not perfect proxies for synapses. A more direct approach to visualize the dynamics of synaptic organization is to use synaptic proteins tagged with genetically encoded fluorescent indicators as ‘synaptic markers’. Postsynaptic densities marked by green fluorescent protein (GFP)tagged PSD-95, a postsynaptic density protein, appear, disappear and move along the dendritic shafts within minutes of imaging in neonatal brain slices34. In cultured hippocampal neurons, more than 20% of the postsynaptic densities turn over in a 24-h period35. These observations likely underestimate the synaptic dynamics in vivo. For comparison, 37% of the presynaptic accumulations marked by GFPtagged VAMP/synaptobrevin form, whereas 27% of the existing VAMP-GFP accumulations disperse, during 2 h of imaging in retinal ganglion cells of X. laevis tadpoles, resulting in a moderate accumulation of new synapses36. Genetically encoded synaptic markers have their own limitations. Whereas fluorescently tagged synaptic proteins are enriched in synapses, aggregations of these proteins are not guaranteed to precisely represent synapses. For example, aggregates of VAMP-GFP may also represent transport packets37,38. A more generic problem is that synaptic proteins are thought to be sequentially recruited to nascent synaptic contacts39. Thus, the presence of no single protein represents the presence of a synapse. Rather, the aggregation of synaptic adhesion proteins may mark the formation of nascent synaptic contacts, whereas recruitment of different neurotransmitter receptor types may mark different stages of synapse maturation and stabilization. For this reason, imaging studies of multiple synaptic proteins will be needed to fully understand synaptic dynamics. Such caveats notwithstanding, time-lapse imaging studies have provided strong evidence that the construction of CNS neural circuits proceeds through concurrent and nearly balanced branch growth and retraction and synapse formation and elimination. No gross temporal division of phases dominated by formation or elimination as proposed by the sequential model of development has been observed. Furthermore, the contrast between neuronal branch and synapse remodeling on the time scales of hours or even minutes and the functional connection refinement over days or weeks implies that the refinement of CNS neural circuits is a ‘trial-anderror’ process consisting of rapid, iterative sampling of a large number of tentative synaptic contacts. Only a fraction of the trial branches and synapses are maintained, resulting in relatively slow accumulation of stable synaptic connections and refinement of the functional connectivity. Determining the quantitative difference between the numbers of synaptic connections formed during development and those maintained into adulthood remains an important task for future experiments. Activity dependence of neural dynamics during development Pioneered by the classic work of Hubel and Wiesel on mammalian visual cortex development, abundant evidence supports that neural

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 1 Two models of neural development. (a) A sequential formationelimination model: neural development proceeds through first a phase of unconstrained and excessive formation of branches and synapses, and then a phase dominated by branch and synapse elimination. (b) A concurrent formation-elimination model: the formation and elimination of synaptic connections both proceed actively during development. As a result, the number of transient branches and synapses formed and eliminated within any given time period exceeds the number of those stabilized at the end of that period.

events may not contribute to the development of geniculocortical connections. Although the early geniculate afferents are mostly confined to columns in the Crowley et al. study, sparse afferents extend outside these columns, in keeping with earlier observations of unconstrained axon branch formation by immature geniculate fibers16. The functional and anatomical data taken together support a concurrent model of neural development: synaptic connections in immature CNS are widespread but sparse, and the elimination of inappropriate connections is coupled to the ramification of appropriate connections to refine the circuit to functional levels as seen in adults. The concurrence of formation and elimination ensures that connections destined for elimination are promptly removed without further elaboration, so that the refinement of neural circuits proceeds efficiently. Given the concurrence of formation and elimination processes during development, it is necessary to ask a quantitative question: How many times more synaptic connections are made and eliminated during development relative to those maintained in the adult neural circuit? In a system undergoing both active formation and elimination, the number of trial branches and synapses formed and then eliminated within a given time period may far exceed those maintained at the end of that period. Therefore, static anatomical approaches are inadequate to address this question; time-lapse imaging is needed. Time-lapse imaging studies increasingly suggest that CNS development is indeed a highly dynamic process of rapid and nearly balanced formation and elimination. Developing neurons undergo a rapid turnover of axon and dendrite branches (Fig. 2a,b). For instance, retinal ganglion cell axons in living Xenopus laevis larvae add 150% and retract 137% of the initial branch tip number in 2 h. Their postsynaptic partners, the tectal neuron dendrites, add 180% and retract 124% of the initial branch tip number during the same period20,21. The higher rate of branch addition leads to a gradual increase in arbor complexity during development. Rapid extension and retraction of branch tips have also been observed during the development of the zebrafish retinotectal system22,23. In addition, many studies have described age-

328

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

REVIEW
activity modulates CNS development40–44. In a system where formation and elimination are concurrent, activity may modulate development by regulating either the rate at which new synaptic connections form or the stability of existing connections. There is evidence supporting both types of modulation. Before discussing such evidence in detail, we should emphasize that it is no trivial matter to determine whether an activity effect on neuronal dynamics reflects an underlying action on the formation of new synaptic connections or an action on the stabilization of existing connections. Time-lapse imaging experiments cannot distinguish these two possibilities if the length of image-sampling intervals exceeds the lifetimes of transient branches and synapses. However, high sampling frequency leads to higher risks in phototoxicity and photobleaching, and encumbers image analysis. As a result, undersampling is a common practice in the field. For example, more than half of the retinal ganglion cell axon branch tips in X. laevis larvae have a lifetime of less than an hour21, whereas most past work on this system used sampling intervals on the scales of hours. The use of two-photon or spinning-disk confocal microscopy to reduce photodamage, and improvement in algorithms for automatic morphological tracing will likely improve our ability to carry out high-frequency time-lapse imaging, and lead to a deeper understanding of the activity-dependence of neural dynamics. Neural activity modulates synapse stabilization Although synaptogenesis occurs in the absence of neural activity45,46, it can also be modulated by neural activity32,47. Neurotransmission, especially NMDA receptor activation by coincident pre- and postsynaptic activity, can promote synapse maturation and stabilization. First, NMDA receptor activation regulates the organization of the postsynaptic cytoskeleton. Moderate NMDA receptor activation slows the turnover of actin filaments within the spine’s actin pool48 and increases the actin filament content in spines49. Such regulation of the actin cytoskeleton is thought to underlie the changes in spine morphology in response to NMDA receptor activation50. Second, NMDA receptor activation recruits AMPA receptors to the postsynaptic membrane51. AMPA receptor activation in mature cultured hippocampal neurons reduces spine motility50 and maintains dendritic spines52. AMPA receptor overexpression has also been observed to increases spine size and density independent of receptor activation53. Third, NMDA receptor activation may regulate the expression of structural proteins to promote synapse maturation54. Homer, an immediate early gene that encodes a synaptic scaffolding protein, promotes the growth and enlargement of mushroom spines when cotransfected with its binding partner Shank in hippocampal culture55. It is possible that similar mechanisms modify the composition of adhesion proteins at the synapse, resulting in stronger synaptic adhesion and higher synapse stability as a result of appropriate synaptic inputs. Finally, NMDA receptor activation induces the synthesis and release of neurotrophins, particularly brainderived neurotrophic factor (BDNF)56,57. BDNF treatment promotes the maturation of presynaptic terminals in dissociated neuronal culture58,59 and increases the number of synapses per axon terminal in vivo36. Thus neurotrophins could serve as mediators for activity-dependent synapse stabilization. Contrary to the stabilizing effects of activity discussed above, Luthi and colleagues observed an increase in spine density in cultured hippocampal neurons induced by long-term NMDA receptor blockage60. How can these findings be reconciled? The long-term pharmacological blockage in the Luthi et al. study may activate homeostatic mecha-

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 2 Time-lapse imaging experiments suggest that neural development is a highly dynamic process of concurrent formation and elimination. (a) Development of a retinal ganglion cell axonal arbor in X. laevis larvae. Top panel, projected confocal images. Lower panel, reconstructed axon arbors. Formation of new branches (red in lower panel) and elimination of existing branches (green in lower panel) occur concurrently. (b) Development of a tectal cell denritic arbor in X. laevis larvae reconstructed from confocal images. The dendritic arbor remodels extensively during 30-min imaging intervals. Branch tip formation (red) and elimination (green) both occur. Arrowhead indicates the axon. (c) Presynaptic remodeling of a X. laevis retinal ganglion cell axon in vivo. Red: DsRed labeled axon. Yellow: VAMP-GFP puncta. Synapse formation (arrows) and elimination (arrowheads) are both observed. (d) Remodeling of the postsynaptic density in a dendrite segment from a cultured hippocampal neuron expressing PSD-95:GFP. Left panels, images of the dendrite branch taken before and after a 24-h interval. Middle panels, binary images showing PSD-95:GFP clusters. Right panel, superimposition of the binary images shows elimination (green) and formation (red) of synaptic clusters. Panel a is reproduced, with permission, from ref. 80; panel b is reproduced, with permission, from ref. 20; panel c is reproduced, with permission, from ref. 36; panel d is reproduced, with permission, from ref. 35.

nisms, which differ from the mechanisms discussed in the previous paragraph based on coincident activity detection. The effects of activity blockage on development have been shown in many cases to be different depending on whether relative activity level or the absolute amount of activity is changed, and whether the change is acute or chronic61–64. To study coincident activity–dependent mechanisms of development, other experimental protocols may be more preferable than global activity blockage, such as selective activity suppression64,65 or patterned synaptic stimuli66,67.

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

329

REVIEW
blockade markedly decreases the branch lifetime of retinal ganglion cell axons and tectal dendrites20,21,71. Neural activity can modulate the organization of the local cytoskeleton to stabilize neuronal branches. Lohmann et al. recorded local, spontaneous rises in calcium in retina explants72. These rises depend on neurotransmission and calcium-induced calcium release (CICR) from intracellular stores in retinal ganglion cell dendrites, and they serve to maintain the retinal ganglion cell dendrites, as perturbing the spontaneous local calcium activity by pharmacological blockage of the CICR signaling pathway leads to dendrite retraction within a few minutes. Furthermore, dendrite retraction can be prevented by focal calcium uncaging, which raises intracellular calcium locally. Because the stabilizing effect of calcium observed by Lohmann et al. is rapid and remains within the dendritic segments where the calcium level is elevated, it likely results from local signaling that modifies the branch cytoskeleton. One molecular pathway by which neural activity may modify the local cytoskeleton is through Rho GTPase signaling. The Rho GTPases are key regulators of the actin cytoskeleton73. In X. laevis tadpoles, optical nerve stimulation catalyzes the switch of Rho GTPases between active and inactive forms within seconds and promote the growth of tectal cell dendrites74. Other evidence suggests that activity may govern branch stability by regulating synapse turnover. A synapse may stabilize the branch bearing it through either providing physical adhesion or activating signalling pathways such as calcium influx-induced cytoskeletal modulation. At the vertebrate NMJ, the activity-dependent differential stabilization of neuromuscular synapses is a key determinant of the eventual motor axon branching pattern1. It may be proposed by analogy that activity-dependent synapse stabilization in the CNS directs axonal and dendritic arbor growth. In support of this idea, Vaughn and colleagues observed predominate dendritic growth of motor neurons in regions of the developing mouse spinal cord where afferents are enriched and synaptic density is high75. In addition, they were able to identify synapses on dendritic filopodia and growth cones. Based on such histological evidence, Vaughn formulated the ‘synaptotropic hypothesis’ of dendrite growth, stating that the maturation and stabilization of a nascent synapse on a given nascent dendrite branch might stabilize that branch, and thus direct arbor growth68 (Fig. 3). According to this hypothesis, new synapses form predominantly on dendritic filopodia. Most filopodia and nascent synapses are promptly eliminated, but a small fraction of nascent synapses receiving appropriate synaptic input persists, leading to the preservation of the corresponding filopodia and their conversion into stable dendrite branch segments. Vaughn further proposed that the synaptotropic hypothesis might also apply to axon development. The relationship between synaptogenesis and axon development was recently examined by Alsina et al.36 by time-lapse imaging of X. laevis retinal ganglion cells expressing VAMP-GFP, a presynaptic marker, and the red fluorescent protein DsRed. The authors observed marked synapse and axon branch remodeling over 24-h recording periods. In comparison to stable branches, branches that were eventually eliminated had, on average, fewer synapses per axon terminal36, suggesting that synapse formation and stabilization leads to the stabilization of axon branches. Thus, regulation of synaptogenesis could be a key link between neurotransmission and the patterning of neuronal arbors. Finally, activity-dependent gene expression potently modulates neuronal arbor growth43, sometimes through stabilizing branches. The expression and enzymatic activity of calcium/calmodulindependent protein kinase II (CaMKII) is regulated by activity76. In an

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 3 Schematic diagram of the growth of a dendrite branch illustrating the synaptotropic hypothesis. The dendrite branch (blue) extends filopodia to form synapses (green dot) with passing axons (brown). Activitydependent synapse stabilization (upper-right panels) leads to the stabilization of the corresponding filopodium and the formation of a new branch; the new branch then grows through successive rounds of selective filopodial stabilization. Synapse elimination (lower-right panel) results in retraction of the corresponding filopodium.

Neural activity modulates synapse formation Because of the lack of high temporal resolution imaging studies, there has been no solid evidence showing that activity directly regulates the rates of synapse formation as opposed to synapse stability. However, activity is widely thought to regulate synapse formation through the regulation of local filopodial dynamics30. Not only axonal, but also dendritic filopodia may initiate synaptic contacts24,33,68; therefore, activity-dependent increases in filopodial dynamics may effectively increase the frequency of synapse formation in developing neurons. Regulation of filopodial dynamics by neurotransmission has been widely observed in slice explant preparations. Kainate receptor activation differentially modulates the motility of axonal filopodia on mossy fibers depending on the developmental stage27. NMDA receptor activation induces the formation of new dendritic protrusions in hippocampal slices69. The growth and maintenance of dendritic filopodia on retinal ganglion cells is regulated by acetylcholine during the peak of cholinergic synapse formation, and subsequently by glutamate during the peak of glutamatergic synapse formation28. Thus neurotransmission mediated by established synapses may promote the growth and maintenance of nearby filopodia and facilitate further synapse formation where functional synaptic contacts exist. Lendvai et al.70 examined the significance of activity-dependent filopodial dynamics in developing rat somatosensory cortex in vivo. Sensory deprivation induced by whisker trimming was found to cause a concomitant reduction of cortical dendritic protrusions and degradation of the sensory receptive field. Interestingly, sensory deprivation did not affect the density or the morphology of spines, suggesting that sensory input does not regulate synapse density per se, but rather biases the sites of synaptogenesis to favor synapse formation where appropriate synaptic inputs have been received. Neural activity modulates branch stability The precise patterning of neuronal arbors contributes to the precise patterning of synaptic connections. Activity is known to regulate neuronal arbor growth through regulating branch stability. In the retinotectal systems of X. laevis and zebrafish larvae, glutamate receptor

330

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

REVIEW
elegant series of in vivo imaging experiments, Wu and colleagues showed that CaMKII expression is necessary and sufficient for the transition of tectal cell dendrites from the immature, dynamic state into the mature, stabilized state77. As another example, overexpression of the activity-regulated candidate plasticity gene 15 (cpg-15) in X. laevis tectal cells concomitantly stabilizes tectal cell dendrites and the retinal ganglion cell axons that contact them78,79. Whether CAMKII and CPG-15 regulate branch stability directly or indirectly through regulating synapse turnover remians a subject for further investigation. Another intriguing theme calling for further investigation is the significance of activity-dependent local protein synthesis in regulating arbor growth80. Could neural activity destabilize branches to cause selective branch retraction? Segregation of retinal ganglion cell axons from opposite eyes in doubly innervated X. laevis tectum is mainly accounted for by the preferential retraction of axon branches from territories dominated by the other eye, and NMDA receptor blockade reduces branch retraction81, suggesting that NMDA receptor activation selectively destabilizes axon branches. Furthermore, the presence of a ‘branch destabilization signal’ and the lack of this signal in neural circuits under impulse activity blockage could underlie the paradoxical axon sprouting induced by tetrodotoxin treatment82. Nitric oxide is released in an activity-dependent fashion83,84 and causes the collapse of growth cones in vivo85, making it an attractive candidate for such a signal. Neural activity modulates neuronal branch formation Definitive evidence for activity-dependent regulation of the rate of neuronal branch formation is absent. But such a role of activity is supported by the stimulatory effects neurotrophins have on branch formation. Nerve growth factor (NGF)-soaked beads induce de novo branch formation from cultured sensory neurons86. Neurotrophin infusion elicits dendritic form changes in cortical slices87 and promotes RGC axon growth in vivo57. Cortical neurons overexpressing BDNF sprout massive numbers of basal dendrites and induce dendrite sprouting from their close neighbors88,89. The newly formed dendrite branches are highly unstable, suggesting that BDNF promotes dendritic arbor growth by accelerating branch dynamics rather than by promoting branch stability. Thus activity-dependent neurotrophin signaling can coordinate the exploratory behavior of axon and dendrite branches to favor the formation of new branches where appropriate synaptic contacts have been established. CONCLUSION Recent advances in biological fluorescence imaging have provided unprecedented opportunities to observe CNS development in real time. One striking feature of CNS development evident from watching neurons develop is that the formation of CNS neural circuits is a highly dynamic process of rapid and concurrent formation and elimination. In immature neural circuits, branches extend and retract; synapses make and break. Only a limited fraction of new connections are maintained in the mature neural circuitry. Neural activity may modulate the formation, as well as the stability, of synapses and neuronal branches to regulate the continual remodeling of synaptic connections in immature neural circuits. Understanding the relationship between synaptogenesis and arbor growth is likely to be a key step in furthering our knowledge of the mechanisms of activity-dependent neural development.
ACKNOWLEDGMENTS We thank J.D. Jontes, L.C. Katz, E.I. Knudsen, L. Luo and J.T. Schmidt for discussions, members of the Smith lab for critical reading of the manuscript, and the US National Institutes of Health and the Vincent Coates Foundation for financial support. Y.H. was supported by a Stanford Graduate Fellowship and a Coates Foundation Fellowship. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 29 August 2003; accepted 21 January 2004 Published online at http://www.nature.com/natureneuroscience/
1. Lichtman, J.W. & Colman, H. Synapse elimination and indelible memory. Neuron 25, 269–278 (2000). 2. Goda, Y. & Davis, G.W. Mechanisms of synapse assembly and disassembly. Neuron 40, 243–264 (2003). 3. Kantor, D.B. & Kolodkin, A.L. Curbing the excesses of youth: molecular insights into axonal pruning. Neuron 38, 849–852 (2003). 4. Crepel, F. Maturation of climbing fiber responses in the rat. Brain Res. 35, 272–276 (1971). 5. Hashimoto, K. & Kano, M. Functional differentiation of multiple climbing fiber inputs during synapse elimination in the developing cerebellum. Neuron 38, 785–796 (2003). 6. Hubel, D.H. & Wiesel, T.N. J. Neurophysiol. 26, 994 (1963). 7. Chen, C. & Regehr, W.G. Developmental remodeling of the retinogeniculate synapse. Neuron 28, 955–966 (2000). 8. Tavazoie, S.F. & Reid, R.C. Diverse receptive fields in the lateral geniculate nucleus during thalamocortical development. Nat. Neurosci. 3, 608–616 (2000). 9. Gaze, R.M., Keating, M.J. & Chung, S.H. The evolution of the retinotectal map during development in Xenopus. Proc. R. Soc. Lond. B Biol. Sci. 185, 301–330 (1974). 10. Jackson, H. & Parks, T.N. Functional synapse elimination in the developing avian cochlear nucleus with simultaneous reduction in cochlear nerve axon branching. J. Neurosci. 2, 1736–1743 (1982). 11. Cajal, R.y. La Textura del Sistema Nerviosa del Hombre y los Vertebrados (Moya,–Madrid, 1899). 12. Huttenlocher, P.R. Synaptic density in human frontal cortex - developmental changes and effects of aging. Brain Res. 163, 195–205 (1979). 13. Huttenlocher, P.R., de Courten, C., Garey, L.J. & Van der Loos, H. Synaptogenesis in human visual cortex—evidence for synapse elimination during normal development. Neurosci. Lett. 33, 247–252 (1982). 14. Rakic, P., Bourgeois, J.P., Eckenhoff, M.F., Zecevic, N. & Goldman-Rakic, P.S. Concurrent overproduction of synapses in diverse regions of the primate cerebral cortex. Science 232, 232–235 (1986). 15. Hamos, J.E., Van Horn, S.C., Raczkowski, D. & Sherman, S.M. Synaptic circuits involving an individual retinogeniculate axon in the cat. J. Comp. Neurol. 259, 165–192 (1987). 16. Antonini, A. & Stryker, M.P. Rapid remodeling of axonal arbors in the visual cortex. Science 260, 1819–1821 (1993). 17. Sretavan, D.W. & Shatz, C.J. Prenatal development of retinal ganglion cell axons: segregation into eye-specific layers within the cat’s lateral geniculate nucleus. J. Neurosci. 6, 234–251 (1986). 18. Campbell, G. & Shatz, C.J. Synapses formed by identified retinogeniculate axons during the segregation of eye input. J. Neurosci. 12, 1847–1858 (1992). 19. Crowley, J.C. & Katz, L.C. Early development of ocular dominance columns. Science 290, 1321–1324 (2000). 20. Rajan, I. & Cline, H.T. Glutamate receptor activity is required for normal development of tectal cell dendrites in vivo. J. Neurosci. 18, 7836–7846 (1998). 21. Rajan, I., Witte, S. & Cline, H.T. NMDA receptor activity stabilizes presynaptic retinotectal axons and postsynaptic optic tectal cell dendrites in vivo. J. Neurobiol. 38, 357–368 (1999). 22. Kaethner, R.J. & Stuermer, C.A. Dynamics of terminal arbor formation and target approach of retinotectal axons in living zebrafish embryos: a time-lapse study of single axons. J. Neurosci. 12, 3257–3271 (1992). 23. Kaethner, R.J. & Stuermer, C.A. Dynamics of process formation during differentiation of tectal neurons in embryonic zebrafish. J. Neurobiol. 32, 627–639 (1997). 24. Ziv, N.E. & Smith, S.J. Evidence for a role of dendritic filopodia in synaptogenesis and spine formation. Neuron 17, 91–102 (1996). 25. Dailey, M.E. & Smith, S.J. The dynamics of dendritic structure in developing hippocampal slices. J. Neurosci. 16, 2983–2994 (1996). 26. Jontes, J.D., Buchanan, J. & Smith, S.J. Growth cone and dendrite dynamics in zebrafish embryos: early events in synaptogenesis imaged in vivo. Nat. Neurosci. 3, 231–237 (2000). 27. Tashiro, A., Dunaevsky, A., Blazeski, R., Mason, C.A. & Yuste, R. Bidirectional regulation of hippocampal mossy fiber filopodial motility by kainate receptors: a two-step model of synaptogenesis. Neuron 38, 773–784 (2003). 28. Wong, W.T. & Wong, R.O. Changing specificity of neurotransmitter regulation of rapid dendritic remodeling during synaptogenesis. Nat. Neurosci. 4, 351–352 (2001). 29. Chang, S. & De Camilli, P. Glutamate regulates actin-based motility in axonal filopodia. Nat. Neurosci. 4, 787–793 (2001). 30. Jontes, J.D. & Smith, S.J. Filopodia, spines, and the generation of synaptic diversity. Neuron 27, 11–14 (2000).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

331

REVIEW
31. Bonhoeffer, T. & Yuste, R. Spine motility. Phenomenology, mechanisms, and function. Neuron 35, 1019–1027 (2002). 32. Trachtenberg, J.T. et al. Long-term in vivo imaging of experience-dependent synaptic plasticity in adult cortex. Nature 420, 788–794 (2002). 33. Fiala, J.C., Feinberg, M., Popov, V. & Harris, K.M. Synaptogenesis via dendritic filopodia in developing hippocampal area CA1. J. Neurosci. 18, 8900–8911 (1998). 34. Marrs, G.S., Green, S.H. & Dailey, M.E. Rapid formation and remodeling of postsynaptic densities in developing dendrites. Nat. Neurosci. 4, 1006–1013 (2001). 35. Okabe, S., Kim, H.D., Miwa, A., Kuriu, T. & Okado, H. Continual remodeling of postsynaptic density and its regulation by synaptic activity. Nat. Neurosci. 2, 804–811 (1999). 36. Alsina, B., Vu, T. & Cohen-Cory, S. Visualizing synapse formation in arborizing optic axons in vivo: dynamics and modulation by BDNF. Nat. Neurosci. 4, 1093–1101 (2001). 37. Ahmari, S.E., Buchanan, J. & Smith, S.J. Assembly of presynaptic active zones from cytoplasmic transport packets. Nat. Neurosci. 3, 445–451 (2000). 38. Ahmari, S.E. & Smith, S.J. Knowing a nascent synapse when you see it. Neuron 34, 333–336 (2002). 39. Ziv, N.E. & Garner, C.C. Principles of glutamatergic synapse formation: seeing the forest for the trees. Curr. Opin. Neurobiol. 11, 536–543 (2001). 40. Zhang, L.I. & Poo, M.M. Electrical activity and development of neural circuits. Nat. Neurosci. 4 (Suppl.), 1207–1214 (2001). 41. Weisel, T.N. Nature 299, 583 (1982). 42. Katz, L.C. & Shatz, C.J. Synaptic activity and the construction of cortical circuits. Science 274, 1133–1138 (1996). 43. Wong, R.O. & Ghosh, A. Activity-dependent regulation of dendritic growth and patterning. Nat. Rev. Neurosci. 3, 803–812 (2002). 44. Cline, H.T. Dendritic arbor development and synaptogenesis. Curr. Opin. Neurobiol. 11, 118–126 (2001). 45. Augustin, I., Rosenmund, C., Sudhof, T.C. & Brose, N. Munc13-1 is essential for fusion competence of glutamatergic synaptic vesicles. Nature 400, 457–461 (1999). 46. Verhage, M. et al. Synaptic assembly of the brain in the absence of neurotransmitter secretion. Science 287, 864–869 (2000). 47. De Paola, V., Arber, S. & Caroni, P. AMPA receptors regulate dynamic equilibrium of presynaptic terminals in mature hippocampal networks. Nat. Neurosci. 6, 491–500 (2003). 48. Star, E.N., Kwiatkowski, D.J. & Murthy, V.N. Rapid turnover of actin in dendritic spines and its regulation by activity. Nat. Neurosci. 5, 239–246 (2002). 49. Fukazawa, Y. et al. Hippocampal LTP is accompanied by enhanced F-actin content within the dendritic spine that is essential for late LTP maintenance in vivo. Neuron 38, 447–460 (2003). 50. Fischer, M., Kaech, S., Wagner, U., Brinkhaus, H. & Matus, A. Glutamate receptors regulate actin-based plasticity in dendritic spines. Nat. Neurosci. 3, 887–894 (2000). 51. Malinow, R. & Malenka, R.C. AMPA receptor trafficking and synaptic plasticity. Annu. Rev. Neurosci. 25, 103–126 (2002). 52. McKinney, R.A., Capogna, M., Durr, R., Gahwiler, B.H. & Thompson, S.M. Miniature synaptic events maintain dendritic spines via AMPA receptor activation. Nat. Neurosci. 2, 44–49 (1999). 53. Passafaro, M., Nakagawa, T., Sala, C. & Sheng, M. Induction of dendritic spines by an extracellular domain of AMPA receptor subunit GluR2. Nature 424, 677–681 (2003). 54. Steward, O. & Schuman, E.M. Protein synthesis at synaptic sites on dendrites. Annu. Rev. Neurosci. 24, 299–325 (2001). 55. Sala, C. et al. Regulation of dendritic spine morphology and synaptic function by Shank and Homer. Neuron 31, 115–130 (2001). 56. Poo, M.M. Neurotrophins as synaptic modulators. Nat. Rev. Neurosci. 2, 24–32 (2001). 57. McAllister, A.K., Katz, L.C. & Lo, D.C. Neurotrophins and synaptic plasticity. Annu. Rev. Neurosci. 22, 295–318 (1999). 58. Collin, C. et al. Neurotrophins act at presynaptic terminals to activate synapses among cultured hippocampal neurons. Eur. J. Neurosci. 13, 1273–1282 (2001). 59. 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). 60. Luthi, A., Schwyzer, L., Mateos, J.M., Gahwiler, B.H. & McKinney, R.A. NMDA receptor activation limits the number of synaptic connections during hippocampal development. Nat. Neurosci. 4, 1102–1107 (2001). 61. Balice-Gordon, R.J. & Lichtman, J.W. Long-term synapse loss induced by focal blockade of postsynaptic receptors. Nature 372, 519–524 (1994). 62. Zhu, J.J. & Malinow, R. Acute versus chronic NMDA receptor blockade and synaptic AMPA receptor delivery. Nat. Neurosci. 5, 513–514 (2002). 63. Zhao, H. & Reed, R.R. X inactivation of the OCNC1 channel gene reveals a role for activity-dependent competition in the olfactory system. Cell 104, 651–660 (2001). 64. Burrone, J., O’Byrne, M. & Murthy, V.N. Multiple forms of synaptic plasticity triggered by selective suppression of activity in individual neurons. Nature 420, 414–418 (2002). 65. Johns, D.C., Marx, R., Mains, R.E., O’Rourke, B. & Marban, E. Inducible genetic suppression of neuronal excitability. J. Neurosci. 19, 1691–1697 (1999). 66. Colicos, M.A., Collins, B.E., Sailor, M.J. & Goda, Y. Remodeling of synaptic actin induced by photoconductive stimulation. Cell 107, 605–616 (2001). 67. Engert, F., Tao, H.W., Zhang, L.I. & Poo, M.M. Moving visual stimuli rapidly induce direction sensitivity of developing tectal neurons. Nature 419, 470–475 (2002). 68. Vaughn, J.E. Fine structure of synaptogenesis in the vertebrate central nervous system. Synapse 3, 255–285 (1989). 69. Maletic-Savatic, M., Malinow, R. & Svoboda, K. Rapid dendritic morphogenesis in CA1 hippocampal dendrites induced by synaptic activity. Science 283, 1923–1927 (1999). 70. Lendvai, B., Stern, E.A., Chen, B. & Svoboda, K. Experience-dependent plasticity of dendritic spines in the developing rat barrel cortex in vivo. Nature 404, 876–881 (2000). 71. Schmidt, J.T., Fleming, M.R. & Leu, B. J. Neurobiol. 58, 328 (2004). 72. Lohmann, C., Myhr, K.L. & Wong, R.O. Transmitter-evoked local calcium release stabilizes developing dendrites. Nature 418, 177–181 (2002). 73. Hall, A. Rho GTPases and the actin cytoskeleton. Science 279, 509–514 (1998). 74. Li, Z., Aizenman, C.D. & Cline, H.T. Regulation of rho GTPases by crosstalk and neuronal activity in vivo. Neuron 33, 741–750 (2002). 75. Vaughn, J.E., Barber, R.P. & Sims, T.J. Dendritic development and preferential growth into synaptogenic fields: a quantitative study of Golgi-impregnated spinal motor neurons. Synapse 2, 69–78 (1988). 76. Lisman, J., Schulman, H. & Cline, H. The molecular basis of CaMKII function in synaptic and behavioural memory. Nat. Rev. Neurosci. 3, 175–190 (2002). 77. Wu, G.Y. & Cline, H.T. Stabilization of dendritic arbor structure in vivo by CaMKII. Science 279, 222–226 (1998). 78. Cantallops, I., Haas, K. & Cline, H.T. Postsynaptic CPG15 promotes synaptic maturation and presynaptic axon arbor elaboration in vivo. Nat. Neurosci. 3, 1004–1011 (2000). 79. Nedivi, E., Wu, G.Y. & Cline, H.T. Promotion of dendritic growth by CPG15, an activity-induced signaling molecule. Science 281, 1863–1866 (1998). 80. Ouyang, Y., Rosenstein, A., Kreiman, G., Schuman, E.M. & Kennedy, M.B. Tetanic stimulation leads to increased accumulation of Ca2+/calmodulin-dependent protein kinase II via dendritic protein synthesis in hippocampal neurons. J. Neurosci. 19, 7823–7833 (1999). 81. Ruthazer, E.S., Akerman, C.J. & Cline, H.T. Control of axon branch dynamics by correlated activity in vivo. Science 301, 66–70 (2003). 82. Chapman, B. Necessity for afferent activity to maintain eye-specific segregation in ferret lateral geniculate nucleus. Science 287, 2479–2482 (2000). 83. Garthwaite, J., Charles, S.L. & Chess-Williams, R. Endothelium-derived relaxing factor release on activation of NMDA receptors suggests role as intercellular messenger in the brain. Nature 336, 385–388 (1988). 84. Brenman, J.E. & Bredt, D.S. Synaptic signaling by nitric oxide. Curr. Opin. Neurobiol. 7, 374–378 (1997). 85. Renteria, R.C. & Constantine-Paton, M. Exogenous nitric oxide causes collapse of retinal ganglion cell axonal growth cones in vitro. J. Neurobiol. 29, 415–428 (1996). 86. Gallo, G., Lefcort, F.B. & Letourneau, P.C. The trkA receptor mediates growth cone turning toward a localized source of nerve growth factor. J. Neurosci. 17, 5445–5454 (1997). 87. 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). 88. Horch, H.W. & Katz, L.C. BDNF release from single cells elicits local dendritic growth in nearby neurons. Nat. Neurosci. 5, 1177–1184 (2002). 89. Horch, H.W., Kruttgen, A., Portbury, S.D. & Katz, L.C. Destabilization of cortical dendrites and spines by BDNF. Neuron 23, 353–364 (1999).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

332

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

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

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Reverse propagation of sound in the gerbil cochlea
Tianying Ren It is commonly believed that the cochlea emits sounds through backward-traveling waves. In the present experiment using a scanning-laser interferometer, I detected forward-traveling but not backward-traveling waves and found that the stapes vibrates earlier than the basilar membrane. These results contradict the current theory and show that the ear emits sounds through the cochlear fluids as compression waves rather than along the basilar membrane as backward-traveling waves. It is widely thought that otoacoustic emissions—sounds generated by the cochlea—propagate along the basilar membrane as backwardtraveling waves to the middle ear and become acoustic emissions in the ear canal1–4(see Supplementary Fig. 1 online). Some preliminary results have been reported5 (Narayan, S.S., Recio, A. & Ruggero, M.A., Abstract 723, Twenty-first Midwinter Research Meeting of the Association for Research in Otolaryngology, St. Petersburg Beach, USA, February 15–19, 1998), however, that are inconsistent with the above theory, and the possibility of an alternative fluid wave has been discussed (see Supplementary Note online). I tested the backward-traveling wave theory by measuring the basilar membrane vibration at the frequency of the otoacoustic emission as a function of longitudinal location, using a newly developed scanninglaser interferometer6. The propagation direction of the basilar membrane vibration was determined by the slope of the phaseversus-longitudinal location curve. A negative slope in the phase data indicates a dominative forward-traveling wave, and a positive slope indicates a backward-traveling wave. The delays of the basilar membrane and stapes vibrations were derived from the phase transfer functions at emission frequencies (Supplementary Methods online). Animal procedures were approved by the Oregon Health & Science University IACUC. Data presented in Figure. 1a were evoked by two 60 dB SPL (0 dB SPL = 20 µPa) tones at 15.455 kHz (f1) and 17.000 kHz (f2), with a frequency ratio (f2/f1) of 1.1. The velocity magnitude curves in Figure 1a (top) clearly show the frequency-dependent longitudinal patterns of basilar membrane vibration. At 17.000 kHz (dashed line), the maximum vibration was located on the basal side, whereas the maximum vibration at 15.455 kHz (dotted line) was on the apex side. The greatest overlap of f1 and f2 was near 2,200 µm. The maximum vibration at 2f1 – f2 (thick solid line in Fig. 1a) was not located at the area of maximum overlap of f1 and f2. Rather, it was at ∼2,700 µm, which is close to the apical end where there is little vibration at f2. At the most overlapped area of f1 and f2 near the basal side (∼2,200 µm),

Figure 1 Magnitudes (shown in logarithmic scale) and phases of basilar membrane vibration at f1, f2 and 2f1 – f2 frequencies. (a–d) The f2/f1 ratios were as follows: 1.1 in a and d, 1.05 in b and 1.2 in c; F2 was 17 kHz in a–c and 12 kHz in d. Despite the change in f2/f1 and f2, all phase curves show a negative relationship with the distance from the cochlear base, indicating forward-traveling waves. All magnitude curves in a–c (top graphs) show the frequency-dependent longitudinal pattern. That is, vibrations at high frequencies occurred on the basal side and those at low frequencies on the apical side.

2f1 – f2 was more than 30 dB below f1 and f2, which is consistent with data measured from a single point on the basilar membrane5,7. Near the apical end, such as at ∼2,800 µm, the 2f1 – f1 magnitude was even greater than f2 by ∼20 dB. In the bottom graph of Figure 1a, phases are plotted against longi-

Oregon Hearing Research Center, Department of Otolaryngology and Head & Neck Surgery, Oregon Health & Science University, 3181 SW Sam Jackson Park Road, NRC 04, Portland, Oregon 97239-3098, USA. Correspondence should be addressed to T.R. ([email protected]). Published online 21 March 2004; doi:10.1038/nn1216

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

333

B R I E F C O M M U N I C AT I O N S
The absence of a detectable backward-traveling wave was confirmed by the finding that the stapes vibrates earlier than the basilar membrane. I recorded magnitude and phase responses of the basilar membrane at the f2 site, the stapes footplate and the emission to two 70 dB SPL tones (f1 and f2) in a sensitive cochlea (Fig. 2). The frequency of f1 was stepped from 9.0 kHz to 16.8 kHz by 200-Hz increments, whereas f2 was fixed at 17 kHz, resulting in a change in 2f1 – f2 from 1 kHz to 16.6 kHz. The emission (thick solid line in Fig. 2a) increased with 2f1 – f2 frequency and reached a maximum near 12 kHz, then decreased near the f2 frequency. The magnitude transfer function of the stapes footplate vibration showed a similar pattern to that of the emission except for the low frequency range. The magnitude of basilar membrane vibration increased with the emission frequency and reached a maximum near the f2 frequency of 17 kHz. The phase curve of the stapes vibration (Fig. 2b) showed the shallowest slope. The calculated delay based on phases (Fig. 2c and Supplementary Methods online) as a function of the emission frequency (Fig. 2d) shows that the stapes vibrates ∼50 µs earlier than the basilar membrane. The facts that the basilar membrane vibration at the emission frequency is dominated by a forward-traveling wave and that the stapes vibrates earlier than the basilar membrane contradict the theory that the cochlea emits sounds through a backward-traveling wave that propagates on the basilar membrane1–4. The results reported here, however, are consistent with Békésy’s paradoxical basilar membrane vibration13, which propagates toward the helicotrema independently of the location of the vibration source along the cochlear length, which has been theoretically demonstrated14. The present data demonstrate that the inner ear emits sound through the cochlear fluids as a compression wave, which vibrates the stapes, launching a forward-traveling wave along the basilar membrane (see Supplementary Fig. 2 and Supplementary Note online).
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS I thank A.L. Nuttall, P. Gillespie and K. Grosh for comments on an earlier version of the manuscript, E.V. Porsov for writing software, and S. Matthews for technical help. Supported by the National Institute on Deafness and other Communication Disorders (NIDCD), and the National Center for Rehabilitative Auditory Research (NCRAR), Portland Veteran’s Administration Medical Center. COMPETING INTERESTS STATEMENT The author declares that he has no competing financial interests.
Received 20 November 2003; accepted 17 February 2004 Published online at http://www.nature.com/natureneuroscience/
1. Kemp, D.T. Hear. Res. 22, 95–104 (1986). 2. Probst, R., Lonsbury-Martin, B.L. & Martin, G.K. J. Acoust. Soc. Am. 89, 2027–2067 (1991). 3. Shera, C.A. & Guinan, J.J. Jr. J. Acoust. Soc. Am. 105, 782–798 (1999). 4. Knight, R.D. & Kemp, D.T. J. Acoust. Soc. Am. 109, 1513–1525 (2001). 5. Robles, L., Ruggero, M.A. & Rich, N.C. J. Neurophysiol. 77, 2385–2399 (1997). 6. Ren, T. Proc. Natl. Acad. Sci. USA 99, 17101–17106 (2002). 7. Cooper, N.P. & Rhode, W.S. J. Neurophysiol. 78, 261–270 (1997). 8. Rhode, W.S. J. Acoust. Soc. Am. 49 (Suppl. 2), 1218–1231 (1971). 9. Khanna, S.M. & Leonard, D.G. Science 215, 305–306 (1982). 10. Russell, I.J. & Nilsen, K.E. Proc. Natl. Acad. Sci. USA 94, 2660–2664 (1997). 11. Narayan, S.S., Temchin, A.N., Recio, A. & Ruggero, M.A. Science 282, 1882–1884 (1998). 12. Olson, E.S. Nature 402, 526–529 (1999). 13. von Békésy, G. Experiments in Hearing (McGraw-Hill, New York, 1960). 14. Zwislocki, J.J. J. Acoust. Soc. Am. 25, 986–989 (1953).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 2 Frequency responses of the basilar membrane, stapes and earcanal sound pressure to constant f2 (17 kHz) and frequency-varied f1. Magnitudes (a), phases (b,c) and delay (d) at 2f1 – f2 of the emission (thick solid line), basilar membrane (BM; dotted line) and the stapes (dashed line) vibrations. (a) The magnitude transfer function of the stapes footplate vibration shows a pattern similar to that of the emission but different from that of the BM vibration at the F2 place. (b,c) The phase curve of the stapes vibration shows the shallowest slope, indicating the shortest delay. (d) The delay of stapes vibration is approximately 50 µs shorter than that of the BM vibration.

tudinal location. The phases of f1 and f2 show a negative relationship with distance from the cochlear base, indicating forward-traveling waves. However, the phase curve at 2f1 – f2 does not show a backward-traveling wave. Indeed, the 2f1 – f2 phase (thick solid line) shows a negative relationship with distance from the base, which closely matches the phase response to an externally given 35 dB SPL tone at 13.909 kHz (thin solid line). The phase data show that the propagation direction of the basilar membrane vibration at the 2f1 – f2 frequency is dominated by a forward-traveling wave. To observe the effects of the f2/f1 ratio (the smaller the ratio, the more overlap between the two traveling waves at f1 and f2) on the propagation direction of the basilar membrane vibration, I used various f2/f1 ratios to evoke the emissions. For example, I recorded responses to 60 dB SPL tones at 16.190 kHz and 17 kHz (f2/f1 ratio of 1.05; Fig. 1b), as well as responses evoked by 60 dB SPL tones at 14.167 kHz and 17 kHz (f2/f1 = 1.2; Fig. 1c). The responses with the smaller 1.05 ratio (Fig. 1b) indicated less separation between f1 and f2, more basal location of 2f1 – f2, and smaller phase slope difference among f1, f2 and 2f1 – f2 (compare bottom graphs in Fig. 1a and b). With the larger ratio (Fig. 1c) opposite effects were found. These data confirm that not only f1 and f2, but also 2f1 – f2, vibrate mainly as forward-traveling waves8–12. It is possible that the putative backward-traveling wave at 2f1 – f2 may be at a longitudinal location basal to the observed region in Figure 1a–c. To test this possibility, I moved the emission generation site toward the apex using low-frequency primary tones. Data in Figure 1d were evoked by 70 dB SPL tones at 10.909 and 12 kHz (f2/f1 = 1.1). The negative phase slope of 2f1 – f2 (9.818 kHz) indicates a forward-traveling wave.

334

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

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

Cholecystokinin-mediated suppression of feeding involves the brainstem melanocortin system
© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience Wei Fan, Kate L J Ellacott, Ilia G Halatchev, Kanji Takahashi, Pinxuan Yu & Roger D Cone Hypothalamic pro-opiomelanocortin (POMC) neurons help regulate long-term energy stores. POMC neurons are also found in the nucleus tractus solitarius (NTS), a region regulating satiety. We show here that mouse brainstem NTS POMC neurons are activated by cholecystokinin (CCK) and feedinginduced satiety and that activation of the neuronal melanocortin-4 receptor (MC4-R) is required for CCK-induced suppression of feeding; the melanocortin system thus provides a potential substrate for integration of long-term adipostatic and short-term satiety signals. Hypothalamic POMC neurons tonically inhibit food intake1 and are regulated by the long-term adipostatic factor leptin2–4. However, the central melanocortin system is also important in the acute regulation of satiety; in particular, central administration of melanocortins reduces food intake by decreasing meal size, a hallmark of satiety5,6. These hypothalamic neurons send fibers to MC4-R target sites in both the hypothalamus and brainstem, and melanocortin agonists administered to either region inhibit feeding7. Notably, in addition to expression in the arcuate nucleus of the hypothalamus (ARC), POMC is also expressed in the caudal aspect of the NTS8, the primary site of synapse of vagal afferent fibers transmitting satiety information from the gastrointestinal system. NTS neurons are activated by either electrical or CCK-induced stimulation of vagal afferent fibers. Furthermore, leptin and CCK act synergistically to inhibit feeding and activate NTS neurons9. Yet regulation of POMC cells in the NTS by metabolic state has not been reported. Here, we test the hypotheses that (i) brainstem POMC neurons are activated by satiety signals and (ii) central melanocortin signaling is required for the action of specific signals that acutely inhibit feeding. Intraperitoneal (i.p.) injection of CCK-8s (the sulfated 8-aminoacid form of cholecystokinin) significantly increased c-Fos immunoreactivity in the NTS (saline 3 ± 1 cells per section, n = 6; CCK-8s 3.5 µg/kg, 54 ± 11 cells per section, n = 4; 10 µg/kg, 80 ± 11 cells per section, n = 4; P < 0.001) (Fig. 1; compare Fig. 1a,d), as shown previously9. Immunohistochemical experiments using a previously characterized transgenic mouse that expresses enhanced green fluorescent protein (EGFP) under the control of the POMC promoter4 showed no significant difference in the number of POMC-EGFP–immunoreactive (IR) neurons in the NTS between saline- (Fig. 1b) and CCK-8streated mice (Fig. 1e). Of the NTS POMC-EGFP neurons, >30% coexpressed c-Fos immunoreactivity after CCK-8s treatment (Fig. 1f,g). c-Fos expression in the ARC did not differ significantly between groups treated with i.p. saline or CCK-8s (data not shown). We also examined a model of feeding-induced satiety. We gave POMC-EGFP mice a 5-d training in which they were allowed access to food for two

Figure 1 CCK-8s activates POMC neurons in the NTS. (a) Saline (i.p.) activates c-Fos (red) in only a few NTS neurons (arrows). Scale bar = 70 µm. (b) Anti-GFP antibodies detect POMC neurons (green) in NTS of the EGFPPOMC mouse. (c) POMC neurons are not activated by saline treatment. (d) CCK-8s (10 µg/kg, i.p.) activates c-Fos (red) in NTS neurons. (e) CCK-8s (10 µg/kg, i.p.) does not alter expression of POMC in NTS (compare b,e). (f) CCK-8s (10 µg/kg, i.p.) activates c-Fos in NTS POMC neurons (red, c-Fos; green, GFP; yellow-orange, c-Fos + GFP). (g) ∼30% of NTS POMC neurons are activated by i.p. CCK-8s (3.5 or 10 µg/kg; ***, P < 0.001 vs. saline, statistical test done by one-way ANOVA with Dunnett’s post hoc test). (h) Receipt of long-term adipostatic signals and acute satiety signals by POMC neurons in ARC and NTS, respectively. Blue, nuclei containing POMC neurons; yellow, nuclei containing MC4-R neurons that may serve to integrate adipostatic and satiety signals. Red arrows, adipostatic signaling; green arrows, satiety signaling. BST, bed nucleus of stria terminalus; CEA, central nucleus of amygdala; PVN, paraventricular nucleus of hypothalamus; LH, lateral hypothalamic area; LPB, lateral parabrachial nucleus; AP, area postrema; DMV, dorsal motor nucleus of vagus. All studies followed the NIH Guide For the Care and Use of Laboratory Animals and were approved by the Oregon Health and Sciences University Animal Care and Use Committee. See Supplementary Methods online for details.

periods totaling 5 h (9:00–10:00 h, 14:00–18:00 h) and examined c-Fos immunoreactivity in the ARC and NTS at 11:00 h (see Supplementary Fig. 1 online; fed n = 8; fasted, n = 3, *** P < 0.001). Feeding activated c-Fos expression in ∼21% of ARC POMC neurons but also in 13% of NTS POMC neurons (Supplementary Fig. 1). Although a majority of c-Fos-IR cells in the ARC were POMC positive, only a few percent of those in the NTS were, showing the complexity of cells in the NTS

Vollum Institute, Oregon Health and Science University, Portland, Oregon 97239-3098, USA. Correspondence should be addressed to R.D.C. ([email protected]) or W.F. ([email protected]). Published online 14 March 2004; doi:10.1038/nn1214

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

335

B R I E F C O M M U N I C AT I O N S a
Cumulative food intake (g) 2.0 1.5 1.0 0.5 0.0
C57 saline n = 7 C57 CCK n = 8 MC3-R – / – saline n = 10 MC3-R – / – CCK n = 9

b
2.0

* * * *

Cumulative food intake (g)

C57 saline n = 8 C57 CCK n = 7 MC4-R – / – saline n = 7 MC4-R – / – CCK n = 7

1.5 1.0 0.5 0.0
30 60 120 180

* * * * *
30

* * * *
60

* * * *

* * *

* * *

* *

120

180

Time (min)

Time (min)

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

c
15 Cumulative food intake (g) 10

Cumulative food intake (g)

Saline saline n = 4 Saline CCK n = 5 SHU9119 CCK n = 5 SHU9119 saline n = 4

d
10

Saline saline n = 6 Saline CCK n = 8 SHU9119 CCK n = 8 SHU9119 saline n = 6

* * * * *
60 120

8 6 4 2 0

* * * * * * * * * *

5

* * * *
40

0

40

60

120

Time (min)

Time (min)

Figure 2 Brainstem MC4-R signaling is required for CCK-8s-induced feeding inhibition. (a) Mc3r–/– mice are fully responsive to CCK-induced inhibition of feeding. After a 16-h fast, 5–10-month wild-type C57BL/6J (C57) and Mc3r–/– mice (MC3-RKO) were injected i.p. with saline or CCK8s (3 nmol/kg); the strains showed a comparable anorexigenic response to CCK-8s 30–180 min after treatment. (b) MC4-R is required for CCKinduced inhibition of feeding. After a 16-h fast, 9-week wild-type and Mc4r–/– mice (MC4-RKO) were injected i.p. with saline or CCK-8s (3 nmol/kg). CCK-8s significantly reduced food intake in wild-type but not Mc4r–/– mice. (c) Pharmacological blockade of central melanocortin receptors in rats partially blocks CCK-induced inhibition of feeding. Rats received 3rd-ventricle injections of a subthreshold dose of SHU9119 (0.375 nmol/4 µl) 10–15 min before i.p. injection of CCK-8s (3 nmol/kg). (d) Pharmacological blockade of brainstem melanocortin receptors in rats fully blocks CCK-induced inhibition of feeding. Rats received 4th-ventricle injections of a subthreshold dose of SHU9119 (0.2 nmol/4 µl) just before i.p. injection of CCK-8s (3 nmol/kg). Data given as mean ± s.e.m. Statistical analyses were done using one-way ANOVA (a,b) or unpaired ttest (c,d). Data presented as mean ± s.e.m. *, P < 0.05; **, P < 0.01; ***, P < 0.001. See Supplementary Methods online for details.

(Fig. 2b), at time points from 30 to 180 min. Next we examined the site of action of endogenous melanocortins. We administered the MC3-R and MC4-R antagonist SHU9119 (ref. 14) to rats via either the 3rd or 4th ventricle to assess the relative contributions of forebrain and brainstem MC4-R target sites in CCK-mediated inhibition of feeding. We used subthreshold doses of SHU9119 previously determined not to stimulate food intake by these routes. Third-ventricular injection of SHU9119, expected to access both forebrain and brainstem MC4-R sites, partially attenuated the inhibition of food intake induced by i.p. injection of CCK-8s (Fig. 2c). Fourth-ventricle injection, which dyeinjection tests had shown primarily accesses brainstem sites, completely attenuated the CCK-8s-induced inhibition of food intake (Fig. 2d). Both CCK-8s and normal food-induced satiety activated a small group of NTS POMC neurons. These brainstem POMC cells are distinct from previously characterized GLP-1-positive and catecholaminergic NTS neurons. CCK-8s-induced inhibition of feeding also seems to depend on MC4-R signaling. These findings support a model in which brainstem MC4-R neurons, and possibly NTS POMC neurons, contribute to the satiety effects of CCK and other mealrelated satiety signals. Recently, electrical activation of cranial visceral afferents in the solitary tract was reported to activate POMC NTS neurons (Appleyard, S.M. et al., Soc. Neurosci. Abstr. 29, 231.11, 2003); however, the role of NTS POMC neurons in the perception of mealrelated satiety has not been established. The distribution of POMC neurons in the ARC, where they are sensitive to the adipostatic hormone leptin, and the NTS, where they are responsive to vagally mediated satiety signals, makes the central melanocortin system ideally suited for the integration of acute regulation of feeding behavior with the long-term control of energy stores (Fig. 1h) Resistance to factors such as CCK may explain, in part, the profound hyperphagia and increased meal size seen in obese subjects with mutations in Mc4r15.
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS Supported by US National Institutes of Health grants DK55819 (R.D.C.) and DK62179 (W.F.), and a grant from the Wellcome Trust (K.L.J.E.). POMC-EGFP mice were a kind gift of M. Low (Oregon Health and Science University). COMPETING INTERESTS STATEMENT The authors declare competing financial interests; see Nature Neuroscience website for details.
Received 15 September 2003; accepted 12 February 2004 Published online at http://www.nature.com/natureneuroscience/
1. Fan, W., Boston, B.A., Kesterson, R.A., Hruby, V.J. & Cone, R.D. Nature 385, 165–168 (1997). 2. Cheung, C.C., Clifton, D.K. & Steiner, R.A. Endocrinol. 138, 4489–4492 (1997). 3. Elias, C.F. et al. Neuron 23, 775–786 (1999). 4. Cowley, M.A. et al. Nature 411, 480-484 (2001). 5. Williams, D.L., Grill, H.J., Weiss, S.M., Baird, J.P. & Kaplan, J.M. Psychopharmacology 161, 47–53 (2002). 6. Azzara, A.V., Sokolnicki, J.P. & Schwartz, G.J. Physiol. Behav. 77, 411–416 (2002). 7. Grill, H.J., Ginsberg, A.B., Seeley, R.J. & Kaplan, J.M. J. Neurosci. 18, 10128–10135 (1998). 8. Joseph, S.A., Pilcher, W.H. & Bennet-Clarke, C. Neurosci. Lett. 38, 221–225 (1983). 9. Wang, L., Martinez, V., Barrachina, M.D. & Tache, Y. Brain Res. 791, 157–166 (1998). 10. Rinaman, L. Am. J. Physiol. 277, R582–R590 (1997). 11. Luckman, S. J. Neuroendocrinol. 4, 149–152 (1992). 12. Butler, A.A. et al. Endocrinol. 141, 3518–3521 (2000). 13. Huszar, D. et al. Cell 88, 131–141 (1997). 14. Hruby, V.J. et al. J. Med. Chem. 38, 3454–3461 (1995). 15. Farooqi, I.S. et al. N. Engl. J. Med. 348, 1160–1163 (2003).

involved in satiety. Previous work has shown that both catecholaminergic and glucagon-like peptide-1 (GLP-1)-positive cells in the NTS are involved in satiety10,11. Dual immunohistochemical analysis showed that although POMC-EGFP–IR cells and tyrosine hydroxylase–IR cells are found in the same region of the NTS, they are not coexpressed in the same neurons (see Supplementary Fig. 2 online). Likewise, POMC and GLP-1 do not colocalize in NTS neurons: POMC-EGFP–IR neurons are focused more medially than GLP-1-IR neurons (Supplementary Fig. 2). To test whether feeding suppression by CCK-8s was dependent on central melanocortin signaling, we examined the ability of CCK-8s to inhibit food intake after a fast in three different mouse lines, two of which carry deletions of the genes encoding melanocortin receptors 3 and 4, respectively: C57BL/6J, C57BL/6J Mc3r–/–12 and C57BL/6J Mc4r–/–13. Administration of CCK-8s i.p. after a 16-h fast produced a ≥50% inhibition of food intake in the first 30 min in both wild-type and Mc3r–/– mice (Fig. 2a) and continued to inhibit food intake for up to 180 min in each strain. We then administered CCK-8s to female Mc4r–/– mice and age-matched female wild-type mice. CCK-8s significantly reduced food intake in wild-type mice, but not in Mc4r–/– mice

336

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

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

A psychophysical test of the vibration theory of olfaction
Andreas Keller & Leslie B Vosshall At present, no satisfactory theory exists to explain how a given molecule results in the perception of a particular smell. One theory is that olfactory sensory neurons detect intramolecular vibrations of the odorous molecule. We used psychophysical methods in humans to test this vibration theory of olfaction and found no evidence to support it.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

A book about the physiologist Luca Turin1, reviewed previously in Nature Neuroscience2 and elsewhere3,4, has generated new interest in the theory that the smell of a molecule is determined by intramolecular vibrations rather than by the molecule’s shape. Vibration theory was introduced in the 1930s5 and was later extended6, but no biological mechanism to convert molecular vibrations into neuronal activation was proposed. As a result, the theory has been largely neglected in the research community. In the 1990s, Turin proposed a transduction mechanism involving inelastic electron tunneling7. Whether because of skepticism or ‘scientific conspiracy’ (as alleged in the book and echoed in most reviews), his predictions have failed to generate empirical tests by other researchers. In the present study, we tested vibration theory’s key psychophysical predictions. All subjects gave informed consent to participate in this study and were tested in a well-ventilated examination room of the Rockefeller University hospital. Procedures were approved by the university’s Institutional Review Board. To minimize observer bias, we used a doubleblind protocol such that neither the subjects nor the test administrator knew the identity of the odorant in a given vial (see Supplementary Methods online). Turin predicts that the smell of a mixture of guaiacol and benzaldehyde has a vanilla character not found in its components because the combined molecular vibrations of benzaldehyde and guaiacol approximate the vibrations of vanillin7. To test this prediction, we asked subjects to rate the vanilla character of benzaldehyde, guaiacol and a 1:1 mixture of both. Subjects were first Figure 1 Additive synthesis and homologous series. (a) Subjects rated (on a 13-point scale13) the familiarized with the individual stimuli at vanilla character of stimuli (1/100 dilutions) presented with an inter-trial interval of 30 s. The two different concentrations under non- benzaldehyde/guaiacol mixture did not have a vanilla character stronger than either of its components (horizontal black line on each bar indicates median, boxed regions indicate 25–75% quantiles, blind conditions. In a subsequent test, whiskers indicate 10–90% quantiles; n = 24 subjects, 12 female; P > 0.05; Newman-Keuls test for vanillin at both concentrations was identi- multiple comparisons after Friedman’s test). The olfactory sensation produced by vanillin is fied with an accuracy of 84%. After being suppressed by trigeminal stimulation14, but at the stimulus concentration used here there was no familiarized with the 13-point rating scale such interference, as is evident by the high score of the three-component mixture. Equivalent results (1 = no vanilla, 13 = extremely vanilla), with the same subjects were obtained at a 1/10,000 dilution (Supplementary Fig. 1a online). Purity of subjects rated the vanilla character of the odors: benzaldehyde >99%, guaiacol 99.7%, vanillin 99.9%. (b) Odor dissimilarity of pairs of aldehydes was rated on a scale from 0 (same) to 10 (very different). Each subject (n = 24, 12 female) individual components and the two- and rated three randomly picked pairs from each of the seven groups (∆0, ∆1, ∆2, ∆3, ∆4, ∆5 and ∆6). three-component mixtures, presented in Odor solutions were 1/10 dilutions. (c) The data shown in b are replotted to compare the median random order. This procedure was done at similarity rating for pairs of aldehydes consisting of two odd, two even, or an odd and an even chain two concentrations: a higher concentration length. No difference between groups was found; see Supplementary Methods online for details.

(1/100 dilution; Fig. 1a) and a lower concentration (1/10,000 dilution; Supplementary Fig. 1 online, panel a). At neither concentration did the mixture of benzaldehyde and guaiacol have a stronger vanilla character than that of its individual components. A similar result was obtained when odor pairs were rated on an odor similarity rating scale (Supplementary Fig. 1 online, panel b). A second prediction of vibration theory as proposed by Turin is that aldehydes with an even number of carbon atoms have a different odor than those with an odd number7. Subjects rated pairs of aldehydes (1/10 dilution) that differed in chain length by up to six carbon atoms. Subjects rated the two aldehydes as smelling more dissimilar as the difference in carbon atom number increased (Fig. 1b). Similar results were obtained with pure aldehydes (Supplementary Fig. 2 online). Contrary to Turin’s prediction, pairs consisting of two odd or two even numbered aldehydes were not perceived as more similar than pairs consisting of an odd and an even numbered aldehyde (Fig. 1c). We found instead, as suggested in previous studies, that the carbon chain length of these molecules is the salient feature sensed by the olfactory system8. A third prediction of Turin’s vibration theory is that acetophenone (AP) and completely deuterated acetophenone (AP-d8), which have the same shape but different molecular vibrations, should have distinguishable smells9. First, subjects rated paired odors for similarity using a 10-point scale (0 = same; 10 = very different). Similarity scores for the AP versus AP-d8 pairing were no different from those of the identical-odorant pairings (Fig. 2a). In addition, we used a triangle test in which subjects were asked to identify the odd stimulus from among three vials (two of which contained the same substance).

Laboratory of Neurogenetics and Behavior, The Rockefeller University, 1230 York Avenue, Box 63, New York, New York 10021, USA. Correspondence should be addressed to A.K. ([email protected]). Published online 21 March 2004; doi:10.1038/nn1215

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

337

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

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 2 Isotope substitution. (a) The similarity between the smells of regular acetophenone (AP) and deuterated acetophenone (AP-d8) was rated on a scale from 0 (same) to 10 (very different) (horizontal black line on each bar indicates median, boxed regions indicate 25–75% quantiles, whiskers indicate 10–90% quantiles; n = 108, 36 and trials, respectively, for the 3 comparisons shown left to right; 36 subjects, 22 female). (b) Subjects easily distinguished r-carvone (r-CAR) from s-carvone (s-CAR) in triangle tests (one test per subject; n = 36 subjects, 22 female), but not AP from AP-d8 (two tests per subject; n = 72). (c) In duo-trio tests, two odors were presented and the subject was asked to identify the one identical to a third reference smell. Each of six subjects (1 female) took this test 30 times over the course of three days (n = 180 trials). (d) Duo-trio tests were performed with different dilutions of both r-carvone/s-carvone and AP/AP-d8 (n = 24 subjects, 12 female). (b–d) The percentage of correct choices and the 95% confidence intervals are shown. The dashed lines indicate chance performance. Chi-square tests were used to compare observed and expected frequencies. ***P < 0.001; *P < 0.05. Purity of odors: AP 99.3%, AP-d8 99.9%, r-carvone/s-carvone >99%.

To verify that subjects understood the task, we included enantiomers (r-carvone and s-carvone) that are readily discriminable10 and differ in shape but not vibration. Subjects easily distinguished the enantiomers but could not distinguish AP from AP-d8 (Fig. 2b). Finally, we used a duo-trio test in which two stimuli were presented and the subject was asked to identify the one identical to a third reference smell. In a separate session, we tested six subjects who had successfully distinguished AP from AP-d8 to determine whether their correct selections reflected chance performance or true discrimination of these two odorants. None of the six subjects was able to distinguish the two smells. The proportion of correct choices ranged from 43% to 67% (mean, 53%; standard error ± 14%; Fig. 2c). To rule out interference of the trigeminal chemosensory system with olfactory perception seen at high stimulus concentrations11, we used duo-trio tests to show that AP and AP-d8 were not distinguished at a wide range of concentrations (Fig. 2d). It has recently been reported that naive subjects perceive a difference between the odors of deuterated and regular benzaldehyde, but this previous study12 was not run double-blind and used an anomalous version of the duo-trio test. Taken as a whole, our results provide no evidence that regular and deuterated acetophenone smell different to naive subjects. We cannot exclude the possibility, however, that olfactory training or experience could alter the outcome of the tests done here. After testing a variety of psychophysical predictions of vibration theory, as formulated by Turin, we conclude that molecular vibrations alone cannot explain the perceived smell of an odorous molecule.

Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS We thank A. Gilbert for expert advice, members of the Vosshall laboratory for comments on the manuscript, and E. Gotschlich, B. Coller and the staff of the Rockefeller University Hospital. A.K. is an M.S. Stoffel Fellow in Mind, Brain and Behavior. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 18 December 2003; accepted 21 January 2004 Published online at http://www.nature.com/natureneuroscience/
Burr, C. The Emperor of Scent (Random House, New York, 2002). Gilbert, A.N. Nat. Neurosci. 6, 335 (2003). Maslin, J. The New York Times February 6 (2003), p. E8. Givhan, R. The Washington Post February 16 (2003), p. X–T8. Dyson, G.M. Chem. Ind. 647–651 (1938). Wright, R.H. J. Theor. Biol. 64, 473–502 (1977). Turin, L. Chem. Senses 21, 773–791 (1996). Laska, M. & Teubner, P. Chem. Senses 24, 263–270 (1999). Turin, L. & Yoshii, F. in Handbook of Olfaction and Gustation (ed. Doty, R.L.) 275–294 (Marcel Dekker, New York, 2003). 10. Laska, M. & Teubner, P. Chem. Senses 24, 161–170 (1999). 11. Livermore, A. & Hummel, T. Int. Arch. Occup. Environ. Health 75, 305–313 (2002). 12. Haffenden, L.J.W., Yaylayan, V.A. & Fortin, J. Food Chem. 73, 67–72 (2001). 13. Watson, W.L., Laing, D.G., Hutchinson, I. & Jinks, A.L. Dev. Psychobiol. 39, 137–145 (2001). 14. Kobal, G. & Hummel, C. Electroencephalogr. Clin. Neurophysiol. 71, 241–250 (1988). 1. 2. 3. 4. 5. 6. 7. 8. 9.

338

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

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

Parietal somatosensory association cortex mediates affective blindsight
Silke Anders1, Niels Birbaumer1, Bettina Sadowski2, Michael Erb3, Irina Mader3, Wolfgang Grodd3 & Martin Lotze1 © 2004 Nature Publishing Group http://www.nature.com/natureneuroscience To investigate the neural substrates underlying emotional feelings in the absence of a conscious stimulus percept, we presented a visual stimulus in the blind field of partially cortically blind patients and measured cortical activity (by functional magnetic resonance imaging, fMRI) before and after the stimulus had been paired with an aversive event. After pairing, self-reported negative emotional valence and blood oxygen level–dependent (BOLD) responses in somatosensory association areas were enhanced, whereby somatosensory activity predicted highly corresponding reported feelings and startle reflex amplitudes across subjects. Our data provide direct evidence that cortical activity representing physical emotional states governs emotional feelings. In this study, we aimed to identify the neural substrate underlying emotional feelings in the absence of a conscious stimulus percept1. Emotional stimuli presented in a cortical visual field defect modulate reflexive emotional responses via extrastriate projections to the amygdala2–5. In the absence of a cortical stimulus representation, emotional experiences might rely on cortical activity representing the physiological state of the organism. Based on clinical6 and neuroimaging studies7, somatosensory association areas within the anterior parietal cortex have been suggested to represent internal states during emotional processing. Thus, we hypothesized that neural activity in this region should be increased when patients report emotional feelings in the absence of a conscious stimulus percept, and that it should furthermore predict the degree of correspondence between self-reported emotional valence and reflexive emotional responses. Nine patients with postgeniculate lesions resulting in partial disconnection or destruction of the left (n = 6) or right (n = 3) primary visual cortex participated in the study. Written informed consent was obtained from all participants. High resolution T1 and diffusion tensor magnetic resonance images confirmed lesions corresponding to visual field defects identified by Tübingen Perimetry in all patients (see Supplementary Fig. 1 and Supplementary Table 1 online). While subjects focused on a central fixation cross, a male face with a neutral expression was randomly presented in the left and right visual hemifield. After 8 habituation trials on either side, 8 of 16 face presentations in each hemifield were paired with an aversive human scream (Fig. 1). Thus, all patients received an equal number of paired and unpaired left and right hemifield stimulations, but only blind-field stimulations of each patient were included in the analyses. The size of the facial stimulus was individually adjusted to assure that blind-field stimulation was confined to the absolute visual field defect (approximately 6 × 8.5°). Average luminance of the visual stimulus and the background were matched. Fixation was controlled during the entire experiment8, and trials with saccades leading to an overlap between
1Institute

Figure 1 Experimental design. After a baseline of 32 ± 8 s, the face was presented for 12 s. Left and right hemifield presentations were randomly intermixed. After 16 consecutive presentations of the face without aversive stimulation, 8 out of 16 presentations of the face in either hemifield were paired with an aversive scream (2.2 s). White-noise startle probes (50 ms) occurred during half of the baseline intervals and half of the face presentations (not shown).

the visual stimulus and the intact part of the visual field were discarded. fMRI data were processed with a linear model approach using SPM99 software9. Conjunction10 in random-effects analysis revealed voxels commonly activated in patients with left and right visual field defects. Bonferroni correction was used to correct for multiple comparisons within the search volume, comprising cortical gray matter voxels of a standard brain11 (19,449 voxels). Startle eyeblink amplitudes and skin conductance responses (SCR) were recorded during scanning, and after blocks of four blind-field stimulations, subjects rated12 the valence of their emotional feelings during two additional blind-field stimulations, intermixed with two blank-screen presentations. For all measures, responses during blind-field stimulation were contrasted with responses during the blank-screen baseline, separately for pre-pairing and pairing blocks, and then subtracted. All patients denied any visual sensation during blind-field stimulation, and analysis of fMRI data showed no activity in the region corresponding to the calcarine sulcus during blind-field stimulation. Startle eyeblink amplitudes of five patients (the remaining four subjects did not tolerate startle probe intensities that reliably elicited the reflex), and negative emotional feelings of all nine patients were significantly enhanced during blind field stimulation, relative to blank screen baseline, after the face had been paired with the aversive scream (startle, t = 2.39; valence, t = –1.93; P < 0.05; Fig. 2). BOLD responses of eight patients (one subject made saccades during 30% of the face presentations during scanning and was not included in the fMRI analysis) were significantly increased in the left anterior parietal cortex (t = 5.04, P < 0.05; MNI coordinates11 (x, y, z) = –42, –42, 45; region SII; Fig. 3a). BOLD responses in the right anterior parietal cortex were also increased, but this effect did not reach statistical significance (t = 1.64; (x, y, z) = 39, –48, 42). No significant changes of skin conductance responses were observed. To further evaluate the relation among anterior parietal activity, peripheral physiological responses and reported feeling, we compared

of Medical Psychology and Behavioral Neurobiology, University of Tübingen, Gartenstrasse 29, 72074 Tübingen, Germany. 2Department of Pathophysiology of Vision and Neuroophthalmology, University Eye Hospital Tübingen, Schleichstrasse 12-16, 72076 Tuebingen, Germany. 3Section for Experimental MR of the CNS, Department of Neuroradiology, University Hospital Tübingen, Hoppe-Seyler-Strasse 3, 72076 Tübingen, Germany. Correspondence should be addressed to S.A. ([email protected]). Published online 14 March 2004; doi:10.1038/nn1213

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

339

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

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 2 Startle eyeblink amplitudes and reported emotional valence during blind-field stimulation, relative to blank screen baseline, before (pre) and after (post) the visual stimulus had been paired with an aversive scream. Symbols distinguish individual patients.

activity in the highest activated voxel with startle potentiation and negative valence ratings across subjects. Parietal activity was more increased in patients who showed a stronger startle potentiation, but a large increase in startle activity did not necessarily lead to strong parietal activity. Although there was a positive relation between parietal activity and reported negative valence (r = 0.60), parietal activity was best predicted by the level of correspondence between reported negative valence and startle reflex potentiation, computed as the rank difference between the two measures (r = 0.79; Fig. 3b). Our data confirm that aversive responses to visual stimuli can be elicited after destruction of the primary visual cortex5,13. These responses have been ascribed to an extrastriate pathway to the amygdala, comprising the superior colliculus of the tectum and the pulvinar of the thalamus3. In humans, amygdala activity has been observed in response to aversive visual stimuli presented in a postgeniculate visual field defect, and this activity covaried with neural activity in the posterior thalamus4. In both animals14 and humans15, startle reflex amplitudes are modulated by direct projections from the central nucleus of the amygdala to the nucleus reticularis pontis caudalis in the brain stem. Neural activity in cortical areas involved in the representation of the internal state of an organism such as the startle response may constitute part of the basis of emotional feelings6. The anterior parietal cortex with its numerous inputs from the internal milieu represents such an area7. The results of the present study provide direct evidence that anterior parietal activity is linked to the level of the correspondence between reported emotional experiences and startle reflex potentiation. We propose that the neural circuit that mediates emotional experience in the absence of conscious stimulus perception incorporates the previously described subcortical pathway to the amygdala2–4. This pathway may modulate reflexive responses and, via peripheral or direct feedback, cortical activity that constitutes part of the neural basis for emotional feelings.
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS We thank H. Flor, K. Mathiak, R. Veit, N. Weiskopf, L. Weiskrantz and D. Wildgruber for helpful discussions, B. Newport, M. Hülsmann and B. Wietek for technical support, and H.O. Karnath, P. Stoerig and U. Schiefer for permitting us to include patients from their wards. This study was partly supported by the Volkswagen Foundation and the Junior Science Program of the Heidelberger Academy of Sciences and Humanities. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.

Figure 3 Cortical activity during blind-field stimulation after the visual stimulus had been paired with the aversive event. (a) A statistical parametric map (SPM) superimposed on coronal (top) and horizontal (bottom) sections of a standard brain11, contrasting BOLD responses during blind field stimulation, relative to blank screen baseline, before and after pairing. BOLD responses were significantly increased in the left anterior parietal cortex. (b) Scatter plots showing the relation between left anterior parietal activity and startle potentiation (top), reported negative valence (middle), and the level of correspondence between reported negative valence and startle reflex potentiation (bottom), computed as rank difference between the two measures. Numbers refer to individual patients.

Received 17 November 2003; accepted 13 February 2004 Published online at http://www.nature.com/natureneuroscience/
1. Weiskrantz, L. Brain 126, 265–266 (2003). 2. LeDoux, J.E. Curr. Opin. Neurobiol. 2, 191 (1992). 3. Linke, R., De Lima, A.D., Schwegler, H. & Pape, H.C. J. Comp. Neurol. 403, 158–170 (1999). 4. Morris, J.S., DeGelder, B., Weiskrantz, L., & Dolan, R.J. Brain 124, 1241–1252 (2001). 5. Hamm, A.O. et al. Brain 126, 267–275 (2003). 6. Adolphs, R., Damasio, H., Tranel, D., Cooper, G. & Damasio, A.R. J. Neurosci. 20, 2683–2690 (2000). 7. Damasio, A.R. et al. Nat. Neurosci. 3, 1049–1056 (2000). 8. Kimmig, H., Greenlee, M.W., Huethe, F. & Mergner, T. Exp. Brain Res. 126, 443–449 (1999). 9. Friston, K.J. et al. Hum. Brain Mapp. 2, 189–210 (1995). 10. Price, C.J. & Friston, K.J. Neuroimage 5, 261–270 (1997). 11. Collins, D.L., Neelin, P., Peters, T.M. & Evans, A.C. J. Comput. Assist. Tomogr. 18, 192–205 (1994). 12. Bradley, M.M. & Lang, P.J. J. Behav. Ther. Exp. Psychiatry 25, 49–59 (1994). 13. Rosen, J.B. et al. J. Neurosci. 12, 4624–4633 (1992). 14. Hitchcock, J.M. & Davis, M. Behav. Neurosci. 105, 826–842 (1991). 15. Pissiota, A., Frans, O., Fredrikson, M., Langstrom, B. & Flaten, M.A. Eur. J. Neurosci. 15, 395–398 (2002).

340

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Dopamine neurons release transmitter via a flickering fusion pore
Roland G W Staal, Eugene V Mosharov & David Sulzer1–3
A key question in understanding mechanisms of neurotransmitter release is whether the fusion pore of a synaptic vesicle regulates the amount of transmitter released during exocytosis. We measured dopamine release from small synaptic vesicles of rat cultured ventral midbrain neurons using carbon fiber amperometry. Our data indicate that small synaptic vesicle fusion pores flicker either once or multiple times in rapid succession, with each flicker releasing ∼25–30% of vesicular dopamine. The incidence of events with multiple flickers was reciprocally regulated by phorbol esters and staurosporine. Thus, dopamine neurons regulate the amount of neurotransmitter released by small synaptic vesicles by controlling the number of fusion pore flickers per exocytotic event. This mode of exocytosis is a potential mechanism whereby neurons can rapidly reuse vesicles without undergoing the comparatively slow process of recycling.

Several studies suggest that small synaptic vesicles (SSVs) release neurotransmitter by full fusion as well as through transient fusion pores (‘kiss-and-run’ exocytosis)1–3. Capacitance recordings that monitor changes in plasma membrane surface area indicate that SSV endocytosis at the calyx of Held occurs 50–100 ms after fusion4 and that 5% of fusion events by pituitary SSV-like microvesicles are followed (within 2 s) by endocytosis5. These relatively rapid instances of vesicle endocytosis are consistent with formation of transient fusion pores. In the neuromuscular junction, the kinase inhibitor staurosporine attenuates the release of the amphipathic fluorescent dye FM1-43 more than the release of acetylcholine during SSV fusion6. This suggests that PKC inhibition reduces the SSV fusion pore aperture and may inhibit full fusion. In hippocampal neurons, kiss-and-run, full fusion and an intermediate mode of endocytosis can be observed by labeling SSVs with a pH-sensitive fluorescent protein7. Also in some hippocampal neurons, some SSV fusion events result in partial loss of FM1-43 fluorescence8, suggesting that the vesicles close before FM1-43 release is complete. Neurotransmitters, however, have far higher diffusion coefficients than FM1-43 (refs. 6,9), and it is not known whether transient pore openings are sufficient for release of the entire neurotransmitter content of an SSV. In addition, the kinetics of kiss-and-run exocytosis cannot be determined due to the insufficient temporal resolution of current approaches. We therefore adapted carbon fiber amperometry to record dopamine release from synaptic terminals of cultured rat ventral tegmental area neurons. This technique directly measures dopamine flux with a time resolution that is 2–5 orders of magnitude greater than capacitance, imaging and postsynaptic recordings (Methods). We found that small synaptic vesicles regulate the release of neurotransmitter via rapid flickering of the fusion pore.

RESULTS Dopamine release from midbrain neurons SSVs are the predominant synaptic vesicles in cultured dopamine neurons from the ventral midbrain of rats (>99%)10,11. Neurons were stimulated with either 40 mM K+ (Fig. 1a) or a combination of 80 mM K+ and 20 nM α-latrotoxin (K+/α-LTX; Fig. 1b). α-Latrotoxin inserts into the plasma membrane in a manner facilitated by neurexin-1 and CIRL/latrophilin receptors and forms a cation channel that enables Ca2+ to enter the cell, thereby increasing the number of exocytotic events12 (Table 1 legend10). Both secretagogues elicited a variety of amperometric peaks (Fig. 1c,d). The average number of dopamine molecules recorded per amperometric event was similar for both secretagogues (K+, 15,800 ± 4,000 molecules; K+/α-LTX, 11,600 ± 1,700 molecules, P > 0.1), although events obtained by K+/α-LTX stimulation had smaller amplitudes (maximum current; Imax) (K+, 35.4 ± 3.4 pA; K+/α-LTX, 18.4 ± 2.5 pA, P < 0.05, Mann-Whitney U-test) and increased durations (width at half-height; t1/2) (K+, 108 ± 12 µs; K+/α-LTX, 178 ± 29 µs, P < 0.05). Simple and complex amperometric events The shapes of 80–85% of amperometric events induced by either secretagogue closely resembled those previously reported for dopamine and serotonin release during SSV exocytosis10,11,13,14 (Fig. 1c). Such peaks, which we refer to as ‘simple’ events, consisted of a single rising and a single falling phase (Fig. 1e; Methods). Our random walk simulations of dopamine release indicated that the minimum fusion pore diameter consistent with the flux of dopamine observed in simple events was 1.5–3.5 nm (Methods). Surprisingly, this calculated diameter of the fusion pore in dopaminergic SSVs is nearly identical to estimates of the initial

Departments of 1Neurology and 2Psychiatry, Black 305, 650 West 168th St, Columbia University, New York, New York 10032, USA. 3Department of Neuroscience, New York State Psychiatric Institute, 722 West 168th Street, New York, New York 10032, USA. Correspondence should be addressed to D.S. ([email protected]). Published online 29 February 2004; doi:10.1038/nn1205

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

341

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 1 Dopamine release from axonal varicosities of rat ventral midbrain dopamine neurons. (a,b) Representative segment of current trace showing dopamine release from neurons stimulated with K+ alone (a) or with K+/α-LTX (b). The stimulus was given earlier in a portion of the trace that has been omitted because of the paucity of events. (c,d) Representative examples of simple events (c) and complex events (d). Simple events each have a single rising and falling slope, whereas complex events have multiple flickers, each with distinct rising and falling phases (Methods). (e,f) The upper panels show examples of amperometric current traces; the lower panels show the first derivative (dI/dt) of the currents. In the amperometric traces, the mean background current is indicated by a solid line (upper panel). To be considered an ‘event,’ the dI/dt must cross a 4.5 × r.m.s. threshold (solid line, lower panel). (e) Events with derivatives that cross the 3.0 × r.m.s. threshold (dotted line) only once in a rising trajectory are ‘simple’. (f) Events that cross the 3.0 x r.m.s. threshold multiple times are ‘complex’. The corresponding flickers (1–3) are indicated in the current trace. (g,h) Histograms of simple versus complex event characteristics obtained from amperometric recordings after K+/ α-LTX stimulation (n = 532 simple events and n = 130 complex events from eight sites; see Table 1 for statistics).

fusion pore formed by large dense-core vesicles (LDCVs) in eosinophils, neutrophils and chromaffin cells15–17 even though the volume of LDCVs is 3–4 orders of magnitude larger. The remaining events, which we refer to as ‘complex’ events (Fig. 1d), contained multiple, well-defined rising and falling phases (Fig. 1f; see Methods for flicker detection protocol). Complex events consisted of 2–5 ‘flickers’ that occurred at a mean frequency of ∼4 kHz (Table 1) and decreased in amplitude from the first to the last flicker (Table 2 and Fig. 2a). Complex events also showed significantly longer durations and released a greater number of molecules than simple events (P < 0.05; Fig. 1g,h and Table 1). As durations of consecutive flickers did not increase (Table 2), both the distance between the site of release and the recording electrode, and the diameter and open time of the fusion pore, were apparently unchanged.

Pharmacological regulation of events Previous studies have suggested that phorbol esters can increase the number of secretory events via activation of Munc-13 (refs. 18,19). In addition, they can enhance the calcium sensitivity of transmitter release20,21 and modulate the kinetics of fusion pore formation22,23 via protein kinase C (PKC). The nonspecific kinase inhibitor staurosporine is reported to promote kiss-and-run exocytosis, possibly through inhibition of PKC6,24. We found that phorbol 12,13-dibutyrate (PDBU; 3 µM, 15–30 min) increased the total number of exocytotic events per stimulus (Table 1 legend), consistent with the ability of phorbol esters to increase the size of the readily releasable pool of SSVs17,18,25. In the presence of staurosporine (5 µM, 15–30 min), fewer amperometric events were recorded upon stimulation with K+/α–LTX, and no events were detected when K+ alone was applied. PDBU decreased the

Figure 2 Amplitudes of flickers within complex events. (a) The amplitude (mean ± s.e.m.) of each flicker (Imax) is plotted against the flicker n. Only flickers from complex events with an Imax > 20 pA are shown because of the increased contribution of background noise to smaller flickers (∼3 pA r.m.s., hatched box; n = 21 complex events for untreated, 23 for PDBUtreated and 9 for staurosporine-treated; all were K+/α-LTX-stimulated events). (b) Dependence of flicker Imax on the fraction of total neurotransmitter released per pore opening. The mean values of the experimental data from untreated neurons (solid line, gray circles) yielded a slope of –4.7 ± 1.7 pA/flicker (mean ± s.d.), corresponding to the decrease predicted for release of 26 ± 9% of transmitter content (mean ± s.d.; r2 = 0.97).

342

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
overlap of simple events could be due to the fusion of clustered SSVs where exocytosis of Simple events a single vesicle within a cluster would trigger t1/2 (µs) Number of Imax (pA) the fusion of other vesicles (Fig. 4b). In this dopamine molecules case, the average Imax for flickers within comK+ control 10,400 ± 1,000 35.2 ± 3.3 92 ± 6 plex events would be identical regardless of PDBU 8,300 ± 1,100 30.3 ± 2.9 76 ± 6* order (that is, Imax of flicker 1 = Imax of K+/α-LTX control 10,200 ± 1,500 17.7 ± 2.3 156 ± 30 flicker 2). In a variation of this model, exocyStaurosporine 14,000 ± 1,900 26.1 ± 5.1 164 ± 38 tosis could occur via the fusion of an SSV to PDBU 8,200 ± 1,000 25.7 ± 4.4 91 ± 4 another SSV that had already fused with the plasma membrane (Fig. 4c); in this case, later Complex events flickers would show an increased duration as complex t1/2 (µs) Inter-flicker Flickers/event Number of Imax (pA) a result of diffusional filtering and the diludopamine molecules interval (µs) tion of neurotransmitter inside the two fused + K control 23,700 ± 8,000‡ 26.9 ± 4.8 380 ± 43‡ 240 ± 27 2.06 ± 0.06 vesicles. As has been demonstrated for PDBU 25,800 ± 6,400*‡ 26.4 ± 2.2 461 ± 83‡ 293 ± 71 2.38 ± 0.22 LDCVs30, multiple SSVs might fuse with K+/α-LTX control 18,200 ± 4,900‡ 20.2 ± 4.2 507 ± 48‡ 261 ± 46 2.32 ± 0.03 each other before exocytosis, allowing the Staurosporine 26,400 ± 5,500‡ 33.2 ± 6.4 573 ± 61‡ 322 ± 24* 2.51 ± 0.19 vesicle contents to mix (compound exocytoPDBU 14,200 ± 2,300‡ 19.1 ± 3.2 291 ± 41**†‡ 165 ± 28** 2.19 ± 0.12 sis). The resulting vesicle would either proFor K+-stimulated neurons, the data are presented from untreated neurons (8 sites, 13 ± 4 events per site; mean ± duce a single peak or, if the SSV matrices or s.e.m.) and PDBU-treated neurons (7 sites, 77 ± 56 events per site). No events were detected from neurons treated cores remained intact after SSVs fusion, mulwith staurosporine when K+ alone was used as a secretagogue. For K+/α-LTX-stimulated neurons, the data are tiple peaks with similar amplitudes and durapresented from untreated (8 sites, 85 ± 50 events per site), staurosporine-treated (11 sites, 15 ± 7 events per site) and PDBU-treated neurons (9 sites, 94 ± 41 events per site). Data in the table are shown as mean ± s.e.m. for each tions (Fig. 4d). All of the above hypotheses recording site: *P < 0.05 and **P < 0.005 for PDBU or staurosporine versus control, †P < 0.05 for PDBU versus were contradicted by our experimentally staurosporine, ‡P < 0.05 for complex versus simple events by two-way ANOVA . Inter-flicker intervals are shown for complex events with Imax > 20 pA. determined Imax and t1/2 values of complex event flickers, which decreased sequentially (Fig. 2a and Table 2). In contrast to the above models of multiple SSV fusion, complex incidence of complex events from 15% to 10% (K+ alone) and from 20% to 6% (K+/α–LTX, Fig. 3), whereas staurosporine doubled the events could result from the fusion of a single SSV that forms a rapincidence of complex events to 40% (K+/α–LTX). The number of idly flickering fusion pore (Fig. 4e). In this case, the Imax of each subflickers per complex event did not significantly change according to sequent flicker would decrease because of the reduced SSV treatment or secretagogue used (Table 1). neurotransmitter concentration after each pore opening (Imax of flicker 1 > Imax of flicker 2). This hypothesis was supported by the DISCUSSION observed decrease in the average Imax of successive flickers within Over a decade ago, the amperometric detection of catecholamines complex events (Fig. 2a and Table 2), which corresponded to the released during exocytosis was first applied to chromaffin cells that release of ∼25–30% of the SSV neurotransmitter content per flicker contain LDCVs26. Subsequent studies used amperometry to detect for neurons in all treatment groups (Fig. 2b). The slight decrease in neurotransmitter release from SSVs in neuronal cell bodies13,14,27,28 the t1/2 of flickers within complex events (Table 2) is consistent with and central synaptic terminals10. Our measurements of dopamine random walk simulations, suggesting that this decrease is due to the released from terminals of cultured midbrain neurons show at least filtering applied to the data (data not shown). Although some studies show transient flickering of the fusion pore two types of amperometric events, which we labeled simple and complex. In contrast to simple events, which consisted of single ampero- in LDCVs31,32, the duration of SSV flickers observed here was considmetric peaks, complex events comprised 2–5 flickers that decreased erably shorter than reported for LDCVs (100–150 µs vs. sequentially in amplitude. Although it has long been remarked that exocytosis is not always an all-or-none event29, flickers have not been previously described for SSV exocytosis because other recording tech*# 40 niques do not provide sufficient time resolution to resolve flickers Control within complex events. PDBU
Table 1 Characteristics of simple and complex events elicited by K+ and K+/α-LTX

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Complex events (%)

Staurosporine 20
16 104

64 160

Mechanisms of SSV exocytosis In Figure 4, we illustrate several possible mechanisms of exocytosis that could produce complex events. One scenario is that two or more SSVs may release their contents simultaneously, but at different distances from the recording electrode, thus producing overlapping simple events (Fig. 4a). Such an overlap would need to be well coordinated, as the incidence of complex events in untreated neurons was 200-fold greater than the probability that any two simple events would occur randomly within the duration of a complex event (Methods). In addition, the apparent duration of the events released farther from the electrode would be longer as a result of diffusional filtering (that is, t1/2 of flicker 1 ≠ t1/2 of flicker 2). Alternatively, an
NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

*
48 499

130 662 52 843

*

0

K+

K+ / α-LTX

Figure 3 Pharmacological regulation of the incidence of complex events. The percentages of complex events are shown for each experimental condition. The numbers of complex events/total number of events are indicted within the bars. *P < 0.05 vs. control, #P < 0.005 vs. PDBU, by chi-square test. No events were detected after K+ stimulation of staurosporine-treated neurons.

343

ARTICLES
Table 2 Characteristics of complex event flickers
Flicker Number of dopamine molecules 10,800 ± 800 7,500 ± 1,200* 6,000 ± 2,400* Imax (pA) t1/2 (µs)

1st 2nd 3rd

18.4 ± 2.2 14.2 ± 1.5* 8.4 ± 1.9†*

129 ± 13 110 ± 9 91 ± 8*

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Data are for flickers 1–3 from complex events with Imax > 20 pA in untreated neurons stimulated with K+/a-LTX (mean ± s.e.m.). *P < 0.05 compared with 1st flicker and †P < 0.05 compared with 2nd flicker.

transmitter content. Whereas full fusion of SSVs has been clearly demonstrated33, data from the present study and others7,8 suggest that some synapses primarily use kiss-and-run exocytosis. The presynaptic terminals of midbrain dopamine neurons contain a relatively small number of SSVs34 with an apparently high probability of exocytosis for any given vesicle35. Fusion pore flickering and kiss-and-run exocytosis may be particularly important for such synapses to prevent the loss of SSVs during full fusion and the relatively slow process of endocytosis and recycling. Relevance for dopamine signaling In contrast to fast-acting neurotransmitter systems with well-defined pre- and postsynaptic structures, dopamine neurons form ‘social’ synapses that often lack well-defined active zones35–38. The dopamine reuptake transporters are located some distance away from the release site, enabling the neurotransmitter to diffuse and act on receptors several microns away37,38. Modeling of the diffusion of dopamine released at social synapses in the striatum suggests that the number of molecules released per exocytotic event (quantal size) determines how far dopamine diffuses and how many receptors are activated37–39. Because the number of molecules released during complex events is greater than in simple events, dopamine will diffuse through a larger volume (Fig. 5a,b), activating more receptors for a longer duration (Fig. 5c). In summary, our data provide evidence that second-messenger systems modulate the mode of SSV exocytosis by regulating the number of fusion pore flickers per exocytotic event. Fusion pore flickering may provide neurons with a means to recycle vesicles more efficiently and to control quantal size, thus regulating the spillover of neurotransmitter from a social synapse.

Figure 4 Mechanisms that may explain complex events (left column) and predicted averaged amperometric event shape (right column). (a) Overlap of simple events with spatial separation of release sites. Either vesicle could release first, resulting in either of the scenarios depicted on the right. The carbon fiber electrode (CFE) is 100 times wider than the diameter of the SSVs, and has been omitted in b–e. (b) Exocytosis of clustered vesicles. (c) Vesicle fusion with another vesicle that has already fused to the membrane. (d) If the vesicular matrices remain intact, compound exocytosis would occur without mixing of vesicular contents. If the matrices are labile, the contents would mix, resulting in a single amperometric peak. (e) Transmitter release from a single SSV via a flickering fusion pore.

10,000–500,000 µs, respectively). We also found that they occurred at a much higher frequency than in LDCVs (4,000 Hz vs. 170 Hz)32 and released a far greater fraction of the vesicle’s neurotransmitter (25–30% vs. <1%)32.

Relevance to vesicle recycling For all treatment groups, the number of molecules in complex amperometric events was significantly greater than in simple events (1.7–3.1 fold; P > 0.05; Table 1 and Fig. 1g). Interestingly, the first flicker within complex events was similar to simple events in amplitude (18.4 ± 2.2 vs. 17.7 ± 2.3 pA), number of molecules (10,800 ± 800 vs. 10,200 ± 1,500 molecules) and t1/2 (129 ± 13 vs. 156 ± 30 µs; Tables 1 and 2). These data suggest that simple events may generally represent neurotransmitter Figure 5 Simulated dopamine spillover in the striatum. (a) Diffusion profiles of dopamine released release through short-lived pores that are not from an SSV following one or three openings of the fusion pore (10,000 and 20,000 dopamine open long enough to release an SSV’s entire molecules released, respectively; Table 1) at 4 µm from the release site as determined by random neurotransmitter content, implying kiss- walk simulations. (b) Maximum dopamine concentrations reached at various distances from the and-run exocytosis. Complex events appear release site. The straight dashed line at 10 nM indicates the EC50 for the activation of dopamine 48 levels > 10 nM. to be exocytotic events in which the fusion receptors (1–20 nM) . (c) The duration that receptors are exposed to dopamine Dopamine spillover from a single flicker activates receptors in a 73,500 µm3 sphere (26 µm radius), pore either flickers (opens and closes) or with nearby receptors activated for 480 ms. If the SSV flickers three times, then the volume of the fluctuates (enlarges and constricts) several sphere is 1.7-fold larger (∼125,000 µm3, 31 µm radius) and the duration of receptor activation is times in rapid succession, resulting in the 620 ms. These calculations are based on vesicular dopamine concentrations in L-DOPA pretreated release of a larger fraction of an SSV’s neuro- neurons, which are 3–5 fold higher than in untreated cultures10.

344

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
METHODS
Rat ventral midbrain neuronal cultures. Postnatally derived ventral midbrain neurons were cultured as previously described40. Neurons were preincubated with 100 µM L-DOPA for 30 min prior to recording11. The secretagogues (92 mM NaCl, 40 mM KCl, 10 mM HEPES, 1 mM Na2HPO4, 2 mM MgCl2, 1.2 mM CaCl2, ∼300 mosm and pH 7.4 or 52 mM NaCl, 80 mM KCl, 10 mM HEPES, 1 mM Na2HPO4, 2 mM MgCl2, 1.2 mM CaCl2 and 20 nM α-LTX) were applied by local perfusion through a glass micropipette (Picospritzer, General Valve) for 6 s at 10 p.s.i. and ∼30 µm from the recording site. cules (n) released during exocytotic events are distributed as a function of vesicle volumes so that the cube roots of n result in a normal distribution (Fig. 1g; for review, see ref. 38). As previously reported10, occasional large events (3 of 772 in untreated cultures) were >5 standard deviations greater than the geometric mean of the cube roots of n and were excluded from the data analysis in Table 1. The data in Figure 4 are nonparametric and were analyzed by chi-square test. Estimation of the expected random overlap of simple events. The probability that complex events resulted from the random overlap of simple events can be estimated from the exponential decay of interspike intervals38. The time constant was 545 ms (R2 = 0.997 from untreated cultures; Supplementary Fig. 1 online). The probability of observing an interspike interval less than 0.5 ms (two overlapping events within the duration of complex events) is P = (1– e–0.5/545) × 100 = 0.09%. Simulation of dopamine release from the vesicle. Random walk simulations38 (finite-difference model) of molecular diffusion to an amperometric (‘consuming’) electrode was performed using Excel software (Microsoft). During each time bin (tbin), the flux (J, molecules/s) of dopamine molecules from the vesicle through a fusion pore was calculated as: Nv [molecules] π · (Rpore [cm])2 · 4 · D[cm2/s] π · (Rv[cm])3 a · Cv · D 3 J= = b b[cm]

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Amperometric recordings. A 5-µm diameter carbon fiber electrode held at +700 mV was positioned over a potential release site (Newport micromanipulator MX300R) and lowered until the tissue was slightly depressed10. At this potential, dopamine is oxidized, resulting in the donation of two electrons to the electrode. Thus, the number of molecules reaching the electrode can be estimated from the current38. The current was filtered using a 4-pole 10 kHz Bessel filter built into an Axopatch 200A amplifier (Axon Instruments), sampled at 100 kHz (PCI-6052E, National Instruments) and digitally filtered using a binomial 10 routine (Igor Pro, Wave Metrics) with a –3 dB cut-off of ∼15 kHz. This yielded an overall –3 dB cut-off frequency of >8 kHz and essentially no time distortion for the t1/2 of amperometric events with durations >50 µs. It also broadened events of shorter duration toward 50 µs41. Traces with root mean square (r.m.s.) noise less than 3 pA r.m.s. were analyzed. The background noise was normally distributed with no maxima for any frequency component between 0.6 and 33 kHz. No events were recorded when the applied voltage was adjusted to 0 mV or when the electrode was transiently lifted from an active recording site. Peak detection and flicker analysis. Raw amperometric data were collected and analyzed using a locally written routine in Igor Pro. The first derivative of the current trace (dI/dt) was used to detect amperometric events. The r.m.s. of the dI/dt noise was first measured in a segment of the trace that did not contain peaks. Then, dI/dt was used to detect events that were 4.5-fold larger than the r.m.s. noise. These spikes represented the total population of amperometric events. The beginning and the maximum of each event were at dI/dt = 0 (Fig. 1e,f). The end of an event was defined as the point when the current returned to the baseline value. If there was more than one maximum within an event and the dI/dt of these maxima (flickers) was three-fold larger than the r.m.s. noise, then the event was classified as ‘complex’ (Fig. 1f). Events that included a single peak with one rising and one falling phase or for which the dI/dt of flickers was less than three times the r.m.s. noise were categorized as ‘simple’ events. This approach was relatively conservative in identifying flickers, but the same rules were applied for each treatment. We found that the same flickers were identified independently of their order within a complex event. Due to the shape of complex events, the typical t1/2 value does not accurately reflect the event’s duration. Thus, the duration of complex events was calculated as: (t(fn) – t(f1) + t1/2(f1) + t1/2(fn) 2

(2)

where a is the area, b is the length and Rpore is the radius of a cylindrical pore43. Cv and Nv are the concentration and the number of molecules of neurotransmitter in the vesicle with radius Rv. D is the diffusion coefficient, which is 6.9*10–6 cm2/s for dopamine in aqueous solution44. We used electron micrographs of tyrosine hydroxylase–immunolabeled cultures to determine SSV diameters under the conditions used in the recordings. SSV diameters were similar to those of previous reports11; 50.7 ± 1.4 nm (mean ± s.e.m., n = 49 vesicles in 7 terminals; data not shown). The length of the fusion pore was estimated as 7.5–15 nm, twice the membrane thickness of 5hydroxydopamine-labeled SSVs in this preparation11. The distance between the release site and electrode was varied from 50–400 nm, beyond which the amperometric currents would be too low in amplitude to be identified. The number of molecules (N) encountering the surface of the amperometric electrode during tbin was converted to units of amperometric current (I) using the following formula38:

I[pA/s] =

N[molecules] · 106[µs] tbin [µs] · 3.121 · 106 [molecules · s/pA] (3)

complex = t 1/2

(1)

where t(f1) and t(fn) are the times at Imax, and t1/2(f1) and t1/2(fn) are the durations of the first and the last flickers of complex events. The number of molecules in the first flicker was estimated by subtracting the integral of the subsequent flickers from the integral of the entire complex event. The baseline for the subsequent flickers was estimated as a line from the beginning of the second event to the end of the complex event. Although this approach may slightly overestimate the number of molecules in the first flicker as a result of the nonlinear decay of events, this error would be <2%. Statistical analysis. The data in Table 1 are reported as averages of the mean values from each recording site42, each of which generally represents a single presynaptic terminal10. Data were analyzed by ANOVA of the means unless indicated otherwise42. As reported in several studies, the numbers of mole-

For simulations, the diameter and the length of the fusion pore as well as the number of molecules inside the vesicle were varied until the amplitude and duration of simulated exocytotic events (after resampling and filtering) were the same as the average t1/2 and Imax of the amperometric peaks that had been recorded experimentally (Table 1). Simulated data were re-sampled at 10-µs intervals and filtered using binomial 10 smoothing to mimic the experimental conditions. Given that dopamine is present in high millimolar concentrations within the SSV, it is possible that dissociation of dopamine from a lumenal core is slower than the diffusion of free molecules through the fusion pore. Additionally, the pore length may be longer than the thickness of the plasma membrane45. Thus, the calculated pore diameters represent a minimum estimate. Simulation of dopamine diffusion in striatum. Dopamine spillover was modeled using the diffusion coefficient of dopamine in the striatum, 2.7*10–6 cm2/s (ref. 46). The dopamine concentration next to the fusion pore was calculated as:

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

345

ARTICLES
19. Rhee, J.S. et al. Beta phorbol ester- and diacylglycerol-induced augmentation of transmitter release is mediated by Munc13s and not by PKCs. Cell 108, 121–133 (2002). 20. Zhu, H., Hille, B. & Xu, T. Sensitization of regulated exocytosis by protein kinase C. Proc. Natl. Acad. Sci. USA 99, 17055–17059 (2002). 21. Yang, Y., Udayasankar, S., Dunning, J., Chen, P. & Gillis, K.D. A highly Ca2+-sensitive pool of vesicles is regulated by protein kinase C in adrenal chromaffin cells. Proc. Natl. Acad. Sci. USA 99, 17060–17065 (2002). 22. Wang, P., Wang, C.T., Bai, J., Jackson, M.B. & Chapman, E.R. Mutations in the effector binding loops in the C2A and C2B domains of synaptotagmin I disrupt exocytosis in a nonadditive manner. J. Biol. Chem. 278, 47030–47037 (2003). 23. Graham, M.E., Fisher, R.J. & Burgoyne, R.D. Measurement of exocytosis by amperometry in adrenal chromaffin cells: effects of clostridial neurotoxins and activation of protein kinase C on fusion pore kinetics. Biochimie 82, 469–479 (2000). 24. Scepek, S., Coorssen, J.R. & Lindau, M. Fusion pore expansion in horse eosinophils is modulated by Ca2+ and protein kinase C via distinct mechanisms. EMBO J. 17, 4340–4345 (1998). 25. Gillis, K.D., Mossner, R. & Neher, E. Protein kinase C enhances exocytosis from chromaffin cells by increasing the size of the readily releasable pool of secretory granules. Neuron 16, 1209–1220 (1996). 26. Wightman, R.M. et al. Temporally resolved catecholamine spikes correspond to single vesicle release from individual chromaffin cells. Proc. Natl. Acad. Sci. USA 88, 10754–10758 (1991). 27. Chen, G., Gavin, P.F., Luo, G. & Ewing, A.G. Observation and quantitation of exocytosis from the cell body of a fully developed neuron in Planorbis corneus. J. Neurosci. 15, 7747–7755 (1995). 28. Jaffe, E.H., Marty, A., Schulte, A. & Chow, R.H. Extrasynaptic vesicular transmitter release from the somata of substantia nigra neurons in rat midbrain slices. J. Neurosci. 18, 3548–3553 (1998). 29. Girod, R., Correges, P., Jacquet, J. & Dunant, Y. Space and time characteristics of transmitter release at the nerve-electroplaque junction of Torpedo. J. Physiol. 471, 129–157 (1993). 30. Hafez, I., Stolpe, A. & Lindau, M. Compound exocytosis and cumulative fusion in eosinophils. J. Biol. Chem. 278, 44921–44928 (2003). 31. Alvarez de Toledo, G., Fernandez-Chacon, R. & Fernandez, J.M. Release of secretory products during transient vesicle fusion. Nature 363, 554–558 (1993). 32. Zhou, Z., Misler, S. & Chow, R.H. Rapid fluctuations in transmitter release from single vesicles in bovine adrenal chromaffin cells. Biophys. J. 70, 1543–1552 (1996). 33. Cremona, O. & De Camilli, P. Synaptic vesicle endocytosis. Curr. Opin. Neurobiol. 7, 323 (1997). 34. Pickel, V.M., Nirenberg, M.J. & Milner, T.A. Ultrastructural view of central catecholaminergic transmission: immunocytochemical localization of synthesizing enzymes, transporters and receptors. J. Neurocytol. 25, 843–856 (1996). 35. Garris, P.A., Ciolkowski, E.L., Pastore, P. & Wightman, R.M. Efflux of dopamine from the synaptic cleft in the nucleus accumbens of the rat brain. J. Neurosci. 14, 6084–6093 (1994). 36. Schmitz, Y., Benoit-Marand, M., Gonon, F. & Sulzer, D. Presynaptic plasticity of dopaminergic neurotransmission. J. Neurochem. 87, 273–289 (2003). 37. Gonon, F. et al. Geometry and kinetics of dopaminergic transmission in the rat striatum and in mice lacking the dopamine transporter. Prog. Brain Res. 125, 291–302 (2000). 38. Sulzer, D. & Pothos, E.N. Presynaptic mechanisms that regulate quantal size. Rev. Neurosci. 11, 159–212 (2000). 39. Cragg, S.J., Nicholson, C., Kume-Kick, J., Tao, L. & Rice, M.E. Dopamine-mediated volume transmission in midbrain is regulated by distinct extracellular geometry and uptake. J. Neurophysiol. 85, 1761–1771 (2001). 40. Pothos, E.N., Przedborski, S., Davila, V., Schmitz, Y. & Sulzer, D. D2-Like dopamine autoreceptor activation reduces quantal size in PC12 cells. J. Neurosci. 18, 5575–5585 (1998). 41. Colquhoun, D. & Sigworth, F.J. Fitting and statistical analysis of single-channel records. in Single-Channel Recording (eds. Sakmann, B. & Neher, E.) 483–587 (Plenum, New York, 1995). 42. Colliver, T., Hess, E., Pothos, E.N., Sulzer, D. & Ewing, A.G. Quantitative and statistical analysis of the shape of amperometric spikes recorded from two populations of cells. J. Neurochem. 74, 1086–1097 (1999). 43. Berg, H.C. Random Walks in Biology 152 (Princeton Univ. Press, Princeton, 1983). 44. Nicholson, C. Diffusion of ions and macromolecules in brain tissue. in Monitoring Molecules in Neuroscience Vol. 8 (eds. Rollema, H., Abercrombie, E., Sulzer, D. & Zackheim, J.) 71–73 (Rutgers Press, Newark, New Jersey, 1999). 45. Jeremic, A., Kelly, M., Choo, S.J., Stromer, M.H. & Jena, B.P. Reconstituted fusion pore. Biophys. J. 85, 2035–2043 (2003). 46. Rice, M.E. et al. Direct monitoring of dopamine and 5-HT release in substantia nigra and ventral tegmental area in vitro. Exp. Brain Res. 100, 395–406 (1994). 47. Schmitz, Y., Lee, C.J., Schmauss, C., Gonon, F. & Sulzer, D. Amphetamine distorts synaptic dopamine overflow: effects on D2 autoreceptors, transporters, and synaptic vesicle stores. J. Neurosci. 21, 5916–5924 (2001). 48. Levant, B. The D3 dopamine receptor: neurobiology and potential clinical relevance. Pharmacol. Rev. 49, 231–252 (1997).

C[M] =

N[molecules] · 6.022 · 10

23[molecules/mole]

4 π · (r [dm])3 bin 3

(4)

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

where rbin is the radius of the sphere next to the fusion pore derived as rbin = √(tbin/2D). Dopamine uptake via the dopamine transporter was assumed to follow Michaelis-Menten kinetics, with Vmax = 4.88 µM/s and Km = 0.77 µM (ref. 47). A tutorial on random walk and finite difference simulations and the Excel spreadsheets used for these calculations are available at our laboratory website (http://www.columbia.edu/∼ds43/pore_RW.html).
Note: Supplementary information is available on the Nature Neuroscience website.

ACKNOWLEDGMENTS We thank Q. Al-Awqati, K. Larsen, M. Nirenberg and Y. Schmitz for critique of the manuscript, and A. Petrenko for α-latrotoxin. Supported by the National Alliance for Research on Schizophrenia and Depression, the Lowenstein Foundation, the Parkinson’s Disease Foundation, the National Institute on Drug Abuse and the National Institute of Neurological Disorders and Stroke. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 23 October 2003; accepted 27 January 2004 Published online at http://www.nature.com/natureneuroscience/
1. Valtorta, F., Meldolesi, J. & Fesce, R. Synaptic vesicles: is kissing a matter of competence? Trends Cell Biol. 11, 324–328 (2001). 2. Heuser, J.E. & Reese, T.S. Evidence for recycling of synaptic vesicle membrane during transmitter release at the frog neuromuscular junction. J. Cell Biol. 57, 315–344 (1973). 3. Ceccarelli, B., Hurlbut, W.P. & Mauro, A. Turnover of transmitter and synaptic vesicles at the frog neuromuscular junction. J. Cell Biol. 57, 499–524 (1973). 4. Sun, J.Y., Wu, X.S. & Wu, L.G. Single and multiple vesicle fusion induce different rates of endocytosis at a central synapse. Nature 417, 555–559 (2002). 5. Klyachko, V.A. & Jackson, M.B. Capacitance steps and fusion pores of small and large-dense-core vesicles in nerve terminals. Nature 418, 89–92 (2002). 6. Henkel, A.W. & Betz, W.J. Staurosporine blocks evoked release of FM1-43 but not acetylcholine from frog motor nerve terminals. J. Neurosci. 15, 8246–8258 (1995). 7. Gandhi, S.P. & Stevens, C.F. Three modes of synaptic vesicular recycling revealed by single-vesicle imaging. Nature 423, 607–613 (2003). 8. Aravanis, A.M., Pyle, J.L. & Tsien, R.W. Single synaptic vesicles fusing transiently and successively without loss of identity. Nature 423, 643–647 (2003). 9. Stiles, J.R., Van Helden, D., Bartol, T.M. Jr., Salpeter, E.E. & Salpeter, M.M. Miniature endplate current rise times less than 100 microseconds from improved dual recordings can be modeled with passive acetylcholine diffusion from a synaptic vesicle. Proc. Natl. Acad. Sci. USA 93, 5747–5752 (1996). 10. Pothos, E.N., Davila, V. & Sulzer, D. Presynaptic recording of quanta from midbrain dopamine neurons and modulation of the quantal size. J. Neurosci. 18, 4106–4118 (1998). 11. Pothos, E.N. et al. Synaptic vesicle transporter expression regulates vesicle phenotype and quantal size. J. Neurosci. 20, 7297–7306 (2000). 12. Bittner, M.A. Alpha-latrotoxin and its receptors CIRL (latrophilin) and neurexin 1 alpha mediate effects on secretion through multiple mechanisms. Biochimie 82, 447–452 (2000). 13. Bruns, D. & Jahn, R. Real-time measurement of transmitter release from single synaptic vesicles. Nature 377, 62–65 (1995). 14. Hochstetler, S.E., Puopolo, M., Gustincich, S., Raviola, E. & Wightman, R.M. Realtime amperometric measurements of zeptomole quantities of dopamine released from neurons. Anal. Chem. 72, 489–496 (2000). 15. Hartmann, J., Scepek, S. & Lindau, M. Regulation of granule size in human and horse eosinophils by number of fusion events among unit granules. J. Physiol. (Lond.) 483, 201–209 (1995). 16. Lollike, K., Borregaard, N. & Lindau, M. The exocytotic fusion pore of small granules has a conductance similar to an ion channel. J. Cell Biol. 129, 99–104 (1995). 17. Albillos, A. et al. The exocytic event in chromaffin cells revealed by patch amperometry. Nature 389, 509–512 (1997). 18. Betz, A. et al. Munc13-1 is a presynaptic phorbol ester receptor that enhances neurotransmitter release. Neuron 21, 123–136 (1998).

346

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Tenascin-R mediates activity-dependent recruitment of neuroblasts in the adult mouse forebrain
Armen Saghatelyan1,3, Antoine de Chevigny1,3, Melitta Schachner2 & Pierre-Marie Lledo1
Neuroblasts arising in the adult forebrain that travel to the olfactory bulb use two modes of migration: tangentially, along the rostral migratory stream, and radially, in the core of the olfactory bulb where they start to ascend to the outer layers. Although the mechanisms of tangential migration have been extensively studied, the factors controlling radial migration remain unexplored. Here we report that the extracellular matrix glycoprotein tenascin-R, expressed in the adult mouse olfactory bulb, initiates both the detachment of neuroblasts from chains and their radial migration. Expression of tenascin-R is activity dependent, as it is markedly reduced by odor deprivation. Furthermore, grafting of tenascin-R-transfected cells into non-neurogenic regions reroutes migrating neuroblasts toward these regions. The identification of an extracellular microenvironment capable of directing migrating neuroblasts provides insights into the mechanisms regulating radial migration in the adult olfactory bulb and offers promising therapeutic venues for brain repair.

The olfactory bulb is one of the few structures in the adult forebrain in which there is a continuous supply of newborn neurons1,2. The neural progenitors, which originate from stem cells located in the subventricular zone (SVZ) of the lateral ventricles, follow an intricate migration path before reaching their final position in the olfactory bulb. First, they move tangentially, in chains, along the entire extent of the rostral migratory stream (RMS); once in the bulb, they turn to move radially out of the RMS into the outer layers, where they differentiate into inhibitory interneurons1,2. Despite increasing knowledge about the origin, proliferation and tangential migration of neuroblasts, how they achieve their radial migration to integrate functionally into the bulbar circuitry remains elusive. Notably, radial glia, which are central to axonal guidance and radial migration during development, are no longer present in the adult olfactory bulb2. This implies that neuroblasts arriving in the rostral extension of the RMS of the adult forebrain follow unique migratory pathways quite distinct from those seen at perinatal stages. A recent report has provided evidence that the olfactory bulb–derived extracellular matrix (ECM) molecule reelin affects detachment of neuroblasts from chains. However, the cues instigating the processes that occur once the neuroblasts reach the olfactory bulb3—halting tangential migration, initiating detachment of neuroblasts and facilitating their radial migration—have not yet been characterized. The effect of sensory experience on these processes also still needs to be examined. Here we investigated the possibility that reoriented migration of neuroblasts in the core of the olfactory bulb is orchestrated by a gradient of extracellular cues surrounding the RMS within the olfactory bulb. We found that the expression pattern of the ECM molecule tenascin-R is potentially compatible with such a functional role: tenascin-R is detectable exclusively in the deep layers of the olfactory
1Laboratory

bulb, around the most anterior extension of the RMS (RMSOB), but not within the RMS itself. Tenascin-R is a member of the tenascin gene family and contains a cysteine-rich amino terminal region, epidermal growth factor–like domains, fibronectin type III homologous repeats and a domain homologous to fibrinogen4. Tenascin-R appears to be restricted to the CNS and is expressed by differentiating oligodendrocytes and some inhibitory interneurons at late embryonic stages4,5. The functions of tenascin-R are manifold: the protein binds to voltagedependent Na+ channels6,7, and the conduction velocity of action potentials is reduced in tenascin-R-deficient mice8. In addition, tenascin-R is an important constituent of perineuronal nets surrounding many, but not all, inhibitory interneurons9 and organization of perineuronal nets is perturbed in tenascin-R-deficient mice. Tenascin-R-deficient mice have alterations in the organization of perisomatic synapses10 as well as synaptic transmission and plasticity in the CA1 region of the hippocampus11. Combining in vitro and in vivo approaches, we identified a previously unknown function of tenascin-R in the adult olfactory bulb. We show here that (i) tenascin-R is a key player in directing neuroblasts into their prospective target area, (ii) the extent of its expression correlates strongly with olfactory sensory activity and (iii) grafting tenascin-R-secreting cells into regions that do not receive progenitor neurons reroutes migrating neuroblasts to these areas. Thus, the activity-dependent recruitment of neuroblasts by tenascin-R represents a fundamental mechanism through which neurogenesis in the adult olfactory bulb is regulated and adapted to the level of sensory input. The tenascin-R signaling pathway might also provide a new approach to cell replacement therapies based on rerouting migrating cells.

of Perception and Memory, CNRS URA 2182, Pasteur Institute, 25 rue du Dr. Roux, 75015 Paris Cedex, France. 2Zentrum für Molekulare Neurobiologie, Universität Hamburg, Martinistrasse 52, D-20246 Hamburg, Germany. 3These authors contributed equally to this study. Correspondence should be addressed to P.M.L. ([email protected]). Published online 14 March 2004; doi:10.1038/nn1211

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

347

ARTICLES
Figure 1 Immunohistological detection of tenascin-R in the SVZ-OB pathway of adult mice. (a) Low-magnification image of SVZ-OB pathway immunostained for tenascin-R (TNR; red) and PSA-NCAM (green). (b–d) Sagittal sections of SVZ (b) and RMS (c) and coronal section of olfactory bulb (OB; d) stained for TNR (left) and PSA-NCAM (right). Note absence of staining for TNR in the SVZ, RMS and RMSOB and its presence in the GCL and IPL. (e) High-magnification images of TNR and PSANCAM staining in the olfactory bulb. Scale bars: a, 500 µm; b,c, 50 µm; d, 200 µm; e, 100 µm. AOB, accessory olfactory bulb; CC, corpus callosum; CTX, cortex; EPL, external plexiform layer; GCL, granule cell layer; IPL, internal plexiform layer; LV, lateral ventricle; RMS, rostral migratory stream; RMSOB, rostral migratory stream of the olfactory bulb; ST, striatum; SVZ, subventricular zone.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

RESULTS Tenascin-R expression in adult olfactory bulb To examine the expression pattern of tenascin-R in the adult mouse subventricular zone–olfactory bulb (SVZ-OB) system, we combined immunofluorescence labeling with confocal microscopy. At various magnifications, scanning through the successive sections from the SVZ to the olfactory bulb (Fig. 1a–d) showed strong tenascin-Rpositive staining that was restricted to the granule cell and internal plexiform layers of the olfactory bulb (Fig. 1d,e). The SVZ and RMS, identified using antibodies against the polysialylated form of the neural cell adhesion molecule (PSA-NCAM), a marker for immature neural cells (Fig. 1a–c), were devoid of tenascin-R labeling in all cases. We did not observe staining for tenascin-R in the olfactory bulb of tenascin-R-deficient mice (data not shown). The immunostaining pattern was consistent with previous in situ hybridization data showing that granule cells are the principal source of tenascin-R synthesis in the adult olfactory bulb5. The pattern of expression exclusively in

the deep layers of the olfactory bulb surrounding the RMS, and not within it, indicates that tenascin-R may be involved in recruiting newborn neurons to the adult olfactory bulb. Fewer newborn cells in tenascin-R-deficient olfactory bulb To investigate the protein’s potential role in bulbar neurogenesis, we first used tenascin-R-deficient mice. We gave adult mutant and wildtype mice four pulses of BrdU, a marker for DNA synthesis, and processed their brains for BrdU immunohistochemistry 21 d later (Fig. 2a–d). Most of the BrdU+ nuclei were found scattered through-

Figure 2 Reduced density of newborn cells in the olfactory bulb of tenascin-R-deficient mice at 21 d after BrdU injection. (a) BrdU+ nuclei in the GCL of control (left) and tenascin-R-deficient mutant mice (right) showed a pronounced reduction in the number of newborn cells in the mutant. (b) Mean density of newborn granule cells in control (+/+) and tenascin-R-deficient mice (–/–). (c) BrdU+ nuclei in the GL of control (left) and tenascin-R-deficient mice (right). (d) Mean density of newborn periglomerular cells in control (+/+) and mutant (–/–) mice. **, P < 0.01. (e) Confocal 3D reconstruction of BrdU+ cells (red) in control (left) and tenascin-R-deficient mice (right) stained for the neuronal marker NeuN (green). Reconstructed orthogonal projections are presented as viewed in the x-z (top) and y-z (right) planes. (f) Percentage of BrdU+ cells doublelabeled with NeuN in control (+/+) and tenascin-R-deficient mice (–/–). (g) Double immunostaining for BrdU (red) and the astrocytic marker GFAP (green) in the GCL of control (left) and tenascin-R-deficient mice (right). (h) Percentage of BrdU+ cells double-labeled with GFAP in control and tenascin-R-deficient mice. In total, 940 (from three control mice) and 972 (from four tenascin-R-deficient mice) randomly chosen BrdU+ cells were inspected for NeuN immunostaining. Similarly, 836 and 1292 BrdU+ cells were checked for GFAP immunopositivity. GCL, granule cell layer; GL, glomerular layer. Scale bars: a,b, 100 µm; e, 10 µm; g, 50 µm. Values in histograms are means ± s.e.m.

348

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
quantified the number of mitotically active cells in the SVZ and the RMS 4 h after a BrdU injection. The distribution of BrdU+ cells in the SVZ-OB pathway did not differ between tenascin-R-deficient mice and control mice (Fig. 3a–c). The densities of BrdU+ cells in the SVZ (Fig. 3b,d) and RMS (Fig. 3c,d) were undistinguishable between the two genotypes (2,414 ± 156 versus 2,721 ± 90 cells/mm2 in the SVZ and 2,992 ± 221 versus 3,298 ± 138 cells/mm2 in the RMS for control and mutant mice, respectively; n = 3, P > 0.05). To assess whether tenascin-R deficiency affects tangential migration, we first examined the cytoarchitecture of the SVZ and RMS. The chains of neuroblasts visualized by PSA-NCAM immunostaining in whole-mount preparations of the SVZ were similar in tenascin-Rdeficient and control mice (Fig. 4a). At both low and high magnifications, an extensive network of tangential pathways was readily visible. Most of the chains in this network formed a longitudinal array that was not altered in tenascin-R-deficient mice. Similarly, the organization of PSA-NCAM+ chains along the migratory pathway to the olfactory bulb was the same in both genotypes (Fig. 4b). To characterize the distribution of newborn cells in these chains, mice were given a BrdU injection and were killed 2 h later. In both genotypes, double labeling for PSA-NCAM and BrdU showed dividing precursor cells integrated in chains along the entire pathway (Fig. 4b,c), thus demonstrating that the organization of neuroblast chains is unaffected in the absence of tenascin-R. We then cultured SVZ explants on Matrigel to assess the rate of neuroblast migration. This technique has previously been adapted for the study of tangential cell migration in vitro3,12,13. When SVZ explants from postnatal day 7 (P7) control and tenascin-R-deficient mice were cultured for 20 h, an extensive network of chains formed around the explants (Fig. 4d). There was no difference in the general organization of the network of chains (Fig. 4d,e) or in their length (210 ± 12 µm versus 209 ± 11 µm, for control and mutant mice, respectively, 12 explants from 4 control mice and 8 explants from 3 mutants, P > 0.05; Fig. 4f). Altogether, these results demonstrate that tenascin-R is not involved in proliferation and tangential migration of neuroblasts. This is consistent with our immunohistological observations that tenascin-R is not detectable in the SVZ and RMS. By contrast, high expression of tenascin-R in the adult olfactory bulb might have a role either in the recruitment of neuroblasts from the RMS to the olfactory bulb or in their survival within the olfactory bulb. To determine whether granule cells in the olfactory bulb of tenascin-R-deficient mice might be dying at a higher rate than in controls, we performed terminal deoxynucleotidyl transferase–mediated biotinylated UTP nick-end labeling (TUNEL) to evaluate the extent of apoptosis and quantified TUNEL+ cells in both the granule cell layer and the RMSOB (Fig. 5a,b). Numbers of TUNEL+ cells did not differ between genotypes (3.4 ± 0.4 versus 3.3 ± 0.4 cells per slice in the granule cell layer and 1.2 ± 0.3 versus 1.7 ± 0.5 in the RMSOB of control and mutant mice, respectively; n = 3, P > 0.05; Fig. 5b). The combined data indicate that the reduced density of BrdU+ cells in the mutant olfactory bulb may result from altered radial migration. If this were the case, the reduced density of newborn cells in the olfactory bulb should be accompanied by an accumulation of neuroblasts in the RMSOB. We therefore examined whether BrdU+ cells would appear ‘trapped’ in the RMSOB of tenascin-R-deficient mice. Mice were killed 2 d after BrdU injection, and BrdU+ cells were quantified throughout the entire SVZ-OB pathway. Newborn cells accumulated exclusively in the RMSOB of mutant mice (Fig. 5c,d), but not in the RMS or the SVZ (Fig. 5d), confirming that tenascin-R does not affect proliferation and tangential migration. The accumulation of neuroblasts in the RMSOB of tenascin-R-deficient mice was greater, to a

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 3 Normal proliferation in the SVZ and RMS of tenascin-R-deficient mice. (a) BrdU immunostaining in sagittal sections through the forebrain of control (+/+, left) and tenascin-R-deficient mice (–/–, right) showing the distribution of BrdU+ cells in the SVZ-RMS pathway. (b,c) Highmagnification images of the SVZ (b) and RMS (c) show similar numbers of mitotically active cells. (d) Quantification of BrdU+ nuclei in the SVZ and the RMS in control (+/+) and tenascin-R-deficient mice (–/–). CC, corpus callosum; LV, lateral ventricle; ST, striatum. Data are presented as means ± s.e.m. Scale bars: a, 500 µm; b,c, 100 µm.

out the granule cell layer (Fig. 2a). The mean density of BrdU+ nuclei in the granule cell layer was significantly lower in mutants than in control mice (241 ± 16 versus 421 ± 42 cells/mm2, respectively; n = 4 for each genotype, P < 0.01; Fig. 2b). A similar effect was also seen in the glomerular layer (90 ± 12 versus 148 ± 2 cells/mm2, P < 0.01; Fig. 2c,d). To test whether the 40% reduction seen in tenascin-R-deficient mice was accompanied by alterations in the fate of newborn cells, we determined the proportion of cells double-labeled for BrdU and either the neuronal marker NeuN (Fig. 2e,f) or the glial marker GFAP (Fig. 2g,h). Orthogonal projections through three-dimensionally (3D) reconstructed BrdU+ cells showed that, in both genotypes, about 80% of newborn cells were neurons (n = 4; Fig. 2f). Similarly, although the numbers of BrdU+ GFAP+ cells were very small (only ∼1% of all BrdU+ cells), no difference was seen between genotypes (n = 4; Fig. 2h). Thus, the reduction in the density of BrdU+ cells was similar in the different layers of the mutant olfactory bulb but the fate of newly generated cells remained unchanged. Altered migration in tenascin-R-deficient mice The reduced number of newborn cells in the olfactory bulb could result from decreased cell proliferation and rate of tangential migration, altered chain organization, distorted radial migration and/or reduced survival of newborn neurons. To assess proliferation, we

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

349

ARTICLES
mice. Tenascin-R-deficient mice had 30.1 ± 7.7 more BrdU+ cells per slice in the RMSOB (105.1 ± 7.7 versus 75 ± 7.1 cells per slice for tenascin-R-deficient and control mice, respectively; P < 0.05) and 20.8 ± 9.4 fewer BrdU+ cells per slice in the olfactory bulb, including glomerular, external plexiform and granule cells layers, (62.5 ± 4 versus 83.3 ± 9.4 cells per slice for tenascin-R-deficient and control mice, respectively; P < 0.05). Notably, the total number of BrdU+ cells counted in the olfactory bulb and RMSOB was not significantly different in tenascin-R-deficient as compared to wild-type mice (167.6 ± 10.5 versus 158.3 ± 15.3 BrdU+ cells per slice for tenascin-R- deficient and control mice, respectively; P > 0.05). These results demonstrate that the excess of BrdU+ cells in the RMSOB of tenascin-R-deficient mice is due to the accumulation of cells in that region. Notably, in both genotypes, the density of BrdU+ cells in the glomerular layer was similar at 2 and 21 d after BrdU injection (compare with Fig. 2d). This contrasts sharply with the density of BrdU+ cells counted in the granule cell layer, which increased 20-fold between 2 and 21 d after injection (compare with Fig. 2c). These results imply that neuroblasts first migrate predominantly to the glomerular layer before populating the granule cell layer. Thus, this might explain why the density of newborn cells in tenascin-Rdeficient mice 2 d after BrdU injection was decreased mainly in the glomerular and external plexiform layers and only slightly in the granule cell layer. As cell proliferation, tangential migration and cell death were not affected in tenascin-R-deficient mice, the reduced density of newborn neurons in the olfactory bulb and increased number of neuroblasts in the RMSOB indicate that newborn progenitors are impeded in leaving the RMSOB. We therefore tested the kinetics of accumulation of newborn cells in the RMSOB. When mice were given a pulse of BrdU and killed 4 h later, the density of BrdU+ cells in the RMSOB was higher in tenascin-R-deficient mice than in the controls (129.2 ± 11.7 versus 100.7 ± 3 cells/mm2; P < 0.05; n = 5; Fig. 5g). This increase was even greater 2 d after BrdU injection (Fig. 5g). Indeed, when the data were expressed as the ratio of mutant to control BrdU+ cell densities in the RMSOB, the increase was significantly smaller 4 h after BrdU injection than 2 d after injection (respectively, 1.3 ± 0.09 versus 1.5 ± 0.04, P < 0.05; Fig. 5h). These results strongly imply that the absence of tenascin-R leads to an accumulation of newborn cells in the RMSOB. To rule out the possibility that elevated numbers of neuroblasts in the RMSOB of tenascin-R-deficient mice result from increased local proliferation of neuroblast precursors, we carried out two sets of experiments. First, we injected BrdU and killed the mice 2 h later to specifically label dividing progenitors in the SVZ-OB bulb pathway. The number of BrdU+ cells in the RMSOB stained with PSA-NCAM antibodies did not differ between genotypes (55.0 ± 10.1 versus 49.5 ± 8 cells/mm2 for control and tenascin-R-deficient mice, respectively; P > 0.05; n = 4; see Supplementary Fig. 1 online), indicating similar local proliferation rates. Second, we assessed the number of cells marked by the endogenous cell division marker Ki67 (ref. 14,15). We found that, in agreement with the first experiment, these numbers did not differ in the RMSOB of control and tenascin-Rdeficient mice (85.5 ± 50 versus 61.3 ± 28.2 Ki67+ cells/mm2; P > 0.05; n = 4; see Supplementary Fig. 1). These results show that the absence of tenascin-R reduces the number of newborn bulbar neurons as a result of their accumulation in the RMSOB. This highlights the pivotal role of tenascin-R in initiating radial migration. Tenascin-R fosters neuroblast detachment and migration Before invading the olfactory bulb, neuroblasts halt their tangential migration, detach from their migrating chains and leave the RMSOB.
VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 4 Normal chain organization and tangential migration of neuroblasts in tenascin-R-deficient mice. (a) Whole-mount immunostaining of PSANCAM+ chains in the SVZ of control (+/+, left) and tenascin-R-deficient mice (–/–, right). Insets, high magnification of the PSA-NCAM+ network of chains. (b,c) PSA-NCAM (green) and BrdU (red) immunostaining in the RMS of control (left) and tenascin-R-deficient mice (right) at low (b) and high magnifications (c). (d) Phase-contrast images of SVZ explants of control (left) and tenascin-R-deficient mice (right) cultured in Matrigel for 20 h. (e) High-magnification images of boxed areas in d, showing chain organization of neuroblasts migrating out of the SVZ explants. (f) Quantification of migration distance, showing no difference between control (+/+) and tenascin-R-deficient explants (–/–). Data are presented as means ± s.e.m. Scale bars: a, 100 µm, inset 40 µm; b,d, 100 µm; c, 30 µm; e, 40 µm.

highly significant degree, than for controls (3,646 ± 97 and 2,470 ± 214 cells/mm2, respectively; n = 4; P < 0.01; Fig. 5d) and was accompanied by a reduced density of BrdU+ cells in the olfactory bulb (Fig. 5e,f). The greatest decreases were measured in the external plexiform layer (18 ± 0.8 versus 29.4 ± 4.7 BrdU+ cells/mm2 for mutant and control mice, respectively; P < 0.05) and the glomerular layer (75.1 ± 11.9 versus 104.7 ± 15.3 BrdU+ cells/mm2; P < 0.05; Fig. 5e). A lesser, nonsignificant decrease occurred in the granule cell layer (20.1 ± 2.8 versus 24 ± 2 BrdU+ cells/mm2; P > 0.05). To estimate whether the excess of BrdU+ cells in the RMSOB of tenascin-R-deficient mice is due to the accumulation of cells that have not migrated further, we compared the number of BrdU+ cells in the RMSOB and the olfactory bulb in tenascin-R-deficient and wild-type

350

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 5 Abnormal radial migration in the olfactory bulb of tenascin-R-deficient mutant mice. (a) TUNEL+ nuclei (arrows) in the GCL of control (+/+) and mutant mice (–/–). (b) Quantification of TUNEL+ nuclei in the GCL and the RMSOB of control and tenascin-R-deficient mice. (c) BrdU immunostaining in coronal sections of control (left) and tenascin-R-deficient mice (right) 2 d after BrdU injection. Note the higher density of BrdU+ nuclei in the RMSOB of tenascin-R-deficient mice. (d) Quantification of BrdU+ nuclei in coronal sections from RMSOB to SVZ (120-µm bin size) in control (black squares) and tenascin-R-deficient mice (gray circles) indicates a significant increase in the density of newborn cells exclusively in the RMSOB of tenascin-R-deficient mice. (e) Conversely, a reduced number of newborn cells is seen in the olfactory bulb of tenascin-R-deficient mice. (f) The total number of BrdU+ nuclei in the olfactory bulb, excluding the RMSOB, is 53 ± 7 versus 38 ± 5 BrdU+ cells/mm2 in control and tenascin-R-deficient mice, respectively. (g) Timedependent accumulation of neuroblasts in the RMSOB of tenascin-R-deficient as compared to control mice 4 h and 2 d after BrdU injection. (h) Significant increase in the ratio of BrdU+ cells in the RMSOB of tenascin-R-deficient versus control mice measured 2 d, as compared to 4 h, after BrdU injection. Data are presented as means ± s.e.m. *, P < 0.05; **, P < 0.01. Scale bars: a, 50 µm; c,e, 100 µm.

To examine which of these steps might be regulated by tenascin-R, we combined in vitro and in vivo approaches. Inspection of PSA-NCAM+ staining at the interface between the RMSOB and granule cell layer showed that most neuroblasts migrated individually in control mice, whereas many clustered in the tenascin-R-deficient mice (Fig. 6a). To assess the role of tenascin-R in halting tangential migration and detaching neuroblasts from chains, we cocultured SVZ explants with tenascin-R-expressing or control BHK cells (Fig. 6b–d). When SVZ explants from P7 mice were cocultured for 20 h with BHK cells, an extensive network of neuroblasts migrated out of SVZ explants (Fig. 6c,d). Quantification of the number of individualized neuroblasts per SVZ explant showed that there were significantly more such cells (90% more; P < 0.001) in the presence of BHK cells secreting tenascin-R (Fig. 6e). Notably, the migration distance did not differ between control explants (248.5 ± 7.1 µm; 24 explants from 8 mice) and those exposed to tenascin-R-secreting cells (260.9 ± 6.5 µm; 29 explants from 8 mice; P = 0.2; Fig. 6f). These results indicate that the expression of tenascin-R is instrumental in detaching neuroblasts from chains but not in halting their tangential migration. This in vitro assay, however, did not allow to determine whether tenascin-R is also involved in the reorientation of tangentially migrating neuroblasts as seen in vivo in the core of the olfactory bulb. To examine whether tenascin-R is necessary and sufficient to reroute tangentially migrating neuroblasts, we introduced tenascin-R into forebrain regions that neither are populated by progenitor cells nor express tenascin-R (Fig. 7a). Tenascin-R-expressing or control BHK cells prestained with PKH26 (red labeling in Fig. 7b,c) were grafted into the striatum (Fig. 7b) or just above the horizontal limb of the RMS (hlRMS) (Fig. 7c). Whereas mice receiving control cells showed unaltered migration (upper panels in Fig. 7b,c; n = 8), all mice grafted with transfected cells (n = 7) had neuroblasts (green labeling in Fig. 7b,c) entering the tenascin-R-containing area (blue staining in

lower panels of Fig. 7b,c). Ectopic expression of tenascin-R resulted in a 4- to 6-fold greater number of neuroblasts migrating out of the SVZ (Fig. 7d,e) and hlRMS (Fig. 7f,g; 24.5 ± 6.2 versus 86.7 ± 19.0 and 15.3 ± 10.5 versus 91.9 ± 16.0 for control and tenascin-R-secreting cells injected, respectively, into the striatum and close to the hlRMS; P < 0.005). Notably, although 28.8 ± 5.8% and 35.5 ± 9.3% of neuroblasts that migrated out of SVZ and hlRMS, respectively, were detached from chains (arrows in Fig. 7d,f), many of the rerouted neuroblasts still assembled in the chains (arrowheads in Fig. 7d,f). These results show that tenascin-R not only promotes detachment of neuroblasts from chains but also has an important role in the reorientation of tangentially migrating neuroblasts. Sensory input regulates tenascin-R expression in the bulb Our results indicated that tenascin-R was not only required, but also sufficient, to initiate radial migration in the adult forebrain. We thus decided to investigate whether this important function is regulated in an activity-dependent manner. Because the olfactory bulb is the first central relay that receives direct inputs from sensory neurons, we used unilateral odor deprivation, achieved by occlusion of one nostril, to test whether expression of tenascin-R is sensitive to the level of sensory activity. Nostril occlusion resulted in a small, but significant, decrease in tenascin-R mRNA in the ipsilateral bulb as early as 1–2 d after occlusion, reaching a maximum at 4–30 d after occlusion (Fig. 8a,b). Notably, the reduction was reversible, because reopening the nostril for 5 d after 20 d of occlusion resulted in upregulation of tenascin-R mRNA to the level seen in the control bulb (Fig. 8b). Concomitantly, unilateral nostril occlusion also resulted in lower levels of tenascin-R protein in the granule cell layer by 27 ± 6%, as assessed 20 d after occlusion (Fig. 8c,d; n = 6). Sham-operated control mice showed no changes in tenascin-R mRNA and protein levels (Fig. 8b,d). Sensory deprivation–induced reduction of both mRNA

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

351

ARTICLES
Figure 6 Tenascin-R acts as a detachment signal for tangentially migrating neuroblasts. (a) PSA-NCAM immunostaining in coronal sections of the GCL of control (+/+, left) and tenascin-R-deficient mice (–/–, right) show that PSA-NCAM+ cells migrate individually (arrows) in control mice, whereas they appear clustered (arrowheads) in tenascin-R-deficient mice. (b–d) Phasecontrast image showing cocultures of SVZ explants (left) with an aggregate of BHK cells (right). Examples of SVZ explants cocultured with BHK cells not secreting (c) and secreting (d) tenascin-R (TNR). Right panels are higher magnifications of boxed areas in left panels. Note the increased number of individualized cells in cocultures of SVZ explants with BHK cells secreting TNR. (e) Significantly higher numbers of individualized cells in cocultures of SVZ explants with BHK cells secreting TNR (41.5 ± 5.8 versus 77 ± 5.9). **, P < 0.001. (f) Quantification of migration distance of neuroblasts from SVZ explants cocultured with BHK cells not secreting (Ctrl) or secreting (TNR) TNR. Data are presented as means ± s.e.m.; 21 and 27 explants from 7 mice were analyzed in cultures without and with TNR, respectively. Scale bars: a,d (right), 50 µm; b,d (left), 100 µm.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

and protein was further confirmed by quantitative PCR and western blotting. Tenascin-R mRNA was reduced by 31.6 ± 15% after 7 d of nostril occlusion and tenascin-R protein by 54.9 ± 12.5% after 20 d of occlusion (see Supplementary Fig. 2 online). These results show that tenascin-R expression is regulated by activity. To assess whether this activity-dependent regulation of tenascin-R expression could have a functional implication for the recruitment of newborn cells to the olfactory bulb, we inspected the density of BrdU+ cells in the RMSOB of control and occluded bulbs 2 d after BrdU injection. Similar to the observations with tenascin-R-deficient mice, downregulation of tenascin-R expression by unilateral odor deprivation for 30 d led to an accumulation of neuroblasts in the RMSOB of the occluded as compared to the control bulbs (4,023.9 ± 209.9 versus 3,258.5 ± 228.3 cells/mm2 for odordeprived and control bulbs, respectively; P < 0.05; n = 5; Fig. 8e,f). These results indicate that regulation of tenascin-R expression by sensory input may be crucial in the recruitment of newborn neurons to the olfactory bulb. Notably, the accumulation of neuroblasts in the RMSOB was less pronounced in occluded mice than in tenascin-R-deficient mice (123.5 ± 6.4% versus 147.5 ± 4.1%, respectively; P < 0.05), showing that the level of tenascin-R expression in the olfactory bulb correlates with the degree of neuroblast accumulation in the RMSOB.

DISCUSSION Bulbar neurogenesis offers a unique model for investigating the mechanisms of neuronal recruitment in the adult brain. An important issue concerns the nature of the molecular cues involved in the correct targeting of newborn neuronal precursor cells. Prompted by the restricted and functionally meaningful expression pattern in the adult olfactory bulb, we have identified the ECM molecule tenascin-R as a molecular cue that induces neuroblasts to detach from chains and begin a radial migration away from the RMS into the outer layers of the olfactory bulb. In addition, we have shown that ectopic expression of tenascin-R in non-neurogenic regions reroutes neuroblasts from their normal migratory pathway, thus offering a promising tool for cell replacement therapies. Tenascin-R enhances or decreases neurite outgrowth and neuronal and glial adhesion depending on the cell type and on its association with other ECM molecules16. Furthermore, tenascin-R defasciculates axons of cerebellar granule cells in vitro, thus possibly allowing inhibitory interneurons to invade the tightly fasciculating bundles of axonal processes17. These observations are pertinent in view of the fact that tenascin-R induces neuroblast detachment and radial migration. In this regard the function of tenascin-R may be similar to that of reelin, which has been reported to influence this process in the adult olfactory bulb3. However, whereas reelin affects only neuroblast detachment from chains3, tenascin-R also acts as a directional cue that reroutes neuronal progenitors from their tangential migratory pathway. In addition, it is not clear whether tenascin-R and reelin share the same signaling pathway for inducing neuroblast detachment. It will therefore be interesting to explore the effects on SVZ explants of tenascin-R alone and in combination with reelin. What tenascin-R receptor(s) allows neuroblasts to migrate radially in the adult brain? Tenascin-R may act directly by interacting with particular cellular receptors present on migrating neuroblasts, or indirectly, along with other ECM-associated molecules, by capturing and presenting some growth factors18 necessary for radial migration. Among cell surface receptors for tenascin-R are contactin (also known as F3)19, acetylated gangliosides that influence phosphorylation of focal adhesion kinases and affect integrin function20, and receptor protein-tyrosine phosphatases belonging to the family of chondroitin sulfate proteoglycans (CSPGs)21,22. Although the involvement of contactin and CSPGs (the latter highly expressed in the adult RMS23) in adult neurogenesis remains to be explored, it was recently shown that integrins are not involved in the radial migration of neuroblasts to the olfactory bulb24. Notably, initiation of radial migration in the adult olfactory bulb correlates with the

352

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 7 Ectopic expression of tenascin-R reroutes migrating neuroblasts. (a) BrdU-immunostained sagittal section showing the place of BHK cell grafts (red circles). (b) Control (upper, Ctrl) or tenascin-R-secreting (lower, TNR) BHK cells prelabeled with PKH26 (red staining) were placed into the striatum neighboring the SVZ. The rerouted neuroblasts were quantified by counting PSA-NCAM+ cells (green staining) in the 400-µm-diameter area calculated from the perimeter of the graft. Note the TNR-immunopositive areas (blue staining) in mice injected with BHK cells secreting TNR (lower). (c) As in b, but control and TNR-transfected BHK cells were grafted into the area just above the horizontal limb of the RMS (hlRMS). Right, high magnifications of the boxed areas shown on the left. Arrowheads indicate chains of neuroblasts diverted from the SVZ (b) or RMS (c). (d) Highmagnification images of striatum injected with BHK cells secreting TNR. Arrows and arrowheads indicate, respectively, the individual neuroblasts and chains of progenitor cells rerouted from their normal migratory pathway. (e) Quantification of the number of neuroblasts rerouted from the SVZ by nontransfected (Ctrl) and TNR-transfected BHK cells (TNR). (f,g) As in d and e, but control and tenascin-R-transfected BHK cells were grafted into the area above hlRMS. CC, corpus callosum; CTX, cortex; LV, lateral ventricle. **, P < 0.001. Values in histograms are means ± s.e.m. Scale bars: b,c, 200 (left) and 50 µm (right); d,f, 20 µm.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

appearance of NMDA receptor–mediated currents25. These receptors are involved in neuronal migration during early developmental stages26. It is thus possible that tenascin-R, through its cell surface receptor(s), trigger(s) the expression of functional NMDA receptors and, as a consequence, radial migration. Alternatively, tenascin-R might directly interact with NMDA receptors through a mechanism similar to that described for GABAB receptors27 and thus control radial migration more directly. Whatever its mode of action, we provide here the first demonstration that tenascin-R has a specific and important role in the initiation of radial migration of newborn neurons in the adult forebrain. It is believed that target structures provide attractive and/or survival factors for developing neuronal networks28,29 and that these factors can be regulated in an activity-dependent manner. For instance,

in the olfactory system, modulation of odor information flow has been reported to affect the survival of newborn neurons: sensory deprivation by nostril occlusion reduces the number of granule cells in the developing30–32 and adult olfactory bulb33, whereas reopening of the nostril after early occlusion34 as well as olfactory enrichment in adults35 increase the number of newly formed bulbar interneurons. Here, we show that as well as causing reduced survival of newborn neurons, odor deprivation also impairs radial migration of neuroblasts from the RMS to the olfactory bulb, correlating with downregulation of tenascin-R expression. We therefore propose that tenascin-R is an important mediator relaying network activity to the recruitment of newborn neurons into the olfactory bulb. The ability to generate neurons in the adult brain is relevant to the development of therapeutic strategies aimed at directing the migra-

Figure 8 Activity-dependent expression of tenascin-R in the olfactory bulb. (a) Pseudocolor images showing expression of tenascin-R mRNA in a coronal section through the olfactory bulb receiving inputs from the open nostril (Ctrl) and the nostril closed for 10 d (Occl). Note the decreased TNR in situ hybridization signal in the odor-deprived olfactory bulb. (b) The effect of sensory deprivation on tenascin-R (TNR) mRNA levels was calculated by relating the mean radioactivity (in cpm/mm2) of the occluded bulb to that of the control bulb. (c) TNR immunostaining of a coronal section through the olfactory bulb receiving inputs from open (Ctrl) and closed (Occl) nostrils 20 d after unilateral odor deprivation. Note decreased immunostaining for TNR in the GCL of the odor-deprived bulb. (d) Ratio between mean TNR immunofluorescence intensity (per mm2) in the GCL of the occluded and control bulbs. (e) BrdU+ cells in the RMSOB of the control and occluded bulbs 2 d after BrdU injection. Mice were subjected to unilateral sensory deprivation 30 d before BrdU injection. The sections were counterstained with methyl green; arrowheads indicate BrdU+ cells in the GCL. (f) Quantification of BrdU+ nuclei in the RMSOB of control (Ctrl) and occluded (Occl) bulbs shows a significant increase in the density of newborn cells after sensory deprivation. AOB, accessory olfactory bulb; GCL, granule cell layer; RMSOB, rostral migratory stream of the olfactory bulb. *, P < 0.05; **, P < 0.01. Values in histograms are means ± s.e.m. Scale bars: c, 200 µm and e, 50 µm.

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

353

ARTICLES
tion of endogenous and/or grafted progenitor cells and their detachment from each other. Recently, it has been elegantly demonstrated that endogenous adult stem cells can regenerate functional neurons in non-neurogenic diseased areas36–38. Neuronal progenitors migrate to the damaged areas from the neurogenic source localized in the SVZ, implying that not only increased proliferative activity following brain damage37,38 but also changes either in the migratory capabilities of neuroblasts and/or the microenvironment of the target regions may recruit the newly generated neurons for repair. Finally, our results have potential therapeutic implications, because they indicate that recruitment of neuronal progenitors in diseased brain areas can be enhanced by exogenous application of tenascin-R. METHODS
Mice. Tenascin-R-deficient mice8 2–4 months of age and age-matched wild-type littermates were derived from heterozygous parents with a mixed C57BL/6J×129Ola background. Mice were kept on a 12-h light-dark cycle at constant temperature (22 °C) with food and water ad libitum. At 3–4 months of age, the mice were subjected to unilateral olfactory deprivation through cauterization of one nostril39. All experimental procedures were in accordance with the Society for Neuroscience and European Union guidelines, and were approved by the institutional animal care and use committees of the Pasteur Institute. BrdU injections. The DNA synthesis marker 5-bromo-2′-deoxyuridine (BrdU; Sigma) was dissolved in a sterile solution of 0.9% NaCl and 1.75% NaOH (0.4 N). This solution was injected intraperitoneally at a concentration of 50 mg per kg body weight. BrdU containing cells were detected by immunohistochemistry after different survival times: a single dose of BrdU was given 2 h before killing the mouse to assess proliferation or 4 h before killing to assess both proliferation and initial stages of migration. Two injections of BrdU spaced by 4 h were done 2 d before the mice were killed to evaluate proliferation and migration of neuroblasts. Finally, four injections repeated every 2 h were administrated to mice 21 d before they were killed to evaluate proliferation, migration and survival of newborn cells. Immunohistochemistry. For all histological analyses requiring tissue fixation by perfusion, mice were deeply anesthetized with an overdose of sodium pentobarbital (100 mg per kg body weight; Sanofi) and perfused intracardially with saline solution (0.9% NaCl) containing heparin (5 × 103 units/ml) at 37 °C, followed by 4% paraformaldehyde in 0.1 M sodium phosphate buffer (pH 7.3). The brains were dissected out and immersed overnight in the same fixative at 4 °C. Immunohistochemistry was carried out on 40-µm-thick freefloating coronal or sagittal sections cut with a vibrating microtome (VT1000S, Leica) and collected in PBS. Sections were first incubated overnight at 4 °C with the following monoclonal antibodies: mouse anti-tenascin-R (clone 619; ref. 40), mouse anti-PSA and mouse anti-NeuN (Chemicon), mouse anti-Ki67 (Novocastra), rabbit anti-GFAP (Dako) and rat anti-BrdU (Accurate Scientific, Harlan Sera-Lab). For BrdU immunostaining, the sections were pretreated with 0.2% Triton X-100 for 2 h and DNA was denatured with 2 N HCl for 30 min at 37 °C. An overnight incubation with anti-BrdU at 4 °C was followed by a 3-h incubation at room temperature (19–24 °C) with biotinylated donkey anti–rat IgG, 1-h incubation with avidin-biotin complex (ABC Kit, Vectastain Elite, Vector Laboratories), and development with diaminobenzidine (DAB, 0.05%) to which 0.005% H2O2 was added. Double- and triplelabeling immunofluorescence was carried out with the following fluorescent secondary antibodies: Alexa 568–labeled goat anti–rat IgG, Alexa 488–labeled goat anti–mouse IgG or IgM, and Cy5 anti–mouse IgG (Molecular Probes). Sections were analyzed using either a standard microscope (BX51; Olympus) for peroxidase staining or a Zeiss confocal microscope (Carl Zeiss S.A.S.) equipped with Ar 488, HeNe1 543 and HeNe2 633 lasers, using the LSM-510 software package for image acquisition and data analysis. TUNEL staining was carried out in 8-µm-thick coronal sections of the olfactory bulb to detect DNA fragmentation in situ. After deparaffinization and rehydration, the tissue was treated with 0.5% Triton X-100 for 10 min and then incubated 15 sec in equilibrium buffer (Serological Corporation). Sections were then incubated for 1 h in a humidified chamber at 37 °C with a solution containing terminal deoxynucleotidyl transferase and digoxigenin nucleotides (Serological Corporation). After vigorous washing, a peroxidase labeled anti-digoxigenin antibody (Serological Corporation) was added for 30 min at room temperature, and staining was revealed by DAB. The forebrains of unilaterally occluded mice were embedded in gelatin and cut in 40-µm-thick coronal sections. Sections containing both bulbs, thus ipsiand contralateral to the occluded nostril, were immunostained for tenascin-R and immunoreactivity was quantified by laser confocal microscopy using identical acquisition parameters for both bulbs. In situ hybridization. The fresh-frozen brains of sham and unilaterally odordeprived mice were cut in 20-µm-thick coronal slices. The sections containing control and occluded bulbs were thawed onto Superfrost Plus slides and stored at –80 °C until used. Hybridization was done with the synthetic oligonucleotide (Eurogentec) 5′-AAG CCC CTC CTT CCT CCT CCA CAG TTT GTC TCT GAG CCC TTT CTG-3′, complementary to nucleotides 720–764 of the mRNA encoding Mus musculus tenascin-R. The sequence of the probe was checked in a GenBank database search to exclude significant homology with other genes. The probe was labeled with [32P]dATP (Perkin Elmer) by the terminal deoxynucleotidyl transferase (Roche Diagnostics) reaction following the manufacturer’s instructions. The sections were hybridized overnight at 42 °C in a hybridization mixture containing 50% formamide, 10% dextran sulfate, 4× SSC (1× SSC is 0.15 M NaCl, 0.015 M sodium citrate). After hybridization, the sections were washed for 30 min in 1× SSC prewarmed to 60 °C, rinsed in 0.1× SSC and dehydrated in an ascending series of ethanol concentrations. The digitalized autoradiograms were obtained by exposing sections in a β-imager (Biospace). This real-time imager provides rapid cartography of 32P labeling (in cpm/mm2) in tissue sections. Real-time PCR. Real-time PCR was carried out in a LightCycler (Roche Diagnostics) using DNA Master SYBR Green I dye (Roche Diagnostics). Seven days after odor deprivation, total RNA was extracted from three control and three occluded bulbs using FastRNA Pro Green Kit (Qbiogene). cDNA synthesis was carried out for 1 h at 37 °C in 50-µl reactions containing 2 µg total RNA, reverse transcriptase and random hexamer primers. The PCR protocol used to amplify the sequences for tenascin-R and actin (used as a housekeeping-gene control) were an initial denaturation at 95 °C for 10 min followed by 40 cycles each consisting of denaturation at 95 °C for 15 s, annealing at 62 °C for 5 s and extension at 72 °C for 10 s, respectively. The oligonucleotide primers used for PCR were as follows: for tenascin-R, 5′-TGCCAGGACTGAACTTGACA-3′ and 5′-CACAGTGACTTCGGAGGAGA-3′; for actin, 5′-CTAAGGCCAACCGTGAAAAGATG-3′ and 5′AGATGGGCACAGTGTGGGTGACC-3′. Melting-curve analysis was performed after each PCR to confirm the specificity of the reaction and identify the peaks of interest in all samples. A relative standard curve for each gene-specific primer pair was generated with tenfold serial dilutions of cDNAs derived from control and manipulated bulbs to validate that the dynamic and amplification efficiency of target and control genes were approximately equal. The dilution curves were then used for a relative quantification of target gene expression based on the individual Ct (the number of cycles to reach threshold). Results were normalized according to the expression level of actin mRNA from the same sample. Western blot analysis. The control and odor-deprived bulbs were homogenized in a glass-Teflon homogenizer in (50 mM Tris-HCl, pH 7.5, 1 mM EDTA, 1 mM EGTA, 1 mM sodium orthovanadate, 50 mM sodium fluoride, 5 mM sodium pyrophosphate, 10 mM sodium β-glycerophosphate, 0.1% 2-mercaptoethanol, 1% Triton X-100) lysis buffer supplemented with Protease Inhibitor Cocktail Set III (Calbiochem). The homogenates were centrifuged at 13,000g at 4 °C for 20 min to remove insoluble material, and the supernatants were assayed for protein concentration. Protein samples (15–20 µg) were separated on NuPage 4–12% Bis-Tris Gel (Invitrogen) and transferred to nitrocellulose membranes (Amersham Biosciences). Tenascin-Rimmunoreactive bands were detected using 619 antibody, horseradish peroxidase–conjugated goat anti-mouse antibodies (Bio-Rad) and an enhanced chemiluminescence substrate mixture (ECL Plus, Amersham Biosciences). The level of expression of tenascin-R in the control and odordeprived bulbs was normalized to that of NeuN. Tenascin-R-secreting cells. A 6.2-kb cDNA fragment containing nucleotides 1–4070 of the coding sequence of the 180-kDa rat tenascin-R was cloned into

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

354

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
the XhoI site of the pcDNA3 vector (Invitrogen) for expression under the control of the CMV promoter. For expression of the protein, the BHK cell line was transfected with 2 µg of this pcDNA3-TNR plasmid per well in a 6-well plate, using the Lipofectamine kit (Invitrogen) according to the manufacturer’s instructions. At 24 h after transfection, tenascin-R was detected by immunocytochemistry at the cell surface of the transfected but not the untransfected cells (data not shown). In addition, culture supernatants were collected and analyzed for secretion of tenascin-R by western blotting. A specific immunoreactive band was seen at 180 kDa in the supernatants of transfected but not untransfected BHK cells (data not shown). z axis, and data are presented as the total number of PSA-NCAM+ cells rerouted from the SVZ or the RMS. To quantify the impact of sensory deprivation on tenascin-R mRNA, hybridized (32P) radioactivity (in cpm/mm2) in the granule cell layer of the occluded bulb was related to that of the control bulb in the same section. Measurements were carried out from 20–30 coronal sections per mouse (2–3 mice per group). To assess the abundance of the tenascin-R protein, mean fluorescence intensity (per mm2) in the granule cell layer of the occluded bulb was related to that of the control bulb of tenascin-R-immunostained sections. All statistical comparisons between groups were made using Student’s t-test.
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS This work was supported by the Pasteur Institute, CNRS, The Annette GrunerSchlumberger Foundation and grants from the French Ministry of Research and Education (ACI ‘Biologie du Développement et Physiologie Intégrative’ and GIS ‘Infections à Prions’) and the Gemeinnuetzige Hertie Stiftung. We are grateful to R. Grailhe and N. Mechawar for help with in situ hybridization, F.-A. Weltzien for help with quantitative PCR analysis, P. Roux at the ‘Plate-form d’Imagerie Dynamique’ of the Institut Pasteur for help with confocal microscopy, A. Cardona for help with the β-imager and M.-M. Gabellec for excellent technical assistance. We thank S. Freitag and F. Morellini for providing tenascin-R-deficient mice and M. Sibbe and M. Kutsche for the pcDNA3-TNR construct. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 10 December 2003; accepted 24 February 2004 Published online at http://www.nature.com/natureneuroscience/
1. Luskin, M.B. Restricted proliferation and migration of postnatally generated neurons derived from the forebrain subventricular zone. Neuron 11, 173–189 (1993). 2. Alvarez-Buylla, A. & Garcia-Verdugo, J.M. Neurogenesis in adult subventricular zone. J. Neurosci. 22, 629–634 (2002). 3. Hack, I., Bancila, M., Loulier, K., Carroll, P. & Cremer, H. Reelin is a detachment signal in tangential chain-migration during postnatal neurogenesis. Nat. Neurosci. 5, 939–945 (2002). 4. Jones, F.S. & Jones, P.L. The tenascin family of ECM glycoproteins: structure, function, and regulation during embryonic development and tissue remodeling. Dev. Dyn. 218, 235–259 (2000). 5. Fuss, B., Wintergerst, E.S., Bartsch, U. & Schachner, M. Molecular characterization and in situ mRNA localization of the neural recognition molecule J1-160/180: a modular structure similar to tenascin. J. Cell Biol. 120, 1237–1249 (1993). 6. Srinivasan, J., Schachner, M. & Catterall, W.A. Interaction of voltage-gated sodium channels with the extracellular matrix molecules tenascin-C and tenascin-R. Proc. Natl. Acad. Sci. USA 95, 15753–15757 (1998). 7. Xiao, Z.C. et al. Tenascin-R is a functional modulator of sodium channel beta subunits. J. Biol. Chem. 274, 26511–26517 (1999). 8. Weber, P. et al. Mice deficient for tenascin-R display alterations of the extracellular matrix and decreased axonal conduction velocities in the CNS. J. Neurosci. 19, 4245–4262 (1999). 9. Bruckner, G. et al. Postnatal development of perineuronal nets in wild-type mice and in a mutant deficient in tenascin-R. J. Comp. Neurol. 428, 616–629 (2000). 10. Nikonenko, A., Schmidt, S., Skibo, G., Bruckner, G. & Schachner, M. Tenascin-Rdeficient mice show structural alterations of symmetric perisomatic synapses in the CA1 region of the hippocampus. J. Comp. Neurol. 456, 338–349 (2003). 11. Saghatelyan, A.K. et al. Reduced perisomatic inhibition, increased excitatory transmission, and impaired long-term potentiation in mice deficient for the extracellular matrix glycoprotein tenascin-R. Mol. Cell. Neurosci. 17, 226–240 (2001). 12. Wichterle, H., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. Direct evidence for homotypic, glia-independent neuronal migration. Neuron 18, 779–791 (1997). 13. Chazal, G., Durbec, P., Jankovski, A., Rougon, G. & Cremer, H. Consequences of neural cell adhesion molecule deficiency on cell migration in the rostral migratory stream of the mouse. J. Neurosci. 20, 1446–1457 (2000). 14. Tanapat, P., Hastings, N.B., Reeves, A.J. & Gould, E. Estrogen stimulates a transient increase in the number of new neurons in the dentate gyrus of the adult female rat. J. Neurosci. 19, 5792–5801 (1999). 15. Kee, N., Sivalingam, S., Boonstra, R. & Wojtowicz, J.M. The utility of Ki-67 and BrdU as proliferative markers of adult neurogenesis. J. Neurosci. Methods 115, 97–105 (2002). 16. Pesheva, P. & Probstmeier, R. The yin and yang of tenascin-R in CNS development and pathology. Prog. Neurobiol. 61, 465–493 (2000). 17. Xiao, Z.C. et al. Defasciculation of neurites is mediated by tenascin-R and its neuronal receptor F3/11. J. Neurosci. Res. 52, 390–404 (1998). 18. Boudreau, N. & Bissell, M.J. Extracellular matrix signaling: integration of form and function in normal and malignant cells. Curr. Opin. Cell Biol. 10, 640–646 (1998). 19. Pesheva, P., Gennarini, G., Goridis, C. & Schachner, M. The F3/11 cell adhesion

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

SVZ explant cultures. Cultures of SVZ explants were prepared as previously described12. Briefly, brains from P7 mice were dissected out and immersed in ice-cold HBSS medium (Gibco). The brains were cut in 200-µm-thick sections and those containing the SVZ were selected for further manipulations. Under a surgical microscope, the SVZ was dissected along the lateral wall of lateral ventricles and then cut into small pieces (100–200 µm in diameter) that were transferred to 70% Matrigel (BD Bioscience). After polymerization of Matrigel at 37 °C for 10 min, Neurobasal medium containing B27 supplement, L-glutamine (0.5 mM) and penicillin-streptomycin (1:1,000) (all from Gibco) was added. Cultures were maintained in a humidified incubator at 5% CO2 and 37 °C. In some experiments, SVZ explants were cocultured with aggregates of control or tenascin-R-transfected BHK cells. Aggregates of BHK cells were prepared under low-serum conditions using the hanging-drop method41. Grafting experiments. Control and tenascin-R-expressing BHK cells were labeled with the PKH26 red fluorescent cell linker kit (Sigma) following the manufacturer’s instructions, resulting in the staining of about 90% of all cells. Cells were grafted to the part of the striatum neighboring the SVZ (from bregma: anterioposterior, 1.5; mediolateral, 1.0; dorsoventral, 2.6) and just above the horizontal limb of RMS (anterioposterior, 3.35; mediolateral, 0.82; dorsoventral, 3.0) using a Kopf stereotaxic apparatus (Harvard Apparatus). On the day of transplantation, BHK cells were resuspended by trypsinization and collected after centrifugation (10 min, 475g, 4 °C). For transplantation, mice were anesthetized with a ketamine-xylazine mixture (Sigma) and approximately 2 × 105 BHK cells in 0.3 µl of solution were injected over a period of 3 min using a very thin glass electrode. The electrode was then left in place for an additional 3 min before being slowly withdrawn. Mice were killed 4 d later and the number of PSA-NCAM+ cells exiting the SVZ or RMS was counted in sagittal sections containing the grafts. The same sections were also processed for tenascin-R immunolabeling. Quantification and statistical analyses. All quantifications were done blind to the experimental conditions. For analysis of cell migration distance in vitro, explants were examined using phase-contrast microscopy after 20 h in culture. Migration distance was quantified by measuring the maximum distance that cells had moved away from the perimeter of each explant. BrdU-immunostained nuclei were quantified in every third 40-µm section along the entire SVZ-OB pathway. To assess the number of newborn neurons in the olfactory bulb, BrdU+ nuclei, observed with ×20 and ×40 objectives, were counted for the entire granule cell and glomerular layers. The numbers of BrdU+ profiles were then related to the surface of granule cell (including mitral cell and internal plexiform layers) and glomerular layers. For RMS and SVZ, the density of BrdU+ cells was evaluated by relating the number of BrdU+ cells to the surfaces occupied by these cells. To count TUNEL+ cells in the granule cell layer and RMSOB, these areas were delineated manually on sections counterstained with Methyl Green (Vector Laboratories). The percentages of BrdU + NeuN and BrdU + GFAP doubly immunostained cells were obtained by analyzing 3D-reconstructed BrdU+ nuclei in the x-z and y-z orthogonal projections for the presence of NeuN or GFAP. Neuroblasts migrating towards the grafted BHK cells were quantified by counting the PSA-NCAM+ cells that had detached from the SVZ or RMS. In mice grafted with BHK cells secreting tenascin-R, tenascin-R-immunoreactive areas were as large as 100–400 µm in diameter. To assess the number of rerouted neuroblasts in BHK-injected mice, PSA-NCAM+ cells outside of their normal migratory pathway were counted in the 400-µm-diameter area, calculated from the perimeter of the graft. Counting was done along the entire

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

355

ARTICLES
molecule mediates the repulsion of neurons by the extracellular matrix glycoprotein J1-160/180. Neuron 10, 69–82 (1993). 20. Probstmeier, R., Michels, M., Franz, T., Chan, B.M. & Pesheva, P. Tenascin-R interferes with integrin-dependent oligodendrocyte precursor cell adhesion by a ganglioside-mediated signalling mechanism. Eur. J. Neurosci. 11, 2474–2488 (1999). 21. Xiao, Z.C. et al. Isolation of a tenascin-R binding protein from mouse brain membranes. A phosphacan-related chondroitin sulfate proteoglycan. J. Biol. Chem. 272, 32092–32101 (1997). 22. Milev, P. et al. High affinity binding and overlapping localization of neurocan and phosphacan/protein-tyrosine phosphatase-zeta/beta with tenascin-R, amphoterin, and the heparin-binding growth-associated molecule. J. Biol. Chem. 273, 6998–7005 (1998). 23. Thomas, L.B., Gates, M.A. & Steindler, D.A. Young neurons from the adult subependymal zone proliferate and migrate along an astrocyte, extracellular matrixrich pathway. Glia 17, 1–14 (1996). 24. Murase, S. & Horwitz, A.F. Deleted in colorectal carcinoma and differentially expressed integrins mediate the directional migration of neural precursors in the rostral migratory stream. J. Neurosci. 22, 3568–3579 (2002). 25. Carleton, A., Petreanu, L.T., Lansford, R., Alvarez-Buylla, A. & Lledo, P.M. Becoming a new neuron in the adult olfactory bulb. Nat. Neurosci. 6, 507–518 (2003). 26. Komuro, H. & Rakic, P. Modulation of neuronal migration by NMDA receptors. Science 260, 95–97 (1993). 27. Saghatelyan, A. et al. Recognition molecule associated carbohydrate inhibits postsynaptic GABAb receptors: a mechanism for homeostatic regulation of GABA release in perisomatic synapses. Mol. Cell. Neurosci. 24, 271–282 (2003). 28. Kennedy, T.E. & Tessier-Lavigne, M. Guidance and induction of branch formation in developing axons by target-derived diffusible factors. Curr. Opin. Neurobiol. 5, 83–90 (1995). 29. Svendsen, C.N. & Sofroniew, M.V. Do central nervous system neurons require targetderived neurotrophic support for survival throughout adult life and aging? Perspect. Dev. Neurobiol. 3, 133–142 (1996). 30. Frazier, L.L. & Brunjes, P.C. Unilateral odor deprivation: early postnatal changes in olfactory bulb cell density and number. J. Comp. Neurol. 269, 355–370 (1998). 31. Brunjes, P.C. Unilateral naris closure and olfactory system development. Brain Res. Brain Res. Rev. 19, 146–160 (1994). 32. Cummings, D.M. & Brunjes, P.C. The effects of variable periods of functional deprivation on olfactory bulb development in rats. Exp. Neurol. 148, 360–366 (1997). 33. Henegar, J.R. & Maruniak, J.A. Quantification of the effects of long-term unilateral naris closure on the olfactory bulbs of adult mice. Brain Res. 568, 230–234 (1991). 34. Cummings, D.M., Henning, H.E. & Brunjes, P.C. Olfactory bulb recovery after early sensory deprivation. J. Neurosci. 17, 7433–7440 (1997). 35. Rochefort, C., Gheusi, G., Vincent, J.D. & Lledo, P.M. Enriched odor exposure increases the number of newborn neurons in the adult olfactory bulb and improves odor memory. J. Neurosci. 22, 2679–2689 (2002). 36. Magavi, S.S., Leavitt, B.R. & Macklis, J.D. Induction of neurogenesis in the neocortex of adult mice. Nature 405, 951–955 (2000). 37. Arvidsson, A., Collin, T., Kirik, D., Kokaia, Z. & Lindvall, O. Neuronal replacement from endogenous precursors in the adult brain after stroke. Nat. Med. 8, 963–970 (2002). 38. Nakatomi, H. et al. Regeneration of hippocampal pyramidal neurons after ischemic brain injury by recruitment of endogenous neural progenitors. Cell 110, 429–441 (2002). 39. Meisami, E. Effects of olfactory deprivation on postnatal growth of the rat olfactory bulb utilizing a new method for production of neonatal unilateral anosmia. Brain Res. 107, 437–444 (1976). 40. Morganti, M.C., Taylor, J., Pesheva, P. & Schachner, M. Oligodendrocyte-derived J1160/180 extracellular matrix glycoproteins are adhesive or repulsive depending on the partner cell type and time of interaction. Exp. Neurol. 109, 98–110 (1990). 41. Fan, C.M. & Tessier-Lavigne, M. Patterning of mammalian somites by surface ectoderm and notochord: evidence for sclerotome induction by a hedgehog homolog. Cell 79, 1175–1186 (1994).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

356

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Stability of dendritic spines and synaptic contacts is controlled by αN-catenin
Kentaro Abe1,2, Osamu Chisaka2, Frans van Roy3 & Masatoshi Takeichi1,2
Morphological plasticity of dendritic spines and synapses is thought to be crucial for their physiological functions. Here we show that αN-catenin, a linker between cadherin adhesion receptors and the actin cytoskeleton, is essential for stabilizing dendritic spines in rodent hippocampal neurons in culture. In the absence of αN-catenin, spine heads were abnormally motile, actively protruding filopodia from their synaptic contact sites. Conversely, αN-catenin overexpression in dendrites reduced spine turnover, causing an increase in spine and synapse density. Tetrodotoxin (TTX), a neural activity blocker, suppressed the synaptic accumulation of αN-catenin, whereas bicuculline, a GABA antagonist, promoted it. Furthermore, excess αN-catenin rendered spines resistant to the TTX treatment. These results suggest that αN-catenin is a key regulator for the stability of synaptic contacts.

Formation of excitatory synapses is established by sequential cellular events, including the protrusion of filopodia from dendrites, the contact between filopodia and axons1,2, and morphological changes of the filopodia into mature spines3,4. Filopodia are highly motile, and this motility is thought to facilitate the initial formation of dendriteaxon contacts. Even after the establishment of these contacts, the spines are still motile, dynamically changing their shape5–7. This structural flexibility of spines is thought to be crucial for remodeling of neural circuits8. Uncovering the cellular and molecular mechanisms that regulate spine motility and stability is thus important for understanding complex synapse functions. Several molecules have been identified as regulators of dendritic spine formation and shape. These include the Rho family of small GTPases9,10, scaffold proteins such as Shank and Homer (ref. 11), SPAR (ref. 12), drebrin (ref. 13), PSD95 (ref. 14) and Syndecan 2 (ref. 15). Furthermore, EphrinB-EphB interaction induces spine formation16, and does GluR2, even on inhibitory neurons17. Concerning the physiological regulation of spine motility and stability, TTX enhances spine motility6; conversely, activation of either AMPA or NMDA receptors inhibits the actin-based protrusive activity of spine heads18. Sensory deprivation due to whisker trimming was found to reduce spine protrusion in the barrel cortex7. These observations suggest that neural activity controls spine motility. Despite such extensive studies, however, the molecular and cellular bases of synapse stability still remain obscure. Classic cadherins and associated proteins are localized in synaptic junctions19–24. Our previous studies showed that when cadherin function is blocked by a dominant-negative form of cadherin, synaptic organization is impaired25. Mutation of αN-catenin, a cadherin-associated protein, also affected dendritic spine morphology; although the mutant spines could establish synaptic contact with axons. Based on these and other observations26, we have postulated that the cad1RIKEN

herin/catenin complex may function as a morphological regulator of synaptic plasticity. In the present study, we analyzed the role of αNcatenin in synapse dynamics and found that, in the absence of αNcatenin, dendritic spine heads became unusually motile and could not maintain their stable contacts with axons. Conversely, overexpression of this molecule caused an increase in the number of mature spines. These results suggest that αN-catenin functions as a critical agent to regulate the stability of synaptic contacts. RESULTS Dendritic spines on αN-catenin-deficient neurons are unstable To examine the role of αN-catenin in synapse dynamics, we cultured hippocampal neurons collected from homozygous mutant mice that lacked the gene encoding αN-catenin (Catna2–/–), as well as from their heterozygous (Catna2+/–) or wild-type (Catna2+/+) littermates. In these cultures, the density of dendritic spines was not statistically different between Catna2–/– and Catna2+/+ neurons (Fig. 1). Immunostaining for a presynaptic marker, synapsin, showed that the density of synapsin puncta present on both dendritic spines and shafts was also not significantly different between these neurons. However, the ratio of synapsin-positive spines to the total number of spines was slightly reduced in Catna2–/– neurons (Fig. 1a), with a small increase in synapsin puncta density on the dendritic shafts. The above hippocampal neurons were transfected with an enhanced green fluorescent protein (EGFP)-actin expression vector at the time of plating to visualize the activities of filopodia and dendritic spines. At 14–15 d.i.v. (days in vitro), fluorescently labeled neurons were subjected to time-lapse observations. In Catna2+/– or Catna2+/+ neurons, dynamic movement of actin-labeled dendritic spines was observed, as reported previously27. Catna2–/– spines were also motile, but unlike heterozygous or wild-type (Fig. 2a) neurons, they showed rapid protrusion and retraction of filopodia from most of the spines (Fig. 2b,

Center for Developmental Biology, 2-2-3 Minatojima-Minamimachi, Chuo-ku, Kobe 650-0047, Japan. 2Graduate School of Biostudies, Kyoto University, Kitashirakawa, Sakyo-ku, Kyoto 606-8502, Japan. 3Department for Molecular Biomedical Research, VIB-Ghent University, B-9000 Ghent, Belgium. Correspondence should be addressed to M.T. ([email protected]). Published online 21 March 2004; doi:10.1038/nn1212

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

357

ARTICLES
Figure 1 Profiles of dendritic spines in Catna2+/+ and Catna2–/– neurons. (a) Mean density of total spines, synapsin-positive puncta and synapsinpositive spines. GFP-transfected neurons were immunostained for synapsin at 20 d.i.v. and subjected to analysis. The density of total spines (P = 0.11) and synapsin-positive puncta (P = 0.34) was not significantly different between the wild-type and mutant samples, but that of synapsin-positive spines was significantly different (P < 0.001). (b) Immunostaining for PSD95 in dendrites of Catna2+/+ or Catna2–/– neurons expressing EGFP at 16 d.i.v. Arrows show filopodial protrusion from spine heads. Scale bar = 20 µm.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

they were equally motile in wild-type, heterozygous and homozygous neurons. These findings indicate that Catna2–/– synaptic contacts are not stable, allowing extra filopodial protrusion from the spine heads. αN-Catenin overexpression causes an increase in spine density To further explore the role of αN-catenin, we examined the effect of its overexpression on synaptogenesis in hippocampal neurons. Freshly isolated neurons were transfected with either αN-catenin and GFP or GFP alone, and then cultured for various periods of time. First we transfected Catna2–/– neurons with αN-catenin, and found that the normal morphology of the spines was restored (data not shown). When wild-type neurons were used for transfection, the density of the spines on dendrites significantly increased in αN-cateninoverexpressing neurons at 20 and 30 d.i.v. (Fig. 3a,b); this tendency was already detectable at 10 d.i.v. (Fig. 3b). The width of spine heads also showed some increase in these neurons (Fig. 3c,e). The length of spines, on the other hand, did not differ between the control and experimental samples, except at 10 d.i.v. (Fig. 3d,f). The difference at 10 d.i.v. suggests that the conversion of filopodia into spines, occurring in such early cultures, was facilitated by αN-catenin overexpression. These effects were observed only in excitatory neurons; that is, no ectopic spine formation was induced on inhibitory neurons. Immunostaining for synapsin showed that the above increase in spine

arrows; see also Supplementary Videos 1 and 2); this movement was recorded as high degrees of oscillation in the length of each spine (Fig. 2d; compare to c). Many of these spines also showed unusually dynamic deformation of their heads (Fig. 2b, arrowhead). The active filopodial protrusions in Catna2–/– neurons likely occurred from the spine heads that were in contact with axons. For example, the spine marked ‘d’ in Figure 2b was highly motile but appears to be associated with an axon (ax) visualized by accidental EGFP-actin labeling. For further confirmation, we stained these neurons for PSD95, which accumulates only in sites of synaptic contact. We found that spines with filopodial protrusions were PSD95-positive ones (Fig. 1b). In Catna2+/– or Catna2+/+ spines (Fig. 2a,c), we could detect some filopodial protrusion and retraction from the already-established synaptic contacts, but movement was restricted mostly to the head body itself, resulting in a narrower range of spine length oscillation (Fig. 2e,f). With regard to filopodia directly protruding from the dendritic shafts,

Figure 2 Enhanced motility in dendritic spines of Catna2–/– neurons. (a,b) Time-lapse movies of 15 d.i.v. neurons transfected with EGFP-tagged actin, obtained from Catna2+/+ or Catna2–/– embryos. Arrows point to representative filopodia repeating protrusion and retraction from the spine heads. Arrowhead, an example of actively deforming spines. Some axons (ax) are also labeled with EGFP-tagged actin; other axons associated with spines are not labeled in these images. See also Supplementary Videos 1 and 2. Scale bars = 5 µm. (c,d) Tracing of the length of six representative spines indicated in a and b during a 95-min incubation period. (e,f) Mean changes in the length of spines undergoing protrusion and retraction during a 95-min period, and cumulative frequency plot for the changes. The length of spines was traced for 5 min, and the minimum length was subtracted form the maximum length for each spine. More than 70 spines from six dendrites were measured. The spine-length change was significantly enhanced in neurons without αNcatenin, and it was suppressed in those overexpressing αN-catenin. *P < 0.001.

358

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 3 Increased spine density in αN-catenin-overexpressing neurons. (a) Neurons at 18 d.i.v. expressing EGFP (GFP) or EGFP-tagged αN-catenin (αN-cat-GFP) immunostained for GFP. Lower panes, close-up views of a representative portion of each neuron. The differences in thickness and branching of dendrites seen between the control and experimental samples are due to a variation in morphology of individual neurons and not to αN-catenin overexpression. (b–d) Mean spine density, head width and length in 10, 20 and 30 d.i.v. neurons transfected with GFP or αN-cat-GFP. More than 820 spines from 12 neurons were measured. (e,f) Cumulative frequency plots for spine length and head width in 30 d.i.v. neurons. (g) Effects of overexpression of α-catenin subtypes, N-cadherin and β-catenin on spine density. More than 470 spines from seven neurons were measured for each construct. (h–j) Triple immunostaining for PSD95, synapsin and GluR2 in 20 d.i.v. neurons transfected with GFP + Flag-tagged αN-catenin (αN-cat-flag) or with GFP only. Scale bars = 5 µm. *P < 0.001, compared with GFP only.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

density in αN-catenin-overexpressing neurons was accompanied by that in the density of synapsin-positive puncta (1.29 ± 0.10 per µm in overexpressing neurons versus 0.75 ± 0.02 per µm in wild type, P < 0.001, n = 13, at 20 d.i.v.), as well as in the ratio of spines immunoreactive for synapsin to the total spines (91.3 ± 1.5% in overexpressing neurons versus 75.5 ± 1.7% in wild-type neurons, P < 0.001, n = 13, at 20 d.i.v.). On the other hand, the staining intensity for synapsin, GluR2 and PSD95 in individual synaptic puncta was not particularly altered by the αN-catenin overexpression (Fig. 3h–j). There are two isoforms of αN-catenin, I and II; both had similar effects on spine density (Fig. 3g). Overexpression of two other subtypes of α-catenin, namely αE-catenin and αT-catenin, also induced excess spine formation, indicating that they share the same spinestabilizing activity. On the other hand, overexpression of N-cadherin or β-catenin had no effect on spine morphology or density (Fig. 3g), which indicates that αN-catenin has a unique activity regarding spine and synapse formation among the molecules constituting the cadherin/catenin complex. We could detect the above effects of αN-catenin only when postsynaptic neurons overexpressed this molecule. We then confined this overexpression to presynaptic sites, that is, to axons. This experiment was technically difficult, however, because transfected αN-catenin molecules were only faintly localized to axons; therefore we could not accurately identify the spines in contact with the labeled axons in crowded cultures. Recent studies show that overexpression of not only β-catenin or N-cadherin, but also that of αNcatenin in hippocampal neurons enhanced their dendritic arborization28. In our transfection protocol, however, we did not observe such an effect of αN-catenin overexpression on dendrite branching, when examined at 7 d.i.v. (Supplementary Fig. 1 online) or at later culture stages. Likewise, Catna2 mutation had no effect on this phenomenon (Supplementary Fig. 1). Neurons that received cDNAs at the levels required for stimulation of dendrite branching might have

not survived during the long culture periods used in our experiments. Thus, our observations were limited to spine morphogenesis. αN-catenin overexpression stabilizes spines The above observations suggest that αN-catenin functions to stabilize spines and synapses. For confirmation, we took time-lapse movies of αN-catenin-overexpressing neurons and found that their spines were indeed more quiescent than those of the controls in terms of filopodiaprotruding activity (Fig. 2e,f). The increase in spine density in αNcatenin-overexpressing neurons suggests the possibility that spine turnover was suppressed in these neurons, resulting in their accumu-

Figure 4 Reduced turnover of dendritic spines in αN-catenin-overexpressing neurons. Time-lapse GFP-fluorescence images were collected from representative dendrites of Catna2+/+ and Catna2–/– neurons transfected with GFP, and also of neurons co-transfected with GFP and αN-catflag, every 3 d during the period from 18 to 27 d.i.v. (a–c) Examples of the spines that disappeared (open triangles), formed anew (dark triangles) or became deformed (arrows) during the observations. The ratio of these spines is also shown (d). Data were collected from more than 1,000 spines on 18 dendrites of seven neurons for each group in the experiment.

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

359

ARTICLES
Figure 5 Requirement of the C-terminal region of αN-catenin for the increase in spine density. (a) Deletion series of αN-catenin. The regions required for association with other proteins, identified for αE-catenin, are depicted. (b) Representative dendrites expressing each construct. Scale bar = 5 µm. (c,d) Mean spine density and length/width in 20 d.i.v. neurons transfected with each construct. Among the deletion mutants, only αN1-870 is effective in increasing spine density. αN-1-681, αN-1-406 and αN-1-262 cause an increase in spine length, suggesting that they have a dominant-negative action. More than 500 spines from seven neurons were measured for each construct. *P < 0.001, as compared to GFP only.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

lation. To test this possibility, we recorded the images of identical dendrites for 9 d from 18 to 27 d.i.v. (Fig. 4a–c), as it has previously been reported that turnover can be observed in this time window29. We analyzed morphological changes in the spines every 3 d during this 9-d period and recorded the means of the data collected from the three different periods in Figure 4d. During each 3-d period, in control neurons expressing GFP only, 15.0% of the spines disappeared, 18.9% formed anew and the other spines remained unchanged, although many showed a change in morphology. In Catna2–/– neurons, 19.8% of the spines disappeared, whereas in αNcatenin-overexpressing neurons, only 9.7% disappeared. These data support the idea that the overexpression of αN-catenin suppressed spine turnover. αN-catenin has various domains that interact with a number of molecules30,31. To identify the domain(s) responsible for the above activities, we designed a series of deletion mutants (Fig. 5a) and used them for transfection (Fig. 5b). An amino (N)-terminal deletion that removed the β-catenin binding site (αN-277-954) abolished the spine density–increasing (SDI) activity, indicating that αN-catenin needs to associate with the cadherin/β-catenin complex for its action. A short carboxy (C)-terminal deletion (αN-1-870) did not affect the SDI activity; whereas molecules with longer C-terminal deletions (i.e., αN-1-681, 1-406 and 1-262) no longer exhibited this activity (Fig. 5c). Nevertheless, these mutants caused elongation of spines (Fig. 5d), as seen in Catna2–/– neurons, suggesting that they had a dominantnegative effect. These results suggest that the C-terminal domain, known to bind various cytoskeletal proteins (Fig. 5a), is essential for the SDI activity of αN-catenin. Involvement of αN-catenin in physiological plasticity of spines Dendritic spines and synapses are structurally and functionally modified under various physiological conditions. We asked whether αNcatenin is involved in such processes. TTX, a neural activity blocker, converts dendritic spines into filopodial processes in hippocampal cultures32. We treated rat hippocampal neurons at 18–20 d.i.v. with 1.5 µM TTX for 3 d, and then stained them for αN-catenin. The fluorescence intensity of αN-catenin signals associated with synapsin puncta significantly decreased in the TTX-treated neurons (Fig. 6a,b). Next, we examined the effect of 40 µM bicuculline, which antagonizes GABA-receptors and thus increases neural activity. Contrary to the TTX treatment, the

bicuculline treatment resulted in an increase in the αN-catenin fluorescence intensity at the synapses. In these cultures, β-catenin signal intensity in synapses was only slightly decreased in the TTX-treated neurons, whereas it was significantly increased in the bicuculline-treated ones (Fig. 6b). On the other hand, synapsin signals were not particularly altered after these treatments, suggesting that presynaptic structures were normally maintained. Under these experimental conditions, the entire expression levels of N-cadherin, αN-catenin and β-catenin did not differ between the non-treated and TTX-treated cultures. In the case of bicuculline treatment, however, the levels of these proteins were slightly upregulated (Fig. 6c). These results suggest that the TTX treatment altered only the distribution of these proteins, whereas bicuculline may have enhanced stabilization or expression of these proteins. There appeared to be a pool of αN-catenin, diffusely distributed and not associated with synapses, as seen in the merged images in Figure 6a. This pool of αN-catenin probably increased in TTX-treated neurons. In the above TTX experiments, it remained unclear whether the reduction in αN-catenin in the synapses had occurred as a result of the TTX-induced morphological changes in the spines or whether TTX-dependent signals actively suppressed the synaptic αN-catenin accumulation, secondarily leading to changes in spine shape. To test these possibilities, we compared αN-catenin-overexpressing and control hippocampal neurons for their response to TTX. In control neurons, their spines were changed into filopodia-like processes after the TTX treatment, whereas αN-catenin-overexpressing neurons did not clearly respond to TTX (Fig. 6d–f). Thus, excess αN-catenin interferes with the TTX action, supporting the idea that αN-catenin may function as a mediator of neural activity–dependent signals for spine morphological changes.

360

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 6 Effects of TTX and bicuculline on αN-catenin distribution and spine morphology in rat hippocampal neurons. (a) Double immunostaining for αN-catenin and synapsin I after 3-d incubation with a control solution, 1.5 µM TTX or 40 µM bicuculline. Close-up views are also shown. (b) Normalized intensity of immunofluorescent signals of synapsin I, αN-catenin and β-catenin that have accumulated in individual synapses. *P < 0.001, as compared to control. (c) Western blots for N-cadherin, αN-catenin and β-catenin after 3-d TTX or bicuculline treatment. Relative intensity of the αN-catenin doublet bands was also quantified (right). The bicuculline-treated samples showed more intense signals than the control (*P < 0.001). (d–f) GFP-transfected or αN-catenin/GFP-cotransfected neurons were treated with 1.5 µM TTX for 3 d, and then stained with antiGFP antibody to visualize their morphology. TTX converted spines into filopodia-like processes in control GFP-transfected neurons, whereas it failed to do so in αN-catenin-overexpressing neurons. Mean spine length after TTX treatment and cumulative frequency plot of spine length are also shown. Scale bar = 5 µm. More than 600 spines from eight neurons were measured for each construct. *P < 0.001, as compared with GFP-only control.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

DISCUSSION The loss of αN-catenin causes deformation of dendritic spines25. Our time-lapse observations revealed more dynamic features of the spines in Catna2–/– neurons than anticipated from fixed samples. These spines were not simply abnormal in their static morphology; they also showed an unusually high activity of filopodial protrusion from the spine heads. Although filopodial protrusion is an essential process for new synapse formation in early neurons, this occurs from dendritic shafts but barely from established synaptic contacts. These observations indicate that αN-catenin is responsible for suppressing the motile activity of filopodia/spines. We also showed that overexpression of αΝ-catenin induced the formation of extra spines and synapses. Synapses are not static; their addition and elimination on a dendrite during prolonged periods have been observed29,33,34. Such turnover of synapses was reduced in αΝ-catenin-overexpressing neurons. Collectively, we can conclude that αΝ-catenin contributes to synapse stability from two aspects: suppression of spine motility activity and suppression of spine turnover. Synapse stability and remodeling are essential for a variety of neural activities35. We propose that αΝ-catenin may function as a regulator for synaptic remodeling, if external signals can modulate αΝ-catenin activities. There are several possible pathways to affect αΝ-catenin function. For example, signals that alter the binding of

αΝ-catenin to β-catenin can affect αΝ-catenin function. The interactions between α-catenin and β-catenin can be altered by various factors, such as Fer and Fyn tyrosine kinases or protein kinase CKII, which affect their binding36,37. Notably, neural activity induces redistribution of β-catenin into synapses38, as was confirmed by our bicuculline experiment. Facilitating the β-catenin binding to cadherin would be expected to recruit more αN-catenin molecules to synapses, and this could be a process to stabilize synapses. However, we found no effect of β-catenin overexpression on synapse stabilization, suggesting that the β-catenin level itself is not sufficient to alter synapse stability. The requirement of the C-terminal domain of αΝcatenin for synapse stabilization suggests other potential regulatory systems. For example, the C-terminal domain is known to bind Factin (ref. 39), ZO-1 (ref. 40) and vezatin (ref. 41), and the interaction of αΝ-catenin with these proteins may be involved in controlling spine motility and stability. It is intriguing to note that profilin, a regulator of actin polymerization, is targeted to spines in an NMDA receptor activity–dependent manner, concomitantly suppressing changes in spine shape. Furthermore, blocking this profilin targeting destabilizes spine structure42. There may be interplay between the profilin/actin and cadherin/αΝ-catenin systems in such a way that the former controls the latter activity, or the latter produces signals for actin reorganization. Given that neural activity can modify any of these molecular interactions, such processes are expected to contribute to changes in spine shape. We found that the treatment of neurons with TTX, which induces the conversion of spines into filopodia-like processes, caused a reduction in the αΝ-catenin concentration in synapses. Furthermore, spines became resistant to TTX when αΝ-catenin was overexpressed. These observations suggest that neural activity blockade directly or indirectly causes a release of αN-catenin from synapses, leading to filopodia-like spine formation, and that αN-catenin overexpression blocks this system. In the TTX experiments, we also detected a subtle reduction in the fluorescence intensity of β-catenin, but this change was not as extensive as the change in αN-catenin, suggesting that the redistribution of αNcatenin had a dominant role in these phenomena. Importantly, overexpression of N-cadherin and β-catenin did not have apparent effects on spine stability. This finding suggests that the size of the cytoplasmic pool of αΝ-catenin, but not that of other molecules of the cadherin/catenin complex, may be a limiting factor for the modulation of synaptic contacts. Thus, synapse stability seems not to be regulated by a simple increase in the number of cadherin molecules recruited to synaptic membranes but rather by an αΝ-catenin-specific signaling

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

361

ARTICLES
system, perhaps including its interaction with cytoskeletal proteins. The overexpression of αΝ-catenin on the postsynaptic side was sufficient for synapse stabilization. Therefore, the above putative signaling system is assumed to reside on the postsynaptic side. We also should stress that αΝ-catenin overexpression was not accompanied by increases in other synaptic components. In synaptic junctions, the cadherin/catenin complex is localized in compartments separable from the active zone20,43. It is thus possible that the αΝ-catenin-dependent system might be independent of other regulatory systems for synapse remodeling, localized in the central portion of the postsynaptic density. Morphological plasticity in synapses often correlates with their physiological plasticity. For example, stimuli that cause long-term potentiation (LTP) concomitantly induce filopodial/spine formation or their shape changes44,45, and synaptic activity regulates the number of spines and synapses8,46,47. Because of the postnatal lethality of our αΝ-catenin knockout mice, we could not examine whether αΝ-catenin is involved in the physiological plasticity of synapses in vivo. By using different genetic methods, we should be able to uncover the in vivo roles of αΝ-catenin in future experiments. METHODS
Molecular construction. αE-catenin expression vector carrying the flag tag at its carboxyl end was constructed as follows: αE-catenin cDNA was amplified by use of a standard PCR method and subcloned into pCA-sal-flag (D. Fushimi and M.T., unpublished data) by use of a SalI linker to generate pCA-αE-catenin-flag, which expresses the C-terminal flag-tag fusion protein under the control of the CAG promoter48. Expression vectors for full-length or mutant αN-catenins were made as for αE-catenin, by using αN-catenin cDNA, pBNCAT1b or pBNCATII (ref. 49) as a template, and the following primers: for full-length αN-catenin, 5′-GATATCGCCACCATGACTTCG GCAACTTCA-3′ (αN-N primer) and 5′-GTCGACGAAGGAATCCATTGC CTTG-3′ (αN-C primer); for αN1-262, αN-N primer and 5′-GTCGACGGA GGTGGCCTGAGCAGCA-3′; for αN277-954, 5′-GATATCGCCACCAT GCAGCCCTGAATGAGT-3′ and αN-C primer; for αNk, αN-N primer and 5′-CGGCGTCGACGAGCACACGGACTTGCTT-5′; for αN1-681, αN-N primer and 5′-CGGCGTCGACTGCTTTCTCCTCCTGTGGT-3′; and for αN1-870, αN-N primer and 5′-CGGCGTCGACTTTGGCTGCCTGGAT GAC-3′. Expression vectors for αT-catenin were constructed as above by using αT-catenin cDNA (PGEMTeasy-mαT-catenin (1-2979) clone7) as a template. All the constructs were checked by sequencing. Cell culture and transfection. Hippocampal neuron cultures were prepared as previously described25. In brief, hippocampal neurons from E17 mice or E18 Sprague-Dawley rats were plated on poly-L-lysine coated cover glasses at the density of 10,000 cells/cm2 and maintained in NeuroBasal medium (Invitrogen) with B27 supplements. To culture neurons from Catna2–/– mice, we collected neurons from E17 pups generated by crossing Catna2+/– parents. Those from individual hippocampi were separately cultured, and genotyping was performed afterwards to determine the genotype in each pool of the cultures. Comparisons of mutant and wild-type samples were generally made among the cultures derived from a single littermate. cDNA transfection was performed by using Effectine (QIAGEN), following the manufacturer’s instructions with optional modifications. Freshly dispersed neurons were plated, and immediately subjected to the transfection treatments. Successfully transfected neurons were detected by EGFP fluorescence. For TTX or bicuculline treatment of neurons, 18 d.i.v. rat cultures were treated with 1.5 µM TTX (Wako BioProducts) or 40 µM bicuculline (Sigma) for 3 d. Time-lapse analysis. For time-lapse imaging, live cells in culture medium were mounted in a chamber at 37 °C through which was passed a continuous flow of 5% CO2. Images were obtained with an LSM510 (Zeiss) confocal microscope using a 40× water immersion lens, and an additional electronic zooming factor of 6×. The laser power was attenuated to 0.1% to reduce phototoxicity. Z-stacked images at 0.8-µm depth intervals were taken every 5 min. For longterm time-lapse imaging, live cell images of the same dendrite were obtained every 3 d by using a 40× water immersion lens with an additional zooming factor of 4×. For cells co-transfected with GFP and another gene, cells were fixed after imaging; and then the transgene expression was verified by immunostaining. Image acquisition and quantification. Confocal images of immunostained neurons were obtained with the LSM510 using a Zeiss 63× objective. Each image was a z-series projection taken at depth intervals of 0.36 µm. More than ten transfected neurons were chosen randomly for quantification from 5–10 coverslips derived from 4–7 independent experiments for each construct. For measurement of spine morphology, spines located within the proximal 75-µm region of two or three of the largest dendrites were chosen, and manually traced. Their length and head width were measured automatically with LSM510 software (Zeiss). To distinguish between filopodia and spines, we triple-stained neurons with anti-PSD95 antibodies together with anti-GFP and anti-flag tag antibodies. Only the processes that contained PSD95 puncta were defined as spines in this study. For quantification of spine-length changes in time-lapse movies, we measured the length of each spine every 5 min during a 95-min period and subtracted the minimum length from the maximal length. For quantification of long-term time-lapse analysis, we collected images of spines every 3 d, and traced their morphologies. For quantification of fluorescence intensities, we triple-stained neurons with synapsin I, αN-catenin and β-catenin. More than 20 coverslips were used for each condition. Confocal images of immunostained neurons were obtained as described above, by using the LSM510 (Zeiss). The same settings for pinhole size, brightness and contrast were used. Fluorescence intensity analysis was conducted as follows: catenin signals overlapping with synapsin signals were collected by using Subtraction Macro, and these signals were defined as the synaptic components of the catenin signals. The threshold was set to a modest one to increase the discrimination of particles26. The same threshold was used within the same experiments. Measured data were exported to Excel software (Microsoft), and the data were compared by using Student’s t-test. Histograms showing the mean ± s.e.m. were constructed. Immunostaining and antibodies. Immunocytochemistry of neurons was performed as described25 by use of the following antibodies: mouse anti-PSD-95 (6G6-1C9, ABR), rabbit anti-synapsin I (Chemicon), mouse anti-GFP (7.1+11.1, Roche), rabbit anti-GFP (Chemicon), rat anti-αN-catenin (NCAT2 or NCAT520,50), mouse anti-β-catenin (5H10, a gift from M.J. Wheelock, University of Nebraska Medical Center), rabbit anti-flag tag (Sigma), mouse anti-GluR2 (MAB397, Chemicon) and mouse anti-N-cadherin (Pharmingen). Western blot analysis. Neurons cultured on each coverslip were individually scraped into a sample buffer and subjected to SDS-PAGE. Proteins were transferred onto a nitro-cellulose membrane, and detection was performed by using ECL Plus (Amersham Biosciences). Signals were quantified by using Scion Image (Scion). Data obtained from 26 coverslips were used for statistical analysis of each experimental condition. Histograms showing the mean ± standard error (s.e.m.) were constructed.
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS We thank T. Manabe for TTX and H. Ishigami for maintaining the mice. This work was supported by the program Grants-in-Aid for Specially Promoted Research of the Ministry of Education, Science, Sports, and Culture of Japan to M.T. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 20 January; accepted 23 February 2004 Published online at http://www.nature.com/natureneuroscience/
1. Ziv, N.E. & Smith, S.J. Evidence for a role of dendritic filopodia in synaptogenesis and spine formation. Neuron 17, 91–102 (1996). 2. Jontes, J.D., Buchanan, J. & Smith, S.J. Growth cone and dendrite dynamics in zebrafish embryos: early events in synaptogenesis imaged in vivo. Nat. Neurosci. 3, 231–237 (2000). 3. Okabe, S., Miwa, A. & Okado, H. Spine formation and correlated assembly of presynaptic and postsynaptic molecules. J. Neurosci. 21, 6105–6114 (2001). 4. Marrs, G.S., Green, S.H. & Dailey, M.E. Rapid formation and remodeling of postsynaptic densities in developing dendrites. Nat. Neurosci. 4, 1006–1013 (2001).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

362

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
5. Dunaevsky, A., Tashiro, A., Majewska, A., Mason, C. & Yuste, R. Developmental regulation of spine motility in the mammalian central nervous system. Proc. Natl. Acad. Sci. USA 96, 13438–13443 (1999). 6. Korkotian, E. & Segal, M. Regulation of dendritic spine motility in cultured hippocampal neurons. J. Neurosci. 21, 6115–6124 (2001). 7. Lendvai, B., Stern, E.A., Chen, B. & Svoboda, K. Experience-dependent plasticity of dendritic spines in the developing rat barrel cortex in vivo. Nature 404, 876–881 (2000). 8. Yuste, R. & Bonhoeffer, T. Morphological changes in dendritic spines associated with long-term synaptic plasticity. Annu. Rev. Neurosci. 24, 1071–1089 (2001). 9. Tashiro, A., Minden, A. & Yuste, R. Regulation of dendritic spine morphology by the rho family of small GTPases: antagonistic roles of Rac and Rho. Cereb. Cortex 10, 927–938 (2000). 10. Nakayama, A.Y., Harms, M.B. & Luo, L. Small GTPases Rac and Rho in the maintenance of dendritic spines and branches in hippocampal pyramidal neurons. J. Neurosci. 20, 5329–5338 (2000). 11. Sala, C. et al. Regulation of dendritic spine morphology and synaptic function by Shank and Homer. Neuron 31, 115–130 (2001). 12. Pak, D.T., Yang, S., Rudolph-Correia, S., Kim, E. & Sheng, M. Regulation of dendritic spine morphology by SPAR, a PSD-95-associated RapGAP. Neuron 31, 289–303 (2001). 13. Hayashi, K. & Shirao, T. Change in the shape of dendritic spines caused by overexpression of drebrin in cultured cortical neurons. J. Neurosci. 19, 3918–3925 (1999). 14. El-Husseini, A.E., Schnell, E., Chetkovich, D.M., Nicoll, R.A. & Bredt, D.S. PSD-95 involvement in maturation of excitatory synapses. Science 290, 1364–1368 (2000). 15. Ethell, I.M. & Yamaguchi, Y. Cell surface heparan sulfate proteoglycan syndecan-2 induces the maturation of dendritic spines in rat hippocampal neurons. J. Cell Biol. 144, 575–586 (1999). 16. Penzes, P. et al. Rapid induction of dendritic spine morphogenesis by trans-synaptic ephrinB-EphB receptor activation of the Rho-GEF kalirin. Neuron 37, 263–274 (2003). 17. Passafaro, M., Nakagawa, T., Sala, C. & Sheng, M. Induction of dendritic spines by an extracellular domain of AMPA receptor subunit GluR2. Nature 424, 677–681 (2003). 18. Fischer, M., Kaech, S., Wagner, U., Brinkhaus, H. & Matus, A. Glutamate receptors regulate actin-based plasticity in dendritic spines. Nat. Neurosci. 3, 887–894 (2000). 19. Yamagata, M., Herman, J.P. & Sanes, J.R. Lamina-specific expression of adhesion molecules in developing chick optic tectum. J. Neurosci. 15, 4556–4571 (1995). 20. Uchida, N., Honjo, Y., Johnson, K.R., Wheelock, M.J. & Takeichi, M. The catenin/cadherin adhesion system is localized in synaptic junctions bordering transmitter release zones. J. Cell Biol. 135, 767–779 (1996). 21. Fannon, A.M. & Colman, D.R. A model for central synaptic junctional complex formation based on the differential adhesive specificities of the cadherins. Neuron 17, 423–434 (1996). 22. Benson, D.L. & Tanaka, H. N-cadherin redistribution during synaptogenesis in hippocampal neurons. J. Neurosci. 18, 6892–6904 (1998). 23. Huntley, G.W. & Benson, D.L. Neural (N)-cadherin at developing thalamocortical synapses provides an adhesion mechanism for the formation of somatopically organized connections. J. Comp. Neurol. 407, 453–471 (1999). 24. Iwai, Y. et al. DN-cadherin is required for spatial arrangement of nerve terminals and ultrastructural organization of synapses. Mol. Cell. Neurosci. 19, 375–388 (2002). 25. Togashi, H. et al. Cadherin regulates dendritic spine morphogenesis. Neuron 35, 77–89 (2002). 26. Bozdagi, O., Shan, W., Tanaka, H., Benson, D.L. & Huntley, G.W. Increasing numbers of synaptic puncta during late-phase LTP: N-cadherin is synthesized, recruited to synaptic sites, and required for potentiation. Neuron 28, 245–259 (2000). 27. Fischer, M., Kaech, S., Knutti, D. & Matus, A. Rapid actin-based plasticity in dendritic spines. Neuron 20, 847–854 (1998). 28. Yu, X. & Malenka, R.C. β-catenin is critical for dendritic morphogenesis. Nat. Neurosci. 6, 1169–1177 (2003). 29. Ebihara, T., Kawabata, I., Usui, S., Sobue, K. & Okabe, S. Synchronized formation and remodeling of postsynaptic densities: long-term visualization of hippocampal neurons expressing postsynaptic density proteins tagged with green fluorescent protein. J. Neurosci. 23, 2170–2181 (2003). 30. Provost, E. & Rimm, D.L. Controversies at the cytoplasmic face of the cadherinbased adhesion complex. Curr. Opin. Cell Biol. 11, 567–572 (1999). 31. Nagafuchi, A. Molecular architecture of adherens junctions. Curr. Opin. Cell Biol. 13, 600–603 (2001). 32. Papa, M. & Segal, M. Morphological plasticity in dendritic spines of cultured hippocampal neurons. Neuroscience 71, 1005–1011 (1996). 33. Trachtenberg, J.T. et al. Long-term in vivo imaging of experience-dependent synaptic plasticity in adult cortex. Nature 420, 788–794 (2002). 34. Grutzendler, J., Kasthuri, N. & Gan, W.B. Long-term dendritic spine stability in the adult cortex. Nature 420, 812–816 (2002). 35. Matus, A. Actin-based plasticity in dendritic spines. Science 290, 754–758 (2000). 36. Bek, S. & Kemler, R. Protein kinase CKII regulates the interaction of β-catenin with α-catenin and its protein stability. J. Cell Sci. 115, 4743–4753 (2002). 37. Piedra, J. et al. p120-catenin-associated Fer and Fyn tyrosine kinases regulate βcatenin Tyr-142 phosphorylation and β-catenin-α-catenin interaction. Mol. Cell Biol. 23, 2287–2297 (2003). 38. Murase, S., Mosser, E. & Schuman, E.M. Depolarization drives β-catenin into neuronal spines promoting changes in synaptic structure and function. Neuron 35, 91–105 (2002). 39. Rimm, D.L., Koslov, E.R., Kebriaei, P., Cianci, C.D. & Morrow, J.S. α1(E)-catenin is an actin-binding and -bundling protein mediating the attachment of F-actin to the membrane adhesion complex. Proc. Natl. Acad. Sci. USA 92, 8813–8817 (1995). 40. Itoh, M., Nagafuchi, A., Moroi, S. & Tsukita, S. Involvement of ZO-1 in cadherinbased cell adhesion through its direct binding to α-catenin and actin filaments. J. Cell Biol. 138, 181–192 (1997). 41. Kussel-Andermann, P. et al. Vezatin, a novel transmembrane protein, bridges myosin VIIA to the cadherin-catenins complex. Embo J. 19, 6020–6029 (2000). 42. Ackermann, M. & Matus, A. Activity-induced targeting of profilin and stabilization of dendritic spine morphology. Nat. Neurosci. 6, 1194–1200 (2003). 43. Mizoguchi, A. et al. Nectin: an adhesion molecule involved in formation of synapses. J. Cell Biol. 156, 555–565 (2002). 44. Maletic-Savatic, M., Malinow, R. & Svoboda, K. Rapid dendritic morphogenesis in CA1 hippocampal dendrites induced by synaptic activity. Science 283, 1923–1927 (1999). 45. Engert, F. & Bonhoeffer, T. Dendritic spine changes associated with hippocampal long-term synaptic plasticity. Nature 399, 66–70 (1999). 46. Hering, H. & Sheng, M. Dendritic spines: structure, dynamics and regulation. Nat. Rev. Neurosci. 2, 880–888 (2001). 47. Nimchinsky, E.A., Sabatini, B.L. & Svoboda, K. Structure and function of dendritic spines. Annu. Rev. Physiol. 64, 313–353 (2002). 48. Niwa, H., Yamamura, K. & Miyazaki, J. Efficient selection for high-expression transfectants with a novel eukaryotic vector. Gene 108, 193–199 (1991). 49. Uchida, N. et al. Mouse αN-catenin: two isoforms, specific expression in the nervous system, and chromosomal localization of the gene. Dev. Biol. 163, 75–85 (1994). 50. Hirano, S., Kimoto, N., Shimoyama, Y., Hirohashi, S. & Takeichi, M. Identification of a neural α-catenin as a key regulator of cadherin function and multicellular organization. Cell 70, 293–301 (1992).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

363

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

The X-linked mental retardation protein oligophrenin-1 is required for dendritic spine morphogenesis
Eve-Ellen Govek1–3, Sarah E Newey1,3, Colin J Akerman1, Justin R Cross1, Lieven Van der Veken1 & Linda Van Aelst1,2
Of 11 genes involved in nonspecific X-linked mental retardation (MRX), three encode regulators or effectors of the Rho GTPases, suggesting an important role for Rho signaling in cognitive function. It remains unknown, however, how mutations in Rho-linked genes lead to MRX. Here we report that oligophrenin-1, a Rho-GTPase activating protein that is absent in a family affected with MRX, is required for dendritic spine morphogenesis. Using RNA interference and antisense RNA approaches, we show that knock-down of oligophrenin-1 levels in CA1 neurons in rat hippocampal slices significantly decreases spine length. This phenotype can be recapitulated using an activated form of RhoA and rescued by inhibiting Rho-kinase, indicating that reduced oligophrenin-1 levels affect spine length by increasing RhoA and Rho-kinase activities. We further demonstrate an interaction between oligophrenin-1 and the postsynaptic adaptor protein Homer. Our findings provide the first insight into how mutations in a Rho-linked MRX gene may compromise neuronal function.

MRX is a disorder characterized by cognitive impairment without any other distinctive clinical features. A major challenge has been to uncover the molecular causes of MRX and the underlying cellular mechanisms responsible for reduced cognitive function. Eleven genes involved in MRX have been identified to date and notably, three of these encode regulators or effectors of the Rho subfamily of small GTP-binding proteins1–3. Members of this family, including RhoA, Rac and Cdc42, are key regulators of the actin cytoskeleton and affect many aspects of neuronal development and morphogenesis3–7. The three Rho-linked MRX genes encode (i) oligophrenin-1, a RhoGTPase activating protein (Rho-GAP)8, (ii) PAK3 (p21-activated kinase-3), a serine/threonine kinase downstream of Rac and Cdc42 (ref. 9) and (iii) ARHGEF6, a Rac GTPase exchange factor also known as αPIX or Cool-2 (ref. 10). The association between mutations in Rho-linked genes and MRX highlights the importance of Rho proteins in neuronal function and has led to the hypothesis that abnormalities in Rho signaling may be a cause of MRX3. Studies examining the effects of these mutations on neuronal signaling and development are therefore critical for the elucidation of cellular mechanisms underlying normal cognitive function and disease. Here we focus on the functional characterization of oligophrenin-1 (encoded by the OPHN1 gene in humans; Ophn-1 in mice), a protein with a Rho-GAP domain shown to negatively regulate RhoA, Rac and Cdc42 in vitro and in non-neuronal cells8,11. OPHN1 was identified by the analysis of a balanced translocation t(X;12) observed in a female patient with mild mental retardation. Its involvement in MRX was established by the identification of a mutation within the OPHN1 coding sequence in a family with MRX (MRX 60). In these two cases, the OPHN1 mutation is associated with a loss of, or dramatic reduction in, mRNA product8. Recent studies show that oligophrenin-1 is

present in neuronal and astroglial cells and that it colocalizes with actin at the tip of growing neurites11. However, the function of oligophrenin-1 in the brain is unknown and it remains to be seen how mutations in OPHN1 affect neuronal development and function, and contribute to MRX. To begin to understand how oligophrenin-1 deficiency affects neuronal function, we examined the effects of reducing oligophrenin-1 levels on the morphology of developing hippocampal (CA1) neurons in organotypic slices. We focused on dendritic spines, the main sites of excitatory synapses in the brain12, because changes in spine dimensions and density have been associated with synaptic plasticity13–16 and learning17, as well as with neurological disorders including mental retardation18–20. Using RNA interference (RNAi) and antisense RNA approaches, we found that downregulation of oligophrenin-1 in CA1 neurons resulted in a significant shortening of dendritic spines. We showed that this spine length phenotype was mediated by the RhoA/Rho-kinase signaling pathway, acting downstream of oligophrenin-1. Furthermore, we identified an interaction between oligophrenin-1 and Homer, placing oligophrenin-1 within a postsynaptic complex that potentially links oligophrenin-1 to glutamate receptor signaling. Our findings suggest how OPHN1 mutations may compromise cognitive function. RESULTS Oligophrenin-1 in the brain We began the characterization of oligophrenin-1 by determining its presence, distribution and subcellular localization in the brain. Immunoblot experiments of rat tissues showed that oligophrenin-1 protein levels were highest in the brain, but were also detectable to variable extents in other tissues (Fig. 1a). Oligophrenin-1 levels were

1Cold Spring Harbor Laboratory, 1 Bungtown Road, Cold Spring Harbor, New York 11724, USA. 2Molecular and Cellular Biology Program, State University of New York at Stony Brook, Stony Brook, New York 11794, USA. 3These authors contributed equally to this work. Correspondence should be addressed to L.V.A. ([email protected]).

Published online 14 March 2004; doi:10.1038/nn1210

364

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 1 Distribution and localization of oligophrenin-1 in the brain. (a) Multiple-tissue western blot from a postnatal day 30 (P30) rat probed with the oligophrenin-1 antibody 1296. Oligophrenin-1 has a molecular mass of approximately 92 kDa. B, brain; SM, skeletal muscle; H, heart; Li, liver; Lu, lung; Ki, kidney. (b) Immunoblot of P30 rat brain regions: Cer, cerebellum; Hip, hippocampus; Thal, thalamus; Flob, frontal lobes; Cor, sensory cortex; Olf, olfactory bulb. (c) Immunoblot of embryonic day 18 (E18), P2, P6, P10, P30 and adult (>8-week old) rat brains. (d) Parasagittal brain sections from a P30 rat double-immunolabeled with antibody 1296 (OPHN1, red staining) and an antibody to the neuronal marker NeuN (green staining). Arrows indicate immunolabeling of blood vessels. Scale bar, 50 µm. (e) Left, hippocampal neuron at 21 d.i.v. immunolabeled with the oligophrenin-1 antibody 1296 (OPHN1). Right, these neurons coimmunostained for actin, PSD-95 and synaptophysin. (f) Left, hippocampal neuron (21 d.i.v.) expressing T7oligophrenin-1 (T7-OPHN1) labeled with a T7specific antibody. Right, these neurons coimmunostained for actin, PSD-95 and synaptophysin. Scale bars (e,f), 10 µm.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

similar in all brain regions examined and the protein was present in both embryonic and adult tissue (Fig. 1b,c). Furthermore, virtually all neuronal populations, including pyramidal neurons of the hippocampus and cerebral cortex, stained positive for oligophrenin-1 (Fig. 1d). In these neurons, oligophrenin-1 was concentrated in the cell bodies and extended out into the dendrites. It was also detected in blood vessels (Fig. 1d, arrows) where it localized to vascular endothelial cells (data not shown). These results show that oligophrenin-1 is present in neurons in major regions of the brain, including those pertinent to learning and memory. To determine the subcellular localization of oligophrenin-1 in neurons, we examined the distribution of endogenous and ectopically expressed (T7-tagged) oligophrenin-1 in cultured hippocampal neurons that have mature synapses. We found high levels of both in the cell body, and in abundant puncta in axons, dendrites and dendritic spines (Fig. 1e,f). The presence of oligophrenin-1 at postsynaptic sites was confirmed by its colocalization with F-actin, which accumulates in spines (Fig. 1e,f), and its overlapping localization with PSD-95, a major component of the postsynaptic density (PSD; Fig. 1e,f). Consistent with a postsynaptic localization, T7-oligophrenin-1 in dendritic spines closely juxtaposed immunolabeling with the presynaptic marker

synaptophysin (Fig. 1f). Further complementary biochemical experiments confirmed that a proportion of endogenous oligophrenin-1 is localized at the synapse. We isolated synaptosomal plasma membranes (SPM) from rat hippocampi, and subsequently prepared PSD fractions, which showed that oligophrenin-1 is concentrated in the insoluble PSD pellets after two rounds of Triton extraction. Further extraction of Triton-insoluble complexes with N-lauroylsarcosine solubilized approximately 50% of the oligophrenin-1 present, leaving 50% in the PSD core. As expected, PSD-95 is concentrated in all insoluble, detergentextracted PSD pellets, whereas synaptophysin is solubilized upon treatment of SPM with Triton (data shown below in Fig. 7b). Together, these data reveal a postsynaptic localization for oligophrenin-1 and its presence in synapses on dendritic spines. Our studies also showed that oligophrenin-1 is present at presynaptic sites. Endogenous oligophrenin-1 immunolabeling was observed as numerous puncta along axons and overlapped with synaptophysin at some presynaptic terminals (Fig. 1e). Furthermore, T7-oligophrenin-1 was concentrated in axonal synaptic boutons that stained positive for synaptophysin (Fig. 1f, bottom row). Our findings that oligophrenin-1 is present both pre- and postsynaptically in neurons suggest an important role for oligophrenin-1 in synaptic function. Knock-down of oligophrenin-1 in neurons Because the mutation in OPHN1 associated with MRX results in a dramatic reduction in OPHN1 mRNA product8, we wanted to assess the effects of oligophrenin-1 knock-down on neuronal development and signaling. We developed two strategies to interfere with oligophrenin-1 expression in primary neurons: RNAi and an antisense approach. For RNAi, two siRNA (small interfering RNA) duplexes were designed, one from the cDNA coding region (Ophn1#1) and the other from the 3′ untranslated region (Ophn1#2) of rat and mouse Ophn-1. The efficacy

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

365

ARTICLES
Figure 2 Knock-down of oligophrenin-1 levels using RNAi and antisense RNA technologies. (a) Western blot of oligophrenin-1 levels in REF52 cells transfected with Ophn-1 siRNAs (Ophn1#1 and Ophn1#2), control siRNA (control) or no siRNA (none), and with either control vector or antisense. The immunoblot was probed with 1296 and anti-ERK2 as a loading control. (b) Western blots of oligophrenin-1 levels in young primary hippocampal neurons transfected with Ophn1#1 siRNA, Ophn1#2 siRNA, control siRNA, β-tubulin siRNA (β-tub) or no siRNA, and with either control vector or antisense. Blots were probed with 1296 and anti-β-tubulin as a loading control. (c) Mean oligophrenin-1 levels in young hippocampal neurons transfected with Ophn-1 siRNAs or antisense expressed as a percentage of control transfections. Oligophrenin-1 levels were normalized to β-tubulin levels in the same sample. Ophn1#1 and Ophn1#2 siRNAs significantly decreased oligophrenin-1 levels (t-test, P < 0.002, n = 5 in both cases), as did the antisense construct (t-test, P = 0.003, n = 5; *denotes statistical significance). Error bars indicate standard error of the mean (s.e.m.). (d) Double immunofluorescence labeling of neurons transfected with Ophn1#2 siRNA (bottom) or control siRNA (top). The right panel shows oligophrenin-1 immunostaining detected with 1724 (antiOPHN1); the left panel shows the same cells labeled for β-tubulin (anti-β-tub) as a cell marker. Scale bar, 10 µm. (e) Western blots of oligophrenin-1 levels in mature hippocampal neurons transfected with Ophn1#2 siRNA or no siRNA. Blots were probed as in b.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

of these siRNAs in reducing oligophrenin-1 levels was first tested in rat embryonic fibroblast (REF52) cells, an easily transfectable cell line, and both were successful in reducing oligophrenin-1 levels compared to a negative control duplex siRNA (Fig. 2a). The Ophn-1 and control siRNAs were subsequently transfected into dissociated hippocampal neurons at 5 or 6 days in vitro (d.i.v.), and 48 h after transfection, both Ophn-1 siRNAs, but not the control siRNAs, dramatically reduced oligophrenin-1 levels in these cells (Fig. 2b,c). These observations were confirmed by immunofluorescence labeling of cultured hippocampal neurons transfected at 6 d.i.v. with either Ophn1#2 siRNA or control siRNA. Neurons were double-labeled for oligophrenin-1 and a cell marker, β-tubulin. Transfection of Ophn1#2 siRNA, but not control siRNA, led to a severe reduction in the intensity of oligophrenin-1 immunolabeling in the cell body and almost completely abolished oligophrenin-1 expression in neuronal processes (Fig. 2e). This effect was specific, as β-tubulin expression was unaffected in both cases. Similarly, we observed a 30% reduction in oligophrenin-1 levels in mature neuronal cultures (15 d.i.v.) transfected with Ophn1#2 siRNA (Fig. 2c). These experiments demonstrate that the selected Ophn-1 siRNAs are effective in reducing oligophrenin-1 levels in primary hippocampal neurons. For the antisense approach, the full-length mouse Ophn-1 cDNA was cloned in the antisense direction into a mammalian expression vector. This construct (termed ‘antisense’) successfully knocked down oligophrenin-1 levels in REF52 cells, as seen in our immuno-blot analysis (Fig. 2a). Subsequent transfection of the antisense construct into hippocampal neurons also resulted in a significant reduction in oligophrenin-1 levels 48 h post-transfection (Fig. 2b,d). Taken together, we established two independent approaches to effectively knock down oligophrenin-1 expression in primary hippocampal neurons. Oligophrenin-1 knock-down affects spine morphology Given the importance of the structure and shape of dendritic spines for synaptic function12–16, and the presence of oligophrenin-1 in spines, we

assessed what effect oligophrenin-1 knock-down has on dendritic spine morphology in CA1 pyramidal cells of rat organotypic hippocampal slices. Hippocampal slices were biolistically transfected with a GFP expression vector alone, or with a GFP expression vector and one of the two Ophn-1 siRNAs, Ophn1#1 or Ophn1#2, or the control siRNA. Using this method, we observed that on average, two or three CA1 neurons, and a similar number of CA3 neurons, are transfected (Supplementary Fig. 1 online). GFP labeled dendrites and spines from transfected CA1 neurons were imaged 48 hours post-transfection, using two-photon laser scanning microscopy. These cells have already acquired their characteristic dendritic branching pattern and display numerous protrusions from their dendrites (Fig. 3a). A minority of these protrusions are filopodial (long and headless), however, a large fraction have a welldefined neck and head structure, characteristic of mature spines21. We measured the length and number of spines (all protrusions including filopodia) on primary and secondary dendrites (see Methods). Both Ophn-1 siRNAs were found to significantly decrease spine length when compared to control siRNA transfected neurons (P < 0.0001 for both). The mean spine lengths for Ophn1#1 and Ophn1#2 siRNA–transfected neurons were 21% and 18% smaller, respectively, than that for control siRNA-transfected neurons (Fig. 3c,d). There was no significant difference in spine length between cells transfected with a GFP expression vector alone and control siRNA (P = 0.63; Fig. 3b). Similar results were obtained for spine lengths analyzed from primary and secondary dendrites separately (Supplementary Table 1 online). Thus, two different siRNAs targeted to two different regions of the Ophn-1 transcript reduced spine length, whereas the control siRNA did not. In contrast to the effect on spine length, we found that reduced oligophrenin-1 levels did not affect the density of spines (including filopodia), nor the density of filopodia (defined as headless protrusions longer than 2 µm) when analyzed separately (Supplementary Table 2 online). To confirm these findings using an independent approach, we biolistically transfected the Ophn-1 antisense construct or an empty plasmid,

366

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 3 Oligophrenin-1 siRNAs cause a reduction in spine length in CA1 pyramidal cells in hippocampal slices. (a) Example of a biolistically transfected hippocampal CA1 pyramidal cell expressing GFP. The first panel shows the typical morphology of the CA1 cells, the second panel is an enlargement of the basal dendrites, and the third panel (the inset) shows a digital zoom of spines (indicated by arrows) used to count spine numbers and lengths. (b–d) The line graphs show the cumulative frequency distribution of spine lengths for neurons transfected with control siRNA (black) and neurons transfected with a GFP expression vector alone, Ophn1#1 siRNA or Ophn1#2 siRNA (white). Insets show the mean spine length for the two groups. Error bars indicate s.e.m. Spine length was significantly shorter for Ophn1#1 siRNA–transfected neurons than for control siRNA–transfected neurons (P < 0.0001, 1,533 control siRNA spines, 2,805 Ophn1#1 siRNA spines), and for Ophn1#2 siRNA–transfected neurons compared to control siRNA–transfected neurons (P < 0.0001, 1,533 control siRNA spines, 3,170 Ophn1#2 siRNA spines). No significant difference in spine length was observed between cells transfected with the GFP expression vector alone and control siRNA–transfected cells (P = 0.63, 1,083 GFP spines, 1,533 control siRNA spines). Scale bar, 5 µm.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

together with a GFP expression vector, into CA1 neurons in hippocampal slices. Consistent with the Ophn-1 siRNA data, neurons expressing antisense RNA showed significantly shorter spines compared to cells transfected with control vector (P < 0.0001). The mean spine length for neurons expressing antisense RNA was 12% smaller than for neurons expressing control vector (Fig. 4a). No significant difference in spine length was observed between cells transfected with the GFP expression vector alone and cells transfected with control vector (P = 0.14; Fig. 4b). Similar results were obtained for spines analyzed from primary and secondary dendrites separately (Supplementary Table 1). As in the siRNA experiments, mean spine and filopodia density did not differ significantly between neurons transfected with control vector and those transfected with antisense or GFP expression vector alone (Supplementary Table 2). Thus, by two independent approaches, we found that knocking down oligophrenin-1 levels in hippocampal CA1 neurons causes a decrease in dendritic spine length. Importantly, a change in spine length of comparable magnitude, albeit in the opposite direction, is reported in a mouse model of fragile X syndrome22,23, indicating that morphological changes of this degree can ultimately lead to deficits in cognitive function. Oligophrenin-1 affects the RhoA/Rho-kinase signaling pathway Oligophrenin-1 contains a Rho-GAP domain that negatively regulates the activity of Rho GTPases8,11. Therefore, a likely mechanism by which

loss of oligophrenin-1 could reduce dendritic spine length is by increasing the activity of one or more of the Rho-GTPase family members. To determine which Rho GTPase family member(s) oligophrenin-1 acts upon in a neuronal context, we expressed full-length oligophrenin-1 in PC12 cells and found that oligophrenin-1 decreased global levels of active GTP-bound RhoA, Rac1 and Cdc42 (Fig. 5a). Although these GTPase pull-down assays suggest that oligophrenin-1 can act as a GAP for all three Rho GTPases, they do no take into account spatial regulation of the Rho GTPases that likely occurs in fully differentiated neurons. The GAP activity of oligophrenin-1 may be restricted in dendrites and spines by the local availability or activity of the target GTPase and/or the presence of additional factors.

Figure 4 Ophn-1 antisense RNA causes a reduction in spine length in CA1 pyramidal cells in hippocampal slices. (a,b) The line graphs show the cumulative frequency distribution of spine lengths for control vector transfected neurons (black) and neurons transfected with antisense or a GFP expression vector alone (white). Insets show the mean spine length for the two groups. Error bars indicate s.e.m. Spine length was significantly shorter for antisense transfected neurons than for control vector transfected neurons (P < 0.0001, 2,508 vector spines, 2,011 antisense spines). No significant difference in spine length was observed between cells transfected with a GFP expression vector alone and control vector transfected cells (P = 0.14; 1,083 GFP spines, 2,508 vector spines). Scale bar, 5 µm.

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

367

ARTICLES
Figure 5 Rho GTPase activation assays and activated Rho GTPase spine phenotypes. (a) Rhotekin RhoA GTPase and Pak Rac/Cdc42 GTPase activation assays show that oligophrenin-1 is capable of acting as a Rho GAP for all three GTPases in PC12 cells. Serum starved cells expressing oligophrenin-1/mycHis (OPHN1) or vector were left untreated or were treated with 1 µM lysophosphatidic acid (LPA) for 3 min or 200 ng/ml epidermal growth factor (EGF) for 1 min as indicated. Pulled-down GTP-bound GTPase was compared to total GTPase in the cell lysates. GTPase activity for oligophrenin-1 transfected cells is represented as a percentage of the GTPase activity of the control vector transfected cells. Error bars indicate standard deviation (s.d.). (b) Activated Rho GTPase spine phenotypes in rat hippocampal CA1 pyramidal cells in slices expressing constitutively active (CA) Rho GTPase mutants. CA RhoA (RhoAV14) results in reduced protrusion length, whereas CA Rac1 (Rac1V12) causes the formation of abnormal lamellipodia/veil-like protrusions. CA Cdc42 (Cdc42V12) did not show any major effects on spine morphology. Scale bar, 5 µm.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

As a first step toward addressing the relationship between oligophrenin-1 and the Rho GTPases in developing hippocampal neurons, we expressed constitutively active Rho GTPases in CA1 neurons in hippocampal slices and compared the resulting spine phenotypes with that obtained upon oligophrenin-1 suppression. Expression of a constitutively active RhoA mutant (RhoAV14) reduced spine length and density, whereas constitutively active Rac1 (Rac1V12) caused the formation of numerous abnormal lamellipodia-like protrusions. A constitutively active Cdc42 mutant (Cdc42V12) did not have an observable effect on spine morphology (Fig. 5b). These findings are consistent with studies by other groups6,7. It is striking that a knockdown of oligophrenin-1 most closely mimicked the effect of an activated RhoA mutant with regard to changes in spine length, supporting the idea that oligophrenin-1 acts predominantly upon RhoA with respect to spine morphogenesis. Although reduced oligophrenin-1 levels did not decrease spine density as seen for constitutively active RhoA6,7 (unpublished data), it is likely that levels of RhoA-GTP are higher in neurons overexpressing a potent activated RhoA mutant than in cells with reduced oligophrenin-1 levels, resulting in a more severe phenotype. Alternatively, oligophrenin-1 may be just one of several negative regulators acting on RhoA in dendritic spines, and therefore loss of only one such regulator may result in a less extreme phenotype. To confirm that oligophrenin-1 affects spine length by acting on the RhoA signaling pathway, we tested whether inhibiting this pathway can rescue the reduced spine length resulting from oligophrenin1 knock-down. For these experiments, we made use of the Rho-kinase inhibitor Y-27632 (ref. 24). Rho-kinase is a major downstream target of RhoA that is involved in neurite retraction and axonogenesis25. Recent studies revealed that treatment of hippocampal neurons with Y-27632 rescued dendritic simplification and reduced spine density induced by RhoAV14, whereas treatment of control neurons with Y-27632 did not have a pronounced effect6. Importantly, we found that Y-27632 treatment of hippocampal slices transfected with an Ophn-1 duplex siRNA largely rescued the oligophrenin-1 knockdown effect on spine length. The mean spine length of neurons transfected with Ophn1#2 siRNA and treated with Y-27632 was 15% larger than that for untreated Ophn1#2 siRNA transfected neurons

(P < 0.0001; Fig. 6a). Y-27632 treatment of hippocampal slices transfected with control duplex siRNA did not significantly increase spine length compared to untreated slices (P = 0.11; Fig. 6a). These findings indicate that the RhoA/Rho-kinase signaling pathway mediates the action of oligophrenin-1 knock-down on dendritic spine length in CA1 pyramidal neurons. The lack of phenotype in control slices treated with Y-27632 implies that the RhoA/Rho-kinase pathway is repressed under physiological conditions6. Examination of the effects of ectopic expression of full-length oligophrenin-1 on spine length yielded further support for this repression model. We found that expression of T7-oligophrenin-1 in CA1 cells in hippocampal slices did not result in an increase in spine length compared to neurons expressing a control vector (P = 0.14; Fig. 6b). Thus we propose that oligophrenin-1 normally acts to repress the RhoA/Rho-kinase pathway to maintain spine length. Upon removal of oligophrenin-1, inhibition of RhoA is relieved, resulting in activation of Rho-kinase and a concomitant decrease in spine length. Oligophrenin-1 interacts with Homer In addition to determining the signaling pathways downstream of oligophrenin-1, we uncovered an interaction that provides a possible connection between oligophrenin-1 and receptor signaling at postsynaptic sites. Within the amino acid sequence of oligophrenin-1, we noticed the sequence PPLEF (residues 4–8) that corresponds to the consensus motif found in proteins that bind to the EVH1 domain of Homer proteins26 (Fig. 7a). Homer proteins are adaptor molecules that link glutamate receptors to multiple intracellular targets and influence dendritic spine morphogenesis and synaptic transmission27,28. The identification of a Homer binding motif, together with our finding that oligophrenin-1 and Homer cofractionate in the PSD core (Fig. 7b), suggested a potential interaction between these two molecules. We therefore used pull-down assays to investigate whether oligophrenin-1 and Homer associate biochemically. Beads loaded with Homer1b- and Homer1c-GST fusion proteins and incubated with lysates from rat hippocampi and cortices successfully pulled down oligophrenin-1 (Fig. 7c). We further used this assay to show that mutation of the con-

368

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 6 Oligophrenin-1 affects the RhoA/Rhokinase signaling pathway. (a) Inhibition of Rhokinase rescues the decrease in spine length resulting from a knock-down of oligophrenin-1 levels in CA1 neurons in hippocampal slices. Left graph shows the mean spine length for control siRNA transfected neurons (con), Ophn1#2 transfected neurons and Ophn1#2 transfected neurons in slices treated with 100 µM Rhokinase inhibitor Y-27632. Error bars indicate s.e.m. Spine length was significantly shorter for Ophn1#2 siRNA transfected neurons than for control siRNA transfected neurons (P < 0.0001; 2,946 control siRNA spines, 1,668 Ophn1#2 siRNA spines) and significantly longer for Ophn1#2 siRNA transfected neurons treated with Y-27632 compared to untreated Ophn1#2 siRNA transfected neurons (P < 0.0001; 1,668 Ophn1#2 siRNA spines, 1,562 Ophn1#2 siRNA + Y-27632 spines). Right graph shows the mean spine length for control siRNA transfected neurons and control siRNA transfected neurons in Y-27632 treated slices. There was no significant difference between the two groups (P = 0.11, 1,551 control siRNA spines, 1,022 control siRNA + Y-27632 spines). Scale bar, 5 µm. (b) Expression of T7-oligophrenin-1 in CA1 neurons does not affect spine length. The line graph shows the cumulative frequency distribution of spine lengths for control vector expressing neurons (black) and neurons expressing T7-oligophrenin-1 (OPHN1; white). Inset shows the mean spine length for the two groups. Error bars indicate s.e.m. No significant difference in spine length was observed between these two groups (P = 0.14, 1,400 control vector spines, 1,328 OPHN1 spines). Scale bar, 5 µm.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

sensus Homer binding motif in oligophrenin-1 disrupts Homer binding. Using HEK 293 cells, we expressed either wild-type or mutant T7oligophrenin-1, which contained mutations P4L and F8C in the Homer binding motif. Mutant oligophrenin-1 showed markedly less binding to immobilized Homer1b than did wild-type oligophrenin-1 (Fig. 7d). Finally, to demonstrate an in vivo association between oligophrenin-1 and Homer, we coimmunoprecipitated the endogenous proteins from rat hippocampal lysates using antibodies against both Homer and oligophrenin-1 (Fig. 7e). Together with our findings above, which implicate the RhoA/Rho-kinase pathway downstream of oligophrenin1, the association between oligophrenin-1 and Homer may provide a connection between glutamate receptor signaling and actin cytoskeletal rearrangements necessary for morphological spine changes. DISCUSSION Mutations in single genes that cause cognitive deficits provide a unique opportunity to uncover the molecular and cellular processes that contribute to normal brain function. To date, no information is available as to how mutations in Rho-linked MRX genes impact the neuronal morphology of affected individuals, and data from knockout mouse models have not been reported. OPHN1 was the first Rholinked MRX gene identified, and mutations in the gene are found to result in reduced OPHN1 expression. Here we show that knocking down oligophrenin-1 significantly decreased dendritic spine length in CA1 pyramidal neurons. Importantly, spine morphological changes of the same magnitude have been reported for a mouse model of fragile X22,23, indicating that such changes can compromise synaptic plasticity29 and potentially lead to learning and memory deficits. We have determined that the RhoA/Rho-kinase signaling pathway mediates the action of oligophrenin-1 knockdown on dendritic spine length. Our data suggest that this pathway, although intact, is repressed in pyramidal neurons under physiological conditions6,30 and that endogenous oligophrenin-1 acts to repress the RhoA signaling pathway to maintain spine length. When this repression is alleviated by loss of oligophrenin-1, there is a subsequent increase in

RhoA and Rho-kinase activities, causing a reduction in spine length. This reduction in spine length is likely brought about by phosphorylation of myosin light chain, either directly by Rho-kinase or indirectly through Rho-kinase phosphorylation and inactivation of myosin light chain phosphatase, leading to an increase in actomyosin contractility25,30. We further uncovered an interaction between oligophrenin-1 and Homer, a protein involved in dendritic spine morphogenesis and synaptic transmission27,28. Homer proteins are key adaptor proteins at the PSD where they organize glutamate receptor signaling complexes. The EVH1 domain of Homer proteins interacts with a conserved Homer binding motif, PPxxF, in a variety of binding partners, including type-I metabotropic glutamate receptors (mGluRs), inositol 1,4,5 tris-phosphate receptors (IP3Rs) and Shank, a scaffolding molecule that indirectly links Homer to NMDA-type glutamate receptors27,28. Interestingly, Homer cooperates with Shank to induce spine enlargement, while a dominant-negative form of Homer (Homer 1a) decreases the size and length of spines and interferes with synaptic transmission31,32. An intriguing possibility is that oligophrenin-1, as a regulator of the RhoA/Rho-kinase pathway in differentiated neurons, provides a crucial link between postsynaptic receptors (via Homer) and the actin cytoskeleton to regulate dendritic spine morphogenesis. This is consistent with extensive evidence that glutamate receptor activation affects the stability of actin, spine morphology and spine maintenance33–36. Additionally, glutamatergic synaptic activity has been shown to regulate RhoA activity, which affects dendritic arbor stabilization37. It is therefore tempting to speculate that oligophrenin-1 may act downstream of glutamatergic receptors to regulate RhoA activity in spines, and thus contribute to their stabilization. Although our data are consistent with a model in which loss of oligophrenin-1 elicits spine length changes by regulating the actin cytoskeleton through the RhoA/Rho-kinase pathway, it is possible that calcium dynamics and signaling contribute to the spine length change we observed. This is particularly relevant given that Homer

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

369

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 7 Oligophrenin-1 interacts with Homer. (a) Sequence alignment of oligophrenin-1 with known Homer-interacting proteins containing the consensus Homer (P)PxxF binding motif. (b) Immunoblots of hippocampal lysates (10 µg per lane), SPM (1.5 µg per lane) and PSD fractions probed with antibodies against oligophrenin-1 (1296), Homer1b/c, PSD-95 and synaptophysin. SPM were extracted in 0.5% Triton X-100 to yield the PSD Triton-1 pellet (P) and supernatant (S). The PSD Triton-1-P was divided and extracted in 0.5% Triton X-100 or 3% N-lauroyl sarcosine to yield PSD Triton-2 and sarcosyl pellets and supernatants, respectively. (c) Homer1b- or Homer1c-GST fusion proteins, or GST alone, were incubated with extracts from rat hippocampi and cortices. Bound oligophrenin-1 was detected by immunoblotting with 1296. Input lane was loaded with 15% of the extract used for the assay. (d) Lysates from HEK 293 cells expressing either wild-type T7-oligophrenin-1 (OPHN1 WT) or mutant T7-oligophrenin-1 containing mutations P4L and F8C (OPHN1 PF) were incubated with Homer1b-GST or GST alone. Immunoblots showing bound oligophrenin-1 were from the same experiment and exposed to film for the same amount of time. Input lane was loaded with 20% of the lysates used for the assay. (e) Extracts from rat hippocampi were immunoprecipitated with 1296 (OPHN1), a polyclonal Homer antibody (Homer), a monoclonal control antibody against CD8 (con Ab) or no antibody (no Ab). Immune complexes were subject to immunoblotting and probed with either a monoclonal Homer antibody or 1296 as indicated. Input lane was loaded with 5% of the total lysate used.

can physically link type-I mGluRs and TRP calcium channels with IP3 receptors27,38, and that calcium release from intracellular stores can affect spine length39. A calcium increase in spines may promote the remodeling of actin filaments and enhance actomyosin contractility, causing alterations in spine structure40. Importantly, spine length changes, whether or not they are dependent on calcium-mediated actin cytoskeletal rearrangements, can themselves influence postsynaptic signaling. Alterations in spine length have been shown to affect the diffusion rate and decay kinetics of calcium in the spine head relative to those of the dendritic shaft41,42, and even small changes in spine length can cause large changes in calcium decay kinetics41. Thus, it is conceivable that interfering with oligophrenin-1 function may cause alterations in calcium dynamics, likely by disrupting its association with Homer and/or RhoA signaling. Given that calcium is a mediator of input-specific forms of synaptic plasticity42, such changes could ultimately affect learning and memory. In summary, this study demonstrates the successful use of siRNAs in organotypic hippocampal slices to examine the function of a disease-related molecule. We found that loss of oligophrenin-1 causes changes in spine morphology, which are believed to be important for synaptic function. Our findings provide a potential explanation as to how loss of this protein may result in cognitive impairment underlying MRX. It should be noted that in addition to the involvement of oligophrenin-1 in MRX, very recent studies implicate a role for oligophrenin-1 in mental impairment in individuals that also have

epilepsy and/or cerebellar hypoplasia43,44. It will therefore be of interest to investigate additional roles for oligophrenin-1 in neuronal development and disease. METHODS
Oligophrenin-1 antibodies. Antibodies 1296 and 1724 were raised in rabbits against the C-terminal peptide CETASRKTGSSQGRLPGDES and amino acids (aa) 635–802, respectively. Constructs and siRNAs. Full-length mouse Ophn-1 cDNA was subcloned in reverse into the T7-epitope tagged expression vector, pCGT45, to generate the antisense construct. pCGT/GTPases were made as described45. GST-Rhotekin (aa 7–89) and GST-Pak3 CRIB (aa 65–137) plasmids were gifts from M. Schwartz (University of Virginia, Charlottesville, Virginia, USA) and R. Cerione (Cornell University, Ithaca, New York, USA). Full-length human OPHN1 cDNA was subcloned into pCGT (pCGT/OPHN1) and pcDNA3.1/MycHisA (pcDNA3.1/MycHisA/OPHN1) (Invitrogen). Full-length OPHN1 P4L/F8C mutant was PCR generated and cloned into pCGT (OPHN1 PF). Homer1b/1c cDNAs were subcloned into pGEX-4T-1 (Amersham). siRNAs were from Dharmacon. siRNA target sequence for Ophn1#1 was 5′ AAAGGGATCAAGACAGAAGGG 3′, and 5′ AAGAGCAGCTCTTTCTGGCCT 3′ for OPHN1#2. Control siRNA was caspase 8 or caspase 2 duplex siRNA (gift from Y. Lazebnik, Cold Spring Harbor Laboratory, Cold Spring Harbor, New York, USA). Cell culture and transfection. REF52 or HEK 293 cells were transfected using Lipofectamine 2000 (Invitrogen) and harvested after 48 h. Medium-density

370

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
primary hippocampal neuron cultures were prepared from E18 rat embryos as described46. For immunofluorescence (IF), hippocampal neurons (19 d.i.v.) were transfected with pCGT/OPHN1 using Lipofectamine 2000, fixed after 48 h and permeabilized. Neurons were transfected before plating with antisense and control vector using the Mouse Neuron Nucleofector kit (Amaxa Biosystems) and harvested 3 d later for western blotting. Transfection efficiencies were 40–70%. For siRNA transfections, 5/6 or 12 d.i.v. neurons were transfected with siRNAs using Lipofectamine 2000 and harvested 48 and 72 h later, respectively, and then subjected to western blotting. For IF, neurons (6 d.i.v.) were transfected with siRNAs and methanol fixed 96 h later. Transfection efficiency was > 90% at 5/6 d.i.v. using a rhodamine-labeled siRNA. Hippocampal slices were made as previously described47. Slices were generated from P4 rat pups and biolistically transfected after 4 d in culture using a Helios Gene Gun (Bio-Rad). The following amounts of plasmids or siRNAs were precipitated onto 12.5 mg of 1.6-µm gold particles: 50 µg of pEGFPN3 (Clontech) and 50 µg of antisense construct, or pCGN; 20 µg of pEGFPN3 alone; 20 µg of pEGFPN3 and 160 µl of 20 µM duplex siRNA; 50 µg of pEGFPN3 and 50 µg of pCGT/OPHN1, or pCGN; and 50 µg of pEGFPN3 and 50 µg of pCGT/GTPase, or pCGN. To assess effective coating of siRNAs onto gold particles, a rhodamine-labeled siRNA was successfully precipitated onto beads as examined by fluorescence microscopy (data not shown). For the Rhokinase inhibitor experiments, 100 µM Y-27632 dihydrochloride (Alexis) was added to the medium at the time of transfection. Slices were incubated 48 h before the 1.5-h fixation in 4% PFA, 4% sucrose, PBS. All animal care protocols were approved by Cold Spring Harbor Laboratory. Western blotting. Rat tissues, REF52 cells and hippocampal neurons were prepared in 75 mM Tris (pH 6.8), 3.8% SDS, 4 M urea, 20% glycerol and subjected to western blotting using standard methods. We used the following primary antibodies: anti-oligophrenin-1 1296, anti-neuronal class III β-tubulin (Covance) and anti-ERK2 (Santa Cruz). Oligophrenin-1 levels were normalized to β-tubulin or ERK2 levels and expressed as a percentage of control transfections. Immunofluorescence. Parasagittal frozen brain sections (10 µm) from a P30 rat were methanol-fixed, and dissociated hippocampal neurons were PFAfixed. We used the following primary antibodies: anti-oligophrenin-1 (1296, 1724), anti-neuronal class III β-tubulin, anti-PSD-95 (IgG2A; Affinity Bioreagents), anti-synaptophysin clone SVP-38 (IgG1; Sigma), anti-T7 tag (IgG2B; Novogen) and anti-NeuN MAB377 (Chemicon). We used the following secondary antibodies and toxins: Alexa Fluor 488 goat anti-mouse (H+L, IgG2a and IgG1 specific), Alexa Fluor 594 goat anti-mouse (IgG2b specific) or anti-rabbit (H+L), Alexa Fluor 488-labeled phalloidin (Molecular Probes). Cells were imaged using an Axioscope fluorescence microscope (Zeiss) with a 63× Plan Apochromat objective. Two-photon imaging and image analysis. Two-photon images were obtained using an Olympus Fluoview laser-scanning microscope with a Ti-Sapphire laser (Mira 900F; Coherent) at 910 nm and a LUMPlanF1/IR 40×, 0.75 NA water immersion lens. Basal dendrites of CA1 pyramidal cells were imaged with 5× zoom. Optical sections were spaced 1.0 µm apart and each was an average of three scans. For the RNAi and antisense experiments, four cells were analyzed for pEGFPN3 alone, five for control siRNA, nine for Ophn1#1 siRNA, twelve for Ophn1#2 siRNA, ten for control vector and eight for antisense construct. For the Rho-kinase inhibitor control experiment, five cells were analyzed for control siRNA and five cells for control siRNA + Y-27632. For the Rho-kinase inhibitor rescue experiment, ten cells were analyzed for control siRNA, five for Ophn1#2 siRNA, and five for Ophn1#2 siRNA + Y27632. For the T7-oligophrenin-1 experiment, six cells were analyzed for control vector and five for pCGT/OPHN1. Dendritic spines (protrusions including filopodia) were measured from an image stack projection using custom NIH Object Image macros. Protrusion lengths were measured from the protrusion’s tip to the point where it met the dendritic shaft. Individual optical section images were used to verify each protrusion, and protrusions of all lengths were measured. Statistical differences remained for all groups when a minimum protrusion length threshold of 0.3 µm was applied. At least 100 µm of primary dendrite and 200 µm of secondary dendrite were analyzed for each cell. Spine and filopodia densities were calculated per cell. Spine statistics were performed using the Kolmogorov-Smirnov two-sample test. Cumulative frequency plots (Figs. 3, 4 and 6) indicate the fraction of spines (y-axis) equal to or less than a certain length (x-axis). For illustrative purposes, data were binned (0.3 µm). Rhotekin and Pak CRIB GTPase activation assays. The Rhotekin and Pak CRIB assays were performed as previously described48, using PC12 cells transfected with pcDNA3.1/MycHisA/OPHN1 or vector. PSD extractions. SPM were prepared from two adult rat hippocampi as described49. PSD fractions were prepared and detergent-extracted, essentially as described50. Equal volumes of PSD pellets and supernatants were subjected to western blotting. We used the following primary antibodies: 1296, Homer1b/c (clone D3; Santa Cruz), PSD-95 (Affinity Bioregents) or synaptophysin (clone SVP-38; Sigma). Pull-down assays and coimmunoprecipitations. Homer1b/1c-GST fusion proteins and GST alone were immobilized onto glutathione-Sepharose beads (Amersham). Rat hippocampi and cortices were homogenized in 50 mM Tris pH 7.4, 1 mM EDTA and 1% CHAPS (TEC); then they were sonicated and centrifuged. Supernatants were incubated with Homer1b/Homer1c-GST fusion proteins or GST. HEK 293 cells expressing OPHN1 WT or OPHN1 PF were extracted in TE, 1% Triton X-100. Lysates were incubated with Homer1b-GST fusion protein or GST. All immunoblots were probed with antibody 1296. For immunoprecipitations, rat hippocampi were homogenized in 50 mM Tris, pH 7.4, 150 mM NaCl and 1% Triton X-100, essentially as described38. Homogenate was centrifuged and supernatant incubated with 2 µg of antibody: 1296, polyclonal Homer FL-354 (Santa Cruz), or monoclonal CD8. Antibody complexes were captured using Protein A agarose (Roche). Immunoblots were probed with 1296 or monoclonal Homer D3 antibody (Santa Cruz).
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS We thank A. Piccini, E. Ruthazer, B. Burbach, C. Kopec, K. Jensen and H. Hsieh for technical assistance. We also thank H. Cline, R. Malinow, J. Skowronski, K. Svoboda, J. Rodriguez and members of the Van Aelst Laboratory for discussions and/or critical reading of the manuscript. This work was supported by the National Institutes of Health and Dana Foundation (to L.V.A.), the Wellcome Trust (to S.E.N. and C.J.A.) and the National Institutes of Health (to E.E.G). COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 28 October 2003; accepted 30 January 2004 Published online at http://www.nature.com/natureneuroscience/
1. Chelly, J. & Mandel, J.L. Monogenic causes of X-linked mental retardation. Nat. Rev. Genet. 2, 669–680 (2001). 2. Frints, S.G., Froyen, G., Marynen, P. & Fryns, J.P. X-linked mental retardation: vanishing boundaries between non-specific (MRX) and syndromic (MRXS) forms. Clin. Genet. 62, 423–432 (2002). 3. Ramakers, G.J. Rho proteins, mental retardation and the cellular basis of cognition. Trends Neurosci. 25, 191–199 (2002). 4. Van Aelst, L. & D’Souza-Schorey, C. Rho GTPases and network signaling. Genes Dev. 11, 2295–2322 (1997). 5. Luo, L. Rho GTPases in neuronal morphogenesis. Nat. Rev. Neurosci. 1, 173–180 (2000). 6. Nakayama, A.Y., Harms, M.B. & Luo, L. Small GTPases Rac and Rho in the maintenance of dendritic spines and branches in hippocampal pyramidal neurons. J. Neurosci. 20, 5329–5338 (2000). 7. Tashiro, A., Minden, A. & Yuste, R. Regulation of dendritic spine morphology by the rho family of small GTPases: antagonistic roles of Rac and Rho. Cereb. Cortex 10, 927–938 (2000). 8. Billuart, P. et al. Oligophrenin-1 encodes a rhoGAP protein involved in X-linked mental retardation. Nature 392, 923–926 (1998). 9. Allen, K.M. et al. PAK3 mutation in nonsyndromic X-linked mental retardation. Nat. Genet. 20, 25–30 (1998). 10. Kutsche, K. et al. Mutations in ARHGEF6, encoding a guanine nucleotide exchange factor for Rho GTPases, in patients with X-linked mental retardation. Nat. Genet. 26, 247–250 (2000). 11. Fauchereau, F. et al. The RhoGAP activity of OPHN1, a new F-actin-binding protein is negatively controlled by its amino-terminal domain. Mol. Cell. Neurosci. 23, 574–586 (2003).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

371

ARTICLES
12. Hering, H. & Sheng, M. Dendritic spines: structure, dynamics and regulation. Nat. Rev. Neurosci. 2, 880–888 (2001). 13. Engert, F. & Bonhoeffer, T. Dendritic spine changes associated with hippocampal long-term synaptic plasticity. Nature 399, 66–70 (1999). 14. Maletic-Savatic, M., Malinow, R. & Svoboda, K. Rapid dendritic morphogenesis in CA1 hippocampal dendrites induced by synaptic activity. Science 283, 1923–1927 (1999). 15. Toni, N., Buchs, P.A., Nikonenko, I., Bron, C.R. & Muller, D. LTP promotes formation of multiple spine synapses between a single axon terminal and a dendrite. Nature 402, 421–425 (1999). 16. Matsuzaki, M. et al. Dendritic spine geometry is critical for AMPA receptor expression in hippocampal CA1 pyramidal neurons. Nat. Neurosci. 4, 1086–1092 (2001). 17. Comery, T.A., Stamoudis, C.X., Irwin, S.A. & Greenough, W.T. Increased density of multiple-head dendritic spines on medium-sized spiny neurons of the striatum in rats reared in a complex environment. Neurobiol. Learn. Mem. 66, 93–96 (1996). 18. Purpura, D.P. Dendritic spine “dysgenesis” and mental retardation. Science 186, 1126–1128 (1974). 19. Kaufmann, W.E. & Moser, H.W. Dendritic anomalies in disorders associated with mental retardation. Cereb. Cortex 10, 981–991 (2000). 20. Fiala, J.C., Spacek, J. & Harris, K.M. Dendritic spine pathology: cause or consequence of neurological disorders? Brain Res. Rev. 39, 29–54 (2002). 21. Harris, K.M. Structure, development, and plasticity of dendritic spines. Curr. Opin. Neurobiol. 9, 343–348 (1999). 22. Comery, T.A. et al. Abnormal dendritic spines in fragile X knockout mice: Maturation and pruning deficits. Proc. Natl. Acad. Sci. USA 94, 5401–5404 (1997). 23. Nimchinsky, E.A., Oberlander, A.M. & Svoboda, K. Abnormal development of dendritic spines in FMR1 knock-out mice. J. Neurosci. 21, 5139–5146 (2001). 24. Uehata, M. et al. Calcium sensitization of smooth muscle mediated by a Rho-associated protein kinase in hypertension. Nature 389, 990–994 (1997). 25. Amano, M., Fukata, Y. & Kaibuchi, K. Regulation and functions of Rho-associated kinase. Exp. Cell Res. 261, 44–51 (2000). 26. Beneken, J. et al. Structure of the Homer EVH1 domain-peptide complex reveals a new twist in polyproline recognition. Neuron 26, 143–154 (2000). 27. Fagni, L., Worley, P.F. & Ango, F. Homer as both a scaffold and transduction molecule. Sci. STKE 137, 1–7 (2002). 28. Xiao, B., Cheng, J. & Worley, P.F. Homer: a link between neural activity and glutamate receptor function. Curr. Opin. Neurobiol. 10, 370–374 (2000). 29. Huber, K.M., Gallagher, S.M., Warren, S.T. & Bear, M.F. Altered synaptic plasticity in a mouse model of fragile X mental retardation. Proc. Natl. Acad. Sci. USA 99, 7746–7750 (2003). 30. Billuart, P., Winter, C.G., Maresh, A., Zhao, X. & Luo L. Regulating axon branch stability: the role of p190 RhoGAP in repressing a retraction signaling pathway. Cell 107, 195–207 (2001). 31. Sala, C. et al. Regulation of dendritic spine morphology and synaptic function by Shank and Homer. Neuron 31, 115–130 (2001). 32. Sala, C. et al. Inhibition of dendritic spine morphogenesis and synaptic transmission by activity-inducible protein Homer1a. J. Neurosci. 23, 6327–6337 (2003). 33. Fischer, M., Kaech, S., Wagner, U., Brinkhaus, H. & Matus, A. Glutamate receptors regulate actin-based plasticity in dendritic spines. Nat. Neurosci. 3, 887–894 (2000). 34. Star, E.N., Kwiatkowski, D.J. & Murthy, V.N. Rapid turnover of actin in dendritic spines and its regulation by activity. Nat. Neurosci. 5, 239–246 (2002). 35. Ackermann, M. & Matus, A. Activity-induced targeting of profilin and stabilization of dendritic spine morphology. Nat. Neurosci. 6, 1194–1200 (2003). 36. McKinney, R.A., Capogna, M., Dürr, R., Gähwiler, B.H. & Thompson, S.M. Miniature synaptic events maintain dendritic spines via AMPA receptor activation. Nat. Neurosci. 2, 44–49 (1999). 37. Sin, W.C., Haas, K., Ruthazer, E.S. & Cline, H.T. Dendrite growth increased by visual activity requires NMDA receptor and Rho GTPases. Nature 419, 475–480 (2002). 38. Yuan, J.P. et al. Homer binds TRPC family channels and is required for gating of TRPC1 by IP3 receptors. Cell 114, 777–789 (2003). 39. Korkotian E. & Segal, M. Release of calcium from stores alters the morphology of dendritic spines in cultured hippocampal neurons. Proc. Natl. Acad. Sci. USA 96, 12068–12072 (1999). 40. Hayashi, K. et al. Modulatory role of Drebrin on the cytoskeleton within dendritic spines in the rat cerebral cortex. J. Neurosci. 16, 7161–7170 (1996). 41. Majewska, A., Tashiro, A. & Yuste, R. Regulation of spine calcium dynamics by rapid spine motility. J. Neurosci. 20, 8262–8268 (2000). 42. Yuste, R., Majewska, A. & Holthoff, K. From form to function: calcium compartmentalization in dendritic spines. Nat. Neurosci. 3, 653–659 (2000). 43. Bergmann, C. et al. Oligophrenin-1 (OPHN1) gene mutation causes X-linked mental retardation with epilepsy, rostral ventricular enlargement and cerebellar hypoplasia. Brain 126, 1537–1544 (2003). 44. Philip, N. et al. Mutations in the oligophrenin-1 gene (OPHN1) cause X linked congenital cerebellar hypoplasia. J. Med. Genet. 40, 441–446 (2003). 45. Joneson, T., McDonough, M., Bar-Sagi, D. & Van Aelst, L. RAC regulation of actin polymerization and proliferation by a pathway distinct from Jun kinase. Science 274, 1374–1376 (1996). 46. Pak, D.T., Yang, S., Rudolph-Correia, S., Kim, E. & Sheng, M. Regulation of dendritic spine morphology by SPAR, a PSD-95-associated RapGAP. Neuron 31, 289–303 (2001). 47. Stoppini, L., Buchs, P.A. & Muller, D. A simple method for organotypic cultures of nervous tissue. J. Neurosci. Methods 37, 173–182 (1991). 48. Yamaguchi, Y., Katoh, H., Yasui, H., Mori, K. & Negishi, M. RhoA inhibits the nerve growth factor-induced Rac1 activation through Rho-associated kinase-dependent pathway. J. Biol. Chem. 276, 18977–18983 (2001). 49. Lau, L.F. et al. Interaction of the N-methyl-D-aspartate receptor complex with a novel synapse-associated protein, SAP102. J. Biol. Chem. 271, 21622–21628 (1996). 50. Cho, K.O., Hunt, C.A. & Kennedy, M.B. The rat brain postsynaptic density fraction contains a homolog of the Drosophila discs-large tumor suppressor protein. Neuron 9, 929–942 (1992).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

372

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Local structural balance and functional interaction of excitatory and inhibitory synapses in hippocampal dendrites
Guosong Liu
Theoretical and experimental studies on the computation of neural networks suggest that neural computation results from a dynamic interplay of excitatory and inhibitory (E/I) synaptic inputs. Precisely how E/I synapses are organized structurally and functionally to facilitate meaningful interaction remains elusive. Here we show that E/I synapses are regulated across dendritic trees to maintain a constant ratio of inputs in cultured rat hippocampal neurons. This structural arrangement is accompanied by an E/I functional balance maintained by a ‘push-pull’ feedback regulatory mechanism that is capable of adjusting E/I efficacies in a coordinated fashion. We also found that during activity, inhibitory synapses can determine the impact of adjacent excitatory synapses only if they are colocalized on the same dendritic branch and are activated simultaneously. These fundamental relationships among E/I synapses provide organizational principles relevant to deciphering the structural and functional basis for neural computation within dendritic branches.

Neural computation is accomplished by interactions between synaptic inputs and membrane channels that translate synaptic input into meaningful temporal patterns of action potentials. Biophysical mechanisms underlying this computation are still largely unknown. Several theoretic and experimental studies of computation in functional neural networks suggest that neural computation results from a concurrent interplay of excitatory and inhibitory (E/I) synaptic inputs1–8. Although E/I interactions at the cell body have been studied9, a large proportion of inhibitory synapses are instead distributed on the dendritic tree10–14. These dendritic E/I synaptic inputs may interact locally to perform neural computation on the dendrite branches themselves1,4–6,15,16. If so, E/I synapses would need to be organized on dendrites to permit their meaningful interaction. The aim of this study was to understand the structural organization and functional interaction of excitatory and inhibitory synapses within individual dendritic branches, as well as to explore the mechanisms that maintain an optimal E/I balance. Our results indicate that E/I synapses are evenly distributed on dendritic trees to maintain a constant ratio among all dendritic branches. Furthermore, we found that E/I inputs to individual dendritic branches were more effective than those outside of dendritic branches. Therefore, this structural arrangement seems to facilitate meaningful E/I interactions on dendritic branches. Finally, we found that the balance of E/I synapses was established and maintained by a push-pull regulatory mechanism. RESULTS Structural balance of E/I synapses in dendrites To explore the rules that control the arrangement of excitatory and inhibitory (E/I) synapses, we analyzed the distribution of functional

E/I synapses along dendritic trees by simultaneously labeling the dendritic surface and active presynaptic terminals in cultured hippocampal neurons (Fig. 1). We found the number of functional synapses per unit length of dendrite to be higher in thicker branches than in thinner ones (Fig. 1a). Since thicker branches have a larger surface area per unit length of dendrite than thinner ones do, maintaining a higher number of synapses in thicker branches might serve to maintain a constant number of synapses across a given area of dendritic surface. To demonstrate this relationship quantitatively, we plotted the number of synapses found along a fixed length of dendrite versus the diameter of the dendrite (Fig. 1b). These two parameters were linearly correlated, suggesting an even distribution of functional synapses across dendritic surfaces17. Next, as FM dye staining cannot distinguish excitatory from inhibitory synapses, we used both excitatory and inhibitory synapse-specific marker proteins to distinguish the two types of synapses (Fig. 1c). Inhibitory synapses were distributed unevenly across the entire neuronal surface, with the highest densities found near somata, similar to the distribution pattern of inhibitory synapses on hippocampal pyramidal neurons in vivo12,13. Despite the lower density of inhibitory synapses in the dendritic tree, the majority of inhibitory synapses (86%; Fig. 1d) were located in the dendritic tree, owing to the substantially larger surface area of dendrites versus somata. Our analysis of the distribution pattern of E/I synapses on the dendritic tree showed that the relative numbers of E/I synapses within dendritic branches were highly correlated (Fig. 1c,e). The ratios of E/I synapses, both among dendritic branches of individual neurons and between neurons at the same stage of development, were constant (Fig. 1f). After synaptic maturation, the ratio of excita-

Picower Centre for Learning and Memory, RIKEN–MIT Neuroscience Research Center, Departments of Brain & Cognitive Sciences and Biology, MIT, Cambridge, Massachusetts 02139, USA. Correspondence should be addressed to G.L. ([email protected]). Published online 7 March 2004; doi:10.1038/nn1206

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

373

ARTICLES
These data suggest that coordinated insertion of E/I synapses resulted in a functional balance of E/I synaptic strength similar to the structural balance of E/I synapses (Fig. 1). Given this coordination in synaptic strength, our results led us to ask how these synapses might interact temporally. Neurons in cultured networks make mutual connections that give rise to periodic bursts of synaptic potentials (Fig. 2d), similar to giant depolarizing potentials observed in hippocampal slices18 and in vivo19. Bursting results from the synchronous activation of hundreds of E/I synapses. To study the composition of these events, we separated the contributions of E/I inputs by alternately clamping the membrane potential at –65 mV or 0 mV to record excitatory or inhibitory postsynaptic currents (EPSCs or IPSCs, respectively; Fig. 2d). To estimate GI/GE during bursts, we varied the clamp membrane potential to determine the burst reversal potential (Vrev) of the compound synaptic inputs (Fig. 2d, middle trace), Figure 1 The structural balance of E/I synapses is maintained throughout hippocampal dendrites. which can then be used to calculate the rela(a) Distribution of functional presynaptic terminals on the dendritic tree. Active synapses were labeled tive strength of GE versus GI (see Methods). by FM4-64 (red), and Alexa 488 delivered during whole-cell recording highlighted the corresponding dendritic surface in cultured hippocampal neurons (13 d.i.v.). (b) Number of synapses/unit length of The Vrev varied from –30 to –45 mV, corredendrite (D) increases linearly with enlarging radius (r) of dendritic branch (D = 0.092 + 0.095r; sponding to a GI/GE ratio that varied from R2 = 0.8911; P < 0.0001). (c) Visualization of E/I synapses arranged across a single dendritic 0.9 to 2.2 with a mean value of 1.3 (n = 4), branch. Excitatory synapses (green) were labeled with the vesicular glutamate transporter VGLUT1, consistent with the GI/GE ratio calculated whereas inhibitory synapses (red) were labeled with the synthetic enzyme GAD65 (20 d.i.v.). (d) Left: from miniature synaptic events (mEPSCs ratio of E/I in somatic and dendritic regions. A high density of inhibitory synapses occurs near the and mIPSCs) described above. These results soma. Right: distribution of inhibitory synapses in somatic and dendritic regions (86% of inhibitory synapses located at dendritic region). (e) Ratios of E/I synapses across individual dendritic branches indicate that a constant ratio of E/I inputs were constant (data from 31 branches in 8 neurons). (f) Ratio of E/I synapses in individual neurons was maintained both in general and dynamduring different stages of development. Each group of symbols represent the ratios of E/I synapses on ically across discrete time intervals. different dendritic branches from a single neuron. The concurrent E/I inputs during a burst is interesting because most excitatory synaptic inputs will be cancelled by concurrent tory to inhibitory synapses was approximately 4:1 across the dendritic inhibitory inputs, such that neural output is determined by the surface. We checked whether such a correlation could result from a dynamic balance of active E/I synapses (Fig. 2d,e). Despite the large random insertion of E and I by simulating insertion of E and I EPSCs during bursts (peak size ∼0.9 nA, mean 0.3 nA), the synaptic synapses with a Poisson model. This simulation placed the chance potential at the soma only reached 10–15 mV, resulting in few action probability of observing the degree of E/I correlation seen in potentials. In contrast, delivering only 0.1 nA of current into the Figure 1e at <0.001. These data suggest that the distribution of E/I soma through a patch pipette induced a similar depolarization and synapses on the dendritic tree is highly organized to maintain a bal- produced the same number of action potentials (Fig. 2e, left). This ance of excitatory and inhibitory inputs on each dendritic branch. suggests that, as is observed in intact preparations3–8, most excitatory inputs were cancelled by concurrent inhibitory inputs. As a result, Functional balance of E/I synaptic inputs neuronal output is controlled not only by the size of excitatory This uniform distribution of E/I synapses along the dendritic tree inputs, but also by the dynamic difference between excitatory and suggested that E/I synaptic function may also be highly balanced. We inhibitory inputs. therefore examined the relationship between the number of functional synapses and the relative overall strength of E/I synapses during Dynamic interplay of E/I inputs on dendritic tree early stages of synaptogenesis. As the overall strength of synaptic con- Given that most E/I synapses are located on the dendritic tree, we posnections is determined by the quantal size (q), defined by the inte- tulate that most E/I cancellations take place locally within dendrite grated conductance during quantal transmission and the frequency branches. To test this hypothesis, it is important to show that inhibitory (f) of quantal transmission, we used G = f*q to represent synaptic synapses on dendritic branches are active during bursting periods and strength (Fig. 2a). As expected, both excitatory and inhibitory synap- make significant contributions to the generation of bursting inhibitory tic strengths (GE and GI) increased in proportion to the number of inputs. To quantify the contribution of dendritic inhibitory synapses, functional synapses (Fig. 2b). The GI/GE balance was maintained we recorded the inhibitory synaptic inputs while blocking functionality among neurons at the same stage of development, although GI/GE of somatic inhibitory synapses with locally applied picrotoxin. The increased from 1 to 3 during neural network maturation (Fig. 2c). pressure of the drug application system was carefully adjusted to limit

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

374

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 2 Functional balance of E/I synaptic inputs in a single neuron. (a) Left, distribution of active synaptic terminals on the dendrite of a recorded neuron (11 d.i.v.). Right, mEPSCs and mIPSCs were recorded from the same neuron by holding the membrane potential at –65 and 0 mV, respectively. Scale bars: 10 pA /0.1 nS (vertical), 1 s/0.1 s (horizontal). (b) The strength of excitatory (GE = qE*fE) and inhibitory inputs (GI = qI*fI) are correlated and scaled linearly against the number of functional synapses (four neurons). (c) The ratio of E/I inputs in neurons with a variety of input strengths is constant (for 11 d.i.v. group: GI/GE = 1.35 (n = 4, R2 = 0.99, P < 0.0001); for 16 d.i.v. group: GI/GE = 2.55 (n = 7, R2 = 0.9, P < 0.01); for 18 d.i.v. group: GI/GE = 3.28 (n = 8, R2 = 0.21, P < 0.0001)). The dash lines are the linear fits to data with constraint to pass zero. 18 d.i.v. data were obtained at 33 °C. (d) Bursting EPSCs and IPSCs recorded by setting membrane potential at the EPSC/IPSC reversal potential. Top trace is the synaptic input and action potential evoked at soma under current clamp. Middle and bottom traces show synaptic current recorded at different holding potentials to record EPSCs (Vm = VI = –65 mV) and IPSCs (Vm = VE = 0 mV). At –35 mV, E/I inputs cancel one another (Vrev). The calculated GI/GE was 1.16. Average GI/GE from four cells was 1.35. Scale bars: 600 pA (vertical), 1 s (horizontal). (e) The relationship between somatic current injection and frequency of resulting action potentials. Top panels are the action potentials evoked by somatic injection of 100 pA (left) and 200 pA (right). Bottom panel is the I/F relationship.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

the picrotoxin to the proximal dendrite and somatic regions; the drug application pipette contained the fluorescent dye FM4-64 to monitor the spatial spread of solution. We confirmed the effectiveness of the somatic blockade by monitoring the size of GABAA receptor–mediated currents induced by the local application of GABA to the somatic region. The local application of picrotoxin to somata completely blocked somatic GABAA currents (Fig. 3a), suggesting the effective removal of somatic inhibitory input. When somatic inhibition was silenced, bursting IPSCs were only reduced by 20.6 ± 2.9% (n = 3). This percentage of reduction is similar to the fraction of inhibitory synapses found at somatic locations (Fig. 1d), suggesting that somatic and dendritic inhibitory synapses are activated in proportion to their availability, and the majority of inhibitory inputs during bursts originate at inhibitory synapses located on the dendritic tree. Hence, active E/I synaptic inputs might interact locally on the dendritic branches. To test this possibility, we compared somatic input conductance (Gin) in the presence and absence of bursting inputs by determining the change in membrane potential in response to a multi-step current injection (Fig. 3b). The membrane potential was recorded under current clamp using the perforated patch technique to preserve the electrotonic properties of the dendritic tree. Gin at rest was 2.8 nS, whereas Gin observed during a bursting volley was 3.7 nS. Thus, Gin at the soma only increased by ∼30% during bursting input. However, the sum of GE and GI during bursts, measured under voltage clamp, was ∼13 nS (measured from a separate set of neurons, n = 10). If all of these synaptic conductances had reached the soma, Gin would have been >15 nS. Thus, only ∼10 % of synaptic conductance induced by activation of E/I synapses at the dendrite appeared to reach the soma (n = 3), suggesting that most E/I synaptic interactions resolve within the dendritic tree. Local interactions of E/I inputs on dendritic branches To understand the biophysical features of E/I interactions, we tried to determine the temporal and spatial interactions of individual excitatory and inhibitory synaptic inputs at dendrite branches. To overcome the difficulty of precisely triggering synchronous E/I synaptic transmission at adjacent synapses, we applied the neurotransmitters glutamate and GABA locally at E/I synapses to effectively mimic the magnitude and time course of E/I synaptic transmission (Fig. 4a). Our previous studies suggest that brief (0.5 ms), highly localized transmitter delivery by iontophoresis can be used to selectively activate receptors from indi-

vidual synapses, thus mimicking endogenous synaptic transmission20,21, although interpreting the results of the present manipulation would still be valid if receptors from more than one synapse were activated. Synaptic potentials evoked by the local application of either excitatory or inhibitory neurotransmitter were recorded at the soma. We first studied the temporal interaction of E/I inputs at two adjacent E/I sites (distance less than 1 µm). The timing of glutamate receptor activation was varied from 120 ms before to 120 ms after GABA receptor activation (Fig. 4b). The size of the resultant excitatory postsynaptic potential (EPSP) was smaller when E/I inputs were stimulated concurrently than when there was a time delay between the activations. The degree of this attenuation decayed with increasing intervals between EPSP and IPSP, resulting in an attenuation time constant of ∼18 ms (Fig. 4b,c; n = 4). The precise match of this constant with IPSC time course (∼20 ms) suggests that a significant portion of EPSP attenuation by IPSPs is associated with the opening of GABAA channels. The characteristics of this inhibition are similar to the shunting inhibition predicted by modeling1. The time window of dendritic inhibition was slightly longer when an inhibitory input followed an excitatory input. We next studied the effects of space on interactions between E/I inputs. Glutamate and GABA were released simultaneously while the

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

375

ARTICLES
Figure 3 Activation of dendritic inhibitory synapses during bursts. (a) Majority of inhibitory inputs during bursts originate at inhibitory synapses located on dendritic tree. Top panel shows the experimental arrangement. Release of picrotoxin from pipette at left side was used to block GABAA receptors at somatic region (region in yellow was the area affected). The somatic GABAA receptors were activated by local application of GABA through electrode at right side. The black trace is the bursting inhibitory inputs flanked by GABA currents initiated by somatic GABA receptor activation. Red trace is the remaining bursting inhibitory inputs after blockade of somatic inhibitory inputs. Scale bars: 0.5 s (horizontal), 100 pA (vertical). (b) Change in input conductance (Gin) induced by synaptic inputs. Currents ranging –50 to –150 pA were injected, and the evoked change in membrane potentials at rest (black trace) and during bursting synaptic input (red trace) were compared. The resulting I/V relationship yields Gin at rest (2.8 nS, black spots) and during the burst periods (3.7 nS, red spots). Scale bars: 0.5 s (horizontal), 25 mV (vertical).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

distance between the glutamate and GABA delivery sites was increased in 1-µm steps (Fig. 4d). The influence of a given inhibitory input over an adjacent excitatory input was attenuated with increasing distance between the E/I synapses, at an overall space constant of ∼10 µm (Fig. 4e,f). Notably, this attenuation did not seem to depend on the relative position of E/I synapses, as inhibitory synapses were capable of attenuating more proximal excitatory inputs. In contrast, when an IPSP was evoked at a different, but adjacent, dendritic branch, the IPSP had a substantially smaller impact on EPSP size, indicating that the influence of inhibitory synapses is local and largely

limited to individual dendritic branches. Such results also support aforementioned models of dendritic shunting1. Overall, it seems that E/I interactions occur locally within a dendritic branch, whereas the uniform spread of E/I synapses discussed earlier may provide a structural arrangement that makes these local interactions possible.
A rule for E/I organization in hippocampal dendrites We began to search for a simple rule to relate our observations of the uniform distribution and balance of E/I synapses throughout the dendritic tree (Figs. 1 and 2) to the localized nature of dendritic inhibition

Figure 4 Spatial and temporal interactions between E/I inputs. (a) Experimental set-up. Potentials evoked by local application of glutamate or GABA were recorded at the cell soma, when relative distance (XE to XI) and timing (TE to TI) of glutamate and GABA release were systematically varied. (b) Temporal interactions between elicited EPSPs and IPSPs. IPSC from the same synaptic site was superimposed to show the relationship between the time course of GABAA receptor opening and the time window of EPSP attenuation (blue, Vm = –90 mV). Scale bars: 2.5 mV (vertical), 50 ms (horizontal). (c) Time window of E/I interaction. Highest attenuation occurs when E/I inputs coincide, and attenuation decays rapidly with a time constant of ∼20 ms. (E + I)/E is the ratio of EPSP + IPSP integral over EPSP integral as measured at the soma. The dash lines are the exponential fits to the data with time constants of 25 (left) and 18 ms (right). (d) Spatial interactions of EPSPs and IPSPs. XI and XE were the locations of GABA and glutamate delivery. Scale bar: 2 µm. (e) Attenuation of EPSPs by concurrent IPSPs occurs only when both E/I synapses are located within the same dendritic branch (black traces: EPSP alone; blue traces: EPSP + IPSP). The size of IPSPs in XI(0) and XI(1) were similar (data not shown). Scale bars: 5 mV (vertical), 100 ms (horizontal). (f) Spatial relationship between the distance of E/I synapses (from soma) and the efficacy of inhibition. Solid dots are percent of inhibition when both synapses on the same dendritic branch. (1 – (E + I)/E). Open circles represent cases where E/I synapses on separate branches were activated.

376

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 5 Feedback regulation of E/I synapses after perturbation of E/I balance. (a) Working model of E/I organization and regulation. (b) Top panels, reducing or enhancing GABAA receptor–mediated inhibition by the antagonist bicuculline (bicu) or potentiator flunitrazepam (fluni). Bottom panels are the compensatory responses of E/I synapses to perturbations. Scale bars: 500/20 pA (vertical), 0.5 s (horizontal). (c) Push-pull regulation of E/I synaptic strength in response to perturbation of E/I balance. Number of neurons in each treatment: 30 (ctrl); 14 (NBQX); 14 (fluni); 14 (bicu); 8 (TTX) (P < 0.01 for all treatments, t-test). (d) The effects of action potential blockade on the E/I balance. (e) Gin increases in compensation for the increase in the functional E/I ratio.

b
Perturbation
EPSCs

Ctrl

Bicu (↓I)

Fluni (↑I)

a


GE

GR

GI

+
VS



+
Response

IPSCs mEPSCs

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

mIPSCs

c 15
GE (nS/s)
10

Excitatory Inhibitory

d
60 40

15 60

e
6 40 4

GE (nS/s)

10

5 5 2 20 20 (Fig. 4). As the interaction of E/I inputs, combined with the resting dendritic conductance (GR), determines the size of the compound 0 0 0 0 0 synaptic potential (VS) on each dendrite, E/I synapses on a dendritic tree might be organ↑E/I ratio during treatment ized to maintain a constant VS within individual dendritic branches. To maintain a proper level of VS, any perturbation to VS would necessitate feedback regulation of GE and GI (Fig. 5a) simultaneously. It enhancement of inhibitory inputs. To quantify these results, we calcuis likely that the robust feedback regulation required under this formula- lated the total GE and GI from the sum of mEPSCs and mIPSCs from tion to maintain a suitable level of excitatory inputs takes many somatic recording (Methods) and plotted them by treatment (Fig. 5c). forms22,23. Candidates include the inverse correlation between excita- In response to the imbalance, E/I synapses adjusted their synaptic tory synaptic density on the dendritic tree and quantal size24, the strength in the opposite direction (Fig. 5c), exhibiting push-pull regulaincrease in quantal size25 and probability of release26 after action poten- tion. The GE and GI responses matched the predictions of our model tial blockade, and the increase of presynaptic transmitter release after (Fig. 5a). We also tested the effects of action potential blockade on E/I postsynaptic potassium channel overexpression27,28. This model also balance (Fig. 5d). Although GE increased slightly, the major effect of adds two predictions: (i) in addition to excitatory inputs, inhibitory neural activity blockade was a dramatic reduction of inhibitory synaptic inputs and resting dendritic conductance are critical in controlling VS transmission, similar to results reported previously29. This result sugand (ii) the balance of E/I synapses within the dendritic tree may be nec- gests that a reduction of inhibitory inputs might be the primary mechaessary for tuning VS to an appropriate level. Furthermore, it predicts that nism of compensating for the loss of excitatory synaptic inputs. It is the absolute level of excitatory inputs should be less critical in control- worth mentioning that because the drugs were applied to the culture ling VS than the balance of E/I inputs (Fig. 2d,e). If our assumptions are medium, the E/I balances of all dendritic branches were perturbed univalid, the strengths of E/I synapses and resting membrane conductance formly. The post-treatment changes in GE and GI represent the averaged should be tuned in a coordinated fashion, with any offset in one direc- responses from synapses in the entire dendritic tree. Thus, these results tion resulting in a restorative change in the other (Fig. 5a). do not provide direct evidence that the various regulatory responses occur in a dendritic branch–specific fashion. To test whether E/I balance Maintaining the balance of E/I input strength can be regulated in a dendrite specific fashion, new experimental To verify this model, we pharmacologically modified the efficacy of approaches need to be developed that allow perturbation and recording either excitatory or inhibitory inputs, and then assessed the extent to of E/I balance at individual dendritic branches. which the neurons compensated for this perturbation by adjusting their We also attempted to determine whether the modification of the E/I efficacies to restore the original balance. The GABAA receptor antag- size of GR is coordinated to maintain VS at a suitable level. Although onist bicuculline was used to block inhibition, and the GABAA receptor it would be desirable to measure changes in GR across treatments, potentiator flunitrazepam was used to prolong the response of the the size of GR at or near individual synapses is difficult to obtain. GABAA receptors, thus enhancing functional inhibition. Similarly, Instead, we determined the change in Gin resulting from each treatNBQX was used to block the binding of glutamate to AMPA receptors, ment to infer the relative strength of GR. Notably, Gin also changed resulting in the elimination of excitation. NBQX and flunitrazepam after treatments, following the same trend as GI (Fig. 5e). However, served to reduce VS, whereas bicuculline had the opposite effect. We the magnitude of change of Gin is much smaller than that of GI, sugrecorded EPSCs and IPSCs under the various drugs used to perturb the gesting that synaptic sites, rather than regulation of dendritic leak functional E/I balance (Fig. 5b, top) and the feedback responses of E/I channels, for example, are the primary regulators of VS. synapses to the perturbations over 48 h (Fig. 5b, bottom). Removing inhibition led to a reduction of excitatory synaptic strength and an DISCUSSION enhancement of inhibitory synaptic strength. The opposite effects were Our results show that E/I synapses were distributed evenly along observed after a reduction of excitatory inputs or, conversely, an dendritic branches, that the functional ratio of E/I inputs was deli-

u

NB Q X

TT X

X

un

tr

tr

tr

ic

Q

C

C

C

B

Fl

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

N

B

B

ic

u

i

l

l

l

GI (nS/s)

GI (nS/s)

GIN (nS)

377

ARTICLES
cately balanced to evenly distribute input along the dendritic surface, and that this E/I balance was actively maintained by push-pull regulatory mechanisms (Fig. 5a). The consequence of this arrangement is that the compound synaptic potential (VS) at individual dendritic branches remains constant, although the exact level of the VS might be different during early stages of development. Since the amount of synaptic input that impacts processing in the soma depends on VS, the ratio of E/I inputs, in conjunction with the number of voltage-dependent and independent channels within a dendrite, ultimately controls neural output. As the firing of a neuron is determined by the dynamic interplay of E/I inputs (Fig. 2d), the constant VS might be the underlying mechanism by which a neuron reduces fluctuations in its firing rate when the level of synaptic input varies25. This new view emphasizes the significant role of inhibitory synapses within dendrites in shaping synaptic integration, and the necessity of E/I balance for neural computation. This idea is further supported by recent studies in vivo. In the barrel cortex of intact rats, for example, when an increase in the number of excitatory synapses is elicited by whisker stimulation, inhibitory synapses are also added to preserve the 4/1 ratio of E/I synapses observed across the dendritic trees of this system30. Inversely, reduction of synaptic input in the visual cortex in lightdeprived animals prevents an increase in both excitatory synaptic strength31 and GABAergic innervation32, thus maintaining the E/I balance when synaptic inputs are reduced. Finally, experimental animal models of temporal lobe epilepsy often involve impaired dendritic inhibition33, suggesting the significance of the E/I balance in preserving the functionality of neural networks. This local regulatory mechanism might act alone or be complementary to an alternative mechanism that relies on action potential back-propagation to normalize the efficacy of dendritic synapses23,34. The presence of local and subthreshold regulatory mechanisms could be important for two reasons: (i) as the back propagation of an action potential cannot reach distal dendritic branches35, not all synapses on the dendritic tree can be normalized by action potential frequency and (ii) for a neuron with a very low firing rate, the number of action potentials might not be the most appropriate parameter for the adjustment of its overall synaptic strength. For example, only 1–2% of CA1 hippocampal neurons fire action potentials during behavior. The average firing rate of these neurons is about 1 Hz36. The latest results from whole-cell patch-clamp recording in awake animals show that the spontaneous firing rate of barrel cortex neurons is only 0.05 Hz, 50-fold less than what has been reported by extracellular unit recordings37. The few milliseconds of membrane depolarization over a 20-s time window might not contain as much information for representation of overall synaptic inputs as do continuous subthreshold synaptic inputs. Further experiments are needed to test whether the local voltage change is sufficient to trigger adjustment of the strength of E/I synapses. Another important component of the present study is the evidence of the localized nature of E/I interactions on dendrites. The efficacy of inhibitory inputs appears to be locally restricted and critically dependent on their spatial and temporal proximity to excitatory inputs (on the same dendritic branch, within ∼20 ms). The local E/I interactions that result are computationally equivalent to the ‘veto’ operation that is thought to occur in dendritic computation1. The uniform spread of E/I synapses could well provide a structural basis for distributing this type of computation over entire dendrite branches. Two elegant studies of direction selectivity in retinal ganglion cells reveal biological evidence for such computation4,6. E/I

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

synapses on dendrites are activated simultaneously, but the degree of their relative activation depends on the direction of movement of visual stimuli. This E/I interaction occurs on dendrites. Similar interactions between E/I synaptic inputs are also observed in neurons of various regions of cortex7,8,38,39. Thus, the ratio of E/I inputs might constitute an elementary unit for neural computation across multiple regions and systems. If this is the case, the activity of inhibitory synapses, both in their intensity and precise timing, must be ‘tuned’ properly to sponsor meaningful patterns of neural activity. This possibility is further supported by recent studies in visual system development showing that inhibitory and excitatory synapses are modified in conjunction40. These cases offer evidence from across the nervous system that E/I interactions in dendrites might be an important form of neural computation. Our results provide new biophysical details of the balance of E/I synapses, how this balance is established throughout the dendritic tree and the regulatory principles that maintain it. This synaptic arrangement instantiated within the dendrite provides a functional scaffold over which to conduct neuronal computation.
METHODS
Whole-cell patch-clamp recording. The procedures for culturing hippocampal neurons, FM1-43 dye staining and whole-cell recording were essentially as described previously21. Whole-cell recordings were obtained at room temperature (∼22 °C) from spindle-shaped pyramidal neurons kept in culture for 11–20 d. The small experimental chamber (0.25 ml) was continuously perfused (0.25 ml min–1) with Tyrode solution containing 145 mM NaCl, 3 mM KCl, 2.6 mM CaCl2, 1.3 mM MgCl2, 10 mM glucose and 10 mM HEPES (pH 7.4 with NaOH). For mEPSC and mIPSC recording from the same cell, the intracellular solution contained 120 mM CsMeSO3, 1 mM CaCl2, 10 mM NaCl, 10 mM EGTA, 2 mM Mg-ATP, 0.3 mM Na-GTP and 10 mM HEPES (pH 7.25 with CsOH). TTX (1 µM) was added to extracellular solution for recording mEPSCs and mIPSCs. We found that damage during the preparation for whole-cell recording could induce a high frequency of mEPSCs and mIPSCs, leading to inaccuracy in the calculated synaptic strength. Therefore, only recordings with low access resistance (< 10 MΩ) and holding current less than –60 pA (Vm = –70 mV) were analyzed. As pyramidal cells and interneurons differ substantially in their synaptic properties and their responses to stimulation that trigger synaptic plasticity, only pyramidal neurons were included in this study. Pyramidal neurons were selected based on the following selection criteria: pyramidal shape of soma with basal and apical dendrites, linear I/V relationship for the AMPA component of EPSCs, and presence of NMDA receptors. For all experiments carried out under current-clamp (Figs. 2d,e and 3b), perforated patch clamp was used to obtain a long-lasting recording with minimum perturbation of the intracellular environment. The patch pipette (1.2–2.2 MΩ resistance) contained 130 mM potassium gluconate, 2 mM KCl, 8 mM NaCl, 0.2 mM EGTA, 1 mM MgCl2 and 10 mM HEPES (pH 7.2). Under these conditions, the normal cable properties of the dendritic tree are undisturbed. This allowed us to monitor the interplay of E/I synaptic inputs under physiological conditions. Calculation of the overall strength of synaptic inputs (GE and GI). GE and GI are determined by

GE =

Σ
0

t

EPSCs

/

t(Vm – VE) and GI =

ΣIPSCs / t(V
0 t

t

m – VI)

during bursting periods, or

GE =

Σ /
0

t

qE

t(Vm – VE) and GI =

Σ q / t(V
0 I

m – VI)

during miniature synaptic events, where q is the charge transfer during quantal transmission. The following formula was used in calculating the ratio of E/I input strength (GI/GE) during bursting synaptic inputs. At the reversal poten-

378

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
tial for bursting inputs, excitatory and inhibitory inputs cancel each other: GE(Vrev – VE) + GI(Vrev – VI) = 0. Thus, the ratio of their conductances can be calculated by GI/GE = (Vrev – VE)/(VI – Vrev). Given the intracellular and extracellular solutions used in the present study, VE will be close to 0 and VI close to –65 mV. Because the kinetic properties of synaptic receptors and synaptic release probability are temperature-sensitive, the ratio of GI/GE determined at room temperature might differ markedly from the ratio at physiological temperature. We have compared the ratios of GI/GE at 22 and 33 °C (Fig. 2c). Raising the temperature in the recording chamber increased the amplitude and sped up the decay of synaptic events. However, the integral of synaptic conductance did not change markedly, resulting in a similar E/I ratio measured at room temperature. Finally, to get an accurate measure of GE and GI, all active synapses were voltage-clamped. Evaluation of the degree of space clamp. Two criteria were used. First, if synaptic events had been attenuated due to poor space clamping, the rise time of synaptic events originating at synapses located at distant locations on the dendritic tree would have been much slower than those from proximal locations. The median values of 20–80% rise time of all of our recordings are <0.5 ms, suggesting that those synapses that generated miniature synaptic events were measured under a reasonable degree of voltage clamp (Supplementary Fig. 1 online, panel a). This analysis could not rule out the possibility that some distant synapses are electronically so far away from somatic recording sites that their synaptic currents are attenuated to the point that they could not be detected. To exclude this possibility, we compared the reversal potential and kinetics of the synaptic currents evoked at various locations on the dendritic tree (Supplementary Fig. 1 online, panels b–d). Neither reversal potentials nor time courses of synaptic current changed significantly with increased distance from the soma, demonstrating that synaptic conductances from distal synapses can be detected at a somatic recording site. As the area of synaptic conductance, rather than peak amplitude, is used to represent quantal synaptic strength, errors associated with inadequate space clamp are further reduced. Perturbation of E/I balance by pharmacological agents. For experiments in Figure 5, neurons were treated with various pharmacological agents (flunitrazepam (Sigma), NBQX (Tocris), bicuculline (Tocris) and TTX (Biotium)) for 48 h before recording. We determined the time course for the restorative E/I response to perturbations. Generally, it took 12 h of treatments to alter E/I synapses, and the responses stabilized after 24 h (data not shown). Synaptic strengths (GE and GI) were determined from mEPSCs and mIPSCs, respectively. Gin was calculated by plotting membrane potential as a function of injected current size.
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS I thank E. Hueske and B. Li for participating in part of the experiments and contributing to Figure 1; M. Wilson, X.J. Wang, N. Wilson, S. Sadeghpour, B. Krupa and T. Emery for comments on the manuscript. This work was supported by grants from National Institutes of Health and the RIKEN–MIT Neuroscience Research Center. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 1 December 2003; accepted 12 February 2004 Published online at http://www.nature.com/natureneuroscience/
1. Koch, C., Poggio, T. & Torre, V. Nonlinear interactions in a dendritic tree: localization, timing, and role in information processing. Proc. Natl. Acad. Sci. USA 80, 2799–2802 (1983). 2. Bush, P.C. & Sejnowski, T.J. Effects of inhibition and dendritic saturation in simulated neocortical pyramidal cells. J. Neurophysiol. 71, 2183–2193 (1994). 3. Ferster, D. & Miller, K.D. Neural mechanisms of orientation selectivity in the visual cortex. Annu. Rev. Neurosci. 23, 441–471 (2000). 4. Fried, S.I., Munch, T.A. & Werblin, F.S. Mechanisms and circuitry underlying directional selectivity in the retina. Nature 420, 411–414 (2002). 5. Gabbiani, F., Krapp, H.G., Koch, C. & Laurent, G. Multiplicative computation in a visual neuron sensitive to looming. Nature 420, 320–324 (2002). 6. Taylor, W.R., He, S., Levick, W.R. & Vaney, D.I. Dendritic computation of direction selectivity by retinal ganglion cells. Science 289, 2347–2350 (2000). 7. Schummers, J., Marino, J. & Sur, M. Synaptic integration by V1 neurons depends on location within the orientation map. Neuron 36, 969–978 (2002). 8. Shu, Y., Hasenstaub, A. & McCormick, D.A. Turning on and off recurrent balanced cortical activity. Nature 423, 288–293 (2003). 9. Chance, F.S., Abbott, L.F. & Reyes, A.D. Gain modulation from background synaptic input. Neuron 35, 773–782 (2002). 10. Beaulieu, C., Kisvarday, Z., Somogyi, P., Cynader, M. & Cowey, A. Quantitative distribution of GABA-immunopositive and -immunonegative neurons and synapses in the monkey striate cortex (area 17). Cereb. Cortex 2, 295–309 (1992). 11. Beaulieu, C. & Somogyi, P. Targets and quantitative distribution of GABAergic synapses in the visual cortex of the cat. Eur. J. Neurosci. 2, 296–303 (1990). 12. Megias, M., Emri, Z., Freund, T.F. & Gulyas, A.I. Total number and distribution of inhibitory and excitatory synapses on hippocampal CA1 pyramidal cells. Neuroscience 102, 527–540 (2001). 13. Miles, R., Toth, K., Gulyas, A.I., Hajos, N. & Freund, T.F. Differences between somatic and dendritic inhibition in the hippocampus. Neuron 16, 815–823 (1996). 14. McBain, C.J. & Fisahn, A. Interneurons unbound. Nat. Rev. Neurosci. 2, 11–23 (2001). 15. Hausser, M., Spruston, N. & Stuart, G.J. Diversity and dynamics of dendritic signaling. Science 290, 739–744 (2000). 16. Poirazi, P. & Mel, B.W. Impact of active dendrites and structural plasticity on the memory capacity of neural tissue. Neuron 29, 779–796 (2001). 17. Larkman, A.U. Dendritic morphology of pyramidal neurones of the visual cortex of the rat: III. Spine distributions. J. Comp. Neurol. 306, 332–343 (1991). 18. Ben-Ari, Y., Cherubini, E., Corradetti, R. & Gaiarsa, J.L. Giant synaptic potentials in immature rat CA3 hippocampal neurones. J. Physiol. 416, 303–325 (1989). 19. Leinekugel, X. et al. Correlated bursts of activity in the neonatal hippocampus in vivo. Science 296, 2049–2052 (2002). 20. Murnick, J. G., Dube, G., Krupa, B. & Liu, G. High-resolution iontophoresis for single-synapse stimulation. J. Neurosci. Methods 116, 65–75 (2002). 21. Liu, G., Choi, S. & Tsien, R. W. Variability of neurotransmitter concentration and nonsaturation of postsynaptic AMPA receptors at synapses in hippocampal cultures and slices. Neuron 22, 395–409 (1999). 22. Davis, G.W. & Goodman, C.S. Genetic analysis of synaptic development and plasticity: homeostatic regulation of synaptic efficacy. Curr. Opin. Neurobiol. 8, 149–156 (1998). 23. Turrigiano, G.G. & Nelson, S. B. Hebb and homeostasis in neuronal plasticity. Curr. Opin. Neurobiol. 10, 358–364 (2000). 24. Liu, G. & Tsien, R.W. Properties of synaptic transmission at single hippocampal synaptic boutons. Nature 375, 404–408 (1995). 25. Turrigiano, G.G., Leslie, K.R., Desai, N.S., Rutherford, L.C. & Nelson, S.B. Activitydependent scaling of quantal amplitude in neocortical neurons. Nature 391, 892–896 (1998). 26. Murthy, V.N., Schikorski, T., Stevens, C.F. & Zhu, Y. Inactivity produces increases in neurotransmitter release and synapse size. Neuron 32, 673–682 (2001). 27. Paradis, S., Sweeney, S.T. & Davis, G.W. Homeostatic control of presynaptic release is triggered by postsynaptic membrane depolarization. Neuron 30, 737–749 (2001). 28. Burrone, J., O’Byrne, M. & Murthy, V.N. Multiple forms of synaptic plasticity triggered by selective suppression of activity in individual neurons. Nature 420, 414–418 (2002). 29. Kilman, V., van Rossum, M.C. & Turrigiano, G.G. Activity deprivation reduces miniature IPSC amplitude by decreasing the number of postsynaptic GABAA receptors clustered at neocortical synapses. J. Neurosci. 22, 1328–1337 (2002). 30. Knott, G.W., Quairiaux, C., Genoud, C. & Welker, E. Formation of dendritic spines with GABAergic synapses induced by whisker stimulation in adult mice. Neuron 34, 265–273 (2002). 31. 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). 32. Morales, B., Choi, S.Y. & Kirkwood, A. Dark rearing alters the development of GABAergic transmission in visual cortex. J. Neurosci. 22, 8084–8090 (2002). 33. Cossart, R. et al. Dendritic but not somatic GABAergic inhibition is decreased in experimental epilepsy. Nat. Neurosci. 4, 52–62 (2001). 34. Rumsey, C.C. & Abbott, L.F. Equalization of synaptic efficacy by activity- and timing-dependent synaptic plasticity. J. Neurophysiol. (2003). 35. Johnston, D. et al. Active dendrites, potassium channels and synaptic plasticity. Philos. Trans. R. Soc. Lond. B Biol. Sci. 358, 667–674 (2003). 36. Wilson, M.A. & McNaughton, B.L. Dynamics of the hippocampal ensemble code for space. Science 261, 1055–1058 (1993). 37. 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). 38. 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). 39. 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). 40. Hensch, T.K. et al. Local GABA circuit control of experience-dependent plasticity in developing visual cortex. Science 282, 1504–1508 (1998).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

379

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Synaptic dynamics mediate sensitivity to motion independent of stimulus details
Harald Luksch1, Reza Khanbabaie2 & Ralf Wessel2
Humans and other animals generally perceive motion independently of the cues that define the moving object. To understand the underlying mechanisms of this generalization of stimulus attributes, we have examined the cellular properties of avian wide-field tectal neurons that are sensitive to a variety of moving stimuli but not to static stationary stimuli. This in vitro study showed phasic signal transfer at the retinotectal synapse and binary dendritic responses to synaptic inputs that interact in a mutually exclusive manner in the postsynaptic tectal neuron. A model of the tectal circuitry predicts that these two cellular properties mediate sensitivity to a wide range of dynamic spatiotemporal stimuli, including moving stimuli, but not to static stationary stimuli in a tectal neuron. The computation that is independent of stimulus detail is initiated by tectal neurons and is completed by rotundal neurons that integrate outputs from multiple tectal neurons in a directionally selective manner.

Objects in our visual environment are normally distinguishable by a number of cues such as luminance, shape, color and texture, yet the percept of motion is qualitatively invariant across different cues1,2. This perceptual invariance, known as form-cue invariance, is paralleled by a similar response invariance of many apparently motionsensitive neurons in primate cortical middle temporal visual area3,4, in primate cortical dorsal middle superior temporal area5 and in avian tectum6–8. Form-cue invariance may be a useful operational principle. To perform optimally in a variable environment, a motion analyzer should generalize across stimulus cues, thus encoding the motion of a stimulus regardless of the cue that enables it to be seen. Despite the apparent importance of form cue–invariant motion processing, the underlying cellular mechanisms for form-cue invariance are poorly understood. The avian tectal slice (Fig. 1a) provides an ideal preparation to study the central cellular mechanisms for the analysis of dynamic spatiotemporal stimuli9. Neurons in deep layers of the avian tectum respond to small moving objects largely independently of object-defining attributes, but they do not respond to static stationary stimuli6,8,10–12. Deep tectal neurons of the morphologically and physiologically identified stratum griseum centrale (SGC-I) type (Fig. 1b) are particularly suitable for cellular studies of spatiotemporal processing. Neurons of this apparently motion-sensitive subpopulation have somata in layer 13, have large circular dendritic fields, extend their dendrites radially and terminate with specialized dendritic endings in layer 5b9,13. Here the SGC-I dendritic endings make monosynaptic contact with axons14 that are derived from a population of small retinal ganglion cells (RGCs)15. These RGC axons form a topographic map on the tectal surface and penetrate the outer tectal layers radially16. The fact that a tectal neuron, which is only one synapse away from the retina, apparently displays form cue–invariant motion sensitivity

suggested to us the hypothesis that form cue–invariant motion sensitivity could be mediated by cellular rather than network mechanisms. To explore this hypothesis, we obtained in vitro whole-cell recordings from SGC-I type neurons in combination with a series of spatiotemporal stimulation experiments with stimulus electrodes at different locations in the retinotectal signal pathway. We then analyzed the functional role of the measured cellular properties with a computer simulation of an SGC-I model in response to assumed retinal representations of various dynamic spatiotemporal visual stimuli. RESULTS SGC-I response to localized synaptic stimulation To investigate the signal transfer from the RGC axon to the SGC-I soma, we locally stimulated a small group of RGC axons with short current pulses that were delivered with a stimulus electrode in layers 2–4 and recorded the response in the SGC-I soma (Fig. 1a,b). In all SGC-I cells tested, single-pulse stimulation resulted in an all-or-none sharp-onset cellular response consisting of either one to three action potentials riding on a broader depolarization or no response (Fig. 1c). A previous investigation9 indicates that the sharp-onset response to synaptic stimulation may be generated remotely from the soma, presumably at the dendritic ending. To characterize this response, we carried out one-site regular pulse train synaptic stimulation experiments (Fig. 2a). The sharp-onset response to each stimulus pulse was probabilistic. For all stimulation intervals tested, the response probability reached a steady state after the second stimulus pulse (Fig. 2b). Therefore, we pooled the data from pulses 2–10 for the same stimulation interval to derive the mean steady-state response probability for that stimulation interval (Fig. 2c). This probability showed an exponential time dependence of the form P(∆t) = Pmax (1 – e–∆t/t0), where ∆t is the stimulation interval

1Institute

of Biology II, Rheinisch-Westfälische Technische Hochschule Aachen, 52074 Aachen, Germany. 2Department of Physics, CB 1105, Washington University, St. Louis, Missouri 63130, USA. Correspondence should be addressed to R.W. ([email protected]). Published online 29 February 2004; doi:10.1038/nn1204

380

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 1 SGC-I morphology, location of stimulus electrodes and response to synaptic stimulation. (a) Schematic view of the tectum slice and the position of the SGC. Cer, cerebellum. (b) Reconstruction of an SGC-I neuron labeled with biocytin after whole-cell patch recording. The characteristics of this cell type include a large dendritic field, the position of the soma in the upper half of the SGC and the arrangement of the bottlebrush dendritic endings in the retinorecipient layer 5b, as indicated by the shaded layer. Note the positioning of the stimulation electrodes above and within the retinorecipient layer 5b. For clarity, only a small subset of RGC axon schematics are shown. (c) Single-pulse stimulation results in either a sharponset response consisting of one to three action potentials riding on a broader depolarization (main graph) or no response at all (inset). The resting membrane potential is indicated below the recording trace.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

and Pmax = 0.87 and t0 = 2,025 ms for the two fitting parameters. Thus, the signal transfer from the RGC axon to the SGC-I soma was phasic in a time-dependent probabilistic manner. The phasic response could have originated either at the synapse or in the dendritic pathway and soma. To localize the site of phasic signal transfer, we stimulated dendritic endings directly with a stimulus electrode in layer 5, thus bypassing the synapse (Fig. 1b). This direct electrical stimulation of dendritic endings led to a sharp-onset response that was essentially identical to the response to synaptic stimulation but with a shorter latency, as described previously9. We carried out one-site paired-pulse direct electrical stimulation of dendritic endings (Fig. 2d) and measured the SGC-I response probability for the second pulse for different stimulation intervals (Fig. 2f). The probability shows an exponential time dependence of the form Pdirect (∆t) = Pmax2 (1 – e (t1 – ∆t)/t2) , where ∆t is the stimulation interval and the fitting parameters are Pmax2 = 1.0, t1 = 4 ms and t2 = 12 ms. We refer to the sum of the time shift, t1, and the exponential time constant, t2, as τdirect = t1 + t2 = 16 ms. In conclusion, signal transfer within SGC-I neurons is tonic at time scales that are two orders of magnitude shorter than signal transfer at the retinotectal synapse. Therefore, the phasic signal transfer (Fig. 2c) originates at the retinotectal synapse. To investigate the cell’s response to spatiotemporal synaptic inputs and to test for potential distance dependence of the interaction, we placed two stimulus electrodes in layers 2–4 at distances of 250–1,500 µm apart, thus stimulating two separate groups of RGC axons (Fig. 2e). Because of the sparse spatial distribution of dendritic endings for one SGC-I neuron, each group of stimulated RGC axons typically activated only one dendritic ending of an SGC-I neuron under consideration9. We recorded from an SGC-I cell that received inputs from both groups of axons and stimulated the two sites in temporal sequence with varying stimulation intervals. The measured response probabilities to the second stimulus pulse for one stimulation interval showed no statistically significant distance dependence for the two-site synaptic stimulation (Fig. 2f). Reversing the sequence in which the sites were stimulated had no effect on the response (data not shown). Thus we pooled the data from the twosite synaptic stimulation from different distances to derive the mean

response probability to the second pulse for each stimulation interval (Fig. 2f). The probability shows an exponential time dependence of the form P2-site (∆t) = Pmax3 (1 – e (t3 – ∆t)/t4), where ∆t is the stimulation interval and the fitting parameters are Pmax3 = 1.0, t3 = 16 ms and t4 = 14 ms. We refer to the sum of the time shift, t3, and the exponential time constant, t4, as the ‘interaction time’, τ = t3 + t4 = 30 ms. Although the exponential time constants for the two stimulus situations are similar, there seems to be a difference in the time shift for direct stimulation at one site, t1 = 4 ms, and for sequential synaptic stimulation at two sites, t3 = 16 ms (Fig. 2f). At present, it is not known whether this difference in time shift is mediated by the retinotectal synaptic latency of 8 ms (ref. 9), by intracellular dendritic events, by the presynaptic horizontal network in layer 5 (refs. 14,17,18) or by combinations thereof. In summary, this series of experiments showed a number of nonlinear cellular properties: (i) the SGC-I cell responded to synaptic stimulation in a binary manner; (ii) the signal transfer from the RGC axon to the SGC-I soma was phasic in a time-dependent probabilistic manner over large time scales; (iii) the site of the phasic signal transfer was the retinotectal synapse; (iv) the phasic synaptic signal transfer largely reached a steady state after the second stimulus pulse; (v) synaptic inputs at two locations typically interacted in a mutually exclusive manner when delivered within the interaction time of approximately 30 ms, a time scale that is two orders of magnitude shorter than the time scale for phasic transfer of synaptic signals and (vi) the postsynaptic interaction was independent of the stimulus electrode distances for the distances examined. Functional role of the nonlinear cellular properties To examine the role of the observed nonlinear cellular properties in shaping the response of SGC-I neurons to assumed retinal representations of dynamic spatiotemporal visual stimuli, we constructed a model of the retinal inputs and the SGC-I cell (Fig. 3). Although a number of mechanisms may contribute to the tectal analysis of dynamic spatiotemporal visual stimuli, in this model we focused exclusively on the observed binary response, the phasic and probabilistic signal transfer and the mutually exclusive response to multiple inputs within the interaction time. This SGC-I model allowed us to determine the limits of what the observed cellular nonlinear properties can explain and thereby establish whether they are important elements in the tectal analysis of assumed retinal representations of dynamic spatiotemporal visual stimuli. First, we investigated the model’s response to the assumed retinal representation (Fig. 4a) of a static stationary luminance-defined bar stimulus (Fig. 3c). An RGC spike arriving at a dendritic ending caused a dendritic spike with probability P(∆t) = Pmax (1 – e–∆t/t0). For physiological RGC spike rates, the time interval, ∆t, between two RGC spikes was typically much smaller than the measured parameter t0 =

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

381

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 2 Response of SGC-I neurons to synaptic and direct stimulation of dendritic endings. (a) Pulse train synaptic stimulation at one location. Top trace: typical response to one trial of ten pulses. The resting membrane potential was –53 mV. Center traces: response raster plot of the same neuron for ten trials. Each tick mark indicates the occurrence of a sharp-onset response. Bottom trace: Pulse train of stimulus pulses delivered to one location in layers 2–4. The stimulation interval was 1,500 ms. (b) Response probability as a function of pulse number for various stimulation intervals. The response probability drops significantly from pulse 1 to pulse 2, but for pulses 2–10 the response probability remains largely constant. (c) Mean response probability to pulse numbers 2 to 10 versus the interval between pulses at one location. The line was obtained by fitting the data points with an exponential function (see text). (d) Direct paired-pulse stimulation of dendritic endings at one location for two different time intervals between stimulus pulses. The response to the second stimulus pulse failed at a stimulus interval of 15 ms. Note the different time scale in a and d. (e) Paired-pulse synaptic stimulation at two locations for two different time intervals between stimulus pulses. The response to the second stimulus pulse failed at a stimulus interval of 15 ms. Note the different time scale in a and e. (f) Response probability to the second pulse versus stimulation interval for paired-pulse direct stimulation at one location (open circle) and paired-pulse synaptic stimulation at two locations (filled square). The lines were obtained by fitting the data points with an exponential function (see text). Inset: response probability to the second pulse versus stimulus electrode distance for paired-pulse synaptic stimulation at two locations for different stimulation intervals.

2,025 ms. Therefore, P(∆t) was small, leading to sparse dendritic spiking (Fig. 4b), which in turn led to sparse SGC-I soma spiking (Fig. 4c). Because of the sparse dendritic spiking, there was typically no dendritic spike interaction. On average, the simulated SGC-I firing rate in response to the stationary bar stimulus was 1.5 ± 0.5 Hz (mean ± s.d.; n = 5 trials of 3-s duration). To evaluate the functional role of the phasic retinotectal synaptic signal transfer, we repeated the simulation with a stationary bar but now with a constant probability, P(∆t) = Pmax (without phasic signal transfer). In this case, more than one dendritic ending generated spikes simultaneously at every time step and, as a result of the nonlinear spike interaction, which was quantified with the interaction time of 30 ms, the SGC-I model responded with a regular rate of 33.3 Hz to a stationary bar (Fig. 4d). The model response to the assumed retinal representation of a moving luminance-defined bar stimulus (Fig. 3c) was completely different. With each time step, the assumed retinal representation (Fig. 4e) of the moving stimulus bar passed over a small number of new dendritic endings. These new dendritic endings initially had a large probability, P(∆t → ∞) = Pmax, of responding with a dendritic spike to the arrival of an RGC spike. Therefore, the sum of dendritic spikes per simulation time step was often one (Fig. 4f). When more than one dendritic ending spiked simultaneously, only one spike was generated in the SGC-I soma and only if the SGC-I had not previ-

ously spiked within the interaction time of 30 ms. As a result of this nonlinear interaction of dendritic spikes, the SGC-I spike train was more regular (Fig. 4g) than the spike train of dendritic endings. On average, the SGC-I firing rate in response to the moving bar stimulus was 18.7 ± 1.1 Hz (mean ± s.d.; n = 5 trials of 3-s duration). Without the phasic synaptic signal transfer, P(∆t) = Pmax, the SGC-I model responded to a moving bar with a rate of 32.6 ± 0.3 Hz (mean ± s.d.; n = 5 trials of 3-s duration; Fig. 4h). The response of the SGC-I model was not limited to moving stimuli alone; other dynamic spatiotemporal stimuli caused strong responses. For instance, the same bar jumping to random locations within the receptive field every 10 ms resulted in a model SGC-I rate of 29.5 ± 0.6 Hz (mean ± s.d.; n = 5 trials of 3-s duration). In conclusion, the SGC-I firing rate in response to a dynamic luminance-defined spatiotemporal stimulus was significantly larger than the firing rate in response to a stationary bar. Thus, the SGC-I model was able to differentiate between a static stationary and a dynamic spatiotemporal stimulus such as a moving small stimulus bar. In contrast, without phasic synaptic signal transfer, the model SGC-I cell responded strongly to both static stationary stimuli and dynamic spatiotemporal stimuli and thus failed to distinguish between these two stimulus classes. This result demonstrates that the phasic signal transfer is a crucial biophysical element for this stimulus classification.

382

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 3 Structure of the SGC-I model and the visual stimuli. (a) Schematic of the model structure in cross-section. A visual stimulus (black square) activates RGC axons. A fraction of the RGC axons (black vertical arrows) terminate on the dendritic endings of one SGC-I neuron. The remaining RGC axons (gray vertical arrows) terminate on other SGC-I neurons and are not considered further in the simulation. The dendritic endings (BBE) are directly connected to the SGC-I soma. (b) Spatial distribution of model SGC-I dendritic endings for one model neuron, corresponding to a top view of a. The spatial extensions of the square dendritic field are 3,000 × 3,000 µm. (c) Representation of the bar stimulus shown in the same spatial dimensions as in b. In the simulations, the stimulus bar was used for static stationary or dynamic spatiotemporal stimuli including first-order motion. (d) Representation of the random-dot stimulus shown in the same spatial dimensions as in b. In the simulations, the random-dot stimulus was used for static stationary or dynamic spatiotemporal stimuli including second-order motion.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Tectal neurons are also sensitive to second-order motion in vivo7, raising the question of whether the SGC-I model contains all the biophysical elements to respond to second-order dynamic spatiotemporal stimuli. Second-order motion stimuli are not luminancedefined but rather correspond to regions of a common moving ‘process,’ for instance in a field of random dots19–21. On any single frame of such a second-order motion stimulus, nothing but a pattern of random dots is present, but in successive frames the pixel values within a moving area, such as a rectangle, are updated according to a rule that leaves the average dot density constant and equal to the stationary background (Fig. 3d). In a second-order motion stimulus, neither average luminance nor individual dots are moving. What is moving is the process of selecting a new set of dots. First, we investigated the response of the SGC-I model to the assumed retinal representation (Fig. 5a) of a static stationary whole-

field pattern of random dots (Fig. 3d). As discussed before, an RGC spike caused an SGC-I dendritic spike with a probability P(∆t) that was small for physiological RGC spike rates. The sparse dendritic spiking (Fig. 5b) caused sparse SGC-I soma spiking (Fig. 5c). On average, the SGC-I firing rate in response to the stationary pattern of random dots was 1.9 ± 0.7 Hz (mean ± s.d.; n = 5 trials of 3-s duration). In contrast, without the phasic synaptic signal transfer, P(∆t) = Pmax, the SGC-I model responded with a regular rate of 33.3 Hz to the stationary dot stimulus (Fig. 5d). Next, we investigated the SGC-I model response to the retinal representation (Fig. 5e) of an uncorrelated second-order motion stimulus (Fig. 3d), which has been used in an electrophysiological study3. On any single frame, a whole-field pattern of random dots is present, but in successive frames the pixel values within a moving rectangle (Fig. 3d) are replaced with uncorrelated random dots. The resultant percept in humans is that of a moving, twinkling rectangle. For most of the dendritic endings that were contacted by excited RGC axons, the probability P(∆t) for dendritic spiking was small. For dendritic endings within the moving rectangle, the situation was different. Because the random dots within the rectangle were replaced by an uncorrelated set at every time step, most dendritic endings had a large probability, P(∆t → ∞) = Pmax. Thus when an RGC spike arrived, dendritic endings within the rectangle were likely to fire, leading to a rate of dendritic spikes that was

Figure 4 Response of the model SGC-I cell to a static stationary (a−d) or moving (e−h) luminance-defined bar stimulus (Fig. 3c). (a) The summed spikes per simulation time step (10 ms) in response to a stationary bar stimulus of those excited RGC axons that terminated on dendritic endings. (b) The summed spikes per simulation time step of all dendritic endings (BBE) of one SGC-I model cell in response to a stationary bar stimulus. (c) The spikes per simulation time step of one SGC-I model cell in response to a stationary bar stimulus. (d) The spikes per simulation time step of one SGC-I model cell in response to a stationary bar stimulus without phasic synaptic signal transfer. (e) The summed spikes per simulation time step of those excited RGC axons that terminated on dendritic endings in response to a moving bar stimulus. (f) The summed spikes per simulation time step of all dendritic endings of one SGC-I model cell in response to a moving bar stimulus. (g) The spikes per simulation time step of one SGC-I model cell in response to a moving bar stimulus. (h) The spikes per simulation time step of one SGC-I model cell in response to a moving bar stimulus without phasic synaptic signal transfer.

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

383

ARTICLES
Figure 5 Response of the SGC-I model to a static stationary (a−d) and uncorrelated second-order motion (e−h) random-dot pattern (Fig. 3d). (a) The summed spikes per simulation time step (10 ms) in response to a static stationary random-dot stimulus of those excited RGC axons that terminated on dendritic endings. (b) The summed spikes per simulation time step of all dendritic endings (BBE) of one SGC-I model cell in response to a static stationary random-dot stimulus. (c) The spikes per simulation time step of one SGC-I model cell in response to a static stationary random-dot stimulus. (d) The spikes per simulation time step of one SGC-I model cell in response to a static stationary random-dot stimulus without phasic synaptic signal transfer. (e) The summed spikes per simulation time step of those excited RGC axons that terminated on dendritic endings in response to uncorrelated second-order motion. (f) The summed spikes per simulation time step of all dendritic endings of one SGC-I model cell in response to uncorrelated second-order motion. (g) The spikes per simulation time step of one SGC-I model cell in response to uncorrelated second-order motion. (h) The spikes per simulation time step of one SGC-I model cell in response to an uncorrelated second-order motion without phasic synaptic signal transfer.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

larger than that in the static stationary stimulus case (Fig. 5b,f). The large dendritic spike rate caused interactions between dendritic spikes and a large SGC-I soma spike rate (Fig. 5g). On average, the firing rate of the SGC-I model in response to the uncorrelated second-order motion stimulus was 14.0 ± 1.9 Hz (mean ± s.d.; n = 5 trials of 3-s duration). Without the phasic synaptic signal transfer, P(∆t) = Pmax, the SGC-I model responded with a regular rate of 33.3 Hz to the uncorrelated second-order motion stimulus (Fig. 5h). In addition, the SGC-I model responded to a variety of other dynamic spatiotemporal second-order stimuli. Smaller responses of the SGC-I model were obtained for the correlated second-order motion stimulus, in which a set of random dots moved across a background of random dots of equal density7. On average, the firing rate of the SGC-I model in response to the correlated second-order motion stimulus of a rectangle of the same size was 9.7 ± 1.1 Hz (mean ± s.d.; n = 5 trials of 3-s duration; data not shown). Responses of the SGC-I model were also obtained from second-order dynamic spatiotemporal stimuli that were not moving. For instance, consider a whole-field pattern of stationary static random dots, but in successive frames the pixel values within a stationary rectangle are replaced with uncorrelated random dots (Fig. 3d). The resultant percept in humans is that of a stationary twinkling rectangle. The resultant response in the SGC-I model was 11.1 ± 0.5 Hz (mean ± s.d.; n = 5 trials of 3-s duration; data not shown). The model response increased to 16.0 ± 1.4 Hz (mean ± s.d.; n = 5 trials of 3-s duration; data not shown) when the same number of dots (six dots) was replaced at random from anywhere within the receptive field rather than just within the rectangle. For a whole-field pattern of random dots in which the set of random dots within the receptive field is replaced with a new set of uncorrelated random dots in successive frames, the SGC-I model response increased to 31.8 ± 0.2 Hz (mean ± s.d.; n = 5 trials of 3-s duration; data not shown). In conclusion, the SGC-I firing rate in response to a dynamic spatiotemporal second-order stimulus, such as second-order motion, was significantly larger than the firing rate in response to a static stationary set of random dots. Thus for the non-luminance-defined object, the SGC-I model was able to differentiate between a static stationary and a dynamic spatiotemporal stimulus but failed to do so without the phasic synaptic signal transfer, that is, a constant probability P(∆t) = Pmax. This demonstrates again that phasic signal transfer is a crucial biophysical element for stimulus classification. Further,

the SGC-I representation of a dynamic spatiotemporal second-order stimulus is similar to the SGC-I representation of a dynamic spatiotemporal first-order stimulus. Hence, the SGC-I differentiation between static stationary and dynamic spatiotemporal stimuli, such as motion, is independent of the details of the stimulus. To what extent do the model results depend on the assumptions made for the retinal representation of the visual stimuli? We assumed that the RGC axons were silent in the absence of a stimulus and fired randomly with an average rate of 80 Hz when stimulated. Although this choice corresponds to biologically plausible RGC responses22,23, there may be large variations in the parameters for different subpopulations of RGCs in different animals24. To quantify to what extent the SGC-I stimulus classification as static stationary or dynamic spatiotemporal stimuli depends on the stimulated mean rate of RGC firing, we repeated the simulations for different stimulated mean RGC firing rates (while keeping the spontaneous rate at 0). Because of the importance of motion stimuli in nature and in the available experimental literature, we restricted this analysis of parameter sensitivity to the static stationary and moving stimuli previously examined (Figs. 4 and 5) and considered the SGC-I model ‘motion-sensitive’ (but see Discussion) when the firing rates of the SGC-I model in response to static stationary or moving stimuli were significantly different. The motion sensitivity of the SGC-I model to first- and second-order motion proved to be largely independent of the stimulated RGC rate over a wide range of physiological RGC firing rates (Fig. 6a). At low mean RGC rates, the model predicted that the SGC-I motion sensitivity breaks down first for second-order motion (<20 Hz) and then for first-order motion (<1 Hz). Given that the stimulated RGC mean firing rate can be modulated by light intensity22, this model result predicts that the SGC-I motion sensitivity was largely independent of the brightness of the stimulus over a wide

384

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 6 Parameter dependence of the SGC-I model response to stationary stimuli and first- and second-order motion. (a) SGC-I rate versus stimulated RGC rate. (b) SGC-I rate versus spontaneous RGC rate. (c) SGC-I rate versus Pmax. The SGC-I rate increases monotonically with increasing Pmax. (d) SGC-I rate versus t0. The SGC-I rate decreases with increasing values of t0 up to approximately t0 = 1,000 ms and for larger values remains largely independent of t0. (e) SGC-I rate versus stimulus speed for a moving bar and a second-order motion stimulus. (f) Number of SGC-I spikes per degree of stimulus movement versus stimulus speed for a moving bar and a second-order motion stimulus. Data points were calculated from simulation results in e using (spikes/deg) = (spike rate [1/s])/(stimulus speed [deg/s]). Inset: same as in f at expanded stimulus speed scale.

range of light intensities. In contrast to the marked independence of the motion sensitivity of the model to the stimulated RGC mean firing rate, it was extremely dependent on the spontaneous RGC mean firing rate (the RGC activity in the absence of any stimulus within its receptive field). Indeed, the motion sensitivity of the SGC-I model broke down at a spontaneous RGC rate of 0.5 Hz or higher (Fig. 6b). Most interestingly, this model prediction is consistent with the experimental observation that avian RGCs have little or no spontaneous activity25, a fact that distinguishes avian RGCs from primate RGCs, which have a spontaneous activity of 20 Hz26. To what extent do the model results depend on the numerical values of Pmax and t0? This question is important for two reasons. First, the numerical values for Pmax and t0 were measured in an in vitro preparation at room temperature, and the in vivo extrapolation may have a large error. Second, the retinotectal synapse in layer 5 is embedded in a dense network of horizontal17,18 and feedback27 connections. Although our knowledge of the cellular details of these connections is limited14, it is biologically plausible that these connections modulate the two parameters of phasic retinotectal signal transfer, Pmax and t0, potentially in a stimulus- or context-dependent manner. To quantify these parameter dependences, we repeated the simulations for different parameter values. The SGC-I model cell remained motion sensitive over a large range of values of Pmax around the measured value of Pmax = 0.87 (Fig. 6c) and over a large range of values of t0 around the measured value of t0 = 2,025 ms (Fig. 6d). The speed tuning of the SGC-I model allowed an interesting comparison with experimental data. Qualitatively, within our model, the following speed tuning to a small moving object was expected. Because of the spike generation in individual dendritic endings9 and

the spatial distribution of dendritic endings13, on average the SGC-I spike rate should increase linearly with increasing speed at low speed. At higher speed, the mutually exclusive interaction of dendritic spikes kicks in and thus the SGC-I spike rate will not increase further with increasing speed. To quantify the speed-tuning curve, we conducted simulations with a moving bar stimulus (Fig. 3c) at various speeds between 0 and 100 deg/s. This produced a linear increase in the SGC-I rate with increasing speed in the range of 0–10 deg/s (Fig. 6e). For higher speeds, the SGC-I rate increased sublinearly and reached its maximum rate of 33 Hz at 60 deg/s. In the range of 0–10 deg/s, the total number of spikes generated by an object moving through the same area in the visual field was independent of the speed but, because of the mutually exclusive interaction, decreased for increasing speed (Fig. 6f). Both the range of linear speed tuning and the constancy of the total number of spikes within this speed range were consistent with in vivo experimental results (N. Troje & B. Frost, Soc. Neurosci. Abstr. 24, 642.9, 1998). The model failed, however, to reproduce the experimentally observed decrease in the SGC-I rate for speeds above 20–40 deg/s7,28, thus suggesting the existence of a presently unknown additional suppressive mechanism for larger speeds that was not included in the model. Of note, the simulation results predict that for a second-order motion stimulus the SGC-I rate is largely independent of the speed (Fig. 6e), but the total number of spikes generated by second-order motion moving through the same area in the visual field decreased with increasing speed (Fig. 6f). Because the spikes were mostly caused by the random update of the dots within the rectangle, the number of spikes decreased with decreasing travel time (increasing speed). At present, no experimental speed tuning data are available for the second-order motion stimulus in avian tectum.

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

385

ARTICLES
Does the ensemble of the model SGC-I neurons, with their randomly distributed dendritic endings, carry information about the direction of motion? This is an important question, because rotundal neurons integrate outputs from multiple tectal neurons29 in a directionally sensitive manner30–32. To address this question, we simulated spike trains in response to moving stimuli for model SGC-I neurons with spatially offset but overlapping receptive fields. We then applied a motion-sensing algorithm that was cross-correlation based33 to extract an estimate of the direction of motion from the ensemble of simulated spike trains. The output of this motion-sensing algorithm is a ‘net motion signal,’ the sign of which indicates the direction of motion. Because of the random distribution of dendritic endings and because of the statistical nature of spike generation, the net motion signal varies from trial to trial, yielding a distribution of net motion signal. A natural measure of the fidelity of the ensemble motion sensitivity is the SNR, which is defined as the mean of the distribution of net motion signals divided by the s.d. of the distribution. For the parameters chosen, the moving bar and the second-order motion yielded an SNR of 3.8 and 2.0, respectively. In contrast, for dynamic spatiotemporal stimuli that were not moving, such as the update of uncorrelated random dots within a stationary rectangle or the whole field, the SNR was 0.0 and 0.1, respectively. The analysis thus shows that the ensemble of model SGC-I neurons (i) carries information to distinguish between moving and stationary dynamic spatiotemporal stimuli and (ii) carries information about the direction of motion for both first- and second-order motion stimuli. DISCUSSION This in vitro study revealed phasic signal transfer at the retinotectal synapse and binary dendritic responses to synaptic inputs that interacted in a mutually exclusive manner in the postsynaptic tectal neuron. A model of the tectal circuitry predicts that the two observed cellular properties mediate sensitivity in the tectal SGC-I neuron to a wide range of dynamic spatiotemporal stimuli, including moving stimuli, but not to static stationary stimuli. Further, the SGC-I model representation of the dynamic spatiotemporal second-order stimulus was similar to the SGC-I representation of the dynamic spatiotemporal first-order stimulus. Hence, SGC-I distinction between static stationary stimuli and dynamic spatiotemporal stimuli, such as motion, is independent of the details of the stimulus. The stimulus classification remained robust over a wide range of model parameters including the brightness of the stimulus. This study indicates that phasic signal transfer at the retinotectal synapse, presumably mediated by synaptic depression, may be a crucial biophysical element of tectal stimulus classification. In general, synaptic depression can endow networks with new and unexpected dynamic properties34–37. It has recently been suggested that thalamocortical synaptic depression, rather than the cortical network, may account for important response properties of cortical neurons38–40. Further, synaptic depression at the input synapses of the nucleus laminaris in chick provides an adaptive mechanism to compensate for intensity variations in the localization of sound41. The present study assigns a similar importance to phasic signal transfer at the retinotectal synapse for the analysis of dynamic spatiotemporal stimuli. This indicates that phasic signal transfer at an incoming synapse may be a conserved feature in the dynamic regulation of neuronal sensitivity during rapid changes in sensory input. It has long been appreciated that it is not the light intensity itself but rather the pattern of local variation in intensity that is typically the exciting factor for neurons in visual pathways42. Often these local variations in light intensity are mediated by motion; however, motion is only a subset of the visual stimuli that lead to spatiotemporal variations in intensity. Perhaps because most previous studies of avian tectal neurons and primate cortical area middle temporal neurons have focused on the coding of motion itself, the neurons were typically tested with static stationary stimuli to which they respond weakly versus moving stimuli to which they respond strongly8,43. Based on these results, the neurons have been interpreted as motion-sensitive. When tested with a broader range of stimuli, however, as was done in this simulation study of model SGC-I neurons and in the more recent in vivo studies on primate middle temporal neurons43–45, these neurons also respond to dynamic spatiotemporal stimuli that are not moving. Based on the examined SGC-I cellular properties, the SGC-I model reproduced the in vivo observations that the tectal SGC-I neuron does not respond to static stationary stimuli but responds to moving stimuli independently of the cues that define the moving object. In addition, the model predicts strong SGC-I responses to stationary dynamic spatiotemporal stimuli that have not previously been tested in vivo. This model prediction suggests an extended interpretation of SGC-I signal processing. Because of the phasic synaptic signal transfer, individual SGC-I neurons do not respond to a static stationary stimulus but do respond vigorously to a dynamic spatiotemporal stimulus that activates new dendritic endings sequentially. Because the SGC-I response is independent of the particular sequence of activation of its dendritic endings, the individual SGC-I neuron responds to most dynamic spatiotemporal stimuli, whether they are moving or not. The interpretation of a single SGC-I response is thus ambiguous in this respect. This is a typical problem in breaking ambiguity by population coding. The individual SGC-I neuron is sensitive to most dynamic spatiotemporal stimuli. The spatiotemporal pattern of spikes in the ensemble of SGC-I neurons, however, contains additional spatiotemporal information about the stimulus, such as motion parameters. In the avian brain, rotundal neurons receive inputs from an ensemble of tectal SGC-I neurons29. Apparently, the tectal population activity is then decoded by postsynaptic rotundal neurons that are sensitive to the direction of motion and looming30–32. In summary, this work indicates that a specific cellular mechanism may exist to distinguish between static stationary stimuli and dynamic spatiotemporal stimuli and further clarifies the functional interpretation of individual SGC-I neurons as form-cue invariant change-sensitive and of ensembles of SGC-I neurons as form-cue invariant motion-sensitive. It is of note that this interpretation does not exclude SGC-I responses to more complex dynamic spatiotemporal stimuli such as relative motion11, responses that are presumably mediated by tectal network properties17,27. The sensitivity of the SGC-I model to second-order motion19 was a direct consequence of its sensitivity to dynamic spatiotemporal stimuli in general. Previous theoretical studies of second-order motion sensitivity have pointed out the necessity of a nonlinearity, such as a rectification, followed by a summation of the preprocessed signal19,46. Both stages of processing second-order signals have long been ascribed to higher-level processing in the visual cortex21. Evidence for low-level processing of second-order motion has been sparse47–49. Here we report that, in the avian tectal circuit, the required rectification stage is implemented by the all-or-none binary response of the SGC-I dendritic endings, which mediate a receptive field with a fine structure of rectifying subunits. The array of rectified responses is subsequently integrated by the SGC-I soma, which in turn projects to the nucleus rotundus13,15,29,50. The computation of the direction of second-order motion could be completed by rotundal cells, which are known to process different aspects of visual information including translational motion30–32.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

386

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
METHODS
Experiment. Thirty-nine White Leghorn chick hatchlings (Gallus gallus; <5 d old) were used in this study. All procedures used in this study were approved by the local authorities and conform to the guidelines of the National Institutes of Health on the Care and Use of Laboratory Animals. Tectal slice preparation, SGC-I soma whole-cell recording, electrostimulation with bipolar tungsten electrodes and SGC-I labeling were carried out as described previously9. We obtained stable whole-cell patch recordings from a total of 67 neurons in the chick SGC-I. The series resistance of the recordings was 10 ± 2 MΩ (mean ± s.d.) and was routinely compensated. We analyzed only neurons that were sufficiently labeled to allow the unequivocal classification as the SGC-I cell type (Fig. 1b)13. SGC-I neurons usually have their somata in the outer aspects of the SGC-I and respond with characteristic rhythmic bursting to somatic current injection9. The cells had a stable resting potential of –66 ± 4 mV (mean ± s.d.) and an input resistance at rest of 72 ± 17 MΩ (mean ± s.d., n = 67). Pulse train stimuli (Fig. 2a–c) were repeated ten times for each stimulation interval in a pseudorandom sequence of stimulation intervals with a waiting time of 5 min between pulse train stimuli. The response probability for a sharp-onset response for each stimulus pulse within the pulse train was derived from the number of responses divided by the number of trials. Pairedpulse stimuli (Fig. 2d–f) were repeated five times for each stimulation interval in a pseudorandom sequence of intervals with a waiting time of 5 min. Model. In the bird, the visual field of each eye is approximately 100° and projects onto a tectal circumference of approximately 10 mm. Assuming for simplicity a homogeneous spatial distribution of RGCs in the retina and a homogeneous spatial distribution of RGC axon terminals on the tectal surface, we estimated that a typical SGC-I dendritic field of 3 mm in diameter corresponds to a visual field of 30 °. In our model, we therefore considered a region of visual space that is 30 × 30 °, which corresponds to a region of tectal surface that is 3,000 × 3,000 µm. Because of this correspondence of visual space and tectal surface, it was more convenient to express the stimulus space in terms of tectal surface units rather than visual space units. In the model, the stimulus space is represented by an array of 300 × 300 squares of 10 × 10 µm, corresponding to 0.1 × 0.1° in visual space. A binary stimulus was represented within this space and moved with a speed in multiples of 10 µm per simulation time step (10 ms). For all simulations (except those for Fig. 6e,f) the stimulus speed was 10 deg/s. We assumed that the described stimulus space is sampled by an array of 100 × 100 RGC axons (Fig. 3a). Thus one RGC samples exclusively 3 × 3 stimulus space units. We assumed no spontaneous activity for the RGC, but the RGC produces a Poisson spike train while a stimulus is present within the 3 × 3 space, with a mean firing rate of 80 Hz, a rate that is typically observed experimentally23. Each SGC-I soma is connected to approximately 500 dendritic endings. These dendritic endings are randomly distributed within the 100 × 100 array (Fig. 3b). Each dendritic ending makes a synaptic contact with one RGC axon. Thus the 100 × 100 RGC axons are sampled sparsely; because this SGC-I model considered only one SGC-I cell, most RGC axons (approximately 9,500 axons) do not terminate on a dendritic ending. The signal transfer from one RGC axon to one dendritic ending is phasic in a time-dependent probabilistic manner. An RGC spike causes one dendritic spike with probability P(∆t) = Pmax (1 – e–∆t/t0), with ∆t the time interval since the previous RGC spike and Pmax = 0.87 and t0 = 2,025 ms the parameters determined from pulse train stimulation experiments (Fig. 2c). For simplicity, we assumed that dendritic spikes from all dendritic endings arrive at the SGC-I soma without delay. Dendritic spikes were counted at the soma. If no dendritic spike was generated in any of the dendritic endings, no SGC-I spike was generated in that simulation time step. If one or more dendritic spikes were generated, one SGC-I spike was generated if the previous SGC-I spike occurred more than 30 ms before (outside the interaction time). No SGC-I spike was generated if the previous SGC-I spike occurred within the interaction time of 30 ms. The SGC-I rate was calculated as the number of spikes during 3-s simulations with five repetitions. The luminance-defined bar stimulus of 5 × 1° in visual space corresponds to 500 × 100 µm on the tectal surface and 50 × 10 space units in the model (Fig. 3c). Because each model RGC axon has a receptive field of 3 × 3 nonoverlapping space units, the bar stimulus excited 56 RGC axons. A model SGC-I neuron has approximately 500 dendritic endings that are randomly distributed within its receptive field (Fig. 3b) and a total of 100 × 100 RGC axons terminating within this field. Therefore, on average, 3 ≅ (56 × 500)/(100 × 100) of the 56 excited RGC axons terminated on an equal number of dendritic endings (one RGC axon per dendritic ending). When the bar moved with a speed of 10 deg/s in the visual space, this corresponded to a speed of one space unit per simulation time step in the model. The stationary whole-field random-dot stimulus is 300 × 300 space units (corresponding to 30 × 30° in visual space and 3,000 × 3,000 µm on the tectal surface). Because each model RGC axon has a receptive field of 3 × 3 nonoverlapping space units, and the dot pattern is of low density, the number of excited RGC axons was approximately equal to the number of dots. This is on average the number of space units (300 × 300) times the dot density (0.002), which yields 180 dots. Using the same values as above, on average 9 = (180 × 500)/(100 × 100) of the 180 excited RGC axons terminated on an equal number of dendritic endings. For second-order motion, we chose a rectangle of 300 × 10 space units (corresponding to 30 × 1° in visual space and 3,000 × 100 µm on the tectal surface). Following the estimate in the previous paragraph, we expected on average 6 dots (= 300 × 10 × 0.002) within the moving rectangle and therefore an equal number of excited RGC axons. A model SGC-I neuron has on average 17 = (300 × 10 × 500)/(300 × 300) dendritic endings within the rectangle. Because the model has a total of 100 × 3.3 = 330 RGC axons within the rectangle, on average only 0.3 = (6 × 17)/330 of the 6 excited RGC axons terminated on an equal number of dendritic endings (one RGC axon per dendritic ending). The rectangle moved with a speed of one space unit per simulation time step (corresponding to 10 deg/s in the visual space). On each temporal frame the random-dot pattern within the moving rectangle was replaced with a different uncorrelated random-dot pattern of equal density and equal mean luminance. For the tectal ensemble decoding analysis, we considered pairs of model SGC-I neurons, labeled A and B, with each neuron having the same properties as described above. The receptive fields of the two neurons were offset, however, by 9° in the visual space (B to the right of A), so the dendritic fields with the randomly distributed dendritic endings were offset by 900 µm at the tectal surface. For each simulation trial we generated spike trains, sA(t) and sB(t), for both model SGC-I neurons in response to a dynamic spatiotemporal visual stimulus. The stimuli considered were a moving bar (Fig. 3c), second-order motion (Fig. 3d) and a dynamic spatiotemporal stimulus that was not moving. The latter stimulus consisted of a pattern of random dots, in which the pixel values within a stationary rectangle (Fig. 3d) or the whole field were replaced with uncorrelated random dots in successive frames. The moving stimuli were moving from left to right. The spike trains, sA(t) and sB(t), were analyzed according to an algorithm that was cross-correlation based33. In brief, the spike trains were convolved with an exponential function, f(t), to create lowpass-filtered responses, rA(t) = sA(t) * f(t) and rB(t) = sB(t) * f(t), where f(t) = e–t/σ for t ≥ 0, f(t) = 0 for t < 0 and σ = 10 ms. A signal, R, indicative of rightward motion was obtained by delaying the filtered response of cell A by an amount δ, multiplying pointwise by the filtered response of cell B and summing the result: R=

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Σ t

rA (t – δ) rB (t),

where the summation is over all the time points in the trial, and rA(t − δ) is circularly shifted to match the duration of rB(t). The delay value, δ, is the spatial receptive field offset divided by the stimulus speed. A signal, L, indicative of leftward motion was obtained correspondingly by: L=

Σ t

rB (t – δ) rA (t).

For one pair of cells and one trial, the net motion signal is N = R − L, the sign of which indicates the direction of motion. Because the problem under investigation was exactly symmetric in space, we considered stimuli that were moving from left to right, thus typically yielding positive values for N. For motion in the opposite direction, the N values would typically be negative. For each trial, we chose a new set of randomly distributed dendritic endings. The repetition of trials (n = 30) yielded a distribution of net motion signals, N. It can be

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

387

ARTICLES
shown that the pairwise computation of the net motion signal described above is equivalent to an approach that combines responses of an ensemble of cells simultaneously33.
ACKNOWLEDGMENTS The authors thank H.J. Karten and D. Kleinfeld for support during the collection of preliminary data, A. Mahani for comments and W.B. Kristan, H. Wagner, G. DeAngelis, M. Ariel, P. Lukasiewicz, J. Sanes and A. Carlsson for critical reading of the manuscript. The work was supported by grants from Deutsche Forschungsgemeinschaft to H.L. and Whitehall Foundation and McDonnell Center for Higher Brain Function to R.W.
20. Lu, Z.L. & Sperling, G. Three-systems theory of human visual motion perception: review and update. J. Opt. Soc. Am. A 18, 2331–2370 (2001). 21. Baker, C.L. Jr. & Mareschal, I. Processing of second-order stimuli in the visual cortex. Prog. Brain Res. 134, 171–191 (2001). 22. Rodieck, R.W. The First Steps in Seeing (Sinauer, Sunderland, Massachusetts, 1998). 23. Meister, M. & Berry, M.J. II. The neural code of the retina. Neuron 22, 435–450 (1999). 24. Tabata, T. & Kano, M. Heterogeneous intrinsic firing properties of vertebrate retinal ganglion cells. J. Neurophysiol. 87, 30–41 (2002). 25. Nalbach, H.O., Wolf-Oberhollenzer, F. & Remy, M. Exploring the image. in Vision, Brain, and Behavior in Birds (eds. Zeigler H.P. & Bischof, H.J.) 25–46 (MIT Press, Cambridge, Massachusetts, 1993). 26. Troy, J.B. & Lee, B.B. Steady discharges of macaque retinal ganglion cells. Vis. Neurosci. 11, 111–118 (1994). 27. Luksch, H. Cytoarchitecture of the avian optic tectum: neuronal substrate for cellular computation. Rev. Neurosci. 14, 85–106 (2003). 28. Frost, B.J., Scilley, P.L. & Wong, S.C.P. Moving background patterns reveal doubleopponency of directionally specific pigeon tectal neurons. Exp. Brain Res. 43, 173–185 (1981). 29. Marin, G. et al. Spatial organization of the pigeon tectorotundal pathway: An interdigitating topographic arrangement. J. Comp. Neurol. 458, 361–380 (2003). 30. Revzin, A.M. Functional localization in the nucleus rotundus. in: Neural Mechanisms of Behavior in the Pigeon (eds. Granda, A.M. & Maxwell, J.H.) 165–175 (Plenum, New York, 1981). 31. Wang, Y. & Frost, B.J. Time to collision is signalled by neurons in the nucleus rotundus of pigeons. Nature 356, 236–238 (1992). 32. Sun, H. & Frost, B.J. Computation of different optical variables of looming objects in pigeon nucleus rotundus neurons. Nat. Neurosci. 1, 296–302 (1998). 33. Chichilnisky, E.J. & Kalmar, R.S. Temporal resolution of ensemble visual motion signals in primate retina. J. Neurosci. 23, 6681–6689 (2003). 34. Abbott, L.F., Varela, J.A., Sen, K. & Nelson, S.B. Synaptic depression and cortical gain control. Science 275, 220–223 (1997). 35. Tsodyks, M.V. & Markram, H. The neural code between neocortical pyramidal neurons depends on neurotransmitter release probability. Proc. Natl. Acad. Sci. USA 94, 719–723 (1997). 36. Chance, F.S., Nelson, S.B. & Abbott, L.F. Synaptic depression and the temporal response characteristics of V1 cells. J. Neurosci. 18, 4785–4799 (1998). 37. Goldman, M.S., Maldonado, P. & Abbott, L.F. Redundancy reduction and sustained firing with stochastic depressing synapses. J. Neurosci. 22, 584–591 (2002). 38. Freeman, T.C.B., Durand, S., Kiper, D.C. & Carandini, M. Suppression without inhibition in visual cortex. Neuron 35, 759–771 (2002). 39. Carandini, M., Heeger, D.J. & Senn, W.A Synaptic explanation of suppression in visual cortex. J. Neurosci. 22, 10053–10065 (2002). 40. Chung, S., Li, X. & Nelson, S.B. Short-term depression at thalamocortical synapses contributes to rapid adaptation of cortical sensory responses in vivo. Neuron 34, 437–446 (2002). 41. Cook, D.L., Schwindt, P.C., Grande, L.A. & Spain, W.J. Synaptic depression in the localization of sound. Nature 421, 66–70 (2003). 42. Lettvin, J.Y., Maturana, H.R., McCulloch, W.S. & Pitts, W.H. What the frog’s eye tells the frog’s brain. Proc. IRE 47, 1940–1951 (1959). 43. Palanca, B.J.A. & DeAngelis, G.C. Macaque middle temporal neurons signal depth in the absence of motion. J. Neurosci. 23, 7647–7658 (2003). 44. Newsome, W.T., Britten, K.H. & Movshon, J.A. Neuronal correlates of a perceptual decision. Nature 341, 52–54 (1989). 45. Croner, L.J. & Albright, T.D. Segmentation by color influences responses of motionsensitive neurons in the cortical middle temporal visual area. J. Neurosci. 19, 3935–3951 (1999). 46. Wilson, H.R. The role of second-order motion signals in coherence and transparency: higher-order processing in the visual system. Ciba Foundation Symposium 184, 227–244 (1994). 47. Orger, M.B., Smear, M.C., Anstis, S.M. & Baier, H. Perception of Fourier and nonFourier motion by larval zebrafish. Nat. Neurosci. 3, 1128–1133 (2000). 48. Roeser, T. & Baier, H. Visuomotor behaviors in larval zebrafish after GFP-guided laser ablation of the optic tectum. J. Neurosci. 23, 3726–3734 (2003). 49. Demb, J.B., Zaghloul, K. & Sterling, P. Cellular basis for the response to secondorder motion cues in Y retinal ganglion cells. Neuron 32, 711–721 (2001). 50. Hellmann, B. & Güntürkün, O. Structural organization of parallel information processing within the tectofugal visual system of the pigeon. J. Comp. Neurol. 429, 94–112 (2001).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 3 November 2003; accepted 9 January 2004 Published online at http://www.nature.com/natureneuroscience/
1. Albright, T.D. Cortical processing of visual motion. in: Visual Motion and its Role in the Stabilization of Gaze (eds. Miles, F.A. & Wallman, J.) 177–201 (Elsevier, Amsterdam, 1993). 2. Croner, L.J. & Albright, T.D. Seeing the big picture: integration of image cues in the primate visual system. Neuron 24, 777–789 (1999). 3. Albright, T.D. Form-cue invariant motion processing in primate visual cortex. Science 255, 1141–1143 (1992). 4. Olavarria, J.F., DeYoe, E.A., Knierim, J.J., Fox, J.M. & VanEssen, D.C. Neural responses to visual texture patterns in middle temporal area of the macaque monkey. J. Neurophysiol. 68, 164–181 (1992). 5. Geesaman, B.J. & Andersen, R.A. The analysis of complex motion patterns by form/cue invariant MSTd neurons. J. Neurosci. 16, 4716–4732 (1996). 6. Jassik-Gerschenfeld, D. & Guichard, J. Visual receptive fields of single cells in the pigeon’s optic tectum. Brain Res. 40, 303–317 (1972). 7. Frost, B.J., Cavanagh, P. & Morgan, B. Deep tectal cells in pigeons respond to kinematograms. J. Comp. Physiol. A 162, 639–647 (1988). 8. Frost, B.J. Subcortical analysis of visual motion: Relative motion, figure-ground discrimination and self-induced optic flow. in: Visual Motion and its Role in the Stabilization of Gaze (eds. Miles, F.A. & Wallman, J.) 159–175 (Elsevier, Amsterdam, 1993). 9. Luksch, H., Karten, H.J., Kleinfeld, D. & Wessel, R. Chattering and differential signal processing in identified motion sensitive neurons of parallel visual pathways in chick tectum. J. Neurosci. 21, 6440–6446 (2001). 10. Jassik-Gerschenfeld, D., Minois, F. & Conde-Courtine, F. Receptive field properties of directionally selective units in the pigeon’s optic tectum. Brain Res. 24, 407–421 (1970). 11. Frost, B.J. & Nakayama, K. Single visual neurons code opposing motion independent of direction. Science 220, 744–745 (1983). 12. Frost, B.J. Moving background patterns alter directionally specific responses of pigeon tectal neurons. Brain Res. 151, 599–603 (1978). 13. Luksch, H., Cox, K. & Karten, H.J. Bottlebrush dendritic endings and large dendritic fields: motion-detecting neurons in the tectofugal pathway. J. Comp. Neurol. 396, 399–414 (1998). 14. Tömböl, T. & Németh, A. Direct connections between dendritic terminals of tectal ganglion cells and glutamate-positive terminals of presumed optic fibres in layers 4–5 of the optic tectum of Gallus domesticus. Neurobiology (Bp) 7, 45–67 (1999). 15. Karten, H.J., Cox, K. & Mpodozis, J. Two distinct populations of tectal neurons have unique connections within the retinotectorotundal pathway of the pigeon (Columba livia). J. Comp. Neurol. 387, 449–465 (1997). 16. Hunt, S.P. & Webster, K.E. The projection of the retina upon the optic tectum of the pigeon. J. Comp. Neurol. 162, 433–445 (1975). 17. Hardy, O., Leresche, N. & Jassik-Gerschenfeld, D. Morphology and laminar distribution of electrophysiologically identified cells in the pigeon’s optic tectum: an intracellular study. J. Comp. Neurol. 233, 390–404 (1985). 18. Luksch, H. & Golz, S. Anatomy and physiology of horizontal cells in the optic tectum of the chick. J. Chem. Neuroanatomy 25, 185–194 (2003). 19. Chubb, C. & Sperling, G. Drift-balanced random stimuli: a general basis for studying non-Fourier motion perception. J. Opt. Soc. Am. A 5, 1986–2007 (1988).

388

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Differential control over cocaine-seeking behavior by nucleus accumbens core and shell
Rutsuko Ito, Trevor W Robbins & Barry J Everitt
Nucleus accumbens (NAc) dopamine is widely implicated in mediating the reinforcing effects of drugs of abuse. However, the precise function of the NAc itself in drug self-administration has been difficult to establish. Here we show a neural doubledissociation of the behavioral processes that underlie cocaine self-administration in rats. Whereas selective excitotoxic lesions of the NAc core had only a minor effect on the acquisition of responding for cocaine under a standard schedule of continuous reinforcement, these lesions profoundly impaired the acquisition of drug-seeking behavior that was maintained by drugassociated conditioned reinforcers and assessed using a second-order schedule of cocaine reinforcement. In contrast, selective excitotoxic lesions of the NAc shell did not impair drug self-administration or the acquisition of cocaine-seeking, but they did attenuate the psychostimulant effects of cocaine. These results further our understanding of how the NAc controls drug-seeking and drug-taking behavior.

Much evidence supports the hypothesis that the dopamine innervation of the nucleus accumbens (NAc) is a key neural substrate mediating the primary reinforcing and psychomotor stimulant effects of drugs of abuse. Intravenous cocaine self-administration is reduced by 6-hydroxydopamine (6-OHDA)-induced dopamine depletion from the NAc1–3. Infusions of dopamine receptor agonists and antagonists directly into the NAc alter rates of intravenous drug self-administration, as if rats are compensating for changes in the reinforcing effects of the drug4,5. Furthermore, extracellular dopamine in the NAc is consistently increased in response to experimenter-delivered or self-administered cocaine, amphetamine, nicotine, opiates and ethanol in rats and in primates6–10. The glutamatergic innervation of the NAc has also been implicated in modulating drug self-administration behavior, though less clearly so11,12. Given the evidence that various neural systems innervating the NAc contribute to drug self-administration, it is perhaps surprising that excitotoxic lesions of the NAc itself, which destroy its medium spiny neuron output and other intrinsic neurons, have variable, and often no, effects on drug self-administration13–16. This leaves some measure of doubt about the NAc having an essential role in drug reinforcement. In resolving this issue, it is important to bear in mind that drug selfadministration probably depends on a complex interaction of several distinct behavioral processes. It involves conditioning as well as the unconditioned effects of the drug itself—these factors may underlie different aspects of drug-seeking and drug-taking behavior17,18. Thus, data from humans and animals indicate that environmental stimuli previously associated with self-administered drugs may potently affect subjective (i.e., craving) as well as behavioral measures of drug-seeking behavior and relapse19–25. Moreover, these

conditioned and unconditioned effects of drugs are neurally dissociable. For example, lesions of the basolateral amygdala do not affect drug self-administration, but they do prevent the acquisition of cocaine-seeking behavior under the control of drug-associated stimuli26 and reinstatement of drug-seeking after extinction27. Distinguishing among these conditioned and unconditioned processes in self-administration protocols is especially important in light of evidence that the NAc itself is a heterogeneous structure with at least two distinct regions, shell and core28,29, that may contribute in different ways to drug self-administration30–33. There is strong evidence that the NAc core is involved in the control of goal-directed behavior by associative processes, consistent with its central position within limbic cortical-ventral striatal circuitry34. For example, selective dopaminergic or excitotoxic lesions of the NAc core, but not the shell, disrupt learnt Pavlovian influences on appetitive behavior35–37. One of the important ways in which Pavlovian conditioned stimuli influence behavior is as conditioned reinforcers that can, by themselves, support instrumental behavior such as drug-seeking23. Excitotoxic lesions of the NAc core disrupt the capacity of foodassociated conditioned reinforcers to control behavior, whereas lesions of the NAc shell specifically impair the mechanism by which drugs such as cocaine and d-amphetamine enhance the effects of such stimuli38. Moreover, intra-NAc shell infusions of amphetamine enhanced the motivational effects of Pavlovian conditioned stimuli on instrumental behavior (Pavlovian-to-instrumental transfer, a form of Pavlovian arousal39). The aim of the present study was to investigate the extent to which the NAc core and shell contribute to conditioned, as well as unconditioned, influences that govern drug-seeking and drug-taking. To this end, we not only studied drug self-administration under standard

Department of Experimental Psychology, University of Cambridge, Downing Street, Cambridge CB1 1BB, UK. Correspondence should be addressed to B.J.E. ([email protected]). Published online 21 March 2004; doi:10.1038/nn1217

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

389

ARTICLES
Figure 1 Excitotoxic lesions of the NAc core and shell. Schematic representation of quinolinic acid lesions of the NAc core (top) and ibotenic acid lesions of the NAc shell (bottom). Areas shaded in gray and black represent the largest and smallest extent of neuronal damage in a single animal, respectively. Coronal sections are +2.2 mm anterior through +0.48 mm posterior to bregma49.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

These findings imply a neural dissociation between the mechanisms underlying the associative control of drug-seeking and those underlying the psychomotor stimulant effects of cocaine. The present results enhance our understanding of how the NAc controls drug-seeking and drug-taking behavior. RESULTS Cocaine self-administration under continuous reinforcement Intravenously catheterized rats with selective excitotoxic lesions of the NAc core or shell regions, and their sham-operated controls (Figs. 1 and 2), were initially trained daily for 2 h to acquire cocaine self-administration under a continuous reinforcement schedule. Response on one of two identical levers (active lever) led to a contingent infusion of cocaine (0.25 mg per infusion). They were deemed to have acquired the task when stable responding with less than 10% variance across three consecutive days was achieved. All three treatment groups (core, shell and sham) reached criterion levels of responding within 10 d of beginning cocaine self-administration under a continuous reinforcement schedule (Fig. 3a). Three-way ANOVA of square-root transformed lever presses during acquisition revealed a significant group × lever × day interaction (F18,315 = 1.62, P < 0.05) and a significant main effect of lesion group (F2,65 = 6.24, P < 0.005). Separate analyses of the pattern of responding on the active and inactive levers using two-way ANOVA followed by Newman-Keuls multiple comparisons revealed that the core-lesioned rats responded at a significantly higher rate on the active lever on days 1 and 2 only (P < 0.01) compared to control rats (group × day interaction, F9,261 = 2.61, P < 0.01; Fig. 3b). There was no difference between sham and shell groups in responding on the active lever (F1,20 = 0.09, P = 0.76). Inactive lever responses in the core-lesioned group were slightly, but significantly, increased overall compared with those in the sham group (group effect, F1,29 = 7.01, P < 0.01; Fig. 3a), whereas there was no significant group effect on inactive lever responding between shell and sham groups (F1,21 = 1.91, P < 0.18). Nevertheless, significantly more responses were made on the active lever, as compared to the inactive lever, in all groups (lever, F1,35 = 276.03, P < 0.0001).

conditions of continuous reinforcement, where every instrumental response is followed by a contingent cocaine infusion, but we also used a procedure in which drug-seeking becomes increasingly under the control of drug-associated conditioned reinforcers (in a so-called second-order schedule of reinforcement23). We found that selective excitotoxic lesions of the NAc core profoundly disrupted the acquisition of cocaine-seeking behavior when this behavior was substantially under the control of drug-associated conditioned reinforcers. Although lesions of the NAc shell did not impair drug self-administration or the acquisition of cocaine-seeking, they did attenuate the response rate–enhancing (or psychostimulant) effects of cocaine.

Figure 2 Representative photomicrographs showing Cresyl Violet-stained and NeuN-stained coronal sections through the NAc in rats with NAc shell or core lesions and sham-operated control subjects. (a) Nissl-stained section through the NAc of a control subject, showing the region of the shell and core and other markers at this antero-posterior level (island of Calleja, lateral ventricle and anterior commissure). (b) Nissl-stained section of a NAc shell lesion, showing the marked loss of staining in the shell region and preservation of neurons in the core region, as well as the infusion cannula tract. (c) Nissl-stained section of a NAc core lesion; marked gliosis can be seen around the medial surface of the anterior commissure. Note the apparent medial shift of the anterior commissure, which is the result of the loss of neurons in the core region. The integrity of the shell region is preserved. (d,e) NeuN-stained sections through the shell region of sham (d) and shell-lesioned subjects. Note the complete disappearance of NeuN-immunoreactivity in the shell region (compare e with d), indicating the loss of neurons in this region of the NAc following infusion of ibotenic acid. Abbreviations: AC, anterior commissure; Core, NAc core; Shell, NAc shell; LV, lateral ventricle; IC, Isle of Calleja.

390

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 3 Acquisition of intravenous cocaine self-administration under a continuous reinforcement schedule. (a) Mean (±s.e.m.) number of responses on the active and inactive lever during each 2 hr session. *P < 0.05, **P < 0.01, compared to sham. (b) The rate of responding (per min) on the active (drug-paired) lever in each session, after sham or excitotoxic lesions of the NAc core and shell regions. **P < 0.01 compared to sham. (c) Representative response records of individual rats from NAc sham, core and shell lesion groups on the last day of acquisition of cocaine selfadministration under a continuous reinforcement schedule. Each bar above the horizontal line represents an individual response on the active lever, whereas each bar below the line represents a response on the inactive lever.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Typical response records on the last day of acquisition (day 10) in sham, core and shell-lesioned rats (Fig. 3c) showed that response patterns in both sham and shell groups were characterized by an initial, rapidly occurring burst of cocaine self-administration (‘loading’ phase) followed by a period of stable ‘titrating pattern,’ evenly spaced responses on the active lever. Responding in core-lesioned rats, however, had the following characteristics: (i) higher response rates, such that core-lesioned rats gained the maximum number of cocaine infusions (50) significantly earlier in the session than the shell and sham lesioned groups across the 10 d of acquisition; (ii) for the postreinforcement pause (PRP; the period after each cocaine infusion and before the first response made subsequently, which provides a measure of the impact of cocaine reinforcement), there were no significant differences, indicating unimpaired control over instrumental responding by cocaine (sham, 2.71 (mean PRP (min) on day 10) ± 0.19 (s.e.m.); core, 2.73 ± 0.26; shell, 3.07 ± 0.25; F2,37 = 0.60, nonsignificant, n.s.) In summary, core-lesioned rats showed only minor changes in the acquisition of cocaine self-administration. They had somewhat higher rates of responding on the active lever, compared with the sham and shell groups, on the first two days of acquisition only. In addition, responding on the inactive lever by core-lesioned rats was inconsistently elevated, thereby reducing discrimination between the active and inactive levers on some days of testing. By contrast, there was no difference between the sham and shell groups in the pattern of acquisition of cocaine self-administration. Cocaine dose-response function In all three groups, variations in the dose of cocaine produced orderly, monotonic changes in the rates of responding on the active lever, with evidently more persistent responding in extinction (i.e., under saline) in the core-lesioned rats (Fig. 4). ANOVA with repeated measures on response rates revealed significant main effects of dose (F4,120 = 41.24, P < 0.0001), lever (F1,30 = 171.21, P < 0.0001) and lesion (F2,30 = 5.59, P < 0.01), with significant group × lever × dose (F8,120 = 2.14, P < 0.04) and dose × lesion (F8,120 = 5.36, P < 0.001) interactions. Separate ANOVA on response rates on the active lever at each dose revealed significant main effects of lesion group in the saline (F2,17 = 12.31, P < 0.001), 0.25 mg/infusion (F2,17 = 7.40, P < 0.005) and 0.5 mg/infusion (F2,17 = 9.05, P < 0.003) conditions. NewmanKeuls pairwise comparisons revealed that core-lesioned rats responded at significantly higher rates than sham and shell group animals on the active lever in extinction (i.e., on saline substitution) and at marginally, yet significantly, higher rates at doses of 0.25 and 0.5 mg/infusion. One-way ANOVAs on response rates on the inactive lever revealed no significant lesion group effect in any of the conditions. Cocaine-seeking under a second-order schedule of reinforcement Once cocaine self-administration under continuous reinforcement had been acquired, training began under the second-order schedules, where the response requirement for cocaine and the conditioned reinforcer was progressively increased (see Methods). It was determined that each rat would have to satisfy a certain criterion in order to move on to the next stage of training. This criterion was to obtain at least ten cocaine infusions per session at each schedule stage for three consecutive days. Significantly more core-lesioned animals failed to reach criterion at each stage beyond FR10(FR4:S), the core group size diminishing from 15 at FR10(FR1:S) to 8 at FR10(FR10:S) (Fig. 5). Responding on the active lever increased in all groups as the schedule requirement was progressively increased across days (day; F1,35 = 713.41, P < 0.0001). However, overall ANOVAs showed that the groups responded at significantly different rates (group, F2,113 = 4.00, P < 0.0001; group × day F2,35 = 10.80, P < 0.0002). Further analyses by lever and training stage separately revealed significantly lower responding on the active lever in the core-lesioned group compared to controls under FR10(FR2:S) (F1,124 = 6.04, P < 0.02), FR10(FR4:S) (F1.116=7.28, P < 0.02), FR10(FR7:S) (F1,104=11.69, P < 0.002) and FR10(FR10:S) (F1,96 = 17.06, P < 0.0004) schedules. Responding on the active lever in the shell-lesioned group was significantly lower than the controls under the FR10(FR10:S) schedule (F1,88 = 10.40, P < 0.004), but not at other stages. By contrast, responding on the inactive lever in the core group was significantly more elevated than that in the sham group (F1,1056 = 7.48, P < 0.01), at all stages of the

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

391

ARTICLES
Figure 4 Cocaine dose-response function. Between-sessions dose-response function and effects of substituting saline for cocaine (responding in extinction) in sham-lesioned and NAc core- and shell-lesioned rats. Error bars, ±s.e.m. *P < 0.05.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

(Fig. 7b). Two-way ANOVA with lesion as the between-group factor and cocaine state (pre-cocaine versus post-cocaine) as the repeated measure showed significant main effects of group (F2,280 = 3.26, P < 0.04) and state (F1,277 = 15.36, P < 0.0001) but no interaction between these factors (F2,277 = 2.56, P < 0.08, n.s.) in the mean PCRP duration of the first 200 responses under the FR10(FR10:S) schedule of cocaine reinforcement in the three groups. Post hoc analysis revealed that PCRP pauses of core-lesioned rats in the drug-free state were significantly shorter compared with the shell and sham groups (P < 0.01). In addition, the PCRP duration during responding for the first infusion was significantly higher compared to that during responding under the influence of cocaine for the second infusion, in shell and sham groups, but not in the core-lesioned group. Thus, core-lesioned rats showed significantly shorter PCRP pauses before a cocaine infusion, compared to the sham- and shell-lesioned rats, but the PCRP duration remained unchanged after a cocaine infusion. second-order schedules of reinforcement. Responding on the inactive lever in the shell group, however, was not significantly different from that of the sham group (F1,786 = 1.24, n.s.). Pattern of responding under FR10(FR10:S) The response patterns of the core-lesioned rats differed from those of the sham- and shell-lesioned rats in three main ways (Fig. 6a,b): (i) whereas sham- and shell-lesioned rats showed rapid burst-like patterns of responding on the active lever, core-lesioned rats exhibited temporally dispersed patterns of responding; (ii) core-lesioned rats took much longer to complete the FR10(FR10:S) response requirement before and even after the first cocaine infusion than did sham- and shell-lesioned rats; (iii) the sham- and shell-lesioned rats showed a regular ‘titrating’ pattern of responding, wherein a period of inactivity followed each cocaine infusion (post-reinforcement pause, PRP); no such regular pauses were observed in core-lesioned rats. Post-reinforcement pause The PRP is the period following the reinforcer delivery and before the start of the next ratio of responding, often taken to represent the rewarding impact of the drug (Fig. 7a). ANOVA revealed significant main effects of group (F2,1836 = 569.39, P < 0.0001) and training stage (F4,1824 = 7.83, P < 0.0001) as well as a significant group × training interaction (F8,1824 = 41.64, P < 0.0001) in the mean duration of the PRP within a 2-h self-administration session at each of the different second-order schedule requirements in sham, core- and shelllesioned rats. In both sham- and shell-lesioned groups, the duration of the PRP increased as the response requirement rose at each stage of the second-order schedule. Post-hoc analysis showed the PRP duration in the core-lesioned rats to be significantly shorter than in the sham controls and shell groups at all second-order schedules of cocaine reinforcement tested (P < 0.01). Post-conditioned reinforcement pause The post-conditioned reinforcement pause (PCRP) is the period between the brief presentation of the conditioned reinforcer and the first response made subsequently. It provides a measure of the impact of the conditioned stimulus acting as a conditioned reinforcer Rate of responding on the active lever The effects of NAc lesions on the mean rate of responding before and after the first cocaine infusion during three sessions under the FR10(FR10:S) schedule were also investigated (Fig. 7c). Two-way ANOVA showed significant main effects of group (F2,28 = 10.3, P < 0.0001) and cocaine state (F1,28 = 103.2, P < 0.0001), and significant interaction between these factors (F2,28 = 7.6, P < 0.002). Separate between-subject one-way ANOVAs revealed that the rate of responding during the pre-cocaine period was not significantly different between the lesion groups (F2,29 = 2.42, P = 0.11), whereas it was for the post-cocaine state (F2,29 = 12.08, P < 0.0001). Separate within-subject one-way ANOVAs showed the difference in the rate of responding pre- and post-cocaine to be significant only for the core and sham groups (core, F1,5 = 14.85, P < 0.006; sham, F1,15 = 303, P < 0.0001; shell, F1,5 = 4.36, P = 0.10, n.s.). Therefore, significant cocaine-induced increases in response rate were evident in sham and core-lesioned, but not in shell-lesioned rats. In summary, core-lesioned rats were profoundly impaired in the acquisition of drug cue–controlled cocaine-seeking behavior. Only half of the core-lesioned group completed the training, and the overall number of responses made on the active lever as well as the rates of responding were significantly reduced at all stages of training under the second-order schedule in these rats, as compared to shamlesioned rats. Moreover, the pattern of responding of core-lesioned rats showed a marked absence of PRPs and significantly shorter PCRP pauses compared to the sham rats. In contrast, lesions of the NAc shell did not significantly affect the acquisition of cocaine-seeking behavior. However, shell-lesioned rats showed significantly smaller increases in the rate of responding in the periods after cocaine intake, as compared to sham- and core-lesioned rats. Cocaine-induced locomotor activity We also recorded locomotor activity during cocaine self-administration sessions under the FR1 schedule on days 1–3 and 10 (Fig. 8). Two-way ANOVA revealed a significant main effect of group (F2,17 = 4.75, P < 0.02) but no significant interaction between the day of acquisition and group. A subsequent post-hoc repeated measures ANOVA showed that cocaine-induced activity measures of the

392

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 5 Acquisition of cocaine self-administration under the second-order schedule. (a) The proportion of sham, core and shell-lesioned rats attaining criterion at each successive stage of acquisition of the second-order schedule of cocaine reinforcement. Abbreviations: yS = FR10(FRy:S). The criterion was set as obtaining at least ten cocaine infusions within a 2-h self-administration session for three consecutive days. (b) Mean (±s.e.m.) responses on the active and inactive lever at each stage of the second-order schedule of intravenous cocaine reinforcement in rats with lesions of the NAc core and shell.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

core-lesioned rats (P < 0.023), but not shell-lesioned rats, were significantly higher than those of the shams. DISCUSSION NAc core and cocaine-seeking The present data provide evidence for an important role of the core region of the nucleus accumbens in cocaine-seeking behavior and help to resolve previous data showing only variable effects of lesions of this structure on drug self-administration behavior. NAc corelesioned rats were impaired in the conditioned control over cocaineseeking, similar to the disruptive effects of NAc core lesions on Pavlovian approach behavior35, Pavlovian-to-instrumental transfer36 and conditioned reinforcement38. The results are also consistent with our previous finding that lesions of the NAc core impair the capacity of a food-related conditioned reinforcer to acquire discriminative control over instrumental behavior38. Taken together with the evidence that lesions of the basolateral amygdala (BLA) also impair the ability of a food-associated conditioned reinforcer to support the acquisition of a new instrumental response40,41, the NAc core may also be a component of the neural circuitry involved in behavioral selection based on reward-related information derived from conditioned reinforcers, mediated via amygdaloid or other limbic cortical afferents to the NAc34. Indeed, selective lesions of the BLA produce a similar pattern of relative sparing of continuously reinforced cocaine self-administration, but a failure to acquire drug cue–controlled cocaine-seeking behavior26. The deficits observed in cocaine-seeking by core-lesioned rats were unlikely to have been due to an inability to form effective stimulusreward associations, as post-training NAc core lesions also impair performance under a second-order schedule of cocaine reinforcement (data not shown), indicating that the deficits observed in the present study are not specific to the acquisition of this behavior. It is important to emphasize that core-lesioned rats in the present study were able to acquire instrumental responding for cocaine under a continuous reinforcement schedule, and indeed they responded more than controls under certain conditions. Similarly, NAc corelesioned rats are also able to learn to respond for food and intravenous heroin15,16,36, as well as to adapt their responding to changes in the dose of cocaine or heroin15 (and present results). These findings indicate intact learning of instrumental action-outcome contingencies and a largely intact efficacy of primary reinforcement, whether drug or food. The lack of a rightward shift in the doseresponse curve is also strong evidence against the simple hypothesis that acquisition impairments in cocaine-seeking are primarily due to an attenuation in drug reinforcement. Nevertheless, there was some evidence of elevated response rates on both the active and inactive lever under continuous reinforcement, which could be taken to indicate a mild attenuation of cocaine reinforcement. However, this generally enhanced operant output could also be attributed to the locomotor hyperactivity known to result from core lesions38,42. The lesion-induced hyperactivity may also contribute to the persistence of

responding in extinction frequently seen in core-lesioned rats15,43,44. However, it is important to note that the hyperactive model of behavior produced by core lesions could not in itself account for the pattern of effects seen in the drug-taking (where responding increased) and drug-seeking (where responding decreased) paradigms, nor could it account for the differential responding on the active and inactive levers in both situations. NAc shell and cocaine-seeking Compared to NAc core lesions, shell lesions had no major effects on the acquisition of cocaine-seeking behavior or self-administration under continuous reinforcement. Only at the most stringent stage of the second-order schedule was there an indication of significantly reduced responding on the active lever. However, there were no differences in the rate of responding in the period prior to cocaine infusion. Therefore, this effect can be entirely attributed to an atten-

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

393

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 6 Effects of NAc lesions on qualitative measures of cocaine selfadministration under the second-order schedule. Representative individual records of (a) cumulative responses and (b) responses in sham, core and shell-lesioned rats under the second-order FR10(FR10:S) schedule of cocaine reinforcement. (a) The triangles represent the points at which the rat obtained a cocaine infusion (0.25 mg/infusion) and the circles represent the presentation of a light CS that was contingent on every tenth lever press. (b) The top panel shows individual records of the first 60 min of a 2-h session, with the arrows depicting the point of cocaine infusion. This overall response record is then truncated into response records for (i) the period up to the first cocaine infusion (middle panel; a period of responding completely unaffected by any selfadministered cocaine) and (ii) the period between the first and second infusion (bottom panel), for each lesion group.

Figure 7 Effects of NAc lesions on quantitative measures of cocaine self-administration under the second-order schedule. (a) Mean duration of PRPs within a 2-h session, during each stage of second-order schedules of cocaine reinforcement. **P < 0.01. Abbreviations: yS = FR10(FRy:S). (b) Mean duration of post-conditioned reinforcement pause (PCRP) during the first 100 responses in a drug-free state (pre-cocaine) and the next 100 responses following the first cocaine infusion (post-cocaine), under the FR10(FR10:S) schedule of reinforcement in sham, core and shell-lesioned rats. **P < 0.01, compared to pre-cocaine value, ++P < 0.01, compared to sham. (c) Mean rate of responding before (pre-cocaine) and after (post-cocaine) the first cocaine infusion of self-administration under the FR10(FR10:S) schedule of reinforcement in core, shell and sham-lesioned rats. **P < 0.01 compared to pre-cocaine rate of responding.

uation of the effect of cocaine itself to increase rates of responding under second-order schedules18,22. A similar, though quantitatively larger, effect of similar NAc shell lesions on the potentiation of responding with conditioned reinforcement by amphetamine as well as other psychomotor stimulant manifestations of the drug, such as locomotor hyperactivity, has previously been reported38. Furthermore, infusions of amphetamine into the NAc shell enhance the ability of non-contingently presented Pavlovian cues to potentiate instrumental lever pressing39. Thus, these data, together with those from the present study, show that the caudomedial NAc shell is not critical for the primary reinforcing effects of cocaine, but it is essential for the invigorating effect of stimulant drugs on condi-

tioned and unconditioned behavioral responses—that is, for their psychomotor stimulant actions. Theoretical implications The deficits observed in core-lesioned rats most likely reflect a loss of the impact of conditioned reinforcement on cocaine-seeking behavior. Highly relevant to such an interpretation is the observation that lesions of the NAc core induce persistent impulsive choice behavior as evidenced by an inability to tolerate delays of food reinforcement45. It should be noted that a second-order schedule of reinforcement necessarily introduces a delay to primary reinforcement, but conditioned reinforcers normally help to mediate this delay by maintaining

394

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
21 °C under a reversed 12-h light/dark cycle (lights off at 09:00). Food (laboratory chow, Purina) and water were available ad libitum but, after recovery from surgery, food was restricted to 20 g of lab chow per day, sufficient to maintain pre-operative body weight and growth. All experiments were carried out during the dark phase, between 09:00 and 18:00 and in accordance with the United Kingdom 1986 Animals (Scientific Procedures) Act Project License No. 80/1324. Surgery. In all surgical procedures, animals were anesthetized with Avertin (10 g of 99% 2,2,2-tribromoethanol (Sigma-Aldrich) in 5 mg tertiary amyl alcohol and 4.5 ml phosphate buffered saline (Dulbecco “A”, Unipath Ltd.) in 40 ml absolute alcohol; 1 ml/100 g body weight, intraperitoneally (i.p.)).
Figure 8 Effects of NAc lesions on cocaine-induced locomotor activity during self-administration sessions. Vertical bars represent the mean ± s.e.m. photocell beam breaks for each group during a 2-h selfadministration session on the first three and last days of acquisition. **P < 0.01, *P < 0.05 compared to the sham.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

instrumental responding34. As core-lesioned rats showed no deficits when responding for cocaine under a continuous reinforcement schedule, the impaired acquisition of cocaine-seeking behavior reported here may depend on an inability of core-lesioned rats to tolerate a delay in cocaine reinforcement. We infer that this reflects a loss of impact of cocaine-associated conditioned reinforcers. The similar effects of BLA lesions26 may implicate both of these structures in a neural system by which conditioned associations help to mediate delays between responses and outcomes. The integrity of the NAc core, unlike that of the caudomedial NAc shell, is not required for the response rate–increasing effects of psychostimulant drugs38. These data thus re-open the issue of the relationship between the stimulant and reinforcing effects of psychomotor stimulants such as cocaine, which have been suggested to be isomorphic46. However, the doubly dissociable effects reported here indicate that the NAc core and shell mediate distinct processes associated with cocaine self-administration behavior. The core seems to mediate control by conditioned reinforcers, whereas the shell seems to mediate the potentiation of that control by cocaine, perhaps reflecting stimulant or motivational effects of the drug. The additional importance of this double dissociation is in demonstrating a selective functional effect produced by the relatively discrete shell lesion (restricted to its caudomedial domain), that contrasts with the qualitatively distinct nature of the deficits produced by the more complete lesion of the nucleus accumbens core. It is significant that this region of the NAc shell is most often implicated in mediating both the reinforcing and response-invigorating effects of psychomotor stimulant drugs30–33,38. How the NAc core and shell interact remains unclear, but recent anatomical evidence suggests that the striatum is organized in a hierarchical fashion, with the shell and its limbic connections capable of influencing the behavioral output of the core, via ‘spiralling’ connections with the midbrain dopamine neurons47. The NAc shell can thus serve to amplify the expression in behavior of information flowing through the NAc core17. Such anatomical relationships may also underlie different factors affecting intravenous drug self-administration behavior. These data emphasize the complex nature of cocaine reinforcement mechanisms, while specifying a particular role for the NAc core subregion that has hitherto not been apparent. METHODS
Animals. Male Lister Hooded rats (Charles River Ltd., UK) were housed in pairs and then individually after surgery, in a room held at a temperature of

Excitotoxic lesions. We used different excitotoxins to damage the core or shell sub-regions of the NAc (quinolinic acid for core; ibotenic acid for shell). This produced selective lesions of these structures with little, if any, overlap between them38. A 1-µl SGE syringe (SGE) was lowered stereotaxically into either the NAc core or shell, and the neurotoxin was infused bilaterally. For NAc core lesions, 0.3 µl of 0.09 M quinolinic acid (Sigma-Aldrich) buffered to pH 7.3–7.4 in 0.1 M sterile phosphate buffer (sterile PB), was infused for 1 min in each hemisphere, using the following coordinates (in mm from bregma); AP: +1.2, L: ±1.8, DV: –7.1 from the skull surface (SS). For NAc shell lesions, three separate infusions of 0.06 M ibotenic acid (Sigma-Aldrich) buffered to pH 7.4 using 0.1 M sterile PB were made at different points along the DV axis in each hemisphere: (i) 0.2 µl at AP = +1.6, L = ±1.1, DV = –7.9 (SS); (ii) 0.1 µl at AP = +1.6, L = ±1.1, DV = –6,9; (iii) 0.1 µl at AP = +1.6, L = ±1.1, DV = –6.4. Sham and lesion groups were treated identically, except that sham controls received injections of sterile PB instead of the toxin. Intravenous catheterization. After a recovery period of at least 5 d with food available ad libitum, rats were then implanted with chronic intravenous jugular catheters as previously described48. Antibiotic treatment (daily subcutaneous administration of 0.1 ml Baytrill; Bayer) was given for 5 d after surgery. Thereafter, before each self-administration session, the animals were flushed with 0.1 ml sterile 0.9% saline and at the end of the session with 0.1 ml heparinised saline (CP Pharmaceuticals Ltd.; 30 units/ml 0.9% sterile saline) to maintain catheter patency. Apparatus. Twelve operant chambers (24 cm wide × 20 cm high × 22 cm deep; Med Associates) contained within a sound-attenuating box with a ventilating fan were used in the experiment. Each chamber was equipped with two retractable levers, a stimulus light above each lever, a house light and three infrared beams (See Supplementary Methods online). Intravenous infusions of cocaine were delivered by a software-operated infusion pump (Semat Technical Ltd.) placed outside the sound-attenuating box, through a counterbalanced single-channel liquid swivel. Animals were tethered to the counterbalanced arm by a metal spring and a skull-mounted plastic-post. The apparatus was controlled by an Acorn Archimedes microcomputer (Acorn Computers Ltd.) running a program written in the BASIC control language, Arachnid (Paul Fray Ltd.). Drugs. Cocaine hydrochloride (McFarlan-Smith) was dissolved in sterile 0.9% saline. The dose of cocaine was calculated as the salt. Behavioral procedures. Cocaine self-administration under continuous reinforcement: Animals were trained to acquire cocaine self-administration under a continuous reinforcement schedule (fixed ratio1) during daily 2-h sessions until stable baseline responding was achieved (defined as 2–3 consecutive d of stable responding with as much as ±10% variation). For each rat, one of the two levers was designated the active or drug lever, and the other was the inactive lever on which responding had no programmed consequence. No drug priming was given at any stage of training. The beginning of the session was marked by illumination of the house light. Subsequent depression of the active lever resulted in the retraction of both levers, extinction of the house light and simultaneous illumination of the drug stimulus light for 20 s, as well as the activation of an infusion pump,

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

395

ARTICLES
delivering 0.1 ml intravenous infusion of cocaine solution (0.25 mg/infusion). On completion of the 20-s CS presentation per time out period, the levers were re-extended, the house light illuminated and the stimulus light extinguished. Throughout training, the maximum number of infusions per session was fixed at 50 to prevent overdosing. Once this number of cocaine infusions had been reached, the session terminated. Second-order schedule of cocaine reinforcement: Once self-administration under a continuous reinforcement schedule had been attained, a second-order FRx(FRy:S) schedule of cocaine reinforcement was introduced. Under this schedule, rats were required to make y responses to obtain a single presentation of a 2 s light CS (or conditioned reinforcer) while completion of x of these response units resulted in the delivery of cocaine, the illumination of the light CS for 20 s, the retraction of both levers and extinction of the house light during a 20 s time out period. In the initial stage of training, x was set at 5 and y was 1. The value for x was then increased from 5 to 10 and remained at this value throughout the training. The value for y was progressively increased from 1 to 10 until stable responding was established at FR10(FR10:S). Animals were allowed to move from one stage to the next when at least ten cocaine infusions within a 2-h session were made over three consecutive days at each stage. Cocaine dose-response function. A separate group of 18 rats (4 or 5 per treatment group) was subjected to a between-sessions cocaine dose-response function once stable acquisition of cocaine self-administration under a CRF schedule at the training dose of 0.25 mg/infusion had been attained. The training dose was substituted by 0.083, 0.125 or 0.50 mg per infusion of cocaine or saline in a 2-h self-administration session on five consecutive days, and the order in which the rats received each dose was counterbalanced. Histological assessment of lesions. Within a week after completion of the testing, all rats were anesthetized with sodium pentobarbitone (1.5 ml/animal, 200 mg/ml Euthatal, Rhone Merieux) and perfused intracardially via the ascending aorta with 0.01 M phosphate-buffered saline (PBS) for 4 min, followed by 4% paraformaldehyde (PFA) in PBS for 6 min. Brains were then removed, stored in PFA and transferred to a 20% sucrose cryoprotectant solution on the day before sectioning (See Supplementary Methods online). For the verification of lesions, coronal sections (60 µm) of the brain were cut using a freezing microtome. Statistical analysis. All behavioral data were analyzed using SPSS version 9. Responses during the acquisition of self-administration under CRF and second-order schedules were square-root transformed to preserve homogeneity of variance and analyzed using a three-factor analysis of variance (ANOVA) with group (core, shell, sham) as the between-group factor and training day and lever (active vs. inactive) as repeated, within-subjects factors. Data collected from the seven core-lesioned animals that failed to complete the second-order schedule training were included in the statistical analyses, as their exclusion did not alter statistical significance of the overall data. Two-way ANOVAs (lesion group as between-subjects factor and training schedule or cocaine state as repeated measures) were conducted for all quantitative data extracted from the response patterns for the second-order schedule contingencies. For all analyses, upon confirmation of significant main effects, differences among individual means were analyzed using the Newman-Keuls post-hoc test. Significant interactions were further analyzed as appropriate using two-way or one-way ANOVAs, with appropriate α adjustments using Sidak’s method (α′ = 1 – (1 – α)1/C, where C is the number of within-experiment analyses).
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS This work was supported by an MRC Programme Grant (G9537855) and conducted within the MRC Centre for Behavioural and Clinical Neuroscience. R.I. was supported by an MRC research studentship. We thank D. Eagle, R. Cardinal and M. Aitken for helpful discussions on statistics. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 16 January; accepted 3 March 2004 Published online at http://www.nature.com/natureneuroscience/
1. Roberts, D.C.S., Corcoran, M.E. & Fibiger, H.C. On the role of ascending catecholaminergic systems in intravenous self-administration of cocaine. Pharmacol. Biochem. Behav. 6, 615–620 (1977). 2. Pettit, H.O., Ettenberg, A., Bloom, F.E. & Koob, G.F. Destruction of dopamine in the nucleus accumbens selectively attenuates cocaine but not heroin self-administration in rats. Psychopharmacology 84, 167–173 (1984). 3. Caine, S.B. & Koob, G.F. Effects of mesolimbic dopamine depletion on responding maintained by cocaine and food. J. Exp. Anal. Behav. 61, 213–221 (1994). 4. Robledo, P., Maldonado-Lopez, R. & Koob, G.F. Role of dopamine receptors in the nucleus accumbens in the rewarding properties of cocaine. Ann. NY Acad. Sci. 654, 509–512 (1992). 5. Caine, S.B., Heinrichs, S.C., Coffin, V.L. & Koob, G.F. Effects of the dopamine D1 antagonist SCH23390 microinjected into the accumbens, amygdala or striatum on cocaine self-administration in the rat. Brain Res. 692, 47–56 (1995). 6. Di Chiara, G.D. & Imperato, A. Drugs abused by humans preferentially increase synaptic dopamine concentrations in the mesolimbic system of freely moving rats. Proc. Natl. Acad. Sci. USA 85, 5274–5278 (1988). 7. Hurd, Y., Weiss, F., Koob, G.F. & Ungerstedt, U. Cocaine reinforcement and extracellular DA overflow in rat nucleus accumbens: an in vivo microdialysis study. Brain Res. 498, 199–204 (1989). 8. Pettit, H.O. & Justice Jr, J.B. Dopamine in the nucleus accumbens during cocaine self-administration as studies by in vivo microdialysis. Pharmacol. Biochem. Behav. 34, 899–904 (1989). 9. Bradberry, C.W., Barrett-Larimore, R.L., Jatlow, P. & Rubino, S.R. Impact of selfadministered cocaine and cocaine cues on extracellular dopamine in mesolimbic and sensorimotor striatum in rhesus monkeys. J. Neurosci. 20, 3874–3883 (2000). 10. Ito, R., Dalley, J.W., Howes, S.R., Robbins, T.W. & Everitt, B.J. Dissociation in conditioned dopamine release in the nucleus accumbens core and shell in response to cocaine cues and during cocaine-seeking behavior in rats. J. Neurosci. 20, 7489–7495 (2000). 11. Pulvirenti, L., Maldonado-Lopez, R. & Koob, G.F. NMDA receptors in the nucleus accumbens modulate intravenous cocaine but not heroin self-administration in the rat. Brain Res. 594, 327–330 (1992) 12. Cornish, J.L., Duffy, P. & Kalivas, P.W. A role for glutamate transmission in the relapse to cocaine-seeking behavior. Neuroscience 93, 1359–1367 (1999) 13. Zito, K.A., Vickers, G. & Roberts, D.C.S. Disruption of cocaine and heroin selfadministration following kainic acid lesions of the nucleus accumbens. Pharmacol. Biochem. Behav. 23, 1029–1036 (1985). 14. Dworkin, S.I., Guerin, G.F., Goeders, N.E. & Smith, J.E. Kainic acid lesions of the nucleus accumbens selectively attenuate morphine self-administration. Pharmacol. Biochem. Behav. 29, 175–181 (1988). 15. Alderson, H.L., Parkinson, J.A., Robbins, T.W. & Everitt, B.J. The effects of excitotoxic lesions of the nucleus accumbens core or shell regions on intravenous heroin self-administration in rats. Psychopharmacology 153, 455–463 (2001). 16. Hutcheson, D.M., Parkinson, J.A., Robbins, T.W. & Everitt, B.J. The effects of nucleus accumbens core and shell lesions on intravenous heroin self-administration and the acquisition of drug-seeking behaviour under a second-order schedule of heroin reinforcement. Psychopharmacology 153, 464–472 (2001). 17. Everitt, B.J., Dickinson, A. & Robbins, T.W. The neuropsychological basis of addictive behaviour. Brain Res. Rev. 36, 129–138 (2001). 18. Robbins, T.W. & Everitt, B.J. Drug addiction: bad habits add up. Nature 398, 567–570 (1999). 19. Grant, S. et al. Activation of memory circuits during cue-elicited cocaine craving. Proc. Nat. Acad. Sci. USA 93, 12040–12045 (1996). 20. Childress, A.R. et al. Limbic activation during cue-induced cocaine craving. Am. J. Psychiatry 156, 11–18 (1999). 21. Stewart, J., de Wit, H. & Eikelboom, R. Role of unconditioned and conditioned drug effects in the self-administration of opiates and stimulants. Psychol. Rev. 91, 256–268 (1984). 22. Arroyo, M., Markou, A., Robbins, T.W. & Everitt, B.J. Acquisition, maintenance, and reinstatement of intravenous cocaine self-administration under a second-order schedule of reinforcement in rats: effects of conditioned cues and continuous access to cocaine. Psychopharmacology 140, 331–344 (1998). 23. Everitt, B.J. & Robbins, T.W. Second-order schedules of drug reinforcement in rats and monkeys: measurement of reinforcing efficacy and drug-seeking behaviour. Psychopharmacology 153, 17–30 (2000). 24. Weiss, F. et al. Control of cocaine-seeking behavior by drug-associated stimuli in rats: Effects on recovery of extinguished operant-responding and extracellular dopamine levels in amygdala and nucleus accumbens. Proc. Natl. Acad. Sci. USA 97, 4321–4326 (2000). 25. See, R.E. Neural substrates of conditioned-cued relapse to drug-seeking behavior. Pharmacol. Biochem. Behav. 71, 517–529 (2002). 26. 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 127, 213–224 (1996). 27. Meil, W.M. & See, R.E. Lesions of the basolateral amygdala abolish the ability of drug associated cues to reinstate responding during withdrawal from self-administered cocaine. Behav. Brain Res. 87, 139–148 (1997).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

396

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
28. Voorn, P., Gerfen, C.R. & Groenewegen, H.J. Compartmental organization of the ventral striatum of the rat: Immunohistochemical distribution of enkephalin, substance P, dopamine and calcium-binding protein. Comp. Neurol. 289, 189–201 (1989). 29. Heimer, L., Zahm, D.S., Churchill, L., Kalivas, P.W. & Wohltmann, C. Specificity in the projection patterns of accumbal core and shell in the rat. Neuroscience 41, 89–125 (1991). 30. Di Chiara G, Tanda G, Frau, R. & Carboni E. On the preferential release of dopamine in the nucleus accumbens by amphetamine: further evidence obtained by vertically implanted concentric dialysis probes. Psychopharmacology 112, 398–402 (1993). 31. Pontieri, F.E., Tanda, G. & Di Chiara, G. Intravenous cocaine, morphine, and amphetamine preferentially increase extracellular dopamine in the “shell” as compared with the “core” of the rat nucleus accumbens. Proc. Natl. Acad. Sci. USA 92, 12304–12308 (1995). 32. Carlezon, W.A. & Wise, R.A. Rewarding actions of phencyclidine and related drugs in nucleus accumbens shell and frontal cortex. J. Neurosci. 16, 3112–3122 (1996). 33. Rodd-Hendricks, Z.A., McKinzie, D.L., Li, T-K., Murphy, J.M. & McBride, W.J. Cocaine is self-administered into the shell but not the core of the nucleus accumbens of Wistar rats. J. Pharmacol. Exp. Ther. 303, 1216–1226 (2002). 34. Cardinal, R.N., Parkinson, J.A., Hall, J. & Everitt, B.J. Emotion and motivation: the role of the amygdala, ventral striatum, and prefrontal cortex. Neurosci. Biobehav. Rev. 26, 321–352 (2002). 35. Parkinson, J.A., Willoughby, P.J., Robbins, T.W. & Everitt, B.J. Disconnection of the anterior cingulate cortex and nucleus accumbens core impairs Pavlovian approach behaviour: Further evidence for limbic cortical-ventral striatopallidal systems. Behav. Neurosci. 114, 42–63 (2000). 36. Hall, J., Parkinson, J.A., Connor, T.M., Dickinson, A. & Everitt, B.J. Involvement of the central nucleus of the amygdala and nucleus accumbens core in mediating Pavlovian influences on instrumental behaviour. Eur. J. Neurosci. 13, 1984–1992 (2001). 37. Dalley, J.W. et al. Nucleus accumbens dopamine and discriminated approach learning: interactive effects of 6-hydroxydopamine lesions and systemic apomorphine administration. Psychopharmacology 161, 425–433 (2002). 38. Parkinson, J.A., Olmstead, M.C., Burns, L.H., Robbins, T.W. & Everitt, B.J. Dissociation in effects of lesions of the nucleus accumbens core and shell on appetitive Pavlovian approach behavior and the potentiation of conditioned reinforcement and locomotor activity by d-amphetamine. J. Neurosci. 19, 2401–2411 (1999). 39. Wyvell, C.L. & Berridge, K.C. Intra-accumbens amphetamine increases the conditioned incentive salience of sucrose reward: enhancement of reward ‘Wanting’ without enhanced ‘Liking’ or response reinforcement. J. Neurosci. 20, 8122–8130 (2000). 40. Cador, M., Robbins, T.W. & Everitt, B.J. Involvement of the amygdala in stimulus reward associations- interaction with the ventral striatum. Neuroscience 30, 77–86 (1989). 41. 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). 42. Maldonado-Irizarry, C.S. & Kelley, A.E. Excitotoxic lesions of the core and shell subregions of the nucleus accumbens differentially disrupt body weight regulation and motor activity in rat. Brain Res. Bull. 38, 551–559 (1995). 43. Annett, L.E., McGregor, A. & Robbins, T.W. The effects of ibotenic acid lesions of the nucleus accumbens on spatial learning and extinction in the rat. Behav. Brain. Res. 31, 231–242 (1989). 44. Reading, P.J. & Dunnett, S.B. The effects of excitotoxic lesions of the nucleus accumbens on a matching to position task. Behav. Brain Res. 46, 17–29 (1991). 45. Cardinal, R.N., Pennicott, D.R., Sugathapala, C.L., Robbins, T.W. & Everitt, B.J. Impulsive choice induced in rats by lesions of the nucleus accumbens core. Science 292, 2499–2501 (2001). 46. Wise, R.A. & Bozarth, M.A. A psychomotor stimulant theory of addiction. Psychol. Rev. 94, 469–492 (1987). 47. Haber, S.N., Fudge, J.L. & McFarland, N.R. Striatonigrostriatal pathways in primates from an ascending spiral from the shell to the dorsolateral striatum. J. Neurosci. 20, 2369–2382 (2000). 48. Caine, S.B., Lintz, R. & Koob, G.F. Intravenous self-administration techniques in animals. in Behavioral Neuroscience: a Practical Approach Vol. 2 (ed. Sahgal, A.) 117–143 (IRL Press, Oxford, UK, 1992). 49. Paxinos, G. & Watson, C. The Rat Brain in Stereotaxic Coordinates (Academic Press, San Diego, 1998).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

397

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Glutamatergic activation of anterior cingulate cortex produces an aversive teaching signal
Joshua P Johansen1–3 & Howard L Fields1,2
Noxious stimuli have motivational power and can support associative learning, but the neural circuitry mediating such avoidance learning is poorly understood. The anterior cingulate cortex (ACC) is implicated in the affective response to noxious stimuli and the motivational properties of conditioned stimuli that predict noxious stimulation. Using conditioned place aversion (CPA) in rats, we found that excitatory amino acid microinjection into the ACC during conditioning produces avoidance learning in the absence of a peripheral noxious stimulus. Furthermore, microinjection of an excitatory amino acid antagonist into the ACC during conditioning blocked learning elicited by a noxious stimulus. ACC lesions made after conditioning did not impair expression of CPA. Thus, ACC neuronal activity is necessary and sufficient for noxious stimuli to produce an aversive teaching signal. Our results support the idea that a shared ACC pathway mediates both pain-induced negative affect and a nociceptor-driven aversive teaching signal.

Pairing a stimulus with intrinsic motivational power (unconditioned stimulus, US; e.g., pain, food) with a neutral sensory stimulus produces changes in neural circuitry such that the previously neutral stimulus becomes capable of generating behavioral responses (conditioned responses, CRs). Thus the previously neutral stimulus becomes a conditioned stimulus (CS). We use the term ‘teaching signal’ to refer to a neural signal that is necessary and sufficient to produce a conditioned response (CR). Accordingly, temporally coincident activation of the pathways transmitting the aversive teaching signal and the initially neutral stimulus produces aversive associative learning by strengthening the connections between the neurons mediating the CS and those whose activity results in the CR. Associative learning using noxious stimuli as the US has been documented behaviorally, and the mediating synaptic change has been elucidated in some systems1–4. However, the neural pathways that mediate nociceptor-driven aversive teaching signals in mammals are not well understood. One candidate pathway for such signals includes projections from the spinal cord dorsal horn to the medial thalamus, and from there to the ACC5–8. This pathway was established by functional imaging studies in humans9 and anatomical and electrophysiological studies in animals10–15. In human imaging studies, the degree of ACC activation is positively correlated with the magnitude of unpleasantness in response to a noxious stimulus16. In addition, the human, primate, rodent and rabbit ACCs contain neurons that respond to noxious stimuli13,17–19. In chronic pain patients, lesions of the ACC or cingulum bundle (an afferent and efferent ACC fiber tract) reduce pain unpleasantness20,21. The ACC has extensive direct interconnections with limbic nuclei including the amygdala, hippocampus, posterior cingulate and ventral striatum22–25, each of which has been implicated in CS-driven aversive behaviors2,26,27.

The observations that ACC neurons respond to noxious stimulation and that ACC activity is correlated with perceived unpleasantness in humans are consistent with the hypothesis that ACC neurons encode and transmit information related to the aversiveness of noxious stimuli and provide the teaching signal required for the acquisition of conditioned aversion. We recently found that excitotoxic lesions of the rostral ACC (r-ACC) selectively prevents avoidance learning elicited by tonic noxious stimuli28. This is consistent with reports that lesions of frontal cortex or of both anterior and posterior cingulate cortices prior to conditioning reduce avoidance learning26,29. Although these studies show that the ACC is required for aversive learning, they do not distinguish between a role for ACC neurons in its acquisition (i.e., in providing an aversive teaching signal) versus expression (i.e., in retrieval). Furthermore, the electrophysiological data are ambiguous on this question. In addition to responding to noxious stimuli (essential if they are to provide an aversive teaching signal and contribute to acquisition of the CR), some ACC neurons respond to pain-predictive sensory stimuli. For example, human imaging and rodent, rabbit and primate electrophysiology studies show activation of ACC neurons in response to a pain-predictive visual CS18,26,30–32. This activation supports the idea that ACC neurons encode and transmit information that generates the motivational properties of the CS after conditioning, rather than generating an aversive teaching signal during learning. In other words, this pattern of activity is more consistent with a role for ACC neurons in the expression rather than the acquisition of learned aversive behaviors. Finally, the facts that neurons responding to both nociceptive (US) and aversive CS are found in the ACC and that, after learning, some ACC neurons respond to both types of stimuli18 raise the intriguing possibility that the ACC is a critical site of plasticity for avoidance learning.

1Departments of Neurology and Physiology and 2The W.M. Keck Foundation Center for Integrative Neuroscience, University of California, San Francisco, 513 Parnassus Avenue, S-784, San Francisco, California 94143-0453, USA. 3Present address: Interdepartmental Ph.D. Program for Neuroscience, UCLA, Los Angeles, California 90095, USA. Correspondence should be addressed to H.L.F. ([email protected]).

Published online 7 March 2004; corrected 12 March 2004 (details online); doi:10.1038/nn1207

398

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Bilateral infusions of the excitotoxin ibotenic acid (IBO) made into r-ACC produced neuronal cell loss and proliferation of small glial cells (data not shown; see ref. 28). All animals included in our analyses met lesion inclusion criterion as described in Methods (Fig. 1a). Mean percent damage calculations for each hemisphere and an overall bilateral mean are as follows: left hemisphere, 66 ± 11%; right hemisphere, 58 ± 11%; mean, 62 ± 9%. Importantly, the lesion extents in this experiment were not different from those in our previous study28. When hindpaw formalin injections were paired with a particular compartment in the place-conditioning apparatus, rats with posttraining r-ACC sham lesions spent less time in the formalin-paired room (i.e., CPA was Figure 1 ACC lesions after training do not affect expression of place aversion. (a) Examples of the largest produced; 389.8 ± 54.8 s pre-conditioning vs. (gray) and smallest (black) lesions among animals in the group. Sections are in the coronal plane, 211.6 ± 90.2 s post-conditioning; Student’s numbers in mm anterior to Bregma in this and subsequent figures. (b) Rats with post-training lesions t-test, P < 0.05). Hindpaw formalin also pro(n = 7) did not differ from those with sham lesions (n = 10). F-CPA scores are shown as mean ± s.e.m. duced CPA in post-training r-ACC lesioned rats (392 ± 131.6 s pre-conditioning vs. 184 ± 94.9 s post-conditioning; Student’s While not mutually exclusive, these hypotheses lead to clearly t-test, P < 0.05). Group comparisons revealed no significant differdifferent predictions of the effect of ACC manipulations on the ence between sham and lesion groups (Fig. 1b; Student’s t-test, acquisition and expression of avoidance learning. Thus, if ACC neu- P > 0.05). Thus, r-ACC lesions made after training have no effect on rons are required to mediate the motivational effect of aversive con- the expression of F-CPA. Two critical conclusions can be drawn from ditioned stimuli, then lesions after conditioning should block the this result. First, the r-ACC is not a significant site of plasticity for expression of avoidance learning. In contrast, if ACC neurons are F-CPA learning and, second, it is not required for retrieval of infornecessary to provide a nociceptive aversive teaching signal, then mation related to the prediction of aversive stimuli by contextual cues. ACC lesions before conditioning (or reversible inactivation during conditioning) should block acquisition, but lesions made after con- r-ACC glutamate receptor blockade prevents F-CPA acquisition ditioning should not affect expression of CPA after it has been The fact that lesions made before28 but not after conditioning block learned. Furthermore, if ACC neuronal activity is sufficient to pro- F-CPA learning strongly supports the hypothesis that the r-ACC is vide an aversive teaching signal, direct activation of these ACC neu- necessary specifically during the acquisition of F-CPA. The existence rons during conditioning, in the absence of a peripheral noxious of a significant spino-thalamo-cingulate nociceptive projection pathstimulus, should produce an aversive teaching signal. Finally, if the way10–15 is also consistent with a major role for the r-ACC in afferent aversive learning is associated with requisite synaptic plasticity nociceptive processing. Assuming that the thalamo-cingulate projecwithin the ACC, lesions before conditioning should block acquisi- tion is glutamatergic33, it is likely that glutamatergic activation of tion, and lesions after conditioning should block expression of r-ACC neurons by a prolonged noxious stimulus is necessary for the avoidance learning. acquisition of CPA. To address this question, we made glutamate We previously examined the functional significance of the ACC receptor antagonist microinjections into the r-ACC during formalin using a nociceptor-driven, associative avoidance-learning assay: conditioning sessions. Microinjections of kynurenic acid (KyA) into the r-ACC before formalin-induced conditioned place aversion (F-CPA)28. However, because the lesions in our earlier study were irreversible and made F-CPA conditioning blocked the acquisition of F-CPA learning before conditioning, one could not distinguish an effect on acquisi- (Fig. 2). There was no difference in the amount of time the r-ACC tion from one on expression. In the current study, to address this KyA animals spent in the formalin-paired context before versus after question, we inactivated or lesioned the ACC in a temporally spe- conditioning (357.5 ± 50.6 s before, 320.6 ± 83.2 s after; Student’s cific manner. In addition, by activating ACC neurons directly, in the t-test, P > 0.05). In contrast, microinjections of vehicle into the absence of a peripheral nociceptive input, we explored whether r-ACC during conditioning had no effect on F-CPA acquisition activity of ACC neurons is sufficient to provide an aversive teaching (388.3 ± 79.4 s before, 234.8 ± 106.5 s after conditioning; Student’s signal. Our results provide direct evidence that ACC neuronal activ- t-test, P < 0.01). For group comparisons, see Figure 2a. Notably, in a ity is sufficient to produce avoidance learning and necessary for separate group of animals, KyA alone did not produce motivational noxious stimuli to elicit an aversive teaching signal. effects. KyA microinjected into the r-ACC in the absence of hindpaw formalin had no effect on room preference (336.6 ± 86.3 s before vs. RESULTS 321.3 ± 137.1 s after conditioning; Student’s t-test, P > 0.05; Fig. 2b). r-ACC lesions do not affect the expression of avoidance learning Furthermore, the reduction of F-CPA by KyA injected into the Excitotoxin-induced r-ACC lesions were made after acquisition of the r-ACC is unlikely to be due to a sedating effect since it did not alter conditioned response to test whether the r-ACC is necessary for the motor activity (data not shown). In further support of this concluexpression of previously learned avoidance behavior. sion, our previous study showed that similarly located IBO-induced

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

399

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 2 Intra-r-ACC microinjection of the ionotropic glutamate receptor antagonist kynurenic acid (KyA) blocks F-CPA. (a–c) Data are represented as mean ± s.e.m. (a) The effect of bilateral intra-r-ACC vehicle (n = 10) or KyA (n = 8) on the magnitude of F-CPA scores. (b) The magnitude of CPA scores for intra-r-ACC KyA in the absence of hindpaw formalin (n = 8). (c) Acute formalin-induced nociceptive scores (rating scale). (d) Injection sites for 50 mM KyA-treated rats. *P < 0.05, Student’s t-test as compared with vehicle-injected rats.

r-ACC lesions had no effect on the place aversion elicited by systemic injection of the kappa opioid agonist U69,593 (ref. 28). As in our earlier lesion study, we found no significant main effect of intracerebral treatment (vehicle vs. KyA) on acute formalin rating scale scores (F1,12 = 0.97; P > 0.05) and no significant interaction between intracerebral treatment and time (F9,108 = 0.72, P > 0.05; Fig. 2c), indicating that KyA reduction of F-CPA is not due to a general decrease in nociceptive processing.
Glutamatergic r-ACC stimulation produces avoidance learning The results of experiments 1 and 2 indicate that activation of r-ACC neurons is necessary for acquisition, but not expression, of F-CPA. However, they do not rule out the possibility that nociceptive activation of r-ACC neurons serves a permissive role during conditioning and that activation of r-ACC neurons alone is not sufficient to produce F-CPA learning. To test this possibility, we directly stimulated the r-ACC by microinjecting an ionotropic glutamate receptor agonist into the r-ACC in the absence of a peripheral nociceptive stimulus. Homocysteic acid microinjected into the r-ACC produced significant, dose-dependent CPA learning. Rats spent significantly less time in the treatment-paired context (366.8 ± 44.8 before vs. 251.2 ± 59.4 after conditioning; Student’s t-test, P < 0.01). Neither intra-

r-ACC vehicle nor low-dose homocysteic acid produced CPA (vehicle, 336.1 ± 58.3 s before vs. 308.4 ± 92.2 s after conditioning; Student’s t-test, P > 0.05; low-dose, 352.1 ± 34.6 s before vs. 365.75 ± 69.8 s after conditioning; Student’s t-test, P > 0.05). Group comparisons of magnitude of CPA scores analyzed using a one-way ANOVA revealed a significant effect of treatment (F2,27 = 6.46; P < 0.01). Further analysis revealed significantly higher CPA scores for the 100 mM HCA treatment group compared to the vehicle group, but no significant difference between vehicle and 5 mM HCA (Newman-Keuls test; P < 0.05 and P > 0.05, respectively; Fig. 3a). To establish the anatomical specificity of our r-ACC microinjections, we used off-site controls (Fig. 3b). High-dose HCA had no motivational effects when injected into a cortical control site lateral to our target r-ACC injections (n = 8; 347.4 ± 43.8 s before vs. 356.5 ± 157.6 s after conditioning; Student’s t-test, P > 0.05).

DISCUSSION Previously we demonstrated that excitotoxic lesions of the r-ACC before conditioning abolish nociceptor-driven learned avoidance behavior (FCPA) without affecting acute nociceptive behaviors or non-nociceptive avoidance behavior28. Our current study extends those findings by showing that activation of r-ACC neurons is required specifically for the acquisition of F-CPA, as lesions made after conditioning have no effect on the expression of F-CPA. In addition, our data implicate r-ACC excitatory neurotransmission specifically in the acquisition of F-CPA, as r-ACC microinjection of a glutamate receptor antagonist during acquisition blocks F-CPA conditioning. Importantly, our data provide critical evidence supporting the hypothesis that r-ACC neuronal activity is sufficient to generate an aversive teaching signal. Thus, microinjection of a glutamate receptor agonist into the r-ACC, but not into an adjacent cortical site, during conditioning produces robust CPA in the absence of a concomitant peripheral noxious stimulus. That selective activation of r-ACC neurons is sufficient to produce avoidance learning in the absence of input Figure 3 CPA is produced by glutamatergic stimulation of the r-ACC. (a) Magnitude of CPA scores in from primary afferent nociceptors is direct animals given intra-r-ACC microinjection of vehicle (n = 9), 5 mM (n = 8) or 100 mM HCA (n = 11). evidence that ACC neuronal activity is causal Data are represented as mean ± s.e.m. *P < 0.05, as compared with vehicle injected rats. (b) Injection sites for 100 mM HCA r-ACC (circles) and off-site injection sites (triangles). rather than permissive for avoidance learning.

400

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Although the evidence is not conclusive, a parsimonious explanation of these results is that formalin injection produces an aversive teaching signal through activation of r-ACC neurons during CPA conditioning.
A model: r-ACC pathway encodes an aversive teaching signal Because dilute intradermal formalin selectively activates nociceptive Aδ and C-fiber primary afferent nociceptors34 and is painful in humans35, F-CPA is, by definition, a nociceptor-driven learned behavior. Nociceptive stimuli reliably activate neurons in the ACC13,17–19. Furthermore, together with the fact that inactivation of the medial thalamus13—an area that receives direct and indirect spinal cord projections and projects to the r-ACC10–12,14,15—reduces this activation, the current results strongly support the idea that the ACC is a major terminus or relay site for a nociceptive afferent pathway. Consistent with the idea that this region of the r-ACC contributes selectively to nociceptor-driven aversive processing, we previously showed that CPA produced by the systemic administration of the kappa opioid agonist U69,593 is unaffected by lesions of the r-ACC28. Importantly, this result implies that r-ACC lesions do not produce a general disruption of associative learning. In addition, we and others have shown that lesions of the rostral28 or caudal ACC36,37 (but see also ref. 38) spare other unconditioned behavioral responses (UR) elicited by noxious stimuli. On the other hand, caudal ACC lesions appear to reduce acute escape responses to noxious heat36. The results of the current experiment and previous work thus demonstrate that a nociceptive pathway through the r-ACC is necessary and sufficient for peripheral noxious stimuli to produce aversive teaching signals. The r-ACC is not necessary for other URs to nociceptive stimuli (e.g., acute formalin behaviors). Thus at some point afferent to the r-ACC, the afferent pathway mediating the aversive teaching signal diverges from that mediating many of the acute behavioral responses elicited by noxious stimuli. Our data also indicate that the neural plasticity underlying the development of avoidance learning occurs in areas of the brain that receive convergent input from the r-ACC neurons encoding the aversive teaching signal and from other sensory pathways whose neurons encode information about initially neutral conditioned stimuli. Working within this model, glutamatergic activation of r-ACC neurons by noxious stimuli is necessary to produce F-CPA, and direct activation of r-ACC neurons is sufficient to serve as a teaching signal for this type of avoidance learning. A teaching signal, in this model, serves to strengthen the CS inputs onto neurons that receive convergent input from both nociceptive (teaching input) and other sensory CS pathways. The strengthening of the CS input such that it becomes capable of eliciting the CR is manifest in the current study as the acquisition of CPA. This type of plasticity elicited by a noxious US has been reported at the synaptic level in other neural systems. For example, using in-vivo intracellular recordings, one study demonstrated enhanced synaptic strength of an olfactory input to an amygdala neuron by temporally coincident activation of a noxious stimulus input onto the same neuron4. Interestingly, a recent report suggests that ACC stimulation is also necessary and sufficient to produce amygdala-dependent aversive conditioning (Tang, T. & Zhuo, M. Soc. Neurosci. Abstr. 293.4, 2003), suggesting that an aversive teaching signal generated by ACC neuronal activity is involved in other forms of aversive learning. Although we have shown that activation of r-ACC neurons is necessary and sufficient to produce an aversive teaching signal, future studies in regions that receive ACC input and convergent contextual sensory inputs are necessary to determine the site and mechanisms of the synaptic plasticity that underlies such learning. CS-responsive neurons in ACC Although there are r-ACC neurons that respond to stimuli (CS) that predict a nociceptive stimulus (in the current experiments, contextual sensory cues in the chamber where the rats received either formalin or intra-ACC HCA), our results do not bear on the function of such CS responses. Although our work does not preclude a role for r-ACC neurons with pain-predictive responses in F-CPA conditioning, it is clear that they are not required for the expression of F-CPA under the conditions of our experiment. One possibility is that different forms of aversive learning recruit the ACC differentially39–41. Another possibility is that r-ACC CS-responsive neurons are involved in a process other than aversive learning. Some studies have implicated the ACC in nociceptive modulation42–45 and also in learned hormonal responses to pain-predictive cues46. Further experiments are necessary to explore these questions and to define other functional roles for CSresponsive neurons in the r-ACC. Implications for chronic pain syndromes Because psychological and emotional dysfunctions are characteristic of chronic pain syndromes, it may be of great clinical importance to understand how the nociceptive pathway through the r-ACC contributes to the long-term behavioral and subjective effects of chronic conditions associated with recurrent and/or prolonged nociceptor activation. Indeed, animal studies report persistent activity47 and plastic changes within the ACC after nerve injury48,49, suggesting that persistent noxious input can lead to local ACC plasticity (sensitization). If the ACC nociceptive system is tonically sensitized under chronic pain conditions, an understanding of the processes that lead to this change and its consequences in downstream projection targets would be of significant clinical importance. In summary, the ACC is part of a nociceptor-activated circuit that, when paired with a contextual CS, can produce a teaching signal resulting in avoidance learning. Since lesions or glutamate receptor blockade of r-ACC neurons reduce acquisition, but postconditioning lesions do not affect expression of such learning, the aversive teaching signal must act on other areas of the brain where nociceptive US and contextual information (CS) converge to produce the synaptic changes underlying the learned avoidance response (CR). Consistent with this idea, excitatory amino acid stimulation of the r-ACC without peripheral noxious stimulation is sufficient to produce CPA learning. Whereas human studies suggest that the ACC processes information relating to the unpleasantness of the stimulus, our data indicate that this signal is necessary to produce avoidance learning. Together, the human and animal studies support the hypothesis that a circuit through the ACC encodes the negative affective quality elicited by noxious stimuli and concomitantly provides an aversive teaching signal. METHODS
Subjects. Subjects were male Long Evans rats (Simonsen Laboratories) weighing 300–350 g at the start of the experiments. Rats were group-housed on a 12-h light-dark schedule with food and water available ad libitum. All experiments were carried out with the approval of the Institutional Animal Care and Use Committee at the University of California, San Francisco. All efforts were made to minimize animal suffering and reduce the numbers of animals used. Drugs. Ibotenic acid (IBO, 1.9 M) was dissolved in 0.1 M PBS and adjusted to pH 7.2–7.4 using 1.0 M NaOH. Stock formaldehyde solution (37% formaldehyde or 100% formalin) was diluted to 2.5% formalin in isotonic saline. The glutamate agonist, homocysteic acid (HCA, 5 or 100 mM) was dissolved in isotonic saline and adjusted to pH 7.2–7.4 using 1.0 M NaOH. The glutamate antagonist

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

401

ARTICLES
kynurenic acid (KyA, 50 mM) was dissolved in a vehicle solution (60% isotonic saline/40% 0.1 M NaOH) and adjusted to pH 7.2–7.4 using 1.0 M HCl. Surgery. Animals were anesthetized with an intraperitoneal (i.p.) injection of sodium pentobarbital (50 mg/kg). Surgery was performed using a Kopf stereotaxic apparatus. For lesion experiments, an injection cannula (30-gauge stainless steel tubing) filled with IBO or 0.1 M PBS was connected to a microinfusion pump (Razel Scientific Instruments) via PE 10 tubing. Surgery details and coordinates for lesion procedures are as previously reported28. For microinjection studies, chronic guide cannulae (33-gauge, Small Parts) were implanted using the stereotaxic procedure described above. Double (1.2 mm spacing between barrels) stainless steel guide cannulae were implanted 1 mm above the ACC injection site (coordinates from Bregma: anterior/posterior (AP), +2.6; dorsal/ventral (DV), –1.6; medial/lateral (ML), 0.6 mm on each side). Single barreled stainless steel guide cannulae were implanted lateral to the ACC injection site for off-site control experiments (coordinates from Bregma: AP, +2.6; DV, –1.5; ML, 2.5 on each side). For both on and off-site experiments, injectors were inserted into the guide cannulae and extended 1 mm beyond the guide tip (see below for microinjection details). Stainless steel dummy cannulae extending to the tip of the guide cannulae were inserted and kept the guide free of debris during the recovery period. All animals (lesion, sham and microinjection) recovered normally from surgery as evidenced by a weight gain on the first test day. Behavioral training and microinjections. All experiments were done as described previously28 using a counterbalanced, unbiased CPA design. The apparatus was exactly as described28: a box with three distinct compartments (a neutral room and two conditioning rooms with distinct olfactory and visual cues) with a removable door to allow room isolation when necessary and photo beams along the floor to record the animal’s position and motor activity. All animals were handled for 3 d prior to testing and habituated to the injection chamber (for microinjection studies). The amount of time the animal spent in the treatment-paired room before vs. after testing was recorded and used for analysis (see below). No initial preferences for any of the compartments in the place-conditioning apparatus were detected before conditioning, indicating that the rats did not prefer any one compartment to the others before conditioning. Lesion study of F-CPA expression. Briefly, experiments began with a pre-test day, during which the animal was allowed to roam freely around all the rooms, and we recorded the amount of time spent in each. This was followed by four conditioning days where the animals were confined to one of the conditioning rooms and received, on alternating days, either nothing in one context or a formalin injection (alternating hindpaws) in the other context (2 UCS pairings total). Conditioning was followed on day 6 by a first post-test day on which the animals were again given free access to all three rooms, and again we recorded the amount of time spent in each room. Surgeries were performed the day after the first post-test, and testing began at least 6 d after surgery. After recovery, the animals were given a second post-test that was identical to the first. Microinjection experiments. For all experiments, injectors were inserted into the guide cannulae after removal of the dummy cannulae, and animals were placed in an injection chamber (injectors protruded 1 mm beyond the guide tip for on- and off-site experiments, so add 1 mm to coordinates given above in “surgery” section for correct DV coordinates). The injectors were attached to a microinfusion pump (Razel Scientific Instruments) via PE 10 tubing. Microinjections of drug or vehicle were made at a rate of 0.5 µl/1.5 min (0.5 µl total volume/side), and the microinjection cannula was left in place for 2 min before and after microinjection. For the glutamate antagonist experiments, microinjection of KyA was made 5 min before the animal received a hindpaw formalin injection and was placed in the box for 50 min. Conditioning was accomplished in 2 d, not 4 d as in the lesion experiment, and included a pre-test and a post-test (4 d total). Thus, all animals received treatment (drug/formalin or just drug) and vehicle context pairings on the same day (counterbalanced by morning or afternoon) and not separated by 1 d as in the lesion experiments. They still received the same number of formalin pairings (2) as in the lesion experiments and no difference in the magnitude of F-CPA was detected between the 4-d and 2-d conditioning regimens (data not shown). Formalin behaviors were also scored using the rating scale method50 on the first or second pairing day (counterbalanced). For experiment 3, intra-ACC or off-site microinjections of a glutamate receptor agonist (HCA) were given without hindpaw formalin injections, and the animals were placed in the conditioning context 5 min after microinjection for 30 min. Experiment 3 was done using the same conditioning regimen as in experiment 2, but three pairings of treatment (drug) and context were made instead of two (5 d total). Pre- and post-tests were identical to the first two experiments. Histology. After completion of the experiments, animals were given a lethal dose of sodium pentobarbital and perfused transcardially with isotonic saline followed by 10% formalin. For microinjection experiments, microinjections of dilute methylene blue were made into the r-ACC just before perfusion. The brains were then removed and fixed first in formalin for 24 h, then in 30% sucrose 24–72 h before slicing. The brains were cut on a sledge microtome at a thickness of 50 µm, stained with cresyl violet and analyzed to assess the extent of the lesion (or injection site) using a light microscope. Using a camera lucida (Nikon), lesions were traced and analyzed using an unbiased stereological method28. Intra-ACC microinjection of IBO produced lesions with clearly definable borders of neuronal cell loss and gliosis as compared with intra-r-ACC microinjection of PBS. Based on past studies, areas of the rodent ACC rich in nociceptive input were targeted (see ref. 28 for detailed region-of-interest). Lesions meeting inclusion criteria had a minimum ‘percentage bilateral damage’ of 50% and at least 30% damage in the least damaged hemisphere within the region of interest. Statistical analyses. For the CPA data, the amount of time spent in the conditioning compartment (i.e., compartment paired with formalin, drug/formalin or drug) on the post-conditioning day (i.e., final test day) was subtracted from the amount of time spent in the same compartment on the pre-conditioning day. This resulted in a ‘magnitude of CPA score’ for each rat. Magnitude of CPA scores between groups were compared using a Student’s t-test when comparing two groups (experiments 1 and 2) or a one-factor ANOVA (intracerebral treatment) followed by a Newman-Keuls post-hoc test when comparing more than two groups (experiment 3). In addition, the absolute amount of time spent in the conditioning compartment on the pre-conditioning day versus the post-conditioning day was compared in sham lesion, lesion, vehicle or drug treated animals using correlated Student’s t-tests. For analysis of formalin behaviors in experiment 2, rating scale nociceptive scores were collected either on day 1 or day 2 (counterbalanced) from formalin-treated rats during each 5-min time bin. The data then were analyzed in separate two-factor ANOVAs (intracerebral treatment × time), with time analyzed as a repeated measure. Post-hoc analyses were performed using the Newman-Keuls test. The accepted level of statistical significance for all experiments was P < 0.05.
ACKNOWLEDGMENTS The authors thank I. Meng for valuable discussions throughout the course of this work. We also thank G. Hjelmstad, J. Levine, J. Mitchell and S. Nicola for reading this manuscript, and C. Evans and C. Bryant for assistance in the completion of this study. Supported by a United States Public Health Service grant NS 21445. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 2 January; accepted 9 February 2004 Published online at http://www.nature.com/natureneuroscience/
1. Kandel, E., Schwartz, J. & Jessel, T. Principles of Neural Science (McGraw-Hill, New York, 2000). 2. LeDoux, J.E. Emotion circuits in the brain. Annu. Rev. Neurosci. 23, 155–184 (2000). 3. Quirk, G.J., Repa, C. & LeDoux, J.E. Fear conditioning enhances short-latency auditory responses of lateral amygdala neurons: parallel recordings in the freely behaving rat. Neuron 15, 1029–1039 (1995). 4. Rosenkranz, J.A. & Grace, A.A. Dopamine-mediated modulation of odour-evoked amygdala potentials during pavlovian conditioning. Nature 417, 282–287 (2002). 5. Fields, H.L. Pain: an unpleasant topic. Pain 6 (Suppl.), S61–S69 (1999). 6. Vogt, B.A. & Sikes, R.W. The medial pain system, cingulate cortex, and parallel processing of nociceptive information. Prog. Brain Res. 122, 223–235 (2000).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

402

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
7. Price, D.D. Psychological and neural mechanisms of the affective dimension of pain. Science 288, 1769–1772 (2000). 8. Treede, R.D., Kenshalo, D.R., Gracely, R.H. & Jones, A.K. The cortical representation of pain. Pain 79, 105–111 (1999). 9. Casey, K.L. Forebrain mechanisms of nociception and pain: analysis through imaging. Proc. Natl. Acad. Sci. USA 96, 7668–7674 (1999). 10. Cliffer, K.D., Burstein, R. & Giesler, G.J. Jr. Distributions of spinothalamic, spinohypothalamic, and spinotelencephalic fibers revealed by anterograde transport of PHA-L in rats. J. Neurosci. 11, 852–868 (1991). 11. Peschanski, M. & Ralston, H.J. 3rd. Light and electron microscopic evidence of transneuronal labeling with WGA-HRP to trace somatosensory pathways to the thalamus. J. Comp. Neurol. 236, 29–41 (1985). 12. Berendse, H.W. & Groenewegen, H.J. Restricted cortical termination fields of the midline and intralaminar thalamic nuclei in the rat. Neuroscience 42, 73–102 (1991). 13. Sikes, R.W. & Vogt, B.A. Nociceptive neurons in area 24 of rabbit cingulate cortex. J. Neurophysiol. 68, 1720–1732 (1992). 14. Hsu, M.M. & Shyu, B.C. Electrophysiological study of the connection between medial thalamus and anterior cingulate cortex in the rat. Neuroreport 8, 2701–2707 (1997). 15. Craig, A.D. Pain mechanisms: labeled lines versus convergence in central processing. Annu. Rev. Neurosci. 26, 1–30 (2003). 16. Rainville, P., Duncan, G.H., Price, D.D., Carrier, B. & Bushnell, M.C. Pain affect encoded in human anterior cingulate but not somatosensory cortex. Science 277, 968–971 (1997). 17. Hutchison, W.D., Davis, K.D., Lozano, A.M., Tasker, R.R. & Dostrovsky, J.O. Painrelated neurons in the human cingulate cortex. Nat. Neurosci. 2, 403–405 (1999). 18. Koyama, T., Tanaka, Y.Z. & Mikami, A. Nociceptive neurons in the macaque anterior cingulate activate during anticipation of pain. Neuroreport 9, 2663–2667 (1998). 19. Yamamura, H. et al. Morphological and electrophysiological properties of ACCx nociceptive neurons in rats. Brain Res. 735, 83–92 (1996). 20. Foltz, E.L. & White, L.E. Pain relief by frontal cingulumotomy. J. Neurosurg. 19, 89–100 (1962). 21. Hurt, R.W. & Ballantine, H.T. Jr. Stereotactic anterior cingulate lesions for persistent pain: a report on 68 cases. Clin. Neurosurg. 21, 334–351 (1974). 22. Allen, G.V. & Hopkins, D.A. Mamillary body in the rat: topography and synaptology of projections from the subicular complex, prefrontal cortex, and midbrain tegmentum. J. Comp. Neurol. 286, 311–336 (1989). 23. Cassell, M.D. & Wright, D.J. Topography of projections from the medial prefrontal cortex to the amygdala in the rat. Brain Res. Bull. 17, 321–333 (1986). 24. Christie, M.J., Summers, R.J., Stephenson, J.A., Cook, C.J. & Beart, P.M. Excitatory amino acid projections to the nucleus accumbens septi in the rat: a retrograde transport study utilizing D[3H]aspartate and [3H]GABA. Neuroscience 22, 425–439 (1987). 25. Reep, R.L. & Corwin, J.V. Topographic organization of the striatal and thalamic connections of rat medical agranular cortex. Brain Res. 841, 43–52 (1999). 26. Gabriel, M., Kubota, Y., Sparenborg, S., Straube, K. & Vogt, B.A. Effects of cingulate cortical lesions on avoidance learning and training-induced unit activity in rabbits. Exp. Brain Res. 86, 585–600 (1991). 27. Salamone, J.D., Cousins, M.S. & Snyder, B.J. Behavioral functions of nucleus accumbens dopamine: empirical and conceptual problems with the anhedonia hypothesis. Neurosci. Biobehav. Rev. 21, 341–359 (1997). 28. Johansen, J.P., Fields, H.L. & Manning, B.H. The affective component of pain in rodents: direct evidence for a contribution of the anterior cingulate cortex. Proc. Natl. Acad. Sci. USA 98, 8077–8082 (2001). 29. Peretz, E. The effects of lesions of the anterior cingulate cortex on the behavior of the rat. J. Comp. Physiol. Psychol. 53, 540–548 (1960). 30. Ploghaus, A. et al. Dissociating pain from its anticipation in the human brain. Science 284, 1979–1981 (1999). 31. Takenouchi, K. et al. Emotional and behavioral correlates of the anterior cingulate cortex during associative learning in rats. Neuroscience 93, 1271–1287 (1999). 32. Nishijo, H. et al. Single neuron responses in the monkey anterior cingulate cortex during visual discrimination. Neurosci. Lett. 227, 79–82 (1997). 33. Gigg, J., Tan, A.M. & Finch, D.M. Glutamatergic excitatory responses of anterior cingulate neurons to stimulation of the mediodorsal thalamus and their regulation by GABA: an in vivo iontophoretic study. Cereb. Cortex 2, 477–484 (1992). 34. Puig, S. & Sorkin, L.S. Formalin-evoked activity in identified primary afferent fibers: systemic lidocaine suppresses phase-2 activity. Pain 64, 345–355 (1996). 35. Dubuisson, D. & Dennis, S.G. The formalin test: a quantitative study of the analgesic effects of morphine, meperidine, and brain stem stimulation in rats and cats. Pain 4, 161–174 (1977). 36. Pastoriza, L.N., Morrow, T.J. & Casey, K.L. Medial frontal cortex lesions selectively attenuate the hot plate response: possible nocifensive apraxia in the rat. Pain 64, 11–17 (1996). 37. Vaccarino, A.L. & Melzack, R. Analgesia produced by injection of lidocaine into the anterior cingulum bundle of the rat. Pain 39, 213–219 (1989). 38. Donahue, R.R., LaGraize, S.C. & Fuchs, P.N. Electrolytic lesion of the anterior cingulate cortex decreases inflammatory, but not neuropathic nociceptive behavior in rats. Brain Res. 897, 131–138 (2001). 39. Morgan, M.A. & LeDoux, J.E. Differential contribution of dorsal and ventral medial prefrontal cortex to the acquisition and extinction of conditioned fear in rats. Behav. Neurosci. 109, 681–688 (1995). 40. Barros, D.M. et al. Molecular signaling pathways in the cerebral cortex are required for retrieval of one-trial avoidance learning in rats. Behav. Brain Res. 114, 183–192 (2000). 41. Kung, J.C., Su, N.M., Fan, R.J., Chai, S.C. & Shyu, B.C. Contribution of the anterior cingulate cortex to laser-pain conditioning in rats. Brain Res. 970, 58–72 (2003). 42. Hardy, S.G. Analgesia elicited by prefrontal stimulation. Brain Res. 339, 281–284 (1985). 43. Hardy, S.G. & Haigler, H.J. Prefrontal influences upon the midbrain: a possible route for pain modulation. Brain Res. 339, 285–293 (1985). 44. Lee, D.E., Kim, S.J. & Zhuo, M. Comparison of behavioral responses to noxious cold and heat in mice. Brain Res. 845, 117–121 (1999). 45. Calejesan, A.A., Kim, S.J. & Zhuo, M. Descending facilitatory modulation of a behavioral nociceptive response by stimulation in the adult rat anterior cingulate cortex. Eur. J. Pain 4, 83–96 (2000). 46. Buchanan, S.L. & Powell, D.A. Cingulate cortex: its role in Pavlovian conditioning. J. Comp. Physiol. Psychol. 96, 755–774 (1982). 47. Mao, J., Mayer, D.J. & Price, D.D. Patterns of increased brain activity indicative of pain in a rat model of peripheral mononeuropathy. J. Neurosci. 13, 2689–2702 (1993). 48. Wei, F., Li, P. & Zhuo, M. Loss of synaptic depression in mammalian anterior cingulate cortex after amputation. J. Neurosci. 19, 9346–9354 (1999). 49. Wei, F. & Zhuo, M. Potentiation of sensory responses in the anterior cingulate cortex following digit amputation in the anaesthetised rat. J. Physiol. 532, 823–833 (2001). 50. Manning, B.H. A lateralized deficit in morphine antinociception after unilateral inactivation of the central amygdala. J. Neurosci. 18, 9453–9470 (1998).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

403

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Prefrontal cortex and decision making in a mixedstrategy game
Dominic J Barraclough, Michelle L Conroy & Daeyeol Lee
In a multi-agent environment, where the outcomes of one’s actions change dynamically because they are related to the behavior of other beings, it becomes difficult to make an optimal decision about how to act. Although game theory provides normative solutions for decision making in groups, how such decision-making strategies are altered by experience is poorly understood. These adaptive processes might resemble reinforcement learning algorithms, which provide a general framework for finding optimal strategies in a dynamic environment. Here we investigated the role of prefrontal cortex (PFC) in dynamic decision making in monkeys. As in reinforcement learning, the animal’s choice during a competitive game was biased by its choice and reward history, as well as by the strategies of its opponent. Furthermore, neurons in the dorsolateral prefrontal cortex (DLPFC) encoded the animal’s past decisions and payoffs, as well as the conjunction between the two, providing signals necessary to update the estimates of expected reward. Thus, PFC might have a key role in optimizing decision-making strategies.

Decision making refers to an evaluative process of selecting a particular action from a set of alternatives. When the mapping between a particular action and its outcome or utility is fixed, the decision to select the action with maximum utility can be considered optimal or rational. However, animals face more difficult problems in a multiagent environment, in which the outcome of one’s decision can be influenced by the decisions of other animals. Game theory provides a mathematical framework to analyze decision making in a group of agents1–4. A game is defined by a set of actions available to each player, and a payoff matrix that specifies the reward or penalty for each player as a function of decisions made by all players. A solution or equilibrium in game theory refers to a set of strategies selected by a group of rational players1,5,6. Nash has proved that any n-player competitive game has at least one equilibrium in which no players can benefit by changing their strategies individually5. These equilibrium strategies often take the form of a mixed strategy, which is defined as a probability density function over the alternative actions available to each player. This requires players to choose randomly among alternative choices, as in the game of rock-paper-scissors during which choosing one of the alternatives (e.g., paper) exclusively allows the opponent to exploit such a biased choice (with scissors). Many studies have shown that people frequently deviate from the predictions of game theory7–21. Although the magnitudes of such deviations are often small, they have important implications regarding the validity of assumptions in game theory, such as the rationality of human decision-makers22–27. In addition, strategies of human decision-makers change with their experience17–21. These adaptive processes might be based on reinforcement learning algorithms28, which can be used to approximate optimal decision-making strategies in a dynamic environment. In the present study, we analyzed the

performance of monkeys playing a zero-sum game against a computer opponent to determine how closely their behaviors match the predictions of game theory and whether reinforcement learning algorithms can account for any deviations from such predictions. In addition, neural activity was recorded from the DLPFC to investigate its role during strategic decision making in a multi-agent environment. The results showed that the animal’s choice behavior during a competitive game could be accounted for by a reinforcement learning algorithm. Individual prefrontal neurons often modulated their activity according to the choice of the animal in the previous trial, the outcome of that choice, and the conjunction between the choice and its outcome. This suggests that the PFC may be involved in updating the animal’s decision-making strategy based on a reinforcement learning algorithm. RESULTS Behavioral performance Two rhesus monkeys played a game analogous to matching pennies against a computer in an oculomotor free-choice task (Fig. 1a; Methods). The animal was rewarded when it selected the same target as the computer that was programmed to minimize the animal’s reward by exploiting the statistical bias in the animal’s choice behavior. Accordingly, the optimal strategy for the animal was to choose the targets randomly with equal probabilities, which corresponds to the Nash equilibrium in the matching pennies game. To determine how the animal’s decisions were influenced by the strategy of the opponent, we manipulated the amount of information that was used by the computer opponent (see Methods). In algorithm 0, the computer selected its targets randomly with equal probabilities, regardless of the animal’s choice patterns. In algorithm 1, the com-

Department of Brain and Cognitive Sciences, Center for Visual Science, University of Rochester, Rochester, New York 14627, USA. Correspondence should be addressed to D.L. ([email protected]). Published online 7 March 2004; doi:10.1038/nn1209

404

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 1 Task and behavioral performance. (a) Free-choice task and payoff matrix for the animal during the competitive game (1, reward; 0, no reward). (b) Recording sites in PFC. Frontal eye field (gray area in the inset) was defined by electrical stimulation50. (c) Frequency histograms for the probability of choosing the right-hand target in algorithms 1 and 2. (d) Probability of the win-stay-lose-switch (WSLS) strategy (abscissa) versus probability of reward (ordinate). (e) Difference in the value functions for the two targets estimated from a reinforcement learning model (abscissa) versus the probability of choosing the target at the right (ordinate). Error bars indicate standard error of the mean (s.e.m.). Histograms show the frequency of trials versus the difference in the value functions. Solid line, prediction of the reinforcement learning model. In all panels, dark and light symbols indicate the results from the two animals, respectively.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

puter analyzed only the animal’s choice history, but not its reward history. In algorithm 2, both choice and reward histories were analyzed. In both algorithms 1 and 2, the computer chose its target randomly if it did not find any systematic bias in the animal’s choice behavior. Therefore, a reward rate near 0.5 indicates that the animal’s performance was optimal. Indeed, the animal’s reward rate was close to 0.5 for all algorithms, indicating that the animal’s performance was nearly optimal. In algorithm 0, the reward rate was fixed at 0.5 regardless of the animal’s behavior, and therefore there was no incentive for the animal to choose the targets with equal probabilities. In fact, both animals chose the right-hand target more frequently (P = 0.70 and 0.90 and n = 5,327 and 1,669 trials, for the two animals, respectively) than the
Table 1 Parameters for the reinforcement learning model.
Algorithm Monkey 1 C E 2 C E α 0.176 (0.130, 0.220) 0.170 (0.143, 0.198) 0.986 (0.983, 0.988) 0.828 (0.801, 0.851) ∆1 0.661 (0.619, 0.704) 0.941 (0.903, 0.979) 0.033 (0.028, 0.039) 0.195 (0.171, 0.218) ∆2 −0.554 (−0.597, −0.512) −1.064 (−1.104, −1.024) 0.016 (0.012, 0.021) −0.143 (−0.169, −0.118)

left-hand target. For the remaining two algorithms, the probability of choosing the righthand target was much closer to 0.5 (Fig. 1c), which corresponds to the Nash equilibrium of the matching pennies game. In addition, the probability of choosing a given target was relatively unaffected by the animal’s choice in the previous trial. For example, the probability that the animal would select the same target as in the previous trial was also close to 0.5 (P = 0.51 ± 0.06 and 0.50 ± 0.04 and n = 120,254 and 74,113 trials, for algorithms 1 and 2, respectively). In contrast, the animal’s choice was strongly influenced by the computer’s choice in the previous trial, especially in algorithm 1. In the game of matching pennies, the strategy to choose the same target selected by the opponent in the previous trial can be referred to as a win-stay-lose-switch (WSLS) strategy, as this is equivalent to choosing the same target as in the previous trial if that choice was rewarded and choosing the opposite target otherwise. The probability of the WSLS strategy in algorithm 1 (0.73 ± 0.14) was significantly higher than that in algorithm 2 (0.53 ± 0.06; P < 10−16; Fig. 1d). Although the tendency for the WSLS strategy in algorithm 2 was only slightly above chance, this bias was still statistically significant (P < 10−5). Similarly, average mutual information between the sequence of animal’s choice and reward in three successive trials and the animal’s choice in the following trial decreased from 0.245 (± 0.205) bits for algorithm 1 to 0.043 (± 0.035) bits for algorithm 2. Reinforcement learning model Using a reinforcement learning model19–21,28,29, we tested whether the animal’s decision was systematically influenced by the cumulative effects of reward history. In this model, a decision was based on the difference between the value functions (that is, expected reward) for the two targets. Denoting the value functions of the two targets (L and R) at trial t as Vt(L) and Vt(R), the probability of choosing each target is given by the logit transformation of the difference between the value functions30. In other words, logit P(R) ≡ log P(R)/(1 − P(R)) = Vt(R) − Vt(L).

α, discount factor; ∆1 and ∆2, changes in the value function associated with rewarded and unrewarded targets selected by the animal, respectively. The numbers in parentheses indicate 99.9% confidence intervals.

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

405

ARTICLES
The value function, Vt(x), for target x, was updated after each trial according to the following: Vt+1(x) = αVt(x) + ∆t(x), where α is a discount factor, and ∆t(x) denotes the change in the value function determined by the animal’s decision and its outcome. In the current model, ∆t(x) = ∆1 if Figure 2 Effects of relative expected reward (i.e., difference in value functions) and its trial-to-trial the animal selects the target x and is changes on the activity of prefrontal neurons. (a) Percentage of neurons with significant correlation rewarded, ∆t(x) = ∆2 if the animal selects the (t-test, P < 0.05) between their activity and the difference in the value functions for the two targets target x and is not rewarded, and ∆t(x) = 0 if (VD = Vτ(R) − Vτ(L)) estimated for the current (τ = t) and previous (τ = t − 1) trials, or between the activity and the changes in the value functions between the two successive trials (VD change = the animal does not select the target x. We ∆t–1(R) − ∆t–1(L)). (b) Correlation coefficient between the VD change and the activity in a given neuron introduced a separate parameter for the (ordinate), plotted against correlation coefficient between the VD in the previous trial (i.e., unrewarded target (∆2) because the proba- Vt−1(R) − Vt−1(L)) and the activity of the same neuron (abscissa). Black (gray) symbols indicate the bility of choosing the same target after losing neurons in which both (either) correlation coefficients were significantly different from 0 (t-test, a reward was significantly different from the P < 0.05). The numbers in each panel correspond to Spearman’s rank correlation coefficient (r) and probability of switching to the other target its level of significance (P). for all animals and for both algorithms 1 and 2. Maximum likelihood estimates31 of the model parameters (Table 1) showed that a frequent use of the WSLS ference in value functions (Fig. 1e). This implies that for most trials, strategy during algorithm 1 was reflected in a relatively small dis- the difference in the value functions of the two targets was relatively count factor (α < 0.2), a large positive ∆1 (> 0.6) and a large negative small, making it difficult to predict the animal’s choice reliably. These ∆2 (< −0.5) in both animals. For algorithm 1, this led to a largely results suggest that during a competitive game, the monkeys might bimodal distribution for the difference in the value functions have approximated the optimal decision-making strategy using a (Fig. 1e). In contrast, the magnitude of changes in value function reinforcement learning algorithm. during algorithm 2 was smaller, indicating that the outcome of previous choices only weakly influenced the subsequent choice of the Prefrontal activity during a competitive game animal. In addition, the discount factor for algorithm 2 was relatively The value functions in the above reinforcement learning model were large (α > 0.8). This suggests that the animal’s choice was systemati- updated according to the animal’s decisions and the outcomes of cally influenced by the combined effects of previous reward history those decisions. To determine whether such signals are encoded in the even in algorithm 2. The combination of model parameters for algo- activity of individual neurons in PFC, we recorded single-neuron rithm 2 produced an approximately normal distribution for the dif- activity in the DLPFC while the animal played the same free-choice task. During the neurophysiological recording, the computer selected its target according to algorithm 2. A total of 132 neurons were examined during a minimum of 128 free-choice trials (mean = 583 trials; Fig. 1b). As a control, each neuron was also examined during 128 trials of a visual search task in

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 3 Percentages of neurons encoding signals related to the animal’s decision. White bars show the percentage of neurons (n = 132) with significant main and interaction effects in a three-way ANOVA (P × R × C). Light gray bars show the same information for the neurons with >256 free-choice trials, which was tested for stationarity in free-choice trials (n = 112). Dark gray bars show the percentage of neurons with significant effects in the three-way ANOVA that also varied with the task (search vs. choice) in a four-way ANOVA (Task × P × R × C). This analysis was performed only for the neurons with >256 free-choice trials for comparison with the control analysis to test stationarity. Black histograms show the percentage of neurons with significant effects in the three-way ANOVA that also have significant non-stationarity in a control 4-way ANOVA across the two successive blocks of 128 free-choice trials.

406

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 4 Example neuron showing a significant effect of the animal’s choice in the previous trial. Top, spike density functions averaged according to the animal’s choice (L and R) and reward (+, reward; −, no reward) in the previous trial and the choice in the current trial. A dotted vertical line indicates the onset of the fore-period, and the two solid lines the beginning and end of the delay period. Bottom, raster plots showing the activity of the same neuron sorted according to the same three factors during the search (gray background) and free-choice (white background) trials.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

which the animal’s decision was guided by sensory stimuli (Methods). During the free-choice trials, activity in some prefrontal neurons was influenced by the difference in the value functions for the two targets (that is, V(R) − V(L)), although the effects in individual neurons were relatively small (Fig. 2). This was not entirely due to the animal’s choice and its outcome in the previous trial, as the value functions estimated for the previous trial produced similar results (Fig. 2a). If individual PFC neurons are involved in the temporal integration of value functions according to the reinforcement learning model described above, differences in the value functions (i.e., V(x)) and their changes (i.e., ∆(x)) would similarly influence the activity of PFC neurons. Interestingly, such patterns were found for the delay and movement periods, but not for the fore-period (Methods; Fig. 2b). These results suggest that some prefrontal neurons might be involved in temporally integrating the signals related to previous choice and its outcome to update value functions. To examine how the activity of individual PFC neurons is influenced by the animal’s choices and their outcomes, we analyzed neural activity by three-way ANOVA with the animal’s choice and reward in the previous trial and its choice in the current trial as main factors. For 39% of PFC neurons, the activity during the fore-period was influenced by the animal’s choice in the previous trial (Fig. 3). For example, in the neuron illustrated in Figure 4, the animal’s choice in

the previous trial exerted a significant influence on the activity before and during the fore-period, as well as during the delay period (3-way ANOVA, P < 0.001). In addition, activity during the movement period was still influenced by the animal’s choice in the previous trial and its outcome, as well as by their interactions with the animal’s choice in the current trial. To determine whether any of these effects could be attributed to systematic variability in eye movements, the above analysis was repeated using the residuals from a regression model in which the neural activity related to a set of eye movement parameters was factored out (Methods). The results were nearly identical, with the only difference found in the loss of significance for the effect of the current choice. During the fore-period, 35% of neurons showed a significant effect of the animal’s choice in the previous trial on the residuals from the same regression model. It is possible that the animal’s choice in the previous trial influenced the activity of this neuron during the next trial through systematic changes in unidentified sensorimotor events, such as licking or eye movements during the inter-trial interval, that were not experimentally controlled. This was tested by comparing the activity of the same neuron in the search and free-choice trials. For the neuron shown in Figure 4, activity during search trials was significantly affected by the position of the target in the previous trial only during the fore-period, and this effect was opposite to and significantly different from that found in the free-choice trials (4-way ANOVA, P < 10−5). The raster plots show that this change occurred within a few trials after the animal switched from search to free-choice trials (Fig. 4). These results suggest that the effect of the animal’s choice in the previous trial on the activity of this neuron did not merely reflect nonspecific sensorimotor events. In 17% of the neurons that showed a significant effect of the animal’s previous choice during the fore-period, there was also a significant interaction between the task type (search vs. free-choice) and the animal’s choice in the previous trial (Fig. 3). This indicates that signals related to the animal’s past choice were actively maintained in the PFC according to the type of decision. It is unlikely that this was entirely due to an ongoing drift in the background activity (i.e., non-stationarity), as the control analysis performed on two successive blocks of free-choice trials did not produce a single case with the same effect during the fore-period (Fig. 3). During the fore-period, 39% of neurons showed a significant effect of the reward in the previous trial. For example, the activity of the neu-

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

407

ARTICLES
Figure 5 Example neuron showing a significant effect of the reward in the previous trial. Same format as in Figure 4.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

ron in Figure 5 was higher throughout the entire trial after the animal was not rewarded in the previous trial, compared to when the animal was rewarded. This effect was nearly unchanged when we removed the eye-movement related activity in a regression analysis, both in this single-neuron example and for the population as a whole. Overall, 37% of neurons showed the effect of the previous reward when the analysis was performed on the residuals from the regression model. The possibility that this effect was entirely due to uncontrolled sensorimotor events is also unlikely, as a substantial proportion of these neurons (21%) also showed a significant interaction between the task type and the previous reward during the fore-period (Fig. 3). To update the value functions in a reinforcement learning model, signals related to the animal’s choice and its outcome must be combined, because each variable alone does not specify how the value function of a particular target should be changed. Similarly, activity of the neurons in the PFC was often influenced by the conjunction of these two variables. In the neuron in Figure 6, for example, there was a gradual buildup of activity during the fore-period, but this occurred

only when the animal had selected the right-hand target in the previous trial, and this choice was not rewarded. During the delay period, the activity of this neuron diverged to reflect the animal’s choice in the current trial (Fig. 6, arrow). The same neuron showed markedly weaker activity during the search trials, suggesting that information coded in the activity of this neuron regarding the outcome of choosing a particular target was actively maintained in free-choice trials (Fig. 6). For the fore-period, 20% of the neurons showed significant interaction between the animal’s choice and its outcome in the previous trial (P < 0.05; Fig. 3). Activity related to eye movements was not an important factor: 90% of these neurons showed the same effect in the residuals from the regression analysis that factored out the effects of eye movements. Furthermore, during the fore-period, 27% of the same neurons showed significant threeway interactions among task type, animal’s choice in the previous trial and outcome of the previous trial. In contrast, the control analysis during the first two blocks of the freechoice task revealed such an effect only in 5% of the neurons (Fig. 3). These results indicate that signals related to the conjunction of the animal’s previous decision and its outcome are processed differently in the PFC according to the type of decisions made by the animal. DISCUSSION Interaction with other intelligent beings is fundamentally different from—and more complex than—dealing with inanimate objects32,33. Interactions with other animals are complicated by the fact that their behavioral strategies often change as a result of one’s own behavior. Therefore, the analysis
Figure 6 Example neuron with a significant interaction between the animal’s choice and its outcome in the previous trial. Same format as in Figure 4. Arrows indicate the time when the animal’s choice in the current trial was first reflected in the neural activity.

408

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
of decision making in a group requires a more sophisticated analytical framework, which is provided by game theory. Matching pennies is a relatively simple zero-sum game that involves two players and two alternative choices. The present study examined the behavior of monkeys playing a competitive game similar to matching pennies against a computer opponent. It is not known whether monkeys treated this game as a competitive situation with another intentional being. Nevertheless, the same formal framework of game theory is applicable to the task used in this study, and as predicted, the animal’s behavior was influenced by the opponent’s strategy. When the computer blindly played the equilibrium strategy regardless of the animal’s behavior, the animals selected one of the targets more frequently. In contrast, when the computer opponent began exploiting biases in the animal’s choice sequence, the animal’s behavior approached the equilibrium strategy. Furthermore, when the computer did not examine the animal’s reward history (algorithm 1), the animals achieved a nearly optimal reward rate by adopting the win-stay-lose-switch (WSLS) strategy. This was possible because this strategy was not detected by the computer. Finally, the frequency of the WSLS strategy was reduced when the computer began exploiting biases in the animal’s choice and reward sequences (algorithm 2). These results also suggest that the animals approximated the optimal strategy using a reinforcement learning algorithm. This model assumes that the animals base their decisions, in part, on the estimates of expected rewards for the two targets and tend to select the target with larger expected reward. During zero-sum games such as matching pennies, strategies of the players behaving according to some reinforcement learning algorithms would gradually converge on a set of equilibrium strategies5–7. However, it is important to update the value functions of different targets by a small amount after each play when playing against a fully informed rational player (such as algorithm 2 in the present study). This is because large, predictable changes in the value functions would reveal one’s next choice to the opponent. In the present study, the magnitude of changes in the value function varied according to the strategy of the opponent and was adjusted through the animal’s experience. Finally, neurophysiological recordings in the PFC revealed a potential neural basis for updating the value functions adaptively while interacting with a rational opponent. Reward-related activity is widespread in the brain34–38. In particular, signals related to expected reward (i.e., value functions) are present in various brain areas39–43, including the DLPFC44–48. Our results showed that neurons in the DLPFC also code signals related to the animal’s choice in the previous trial. Such signals might be actively maintained and processed differently in the DLPFC according to the type of information required for the animal’s upcoming decisions. Furthermore, signals related to the animal’s past choices and their outcomes are combined at the level of individual PFC neurons. These signals might then be temporally integrated according to a reinforcement learning algorithm to update the value functions for alternative actions. Many neurons in the PFC show persistent activity during a working memory task, and the same circuitry might be ideally suited for temporal integration of signals related to the animal’s previous choice and its outcome49. Although the present study examined the animal’s choice behavior in a competitive game, reinforcement learning algorithms can converge on optimal solutions for a wide range of decision-making problems in dynamic environments. Therefore, the results from the present study suggest that the PFC has an important role in optimizing decision-making strategies in a dynamic environment that may include multiple agents. METHODS
Animal preparations. Two male rhesus monkeys were used. Their eye movements were monitored at a sampling rate of 250 Hz with either a scleral eye coil or a high-speed video-based eye tracker (ET49, Thomas Recording). All the procedures used in this study conformed to National Institutes of Health guidelines and were approved by the University of Rochester Committee on Animal Research. Behavioral task. Monkeys were trained to play a competitive game analogous to matching pennies against a computer in an oculomotor free-choice task (Fig. 1a). During a 0.5-s fore-period, they fixated a small yellow square (0.9 × 0.9°; CIE x = 0.432, y = 0.494, Y = 62.9 cd/m2) in the center of a computer screen, and then two identical green disks (radius = 0.6°; CIE x = 0.286, y = 0.606, Y = 43.2 cd/m2) were presented 5° away in diametrically opposed locations. The central target disappeared after a 0.5-s delay period, and the animal was required to shift its gaze to one of the targets. At the end of a 0.5-s hold period, a red ring (radius = 1°; CIE x = 0.632, y = 0.341, Y = 17.6 cd/m2) appeared around the target selected by the computer, and the animal maintained its fixation for another 0.2 s. The animal was rewarded at the end of this second hold period, but only if it selected the same target as the computer. The computer had been programmed to exploit certain biases displayed by the animal in making its choices. Each neuron was also tested in a visual search task. This task was identical to the free-choice task, except that one of the targets in the free-choice task was replaced by a distractor (red disk). The animal was required to shift its gaze toward the remaining target (green disk), and this was rewarded randomly with 50% probability. This made it possible to examine the effect of reward on the neural activity. The location of the target was selected from the two alternative locations pseudo-randomly for each search trial. Algorithms for computer opponent. During the free-choice task, the computer selected its target according to one of three different algorithms. In algorithm 0, the computer selected the two targets randomly with equal probabilities, which corresponds to the Nash equilibrium in the matching pennies game. In algorithm 1, the computer exploited any systematic bias in the animal’s choice sequence to minimize the animal’s reward rate. The computer saved the entire history of the animal’s choices in a given session, and used this information to predict the animal’s next choice by testing a set of hypotheses. First, the conditional probabilities of choosing each target given the animal’s choices in the preceding n trials (n = 0 to 4) were estimated. Next, each of these conditional probabilities was tested against the hypothesis that the animal had chosen both targets with equal probabilities. When none of these hypotheses was rejected, the computer selected each target randomly with 50% probability, as in algorithm 0. Otherwise, the computer biased its selection according to the probability with the largest deviation from 0.5 that was statistically significant (binomial test, P < 0.05). For example, if the animal chose the right-hand target with 80% probability, the computer selected the left-hand target with the same probability. Therefore, to maximize reward, animals needed to choose both targets with equal frequency and select a target on each trial independently from previous choices. In algorithm 2, the computer exploited any systematic bias in the animal’s choice and reward sequences. In addition to the hypotheses tested in algorithm 1, algorithm 2 also tested the hypothesis that the animal’s decisions were independent of prior choices and their payoffs in the preceding n trials (n = 1 to 4). Thus, to maximize total reward in algorithm 2, it was necessary for the animal to choose both targets with equal frequency and to make choices independently from previous choices and payoffs. Neurophysiological recording. Single-unit activity was recorded from the neurons in the DLPFC of two monkeys using a five-channel multi-electrode recording system (Thomas Recording). The placement of the recording chamber was guided by magnetic resonance images, and this was confirmed in one animal by metal pins inserted in known anatomical locations. In addition, the frontal eye field (FEF) was defined in both animals as the area in which saccades were evoked by electrical stimulations with currents <50 µA (ref. 50). All the neurons described in the present study were anterior to the FEF. Analysis of behavioral data. The frequency of a behavioral event (e.g., reward) was examined with the corresponding probability averaged across recording

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

409

ARTICLES
sessions and its standard deviation. The values of mutual information were corrected for the finite sample size. Null hypotheses in the analysis of behavioral data were tested using a binomial test or a t-test (P < 0.05). Parameters in the reinforcement learning model were estimated by a maximum likelihood procedure, using a function minimization algorithm in Matlab (Mathworks Inc.), and confidence intervals were estimated by profile likelihood intervals31. Analysis of neural data. Spikes during a series of 500-ms bins were counted separately for each trial. The effects of the animal’s choice (P) and reward (R) in the previous trial and the choice in the current trial (C) were analyzed in a 3-way ANOVA (P × R × C). The effect of the task (search versus free-choice) was analyzed in a four-way ANOVA (Task × P × R × C). As a control analysis to determine whether the task effect was due to non-stationarity in neural activity, the same four-way ANOVA was performed for the first two successive blocks of 128 trials in the free-choice task (Fig. 3). To determine whether eye movements were confounding factors, the above analysis was repeated using the residuals from the following regression model: S = a0 + a1 Xpre80 + a2 Ypre80 + a3 XFP + a4 YFP + a5 XSV + a6 YSV + a7 SRT + a8 PV + ε where S indicates the spike count, Xpre80 (Ypre80) the horizontal (vertical) eye position 80 ms before the onset of central fixation target, XFP (YFP) the average horizontal (vertical) eye position during the fore-period, XSV (YSV) the horizontal (vertical) component of the saccade directed to the target, SRT and PV the saccadic reaction time and the peak velocity of the saccade, and ε the error term.
ACKNOWLEDGMENTS We thank L. Carr, R. Farrell, B. McGreevy and T. Twietmeyer for their technical assistance, J. Swan-Stone for programming, X.-J. Wang for discussions, and B. Averbeck and J. Malpeli for critically reading the manuscript. This work was supported by the James S. McDonnell Foundation and the National Institutes of Health (NS44270 and EY01319). COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 23 December 2003; accepted 12 February 2004 Published online at http://www.nature.com/natureneuroscience/
1. Von Neumann J. & Morgenstern O. Theory of Games and Economic Behavior (Princeton Univ. Press, New Jersey, 1944). 2. Fudenberg, D. & Tirole, J. Game Theory (MIT Press, Cambridge, Massachusetts, 1991). 3. Dixit, A. & Skeath, S. Games of Strategy (Norton, New York, 1999). 4. Glimcher, P.W. Decisions, Uncertainty, and the Brain (MIT Press, Cambridge, Massachusetts, 2003). 5. Nash, J.F. Equilibrium points in n-person games. Proc. Natl. Acad. Sci. USA 36, 48–49 (1950). 6. Robinson, J. An iterative method of solving a game. Ann. Math. 54, 296–301 (1951). 7. Binmore, K., Swierzbinski, J. & Proulx, C. Does minimax work? An experimental study. Econ. J. 111, 445–464 (2001). 8. Touhey, J.C. Decision processes, expectations, and adoption of strategies in zerosum games. Hum. Relat. 27, 813–824 (1974). 9. O’Neill, B. Nonmetric test of the minimax theory of two-person zerosum games. Proc. Natl. Acad. Sci. USA 84, 2106–2109 (1987). 10. Brown, J.N. & Rosenthal, R.W. Testing the minimax hypothesis: a re-examination of O’Neill’s game experiment. Econometrica 58, 1065–1085 (1990). 11. Rapoport, A. & Boebel, R.B. Mixed strategies in strictly competitive games: a further test of the minimax hypothesis. Games Econ. Behav. 4, 261–283 (1992). 12. Rapoport, A. & Budescu, D.V. Generation of random series in two-person strictly competitive games. J. Exp. Psychol. Gen. 121, 352–363 (1992). 13. Budescu, D.V. & Rapoport, A. Subjective randomization in one- and two-person games. J. Behav. Dec. Making 7, 261–278 (1994). 14. Ochs, J. Games with unique, mixed strategy equilibria: an experimental study. Games Econ. Behav. 10, 202–217 (1995). 15. Sarin, R. & Vahid, F. Predicting how people play games: a simple dynamic model of choice. Games Econ. Behav. 34, 104–122 (2001). 16. Shachat, J.M. Mixed strategy play and the minimax hypothesis. J. Econ. Theory 104, 189–226 (2002). 17. Walker, M. & Wooders, J. Minimax play at Wimbledon. Am. Econ. Rev. 91, 1521–1538 (2001). 18. Fudenberg, D. & Levine, D.K. Theory of Learning in Games (MIT Press, Cambridge, Massachusetts, 1998). 19. Mookherjee, D. & Sopher, B. Learning behavior in an experimental matching pennies game. Games Econ. Behav. 7, 62–91 (1994). 20. Erev, I. & Roth, A.E. Predicting how people play games: reinforcement learning in experimental games with unique, mixed strategy equilibria. Am. Econ. Rev. 88, 848–881 (1998). 21. Camerer, C.F. Behavioral Game Theory (Princeton Univ. Press, Princeton, New Jersey, 2003). 22. Simon, H.A. Models of Man (Wiley, New York, 1957). 23. Kahneman, D., Slovic, P. & Tversky, A. Judgement Under Uncertainty: Heuristics and Biases (Cambridge Univ. Press, Cambridge, UK, 1982). 24. O’Neill, B. Comments on Brown and Rosenthal’s reexamination. Econometrica 59, 503–507 (1991). 25. Rapoport, A. & Budescu, D.V. Randomization in individual choice behavior. Psychol. Rev. 104, 603–617 (1997). 26. Chen, H-C., Friedman, J.W. & Thisse, J-F. Boundedly rational Nash equilibrium: a probabilistic choice approach. Games Econ. Behav. 18, 32–54 (1997). 27. Rubinstein, A. Modeling Bounded Rationality (MIT Press, Cambridge, Massachusetts, 1998). 28. Sutton, R.S. & Barto, A.G. Reinforcement Learning: an Introduction (MIT Press, Cambridge, Massachusetts, 1998). 29. Littman, M.L. Markov games as a framework for multi-agent reinforcement learning. Machine Learning: Proc. 11th Int. Conf. pp. 157–163 (Morgan Kaufmann, San Francisco, California, 1994) 30. Christensen, R. Log-linear Models and Logistic Regression edn. 2 (Springer-Verlag, New York, 1997). 31. Burnham, K.P. & Anderson, D.R. Model Selection and Multimodel Inference 2nd edn. (Springer-Verlag, New York, 2002). 32. Byrne, R.W. & Whiten, A. Machiavellian Intelligence: Social Expertise and the Evolution of Intellect in Monkeys, Apes and Humans (Oxford Univ. Press, Oxford, UK, 1988). 33. Whiten, A. & Byrne, R.W. Machiavellian Intelligence II: Extensions and Evaluations (Cambridge Univ. Press, Cambridge, UK, 1997). 34. Schultz, W. Predictive reward signal of dopamine neurons. J. Neurophysiol. 80, 1–27 (1998). 35. Rolls, E.T. Brain and Emotion (Oxford Univ. Press, Oxford, UK, 1999). 36. Amador, N., Schlag-Rey, M. & Schlag, J. Reward-predicting and reward detecting neuronal activity in the primate supplementary eye field. J. Neurophysiol. 84, 2166–2170 (2000). 37. Stuphorn, V., Taylor, T.L. & Schall, J.D. Performance monitoring by the supplementary eye field. Nature 408, 857–860 (2000). 38. Ito, S., Stuphorn, V., Brown, J.W. & Schall, J.D. Performance monitoring by the anterior cingulate cortex during saccade countermanding. Science 302, 120–122 (2003). 39. Kawagoe, R., Takikawa, Y. & Hikosaka, O. Expectation of reward modulates cognitive signals in the basal ganglia. Nat. Neurosci. 1, 411–416 (1998). 40. Platt, M.L. & Glimcher, P.W. Neural correlates of decision variables in parietal cortex. Nature 400, 233–238 (1999). 41. Shidara, M. & Richmond, B.J. Anterior cingulate: single neuronal signals related to degree of reward expectancy. Science 296, 1709–1711 (2002). 42. Ikeda, T. & Hikosaka, O. Reward-dependent gain and bias of visual responses in primate superior colliculus. Neuron 39, 693–700 (2003). 43. McCoy, A.N., Crowley, J.C., Haghighian, G., Dean, H.L. & Platt, M.L. Saccade reward signals in posterior cingulate cortex. Neuron 40, 1031–1040 (2003). 44. Watanabe, M. Reward expectancy in primate prefrontal cortex. Nature 382, 629–632 (1996). 45. Leon, M.I. & Shadlen, M.N. Effect of expected reward magnitude on the response of neurons in the dorsolateral prefrontal cortex of the macaque. Neuron 24, 415–425 (1999). 46. Kobayashi, S., Lauwereyns, J., Koizumi, M., Sakagami, M. & Hikosaka, O. Influence of reward expectation on visuospatial processing in macaque lateral prefrontal cortex. J. Neurophysiol. 87, 1488–1498 (2002). 47. Roesch, M.R. & Olson, C.R. Impact of expected reward on neuronal activity in prefrontal cortex, frontal and supplementary eye fields and premotor cortex. J. Neurophysiol. 90, 1766–1789 (2003). 48. Tsujimoto, S. & Sawaguchi, T. Neuronal representation of response outcome in the primate prefrontal cortex. Cereb. Cortex 14, 47–55 (2004). 49. Wang, X.-J. Probabilistic decision making by slow reverberation in cortical circuits. Neuron 36, 955–968 (2002). 50. Bruce, C.J., Goldberg, M.E., Bushnell, M.C. & Stanton, G.B. Primate frontal eye fields. II. Physiological and anatomical correlates of electrically evoked eye movements. J. Neurophysiol. 54, 714–734 (1985).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

410

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Men and women differ in amygdala response to visual sexual stimuli
Stephan Hamann1, Rebecca A Herman1, Carla L Nolan1 & Kim Wallen1,2
Men are generally more interested in and responsive to visual sexually arousing stimuli than are women. Here we used functional magnetic resonance imaging (fMRI) to show that the amygdala and hypothalamus are more strongly activated in men than in women when viewing identical sexual stimuli. This was true even when women reported greater arousal. Sex differences were specific to the sexual nature of the stimuli, were restricted primarily to limbic regions, and were larger in the left amygdala than the right amygdala. Men and women showed similar activation patterns across multiple brain regions, including ventral striatal regions involved in reward. Our findings indicate that the amygdala mediates sex differences in responsiveness to appetitive and biologically salient stimuli; the human amygdala may also mediate the reportedly greater role of visual stimuli in male sexual behavior, paralleling prior animal findings.

Functional neuroimaging studies have identified a growing number of sex differences in human brain function. In addition to cognitive differences1–3, men and women also differ markedly in aspects of sexual behavior, such as the reportedly greater male interest in and response to sexually arousing visual stimuli4–6. Animal studies have identified several sex differences in limbic brain regions that mediate reproductive behavior, which may provide clues to brain regions underlying sex differences in human sexual response. In rats, for example, male but not female appetitive responses to distal olfactory and visual sexual signals are critically mediated by the medial amygdala7. Lesions to the medial amygdala in male but not female rats disrupt appetitive, sexual behaviors involved in gaining access to a receptive mate, but these lesions leave consummatory, copulationrelated behaviors intact7. In addition, male and female rats’ reproductive functions are controlled by different hypothalamic regions7. Although the amygdala and hypothalamus have also been linked to male responses to sexually arousing stimuli in neuroimaging studies8,9 and these structures are considerably influenced by sex hormones10,11, it remains unclear whether sex differences in the function of these regions also exist in humans. Here we examined human sex differences in reactions to visual sexual stimuli using fMRI, contrasting neural responses of healthy men and women to sexually arousing photographs and control stimuli. To permit sex differences in neural responses to be characterized while controlling for possible differences in brain response related to typically higher arousal levels for males, we selected stimuli through prior testing that yielded equivalent sexual attractiveness and physical arousal ratings from both sexes. On the basis of converging evidence from earlier studies7–11, we were particularly interested in whether males would show greater activation in the amygdala and hypothalamus.

Twenty-eight young adults (14 female) passively viewed alternating short blocks of four types of photographic stimuli via video goggles: two types of sexually arousing stimuli, including heterosexual couples engaged in sexual activity (‘couples’ stimuli) and sexually attractive opposite-sex nudes (‘opposite-sex’ stimuli), and two types of control stimuli, including pleasant scenes depicting non-sexual male-female interaction, such as therapeutic massage (neutral stimuli), or a fixation cross (fixation). Subjects were screened to verify that they were heterosexual and found visual erotica sexually arousing. Each block contained five stimuli of the corresponding type. Two runs were presented, each containing four blocks of each type, and ratings of sexual attractiveness and physical arousal were assessed after scanning. We found that the amygdala and hypothalamus were more activated in men than in women when viewing identical sexual stimuli, even when females reported greater arousal. RESULTS Females and males rated the sexual stimuli as equivalently sexually attractive and physically arousing, and both sexes reported the couples stimuli as more attractive and arousing than the opposite-sex stimuli (Fig. 1). A two-factor analysis of variance (ANOVA; sex × stimulus type) conducted separately for attractiveness ratings and physical arousal ratings showed a main effect of stimulus type, with couples stimuli rated higher in attractiveness (F1,26 = 8.08, P < 0.01) and physical arousal (F1,26 = 17.88, P < 0.001) than opposite-sex stimuli. However, there was no difference between females and males in overall ratings for either attractiveness (F1,26 = 1.21, P > 0.28) or physical arousal (F1,26 = 1.01, P > 0.32), and there was no interaction between sex and stimulus type for either attractiveness (F1,26 = 1.06, P > 0.31) or physical arousal (F1,26 = 2.04, P > 0.17). Thus, although the females seemed to show a somewhat larger dif-

1Department

of Psychology, 532 North Kilgo Circle, Emory University, Atlanta, Georgia 30322, USA. 2Yerkes National Primate Research Center, 954 Gatewood Road, Emory University, Atlanta, Georgia 30322, USA. Correspondence should be addressed to S.H. ([email protected]). Published online 7 March 2004; doi:10.1038/nn1208

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

411

ARTICLES
Couples Opposite-sex

3

Attractiveness rating

2.5 2 1.5 1 0.5 0 3

Figure 1 Women’s (n = 14) and men’s (n = 14) ratings of visual stimuli according to attractiveness and physical arousal. Each subject rated 40 couples stimuli and 40 opposite-sex stimuli. A Likert-type rating scale was used: 0 (lowest) to 3 (highest). Top, immediate post-scan ratings of sexual attractiveness. Bottom, post-experiment ratings of experienced physical arousal. Error bars indicate the standard error of the mean (s.e.m.).

2.5 2 1.5 1 0.5 0

Women

Men

ference between their ratings for opposite-sex and couples stimuli (Fig. 1), this difference was not significantly greater than the difference shown by the males. Following spatial pre-processing of the functional images, activation contrasts between conditions were estimated for each subject at each voxel using linear regression, producing statistical parametric t-statistic maps12. Sex differences in activation were assessed using second-level, mixed-effects t-tests. We focused on responses to the couples stimuli because these stimuli elicited the highest arousal and allowed us to directly compare female and male responses to identical stimuli. The entire brain was examined for regions where differential activity surpassed a statistical threshold of P < 0.001 (uncorrected for multiple comparisons) and spanned a minimum of five contiguous 64-mm3 voxels. Men had greater neural responses in the bilateral amygdala and hypothalamus than did women to the couples stimuli (P < 0.001; Fig. 2a). Sex differences were restricted to these regions, with the exception of the right cerebellum (Fig. 2a) and right posterior thalamus (data not shown; P < 0.001). Within the a priori regions of interest (ROIs), the differential activations also survived a more stringent statistical correction for multiple spatial comparisons: left amygdala, P < 0.001, corrected; maxima at –20, –4, –20 (x, y, z in MNI space, see Methods; Z = 3.95) and –16, 0, –16 (Z = 3.77); right amygdala, two clusters, P < 0.05, corrected; maxima at 24, –4, –24 (Z = 3.43) and 16, 0, –16 (Z = 3.32); hypothalamus, P < 0.001, corrected; maximum at 4, 0, –16 (Z = 3.58). Notably, in no region did females show significantly greater activation than males at this statistical threshold. These sex differences were also evident when the couples vs. fixation contrast was examined separately in each group in these same regions (Fig. 2b,c). For males, the left and right amygdala and the hypothalamus were significantly activated (left amygdala, P < 0.001, corrected; right amygdala, P < 0.001, corrected; hypothalamus, P < 0.01, corrected; one group t-test), whereas females showed no significant activations in these regions. To establish that the observed sex differences were related specifically to the sexual aspects of the couples stimuli, we contrasted the responses to the couples stimuli with responses to the more closely matched neutral, non-sexual stimuli that depicted male-female interaction. This contrast controlled for non-sexual attributes of the cou-

ples stimuli, including pleasant, non-sexual physical interaction between males and females. We further restricted this contrast to those regions where sex differences were previously identified in the sex-differences analysis for the couples stimuli vs. fixation contrast (masked at P < 0.01). This served to isolate group differences that resulted from greater increases for the sexual stimuli relative to the neutral stimuli as well as those that resulted from increases relative to the fixation baseline. This contrast revealed more focal differential activations (men > women) at a lower statistical threshold (P < 0.005, at least five contiguous voxels) in left amygdala and right amygdala (Fig. 2d; the same contrast without masking identified similar but more extensive regions of differential activation, see Supplementary Fig. 1 online), as well as in the hypothalamus, bilateral posterior thalamus and left hippocampus (data not shown). As before, no regions were observed in which females showed significantly greater activation than males at the same threshold. The left amygdala differential activation (men > women) survived a more stringent correction for multiple spatial comparisons (left amygdala, P < 0.05, corrected; maximum at –16, 0, –20; Z = 3.23), the right amygdala activation was marginally significant (P = 0.06; maximum at 16, 0, –16; Z = 2.89), and the hypothalamic activation did not reach significance. The absence of differential hypothalamic activation in this latter analysis stemmed largely from low-level activation (at P < 0.05, uncorrected) detected in the hypothalamus for non-sexual stimuli, in men but not women, possibly related to a greater propensity for the males to appraise the nominally non-sexual scenes as weakly sexually appetitive. Sex differences were also evident when the couples stimuli versus neutral stimuli contrast was examined separately in each group in these same regions (Fig. 2e,f). For males, the left and right amygdala and the hypothalamus were significantly activated: left amygdala, P < 0.01, corrected; maximum at –16, 0, –24 (Z = 3.64); right amygdala, P < 0.05, corrected, maximum at –20, 0, –20 (Z = 3.31); hypothalamus, P < 0.05, corrected; maximum at –4, 0, –12 (Z = 3.66); in contrast, females showed no significant activations in these same regions. To compare differences in fMRI signal change across all stimulus conditions and between brain regions, we calculated the average fMRI signal change relative to the fixation baseline for each subject for ROIs centered on the left and right amygdala and the hypothalamus. Males showed significantly greater activations in the left amygdala than did females for the couples stimuli (P < 0.001) and marginally greater activations in the right amygdala (P = 0.11) and the hypothalamus (P = 0.11; Fig. 3). The two sexes did not differ significantly in any ROI for the opposite-sex or neutral stimuli (see Supplementary Table 1 online for a complete listing of all ROI contrast statistics). The sex difference in responses to the couples stimuli was marginally greater in magnitude (P = 0.11, F1,26 = 2.71) in the left amygdala than the right amygdala, and was larger in spatial extent in the left amygdala (Fig. 2d). Men showed marginally greater activation for the couples stimuli versus opposite-sex stimuli in the left amygdala (P = 0.10) but not in the right amygdala or hypothalamus. Men also showed greater activation for the couples stimuli relative to neutral stimuli in the left amygdala (P < 0.01), right amygdala (P < 0.05) and hypothalamus (P < 0.01), but activations for opposite-sex stimuli relative to neu-

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Physical arousal rating

412

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
Figure 2 Regional activation maps. Activation contrast for the couples stimuli versus fixation (a–c) and the couples stimuli versus neutral contrast (d–f) (P < 0.005, minimum five contiguous voxels). White circles indicate the approximate location of the ROIs; the left and right circles and upper and lower circles show the left and right amygdala ROIs on the coronal and axial views, respectively. The medial circles show the hypothalamic ROI, which is not visible on the axial views at z = –20. Color bar indicates maximal Z values. Note that color scale bars vary from image to image. The right hemisphere is on the right of the coronal images and bottom of the axial images. (a) Left, coronal image (y = 0) showing greater bilateral amygdala and hypothalamic activations for males versus females for the couples versus fixation contrast. Right, axial view (z = –16) of the same contrast, showing additional right cerebellar activation. (b) Couples versus fixation contrast for males, at the same coronal and axial views. (c) The same contrast and views for females. (d) Left, coronal image (y = 0) showing greater bilateral amygdala activations for males versus females for the couples versus neutral stimuli contrast, within those regions showing greater activity for males versus females for the couples versus fixation contrast (at P < 0.10). The region of greater hypothalamic activation for males is not visible at this coronal level. Right, axial view (z = –20) of the same contrast, showing primarily left-sided amygdala activation. (e) Couples versus neutral stimuli contrast for males, at the same coronal and axial views. (f) The same contrast and views for females, showing an absence of differential activity in the ROIs.

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

tral stimuli did not reach significance for any ROI. Opposite-sex stimuli depicted isolated nudes, whereas couples stimuli depicted varied, explicit sexual activity. Because cross-cultural studies13 report that males prefer sexual variety more than females do, greater male habituation14,15 and lower arousal may have attenuated potential sex differences for the opposite-sex stimuli. Women showed the reverse pattern from males in the left amygdala, with greater activation for the less-arousing opposite-sex stimuli than for couples stimuli (P < 0.05); no corresponding differences were observed in the right amygdala or hypothalamus. Across all activation contrasts with the neutral stimulus condition, women showed greater activation for the couples stimuli (marginally, at P = 0.07) only in the hypothalamus ROI, whereas men showed greater activation for the couples stimuli vs. neutral stimuli in all ROIs. In summary, the pattern of results from the ROI analysis was consistent with the wholebrain analysis and revealed a left-sided lateralization of amygdala response to sexual stimuli for men. Because reported sexual attractiveness and experienced physical arousal was equivalent in females and males, the greater activations for males were unlikely to be attributable to greater subjective arousal. Moreover, when one female subject in the current study who reported low arousal ratings for the couples stimuli was excluded from analysis, reported arousal was greater for females than males (P < 0.005), yet the activation differences favoring males remained unchanged. In addition to characterizing sex differences, we also examined activations that women and men shared in common by computing the statistical conjunction between activation maps for the two groups for the couples stimuli versus neutral stimuli contrast (Fig. 4). Three regions of overlap were observed: (i) a large, bilateral parieto-temporal-occipital activation spanning regions associated with visual processing, attention, and motor and somatosensory function (Fig. 4a,b; P < 0.0001; corrected maxima at 36, –84, 12; 28, –52, 60; –28, –56, 52; Z = 9.21), (ii) the anterior cingulate, which is linked to emotion,

attention and sexual motivation9 (Fig. 4a; P < 0.001, corrected; maximum at 0, 40, 8; Z = 5.91); and (iii) the nucleus accumbens/ventral striatum (Fig. 4a; P < 0.01 corrected, maxima at 0, 16, –4 (Z = 5.37); –8, 24, 0 (Z = 5.57); 8, 20, –8 (Z = 5.32)). Because of the close association of the ventral striatum with reward processes16–18, coactivation in this region suggests that the sexual stimuli were rewarding to a similar degree for both groups, corroborating the subjective reports from both groups that they found sexual stimuli significantly rewarding. The coactivation and lack of sex differences in these broadly distributed regions contrasts with the marked and regionally localized sex differences observed in the amygdala and hypothalamus. Thus, the sex differences we observed in the processing of visual sexual stimuli
0.4 Males Females 0.3

Percent signal change

0.2

0.1

0

–0.1 Couples O.S. Neutral Left amygdala Couples O.S. Neutral Right amygdala Couples O.S. Neutral Hypothalamus

Figure 3 Average fMRI signal change for males and females for couples, opposite-sex and neutral stimuli (vs. fixation baseline), for ROIs in the left amygdala, right amygdala and hypothalamus. Couples = couples stimuli; O.S. = opposite-sex stimuli; Neutral = neutral stimuli. Error bars indicate s.e.m.

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

413

ARTICLES
ject in the current study who reported low arousal ratings for the couples stimuli was excluded from analysis, reported arousal was greater for females than males, yet the activation differences favoring males remained unchanged. We note, however, that because sexual arousal has multiple psychological and physiological aspects, further study will be required to determine to what extent these other aspects may contribute to observed sex differences22. Strong positive correlations between emotional arousal and amygdala activity have been reported for both appetitive and aversive stimuli23,24, and arousal has been suggested as the primary factor influencing amygdala activity in response to olfactory and visual stimulation. Here, however, amygdala responses to appetitive visual sexual stimuli were not solely determined by arousal, but instead were strongly influenced by the sex of the viewer. The amygdala has multiple functions, however, and although processes related to emotional arousal are clearly of prime importance, in specific contexts these other roles may take precedence in determining amygdala activity. For example, considerable evidence from humans and other animals points to a critical role for the amygdala in appetitive incentive motivation, whereby the amygdala mediates the acquisition of high motivational value by stimuli, which in turn drives instrumental behavior17,18,25–27. In this context, the greater amygdala activation in males observed here may in part reflect a greater appetitive incentive value of visual sexual stimuli, either intrinsic or learned, rather than greater emotional arousal. This would be consistent with the greater male motivation to seek out and interact with such stimuli. Notably, it has recently been reported that larger amygdala size is related to higher sexual drive in humans, further supporting a role of the human amygdala in sexual motivation28. It is also possible that sexual stimuli could represent a specific type of biologically salient stimulus that is processed differently from other types of appetitive visual stimuli, and for which the relation between arousal and amygdala activation is more complex. The amygdala has an established role in processing biologically salient appetitive and aversive stimuli, and initiating rapid adaptive responses via activation of other brain regions including the hypothalamus1,15,29–34. The current results extend understanding of amygdala function by showing that the amygdala acts to mediate sex differences in responses to appetitive, emotionally positive stimuli. This accords with two reports that the amygdala mediates sex differences in memory for emotional visual stimuli, each of which found greater left-sided activation related to subsequent emotional memory in women but greater right-sided activation in males1,2 and suggests that the amygdala may be implicated in a variety of sex differences in emotion processing. Here, sex differences between men and women were greater for the left amygdala than the right amygdala, consistent with the predominantly left-sided amygdala activations elicited by pleasant and unpleasant visual stimuli in previous reports1,19. The possible differential roles of the left and right amygdala in emotion processing have been discussed extensively in the context of aversive stimuli34–36. However, the differential roles of the left and right amygdala in processing appetitive emotional stimuli, and in mediating sex differences, remain unclear, in part because few studies have examined these issues to date. A parallel with the differential roles of the amygdala in male appetitive versus consummatory sexual responses highlighted in previous animal studies is suggested by a recent positron emission tomography (PET) study of brain activity in men during consummatory sexual behavior elicited by tactile stimulation by a female partner37. Relative to a resting baseline, consummatory male sexual behavior (erection and orgasm) elicited decreased activity in only one brain region, the amygdala,

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Figure 4 Regions of significant overlap (conjunction) between group activations for males and females. The statistical conjunction between the activation maps for males and females (P < 0.05 corrected, ≥10 contiguous voxels) for the couples stimuli versus neutral, non-sexual stimuli. (a) Axial view (z = –4) showing common activation in ventral striatum and occipital cortex. Color bar indicates maximal Z values. (b) Brain-surface rendered view of the same map, showing parietotemporal-occipital and frontal activations spanning regions associated with visual processing, attention, and motor and somatosensory function. Regions in red surpassed a P < 0.05 corrected threshold. The right hemisphere is on the bottom of the image.

occurred against a background of considerable similarity in the processing of such stimuli by men and women. DISCUSSION Sex differences in activations to sexual stimuli could arise from differences in processing mode between men and women (e.g., different cognitive styles or neural pathways), from activations related to higher arousal, irrespective of biological sex, or from a combination of these factors1. By the arousal hypothesis, when men and women are matched on levels of elicited arousal, sex differences in brain activation should be eliminated. In contrast, the processing-mode hypothesis predicts that men should still show greater brain activation than women in specific regions after controlling for arousal. Our present results support the second hypothesis. In addition, the highly localized nature of the sex differences is more consistent with the processing-mode hypothesis. Previous studies contrasting brain responses to affectively positive visual stimuli with those to less arousing stimuli consistently report arousal-related activations distributed across multiple regions19,20. This stands in marked contrast to the localized differences found here. A previous neuroimaging study examined sex differences in responses to sexual stimuli21, but did not observe sex differences in the amygdala, possibly because its design rendered it less sensitive to fMRI signal changes in this structure (see Supplementary Note online). Arousal was also substantially higher for males than females in this earlier study. After controlling for arousal, the only sex difference observed (in the hypothalamus) was eliminated21 (see Supplementary Note online). In our present study, reported arousal was equated for females and males. Moreover, when one female sub-

414

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

ARTICLES
bilaterally during erection and in the left amygdala during orgasm. Thus, whereas viewing appetitive sexual stimuli by males in the current study elicited highly localized increases in amygdala activation, consummatory sexual behavior elicited correspondingly focal deactivations in the amygdala. Further investigation will be required to determine whether such parallels indeed reflect a conservation of amygdala function across species. For males, more than for females, visual sexual stimuli seem to preferentially recruit an amygdalo-hypothalamic pathway. This accords with previous speculations that the amygdala is a critical initial structure in a processing pathway recruited in human males during the processing of sexually arousing visual stimuli8. It is also in line with findings from animal studies7,18,25 and with the efferent connectivity from the amygdala to the hypothalamus, which controls physiological reactions associated with sexual arousal7. In summary, the current findings suggest a possible neural basis for the greater role of visual stimuli in human male sexual behavior3–6. Whether the sex differences observed here reflect inherent differences in neural function or stem from differential experience is a matter for further study. METHODS
Subjects and task. Twenty-eight healthy subjects took part in the study, 14 female (mean age 25.0 years) and 14 male (mean age 25.9 years). All subjects gave informed consent to participate, and the study was approved by the Emory University human investigations committee. Subjects were prescreened to verify that they were heterosexual (self-reported as having only opposite-sex sexual desire and sexual experiences), had experience viewing stimuli similar to those used in the study, and found such materials significantly sexually arousing. Thirty-four males were pre-screened: four (12%) were excluded because they reported same-sex desire or experience; no males were excluded because of insufficient response to erotica. Forty-five females were pre-screened: 16 (36%) were excluded because they reported same-sex desire or experience and 7 (16%) were excluded because of insufficient response to erotica. The remaining subjects who were not included in the analysis were excluded either because of technical difficulties with the scanner or video goggle system, fMRI signal drop-out, or because a sufficient number of subjects had already been tested. Subjects viewed alternating 20.125-s blocks of four types of stimuli: heterosexual couples engaged in explicit sexual activity (couples stimuli), attractive opposite-sex nudes in modeling poses (opposite-sex stimuli), pleasant social interaction between partially or fully clothed males and females with minimal or no overt sexual content (neutral stimuli; therapeutic massage, dancing, weddings) or a visual fixation cross. Sexual stimuli were pre-selected so that they would be maximally attractive to females, in an effort to match females and males on elicited arousal. Selection was conducted via computerized anonymous ratings with a separate group of female subjects. Only sexual stimuli rated as highly sexually attractive and physically arousing were selected for use in the primary experiment; stimuli that elicited weak arousal or were rated as aversive or humorous were eliminated. Stimuli were presented via MRIcompatible goggles (Resonance Technology, Inc.). We presented stimuli rapidly, in alternating blocks of five stimuli of each type. Each block contained five stimuli of the corresponding type, with each stimulus presented for 3,750 ms followed by a fixation cross for 275 ms. Two runs were presented, each containing four blocks of each type presented in a pseudorandom order. Ratings of sexual attractiveness were assessed immediately after each scan; physical sexual arousal was assessed retrospectively immediately after all scanning had concluded. A Likert-type rating scale was used: 0 (lowest) to 3 (highest). For the re-analysis that omitted one female subject who reported very low arousal ratings, physical arousal ratings for females and males were 2.85 ± 0.10 and 2.31 ± 0.12 (P < 0.005), respectively. Physiological measurement of sexual response was not conducted because of incompatibility of female genital plethysmography with MRI scanning and the lack of commonly accepted methods for comparing magnitudes of male and female genital responses. Subjects were instructed to view each stimulus attentively and to experience whatever reactions the stimuli might elicit. Overt responses were not required, to avoid possible interference with emotional processes elicited by the stimuli. No stimuli were repeated during the experiment. Imaging and data analysis. MRI scanning was performed on a 1.5-tesla Philips Intera scanner. After acquisition of a high-resolution T1-weighted anatomical scan, subjects underwent whole-brain functional runs (echo-planar imaging, gradient recalled echo; TR = 3,000 ms; TE = 40 ms; flip angle, 90°; 64 × 64 matrix, 25 5-mm axial slices) for measurement of blood oxygen level–dependent (BOLD) effects. The first four volumes were discarded to allow for T1 equilibration effects. Data were analyzed using SPM99 software (http:// www.fil.ion.ucl.ac.uk/spm). Functional EPI volumes were realigned to the first volume and normalized to a standard EPI template volume using 4 mm × 4 mm × 4 mm voxels. Images were subsequently smoothed with an 8-mm isotropic Gaussian kernel and band-pass filtered in the temporal domain. Images were carefully inspected for regions of magnetic susceptibility induced signal dropout in the amygdala and hypothalamus. Three males and one female were scanned but had significant signal dropout in the amygdala or hypothalamus and were replaced by newly tested subjects. Thus, all 28 subjects reported here had minimal signal dropout in the regions of interest. Because only those voxels with sufficient signal in all subjects were included for analysis, any signal dropout would have tended to decrease the spatial extent of observed sex differences in activation. Condition effects for the stimulus conditions were estimated using box-car regressors convolved with a canonical hemodynamic response function, separately for each subject at each voxel according to the general linear model (GLM), and regionally specific effects were compared using linear contrasts12. Contrasts between conditions produced statistical parametric maps for each subject of the t-statistic at each voxel. Sex differences in activation were assessed with a second-level, mixed-effects analysis with subjects as the random-effects factor, using a two-group unpaired t-test on the individual subject-specific contrast images, yielding statistical parametric maps. The second-level mixed-effects conjunction analysis was conducted with the individual subject-specific contrast images contrasting the couples condition with the neutral condition, using linear regression (P < 0.05, corrected for spatial comparisons across the whole brain, extent threshold ≥10 voxels). For the whole-brain analysis, we thresholded these summary statistical maps at a voxel-wise intensity threshold of P < 0.001 (uncorrected for multiple comparisons) with a spatial extent threshold of ≥5 contiguous voxels. For the comparison between males and females on the couples versus neutral stimuli contrast (Fig. 2d), a P < 0.005, ≥5 contiguous voxels threshold was used, masked inclusively by the group comparison for the couples versus fixation contrast at a lenient P < 0.01 threshold (Fig. 2a). Based on previous studies, we defined the amygdala and hypothalamus as ROIs. We did a correction for multiple spatial comparisons within each region, as a more stringent test of our a priori hypotheses. The amygdala region was defined as an 8-mm sphere centered on the following coordinates: left amygdala, –20, –4, –20; right amygdala, 20, –4, –20. The hypothalamic ROI was an 8-mm sphere centered on the coordinates 0, –4, –8. For visualization of activation extent, the group activation maps were thresholded at P < 0.005 uncorrected, with a five-voxel extent threshold, and they were overlaid on a representative high-resolution structural T1-weighted image from a single subject from the SPM99 canonical image set, coregistered to Montreal Neurological Institute (MNI) space—a widely used approximation of canonical Talairach space38. All coordinates are reported in MNI space, and may be converted to Talairach space using the freely available MNI2TAL program (http://www.mrc-cbu.cam.ac.uk/Imaging/ Common/mnispace.shtml). Anatomical localization of group activations was assisted by reference to the atlas of Duvernoy39. For the ROI-averaged analysis, to examine differences in the magnitude of fMRI signal change across all stimulus conditions and between different brain regions, we calculated the average fMRI signal change relative to the fixation baseline for each subject. We did this for each stimulus condition for 8-mmradius spherical ROIs centered on the left amygdala, right amygdala and hypothalamus. Specifically, for each subject, hemodynamic response functions for each condition type were estimated across each ROI using a finite impulse response formulation of the GLM, partialling out the modeled effects of the other conditions, as implemented in R. Poldrack’s SPM ROI Toolbox (http://spm-toolbox.sourceforge.net). Parameter estimates for this model are

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 4 | APRIL 2004

415

ARTICLES
estimates of the temporally evolving response magnitude across each point in peristimulus time, averaged across all occurrences of that peristimulus time interval. Response estimates were averaged across the five peristimulus time points occurring from 6 to 18 s after the onset of each 20.125-s block (at 3-s intervals). This is the temporal window where the average signal was predicted to be maximal based on the canonical hemodynamic lag. Statistical comparisons between conditions, regions and groups were conducted using these averaged values from each subject using t-tests and ANOVAs using a criterion of P < 0.05 (two-tailed).
16. Kampe, K.K., Frith, C.D., Dolan, R.J. & Frith, U. Reward value of attractiveness and gaze. Nature 413, 589 (2001). 17. Gottfried, J.A., O’Doherty, J. & Dolan, R.J. Encoding predictive reward value in human amygdala and orbitofrontal cortex. Science 301, 1104–1107 (2003). 18. Everitt, B.J. Sexual motivation: a neural and behavioral analysis of the mechanisms underlying appetitive and copulatory responses of male rats. Neurosci. Biobehav. Rev. 14, 217–232 (1990). 19. Hamann, S.B., Ely, T.D., Hoffman, J.M. & Kilts, C.D. Ecstasy and agony: activation of the human amygdala in positive and negative emotion. Psychol. Sci. 13, 135–141 (2002). 20. Garavan, H., Pendergrass, J.C., Ross, T.J., Stein, E.A. & Risinger, R.C. Amygdala response to both positively and negatively valenced stimuli. Neuroreport 12, 2779–2783 (2001). 21. Karama, S.M. et al. Areas of brain activation in males and females during viewing of erotic film excerpts. Hum. Brain Mapp. 16, 1–13 (2002). 22. Bradley, M.M., Codispoti, M., Sabatinelli, D. & Lang, P. Emotion and motivation. II: Sex differences in picture processing. Emotion 1, 300–319 (2001). 23. Anderson, A.K. et al. Dissociated neural representations of intensity and valence in human olfaction. Nat. Neurosci. 6, 196–202 (2003). 24. Canli, T., Zhao, Z., Brewer, J., Gabrieli, J.D. & Cahill, L. Event-related activation in the human amygdala associates with later memory for individual emotional experience. J. Neurosci. 20, 1–5 (2000). 25. Everitt, B.J., Cardinal, R.N., Parkinson, J.A. & Robbins, T.W. Appetitive behavior: impact of amygdala-dependent mechanisms of emotional learning. Ann. NY Acad. Sci. 985, 233–250 (2003). 26. Arana, F.S. et al. Dissociable contributions of the human amygdala and orbitofrontal cortex to incentive motivation and goal selection. J. Neurosci. 23, 9632–9638 (2003). 27. Gallagher, M. The amygdala and associative learning. in The Amygdala: a Functional Analysis edn. 2 (ed. Aggleton, J.P.) 311–330 (Oxford Univ. Press, Oxford, UK, 2001). 28. Baird, A.D. et al. The amygdala and sexual drive: insights from temporal lobe epilepsy surgery. Ann. Neurol. 55, 87–96 (2004). 29. Hamann, S.B., Ely, T.D., Grafton, S.T. & Kilts, C.D. Amygdala activity related to enhanced memory for pleasant and aversive stimuli. Nat. Neurosci. 2, 289–293 (1999). 30. Anderson, A.K. & Phelps, E.A. Lesions of the human amygdala impair enhanced perception of emotionally salient events. Nature 411, 305–309 (2001). 31. Adolphs, R., Tranel, D. & Damasio, A.R. The human amygdala in social judgment. Nature 383, 470–473 (1998). 32. Davis, M. & Whalen, P.J. The amygdala: vigilance and emotion. Mol. Psychiatry 6, 12–34 (2001). 33. Calder, A.J., Lawrence, A.J. & Young, A.W. Neuropsychology of fear and loathing. Nat. Rev. Neurosci. 2, 352–363 (2001). 34. Dolan, R.J. & Morris, J.S. The functional anatomy of innate and acquired fear: perspectives from neuroimaging. in Cognitive Neuroscience of Emotion (eds. Lane, R.D. & Nadel, L.) 225–241 (Oxford Univ. Press, New York, 2000). 35. Funayama, E.S., Grillon, C., Davis, M. & Phelps E.A. A double dissociation in the affective modulation of startle in humans: effects of unilateral temporal lobectomy. J. Cogn. Neurosci. 13, 721–729 (2001). 36. Morris, J.S., Ohman, A. & Dolan, R.J. Conscious and unconscious learning in the amygdala. Nature 393, 467–470 (1998). 37. Holstege, G. et al. Brain activation during human male ejaculation. J. Neurosci. 23, 9185–9193 (2003). 38. Talairach, J. & Tournoux, P. Co-planar Stereotaxic Atlas of the Human Brain (Thieme Medical, New York, 1988). 39. Duvernoy, H.M. The Human Brain: Surface, Three-dimensional Sectional Anatomy and MRI (Springer, New York, 1991).

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS This research was supported by the Center for Behavioral Neuroscience, a Science and Technology Center Program of the National Science Foundation, under agreement IBN-9876754. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.
Received 21 November 2003; accepted 6 February 2004 Published online at http://www.nature.com/natureneuroscience/
1. Canli, T., Desmond, J.E., Zhao, Z. & Gabrieli, J.D.E. Sex differences in the neural basis of emotional memories. Proc. Natl Acad. Sci. USA 99, 10789–10794 (2002). 2. Cahill, L. et al. Sex-related difference in amygdala activity during emotionally influenced memory storage. Neurobiol. Learn. Mem. 75, 1–9 (2001). 3. Gur, R.C. et al. An fMRI study of sex differences in regional activation to a verbal and spatial task. Brain Lang. 74, 157–170 (2000). 4. Symons, D. The Evolution of Human Sexuality (Oxford Univ. Press, Oxford, UK, 1979). 5. Laumann, E.O., Gagnon, J.H., Michael, R.T. & Michaels, S. The Social Organization of Sexuality (University of Chicago Press, Chicago,1994). 6. Herz, R.S. & Cahill, E.D. Differential use of sensory information in sexual behavior as a function of gender. Hum. Nature 8, 275–286 (1997). 7. Newman, S.W. The medial extended amygdala in male reproductive behavior: a node in the mammalian social behavior network. Ann. NY Acad. Sci. 877, 242–257 (1999). 8. Beauregard, M., Levesque, J. & Bourgouin, P.J. Neural correlates of conscious selfregulation of emotion. J. Neurosci. 21, 1–6 (2001). 9. Redoute, J. et al. Brain processing of visual sexual stimuli in human males. Hum. Brain Mapp. 11, 162–177 (2000). 10. Arnold, A.P. & Gorski, R.A. Gonadal steroid induction of structural sex differences in the central nervous system. Annu. Rev. Neurosci. 7, 413–442 (1984). 11. Roselli, C.E., Klosterman, S. & Resko, J.A. Anatomic relationships between aromatase and androgen receptor mRNA expression in the hypothalamus and amygdala of adult male cynomolgus monkeys J. Comp. Neurol. 439, 208–223 (2001). 12. Friston, K.J. et al. Statistical parametric maps in functional imaging: a general linear approach. Hum. Brain Mapp. 2, 189–210 (1995). 13. Schmitt, D.P. et al. Universal sex differences in the desire for sexual variety: tests from 52 nations, 6 continents, and 13 islands J. Pers. Soc. Psychol. 85, 85–104 (2003). 14. Rauch, S.L. et al. Neural activation during sexual and competitive arousal in healthy men. Psychol. Res. Neuroimag. 91, 1–10 (1999). 15. Stoleru, S. et al. Neuroanatomical correlates of visually evoked sexual arousal in human males. Arch. Sex. Behav. 28, 1–21 (1999).

416

VOLUME 7 | NUMBER 4 | APRIL 2004 NATURE NEUROSCIENCE

Sponsor Documents

Or use your account on DocShare.tips

Hide

Forgot your password?

Or register your new account on DocShare.tips

Hide

Lost your password? Please enter your email address. You will receive a link to create a new password.

Back to log-in

Close